Datasets:

dongqi-me

/

SciNews

like 4

Physics

New sensor material could enable more sensitive readings of biological signals

Alexander Giovannitti et al, N-type organic electrochemical transistors with stability in water, Nature Communications (2016). DOI: 10.1038/NCOMMS13066 Journal information: Nature Communications

http://dx.doi.org/10.1038/NCOMMS13066

https://phys.org/news/2016-10-sensor-material-enable-sensitive-biological.html

Abstract Organic electrochemical transistors (OECTs) are receiving significant attention due to their ability to efficiently transduce biological signals. A major limitation of this technology is that only p-type materials have been reported, which precludes the development of complementary circuits, and limits sensor technologies. Here, we report the first ever n-type OECT, with relatively balanced ambipolar charge transport characteristics based on a polymer that supports both hole and electron transport along its backbone when doped through an aqueous electrolyte and in the presence of oxygen. This new semiconducting polymer is designed specifically to facilitate ion transport and promote electrochemical doping. Stability measurements in water show no degradation when tested for 2 h under continuous cycling. This demonstration opens the possibility to develop complementary circuits based on OECTs and to improve the sophistication of bioelectronic devices. Introduction Interest in mixed conduction, as in recent bioelectronic and energy applications, has led to a surge in novel organic electronic materials and devices. One characteristic example is an organic electrochemical transistor (OECT), in which ions from an electrolyte penetrate a polymer film and modulate its conductivity. As a result, OECTs can efficiently transduce ionic signals into electronic ones, making them ideal biological sensing elements. OECTs can be fabricated from biocompatible materials 1 , 2 and operate in aqueous environments, which enables recordings in vivo and in vitro 1 , 3 , 4 , 5 . The advantage of OECT-based sensors compared to organic field-effect transistor (OFET)-based sensors is that ions in the former interact with the whole volume of the active material, giving rise to lower impedance and higher transconductance 6 . As a result, the performance of an OECT, defined by the efficiency with which it transduces a voltage modulation at the gate ( V G ) into a current modulation in the channel ( I D ), depends on the thickness of the active layer. This is contrary to OFETs, which rely on the interfacial accumulation of charges and are thus limited by the double layer capacitance 5 , 6 . In particular, the transconductance ( g m =∂ I D /∂ V G ) of OECTs operated in an aqueous environment can reach more than 3.0 mS (refs 3 , 7 , 8 ) at low biases, enabling their use in clinical neuroscience applications, including electrocardiography 3 , 9 , electroencephalography 6 , 9 or neural stimulation 4 . The current state-of-the-art active material for OECTs is the conducting polymer blend poly(3,4-ethylenedioxythiophene):poly(styrenesulfonate) PEDOT:PSS, where the material is doped in its native state, requiring the OECT to be operated in depletion mode 3 , 8 . The operation of an OECT in accumulation mode allows for low power consumption devices with high ON/OFF ratios. The ability of a polythiophene-based polyelectrolyte to operate as an accumulation mode OECT has previously been demonstrated with transconductances of up to 2.0 mS (ref. 10 ). Intermediate materials, where conducting and semiconducting polymers were blended together, have also been reported 11 . To date, however, all reported OECTs have relied on hole transport (p-type), while development of electron transporting (n-type) or ambipolar OECTs has been ignored. Such OECTs would allow the development of complementary circuits and will dramatically improve the sophistication of bioelectronics devices. However, the development of appropriate semiconductor materials is a major challenge: It requires that a material be both stable in an aqueous electrolyte, and that it be reversibly reduced and oxidized within the electrochemical window imparted by that electrolyte. This requires the design of materials that concurrently have a high electron affinity (EA), a low ionization potential and the capacity for facile ion penetration. In addition to their ability to be oxidized and reduced efficiently, these materials should also show high electron and hole mobilities in order to sustain large electronic currents and yield efficient current modulation in an OECT format. Recently, the electron mobility of n-type polymers has increased rapidly, reaching values of more than 1.0 cm −2 V −1 s −1 in OFETs, 12 , 13 , 14 , 15 , 16 , 17 thus enabling n-type OECTs operating in accumulation mode. While air stable n-type materials for OFETs have been reported 18 , 19 , these materials usually degrade when operated in water. In this work, we report the first n-type and ambipolar OECT which operates in water and shows a high stability during pulse measurements over 2 h. This work paves the way for the fabrication of OECT complementary circuits. Results Materials synthesis The synthetic design for developing n-type OECT materials required a narrow band gap donor acceptor copolymer with polar side chains. To this end, we focus on the highly electron-deficient 2,6-dibromonaphthalene-1,4,5,8-tetracarboxylic diimide (NDI) monomer which can easily be copolymerized with electron-rich thiophene based co-monomers. The synthesis of linear glycol chains with several ethylene glycol repeating units and an amine end-group which is needed for the imide formation is usually time consuming and involves several reaction steps 20 . Here, we present a simple reaction where the NDI monomer gNDI-Br 2 ( 2 , Fig. 1 ) with long linear ethylene glycol-based side chains is synthesized in a one-pot reaction from commercial reagents. The amino alcohol forms an ester with 2-[2-(2-methoxyethoxy)ethoxy]acetic acid while the dianhydride 1 is converted to the diimide monomer 2 . Stille polymerization of 2 in chlorobenzene with 5,5′-bis(trimethylstannyl)-2,2′-bithiophene 3b and the (2-(2-(2-methoxyethoxy)ethoxy)ethoxy) analogue 3a with 2 mol% Pd 2 (dba) 3 and P( o -tol) 3 was carried out to synthesize polymers p(gNDI-T2) and p(gNDI-gT2), respectively, following a procedure for semiconducting polymers with polar side chains 21 . The solubility of the polymer in polar solvents increases with the amount of glycol side chains. Polymer p(gNDI-gT2) is soluble in chloroform, 1,1,2,2-tetrachloroethane and dimethylformamide while p(gNDI-T2), with lower glycol side chain density, can only be dissolved in hot 1,1,2,2-tetrachloroethane. End-capping of p(gNDI-gT2) with mono-functionalized thiophenes was performed to remove bromo or organo-tin polymer endgroups. Due to a low solubility of p(gNDI-T2) in chlorobenzene it was not possible to carry out the end-capping procedure. The molecular weight distribution of p(gNDI-gT2) was measured by matrix-assisted laser desorption/ionization time-of-flight spectrometry (MALDI-TOF) where sequences of (gNDI-gT2) n up to seven repeating units ( n =7) were detected ( Supplementary Fig. 16 ). In addition, 1 H NMR end-group analysis yielded an average repeating unit of seven repeating units and is in agreement with the MALDI-TOF measurements ( Supplementary Fig. 17 ). Figure 1: Monomer and polymer synthesis. ( a ) One-pot synthesis of monomer 2 and ( b ) Stille polymerizations affording p(gNDI-T2) and p(gNDI-gT2). Full size image Electrochemical and optical properties Figure 2a presents the cyclic voltammetry (CV) measurements of thin films comprising p(gNDI-T2) and p(gNDI-gT2) on indium tin oxide (ITO)-coated glass substrates in acetonitrile solution. The results of the CV measurements are summarized in Table 1 . Polymer p(gNDI-T2) has an electrochemical bandgap of 1.68 eV and polymer p(gNDI-gT2) of 0.71 eV where the substitution in the 3,3′-positions of the bithiophene from a hydrogen atom to a methyl end-capped triethylene glycol chain decreases the ionization potential (IP) from 5.53 to 4.83 eV and increases the EA from 3.85 to 4.12 eV. The thin film UV-Vis-NIR absorption of p(gNDI-T2) and p(gNDI-gT2) is presented in Fig. 3 . In the solid state, the absorption onset for polymer p(gNDI-gT2) is at 1,550 nm ( Supplementary Fig. 1 ) and for p(gNDI-T2) at 817 nm which is consistent with an increased donor strength of the glycol-bithiophene unit compared to bithiophene unit and in agreement with literature 22 , 23 . The initial UV-Vis spectrum of p(gNDI-gT2) has two absorption peaks, a π - π * transition located between 350 and 550 nm and a broad absorption band from an intramolecular charge transfer complex (ICT) between 600 and 1,550 nm. Time-dependent density functional theory calculations at WB97XD/6-31G(d,p) level of theory were carried out to simulate the UV-Vis spectra of both polymers and compare them to the experimental absorption spectra. A similar trend for the band gap was found where polymer p(gNDI-gT2) shows an ICT transition at lower energies compared to p(gNDI-gT2) in the gas phase ( Supplementary Fig. 5 ). As shown in Fig. 2b , the attachment of polar glycol chains at the polymer backbone enables reversible electrochemical switching between the reduced and neutral states in 0.1 M NaCl aqueous solution for both polymers. During the reduction (n-type doping process), sodium ions drift into the thin film to stabilize the negative charge on the polymer backbone. In the case of the narrower bandgap polymer p(gNDI-gT2), which has a lower ionization potential, oxidation (p-type doping) takes place at relatively low positive voltages where chloride ions are driven into the film to stabilize the positive charge on the backbone. With these synthetic modifications, we are able to achieve stable redox reactions for the conjugated polymers p(gNDI-T2) and p(gNDI-gT2) in 0.1 M NaCl aqueous solution and also in phosphate-buffered saline solution ( Supplementary Fig. 11 ) where the measurements show reversible doping and de-doping over 40 cycles. Remarkably, all CV measurements in aqueous solutions were performed under ambient conditions, without removing oxygen. Figure 2: Cyclic voltammetry of thin films in acetonitrile and aqueous electrolyte solutions. The polymers p(gNDI-T2) and p(gNDI-gT2) were deposited by spin coating on ITO-coated glass substrates and measured with a scan rate of 100 mV s −1 in ( a ) 0.1 M TBAPF 6 acetonitrile solution (ferrocene is shown as the reference, three cycles) and ( b ) 0.1 M NaCl aqueous solution (40 cycles) with a background measurement of a neat ITO substrate. Full size image Table 1 Properties of p(gNDI-T2) and p(gNDI-gT2). Full size table Figure 3: Normalized UV-Vis absorption spectra in solid state. p(gNDI-T2) (blue line) and p(gNDI-gT2) (black line). Full size image In general, n-type doping processes of organic semiconductors are usually highly sensitive to traces of oxygen or water due to the formation of electron-rich doped intermediates 24 , 25 , 26 , 27 . To study the n-type doping process of the synthesized polymers and to evaluate the polymer charge carrier density by electrochemical doping, spectroelectrochemical measurements were carried out. The results are depicted in Fig. 4 where UV-Vis measurements are shown for a series of negative potentials applied to a thin film of p(gNDI-gT2). Figure 4: Spectroelectrochemical measurements of p(gNDI-gT2). UV-Vis measurements were carried out while applying a voltage between 0 and –0.7 V versus Ag/AgCl in 0.1 M NaCl aqueous solution. Full size image As shown in Fig. 4 , a new absorption feature can be observed between 550 and 650 nm when a negative potential is applied; the intensity of the feature increases incrementally on applying higher negative voltages. At higher doping levels ( V =−0.7 V), two new absorption features within the ICT absorption band can be observed with a new absorption peak at 810 nm ( λ max ) and a broad band evolving at 900–1,100 nm, while the intensity of the initial ICT transition decreases correspondingly. Upon applying a negative potential during the spectroelectrochemical measurements, electrons are injected into the lowest unoccupied molecular orbital (LUMO) of the polymers to form a single occupied molecular orbital (SOMO) that enable new optical transitions similar to previously reported doping studies of P(NDI2OD-T2) (Polyera N2200) 28 , 29 . Consequently, the intensity of the charge transfer complex decreases while new absorption peaks arise from transitions between the HOMO and SOMO or SOMO and higher states. Polymer p(gNDI-T2) shows a similar trend where the changes in absorption occur at slightly higher voltages, resulting from a lower EA of p(gNDI-T2) ( Supplementary Fig. 6 ). It is likely that the n-type doping in aqueous solution only occurs to a certain degree before the polymers decompose, similar to PEDOT:PSS where over-oxidation degrades the material resulting in a lower conductivity 30 . For both polymers presented here, the resulting UV-Vis spectra are convolutions of un-doped and n-type doped polymer absorption bands, and a reversible n-type doping can be observed up to voltages of −0.5 V for p(gNDI-gT2) and −0.8 V for p(gNDI-T2) versus Ag/AgCl. Transistor characterization The OECT is illustrated in Fig. 5a . Gold contacts (source and drain electrodes) and interconnects were patterned on a glass substrate where an additional layer of Parylene C insulates the gold interconnects. The conjugated polymer is deposited by spin coating without additives or further annealing steps and the device is operated in an aqueous 0.1 M NaCl solution with a Ag/AgCl gate electrode 6 , 8 . In accumulation mode, the electrochemical redox reactions are triggered by applying a positive (n-type doping) or negative (p-type doping) gate voltage. These reduction or oxidation reactions of the active material change the doping level of the semiconducting polymer and increase the conductivity of the polymer. Fig. 5b presents the limits of the gate potentials that can be applied versus Ag/AgCl for an OECT in water 31 . Figure 5: OECT structure and reactions governing operation. ( a ) Cross-sectional representation of the OECT architecture and ( b ) redox reactions of the polymer (P) in aqueous solutions. The OECT is operated in the common source configuration with a Ag/AgCl gate electrode. Electrochemical redox reactions of a polymer including the voltage limits of a stable operation in aqueous solution versus Ag/AgCl 26 , 31 . Full size image After demonstrating that the n-type doping and de-doping processes are fully reversible in aqueous solution, the performance of p(gNDI-T2) and p(gNDI-gT2)-based OECTs were measured. For polymer p(gNDI-gT2), it is possible to n-type dope the polymer with a channel length of 10 to 50 μm; the OECT performance is presented in Fig. 6a,b . Polymer p(gNDI-T2) on the other hand afforded currents that were below the measurement sensitivity for the given device geometry. Owing to its lower ionization potential, polymer p(gNDI-gT2) also performs as a p-type OECT ( Fig. 6c,d ) while the n-type device performance (current and transconductance) is found to be better than p-type operation. Table 2 summarizes the results of the transistor performance with two different device dimensions. Here, the best performing n-type device with a channel length of 10 μm and a width of 100 μm has a peak current of 3.85 μA (at V G =0.6 V) and a peak transconductance of 21.7 μS (at V G =0.5 V) with an ON/OFF ratio of 3.2 × 10 3 . Figure 6: Performance of the ambipolar OECT. ( a ) n-type output characteristics, ( b ) n-type transfer curve (black) and transconductance (blue dotted), ( c ) p-type output characteristics, ( d ) p-type transfer curve (black) and transconductance (blue dotted) measured in 0.1 M NaCl aqueous solution. Output characteristics for n-type operation ( a ) are measured over the range of V G =−0.2 V to 0.55 V, with a V G step size of 0.05 V (diode-like behaviour −0.2≤ V G <0 is shown in grey (dashed)). For p-type operation ( c ), V G =0V to −0.8 V, with a V G step size of 0.05 V (diode-like behaviour 0≥ V G >−0.3 is shown in grey (dashed)). The device shown has channel dimensions of length 100 μm, width 10 μm and thickness 200 nm. Full size image Table 2 OECT performance of p(gNDI-gT2). Full size table One of the major requirements for OECT recordings of biological processes 1 , 4 is stable operation without degradation of the active material. Fig. 7a presents the stability of the current generated in the n-type OECT ( V D =0.5 V) upon successive gate voltage pulses (Δ V G =0.5 V, pulse length=5 s) for five minutes. The device performance is extremely stable and only at gate voltages higher than V G =0.5 V, a slow device degradation can be observed, which is in agreement with the spectroelectrochemical stability measurements presented above. Long term stability measurements are presented in Fig. 7b where a 2 h stable device operation is demonstrated with no sign of degradation. To demonstrate the versatile applicability of the OECT, the electrolyte is changed to a phosphate-buffered saline solution and equivalent performance is observed compared to a 0.1 M NaCl aqueous solution ( Supplementary Fig. 10 ). Figure 7: Long-term stability measurement of the transistor in 0.1 M NaCl aqueous solution. ( a ) Drain current and applied gate voltage pulses during the first 5 min of a 2 h operation, and ( b ) stable 2 h operation. A gate voltage pulse ( V G =0.5 V) was applied for 5 s with an interval time of 5 s between the successive pulses ( V D =0.5 V), the device shown has channel dimensions of length 100 μm, width 10 μm and thickness 350 nm. The response times of the device are ∼ 5 ms for both device turn on and turn off. Frequency-dependent transconductance is included in the Supplementary Information ( Supplementary Fig. 12 ). Full size image When this device is compared to an accumulation mode p-type OECT based on a conjugated polyelectrolyte 10 , current and transconductance values of p(gNDI-gT2) are an order of magnitude lower. The conductivity ( σ ) is the product of the charge of an electron/hole q , charge carrier concentration n i ( n i = n (electron) or p (holes)) and the electronic mobility μ of the material. The reason for low conductivities could be either a limited increase in the charge carrier density or a low electron mobility. Since the CV and UV-Vis measurements showed an increase of the charge carrier density, we seek to compare electron mobilities by investigating OFETs with these polymers. OFETs were tested with bottom gate bottom contact architecture and the film casting/processing was kept the same as with OECTs, such that no additional OFET device optimization (contact treatments or annealing) was performed ( Supplementary Figs 14 and 15 ). The electron mobilities ( μ e ) were measured to be 1.0 × 10 −4 cm −2 V −1 s −1 for p(gNDI-T2) and 1.0 × 10 −5 cm −2 V −1 s −1 for p(gNDI-gT2). Although charge transport occurs at the interface of the semiconductor and dielectric in a field effect transistor and in the bulk in an electrochemical transistor, we seek to compare the relative mobilities of the polymers and relate them to the device performance. We believe that the low electron mobility of p(gNDI-gT2) is the limiting factor to obtain high currents and transconductances as compared to previous accumulation mode OECTs 10 . X-ray scattering Grazing-incidence wide-angle X-ray scattering measurements of spin cast thin films reveal that both polymers show a propensity for the conjugated backbones to lie parallel to the substrate, with mixed lamellar/ π -stacking texture. The general microstructure is evocative of the well-studied alkylated analogue p(NDI2OD-T2) 32 . In comparing p(gNDI-gT2) and p(gNDI-T2), the polymer with the higher glycol side chain density, p(gNDI-gT2), shows more readily observable higher order diffraction in the lamellar direction, and stronger scattering from the backbone and π -stacks as presented in Fig. 8 . However, the crystallites/aggregates of p(gNDI-T2) show generally tighter packing in all directions, most notably in the lamellar stacking ( d lam =2.07 nm [ q lam =0.302 Å− −1 ] for p(gNDI-T2) versus d lam =2.38 nm [ q lam =0.263 Å −1 ] for p(gNDI-gT2)) and π -stacking directions ( d π =3.53 Å [ q π =1.78 Å −1 ] for p(gNDI-T2) versus d π =3.73Å [ q π =1.68 Å −1 ] for p(gNDI-gT2)). The reason for the closer packing of p(gNDI-T2) is most likely related to the lower glycol side chain density and may contribute to the higher FET mobility observed. Figure 8: Two-dimensional grazing incidence X-ray scattering. 2D-GIXD line cuts of ( a ) p(gNDI-T2) and ( b ) p(gNDI-gT2). Cuts along the Q xy direction (red) represent scattering in the plane of the substrate, while the scattering in the Q z direction (blue) results from out-of-plane scattering. The associated lamellar (h00), and pi-stacking (0k0) peaks are indicated. Full size image Discussion A stable operation of an n-type electrolyte gated organic field effect transistor (EGOFET) was already reported where the operational voltage was reduced by changing from a solid state dielectric gated to an electrolyte gated device 33 . OECT performance, in comparison to EGOFET, relies on a large capacitance in the channel. From the measurements of electrochemical impedance spectroscopy of the polymers in 0.1 M NaCl aqueous solution ( Fig. 9 ), it is possible to extract the effective capacitance per unit area. This is a quantitative means to determine whether doping of the polymer is largely electrostatic (and limited to the polymer/electrolyte interface—typical of EGOFET 34 ) or whether the doping permeates the thickness of the channel, a defining characteristic of OECTs. A potential was applied to the polymer-coated gold electrode (working electrode (WE)) with respect to the Ag/AgCl reference electrode in 0.1 M NaCl aqueous solution, and the change in impedance was monitored. The effective capacitance per unit area of p(gNDI-gT2) is 9.9 × 10 −3 F cm −2 at an offset potential of —0.4 V and is more than three orders of magnitude higher than for a P3HT-based EGOFET where a capacitance per area of 3–6 × 10 −5 F cm −2 is measured on gold electrodes (and is relatively insensitive to the applied offset) 34 . Figure 9: Electrochemical impedance spectroscopy. Bode plot and phase angle of p(gNDI-T2) with d =165 nm ( a , b ) and p(gNDI-gT2) with d =250 nm ( c , d ) in 0.1 M NaCl aqueous solution including the potentials applied at the working electrode (WE). Both polymers are cast on gold electrodes, and are 3.48 × 10 −3 cm −2 in area. Full size image The device presented herein can therefore be categorized as a working OECT, due to the fact that (a) the turn on voltages of the OECT during n- and p-type operation are in agreement with the reduction and oxidation values obtained from CV measurements, (b) bulk electrical doping was verified by spectroelectrochemical measurements, and (c) the ability to store charge extends into the bulk of the film, yielding high capacitance values (with areal capacitances far greater than those of an EGOFET). In comparing OECTs prepared with p(gNDI-gT2) to those with p(gNDI-T2), it is interesting to note that the material with the higher FET mobility (and tighter π -stacking) happens to be the material that did not work as an electrochemical transistor. Previous work has discussed the performance of PEDOT-based OECTs on the grounds of both electronic mobility, and capacity for charge due to electrochemical (de)doping upon ion injection 6 . The improved energy levels when comparing p(gNDI-gT2) to p(gNDI-T2), mainly a higher EA, allow for a decrease in the turn on voltage for polymer p(gNDI-gT2) and must lead to more facile cation injection/n-type doping. The enhanced ion injection is likely facilitated by higher glycol side chain density, and the resulting less densely packed conjugated backbones. Accounting for the thickness of the polymer layer, the estimated volumetric capacitance ( C* ) extracted from both the effective capacitance at 1 Hz as well as an equivalent circuit model fit to the impedance data at a bias of −0.4 V is 190 F cm −3 and 397 F cm −3 for p(gNDI-T2) and p(gNDI-gT2), respectively (model fits, Supplementary Fig. 13 ). Therefore, a combination of the doping energetics within the water-imposed stability window, and the higher capacity for electrochemical charge allow for p(gNDI-gT2) to operate as a stable, n-type (even ambipolar) accumulation mode OECT. In comparison to PEDOT:PSS, the volumetric capacitance of p(gNDI-gT2) is one order of magnitude higher 6 , demonstrating the potential for ethylene glycol chains to become the side chain of choice for OECT materials. We demonstrate the successful development of an ambipolar OECT, opening a new direction for n-type conjugated polymers. This device was developed utilizing semiconducting copolymers comprising naphthalene-1,4,5,8-tetracarboxylic diimide and bithiophene units to form materials with hybridized energy levels and therefore high electron affinities and low ionization potentials. During the doping and de-doping processes, ions drift into and out of the active layer to compensate positive and negative charges on the polymer backbone. Glycol side chains show a strong tendency to interact with hydrated ions and water and therefore the polymers were functionalized with ethylene glycol-based side chains, which facilitated electrochemical switching between the reduced, neutral and oxidized states in aqueous solution. Spectroelectrochemical measurements were carried out to demonstrate high stability for p- and n-type doping in aqueous solution. A remarkably stable OECT operation was achieved, where the device was operated for 2 h without degradation. Methods Materials characterization Column chromatography was carried out with silica gel for flash chromatography from VWR Scientific. Microwave experiments were performed in a Biotage Initiator V 2.3. 1 H and 13 C NMR spectra were recorded on a Bruker AV-400 spectrometer at 298 K and are reported in ppm relative to TMS. Deuterated solvents were purchased from Sigma Aldrich. UV-Vis absorption spectra were recorded on UV-1601 (1,100 nm) and UV-2600 (1,400 nm) UV-VIS Shimadzu UV-Vis spectrometers. Electrospray (ESI-TOF) mass spectrometry was performed with a Micromass LCT Premier. MALDI-TOF spectrometry was conducted in negative linear mode on a Micromass MALDImxTOF with trans-2-[3-(4-tert-butylphenyl)-2-methyl-2-propenylidene]-malononitrile (DCTB) as the matrix. Electrochemical characterization Cyclic voltammograms were recorded using an Autolab PGSTAT101 with a standard three-electrode setup with ITO-coated glass slides as the working electrode, a platinum mesh counter electrode and a Ag/AgCl reference electrode calibrated against ferrocene (Fc/Fc + ). The measurements were carried in an anhydrous, degassed 0.1 M tetrabutylammonium hexafluorophosphate TBAPF 6 acetonitrile solution or a 0.1 M NaCl aqueous solution as the supporting electrolyte at a scan rate of 100 mV s −1 . For the CV measurements in acetonitrile, the glassware was dried at 100 °C and the cell was purged with argon during the measurement to prevent oxygen contamination. Ionization potentials were obtained using the equation: IP=( E ox – E Fc + 4.8 V). Electrochemical impedance spectroscopy was performed with a three-electrode configuration using a potentiostat (Metrohm Autolab) with platinum and Ag/AgCl counter and reference electrodes, respectively. The polymer-coated gold electrode was the working electrode and the electrolyte was a 0.1 M NaCl aqueous solution. Effective capacitance was determined from C ∼ 1/(2 πf Im(Z)) where f is the frequency, and Z is the complex impedance; this capacitance was confirmed for doped spectra from a fit to a R ( R || C ) equivalent circuit in order to extract both capacitance per unit area, and C* . Analysis was performed with Metrohm NOVA software and custom MATLAB tools. Spectroelectrochemical measurements were performed using a PGSTAT101 potentiostat. ITO-coated glass slides with spun cast polymer was used as the working electrode immersed in a cuvette. A UV-1601 UV-VIS Shimadzu UV-Vis spectrometer was employed with the beam path passing through the cuvette filled with a 0.1 M NaCl aqueous solution and the polymer/ITO/glass sample. A background spectrum with the cuvette/electrolyte/ITO/glass was recorded for baseline correction before the experiment was started. Potential were applied for 5 s before the UV-Vis spectra were recorded. Transistor fabrication and characterization OECTs were fabricated as previously reported 8 , except that the polymers p(gNDI-T2) and p(gNDI-gT2) were spun cast from chloroform before sacrificial peel off of Parylene C for the dry patterning processes. The completed samples were not annealed or treated after deposition, the samples were briefly rinsed in deionized water before testing. OECT IV curves (transfer and output), as well as repetitive pulsing were performed with a Keithley 2400 source-measure unit, and custom LabView scripts. Analysis was performed with MATLAB. Data availability The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information files. Additional information How to cite this article: Giovannitti A. et al . N-type organic electrochemical transistors with stability in water. Nat. Commun. 7 , 13066 doi: 10.1038/ncomms13066 (2016). Change history 15 December 2016 A correction has been published and is appended to both the HTML and PDF versions of this paper. The error has not been fixed in the paper.

Scientists have created a material that could make reading biological signals, from heartbeats to brainwaves, much more sensitive. Organic electrochemical transistors (OECTs) are designed to measure signals created by electrical impulses in the body, such as heartbeats or brainwaves. However, they are currently only able to measure certain signals. Now researchers led by a team from Imperial College London have created a material that measures signals in a different way to traditional OECTs that they believe could be used in complementary circuits, paving the way for new biological sensor technologies. Semiconducting materials can conduct electronic signals, carried by either electrons or their positively charged counterparts, called holes. Holes in this sense are the absence of electrons - the spaces within atoms that can be filled by them. Electrons can be passed between atoms but so can holes. Materials that use primarily hole-driven transport are called 'p-type' materials, and those that use primarily electron-driven transport are called, and 'n-type' materials. An 'ambipolar' material is the combination of both types, allowing the transport of holes and electrons within the same material, leading to potentially more sensitive devices. However, it has not previously been possible to create ambipolar materials that work in the body. Credit: Imperial College London The current most sensitive OECTs use a material where only holes are transported. Electron transport in these devices however has not been possible, since n-type materials readily break down in water-based environments like the human body. But in research published today in Nature Communications, the team have demonstrated the first ambipolar OECT that can conduct electrons as well as holes with high stability in water-based solutions. The team overcame the seemingly inherent instability of n-type materials in water by designing new structures that prevent electrons from engaging in side-reactions, which would otherwise degrade the device. These new devices can detect positively charged sodium and potassium ions, important for neuron activities in the body, particularly in the brain. In the future, the team hope to be able to create materials tuned to detect particular ions, allowing ion-specific signals to be detected. Lead author Alexander Giovannitti, a PhD student under the supervision of Professor Iain McCulloch, from the Department of Chemistry and Centre for Plastic Electronics at Imperial said: "Proving that an n-type organic electrochemical transistor can operate in water paves the way for new sensor electronics with improved sensitivity. "It will also allow new applications, particularly in the sensing of biologically important positive ions, which are not feasible with current devices. For example, these materials might be able to detect abnormalities in sodium and potassium ion concentrations in the brain, responsible for neuron diseases such as epilepsy."

10.1038/NCOMMS13066

Biology

CRISPRing trees for a climate-friendly economy

Barbara De Meester et al. Tailoring poplar lignin without yield penalty by combining a null and haploinsufficient CINNAMOYL-CoA REDUCTASE2 allele, Nature Communications (2020). DOI: 10.1038/s41467-020-18822-w Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-020-18822-w

https://phys.org/news/2020-10-crispring-trees-climate-friendly-economy.html

Abstract Lignin causes lignocellulosic biomass recalcitrance to enzymatic hydrolysis. Engineered low-lignin plants have reduced recalcitrance but often exhibit yield penalties, offsetting their gains in fermentable sugar yield. Here, CRISPR/Cas9-generated CCR2 (−/*) line 12 poplars have one knockout CCR2 allele while the other contains a 3-bp deletion, resulting in a 114I115A-to-114T conversion in the corresponding protein. Despite having 10% less lignin, CCR2 (−/*) line 12 grows normally. On a plant basis, the saccharification efficiency of CCR2 (−/*) line 12 is increased by 25–41%, depending on the pretreatment. Analysis of monoallelic CCR2 knockout lines shows that the reduced lignin amount in CCR2 (−/*) line 12 is due to the combination of a null and the specific haploinsufficient CCR2 allele. Analysis of another CCR2 (−/*) line shows that depending on the specific CCR2 amino-acid change, lignin amount and growth can be affected to different extents. Our findings open up new possibilities for stably fine-tuning residual gene function in planta . Introduction The lignin polymer provides strength and hydrophobicity to the plant cell wall and is generally derived from the monolignols coniferyl and sinapyl alcohol and low levels of p -coumaryl alcohol. Depending on the plant species, other monomers or derivatives may also contribute to the lignin polymer 1 . After polymerization in the cell wall, the monolignols produce guaiacyl (G), syringyl (S), and p -hydroxyphenyl (H) units, respectively 1 . Engineering plants to deposit less lignin is a promising strategy to enable improved biomass processability. However, hurdles need to be overcome for the development of low-lignin elite clones for forestry applications. One hurdle is to reduce lignin amount in a stable way. For example, RNA interference (RNAi) was frequently used to downregulate the expression of lignin biosynthesis genes in poplar 2 , 3 , 4 , 5 . However, this method often results in unstable downregulation of the targeted genes 3 , 4 , 5 . As an illustration, the red xylem phenotype caused by reductions in CINNAMOYL-CoA REDUCTASE (CCR) activity, appeared in patches on debarked CCR2 -downregulated poplar stems, as a consequence of the unequal levels of gene silencing in red versus white regions 3 . Unequal gene silencing levels even appeared between individual clones of the same CCR2 -downregulated line 3 . A second hurdle is to reduce lignin amount without affecting plant development and biomass yield. For example, the CCR2 -downregulated poplars with the highest levels of CCR2 downregulation had up to 24% less lignin and an up to 104% increased enzymatic cellulose-to-glucose conversion without pretreatment 3 . Unfortunately, similar to many other plants yielding higher cellulose-to-glucose conversion levels 2 , 6 , 7 , 8 , these CCR2 -downregulated poplars suffered from a reduction of up to 51% in biomass, (entirely) offsetting their gains in fermentable sugar yield 3 . Hence, for applications, a method is desired to make plants with a stable and fine-tuned lignin amount to still achieve higher sugar yields in all replicates, but without affecting growth. To evaluate the specific role of CCR2 in poplar, CCR2 (−/−) null mutants are generated using CRISPR/Cas9. In addition to severely dwarfed CCR2 (−/−) plants, a biallelically modified line has normal growth. Here, we show that this line, named CCR2 (−/*) line 12, contains a knockout and a specific haploinsufficient CCR2 allele (114I115A-to-114T amino-acid change in the corresponding CCR2 protein sequence) that results in a uniformly distributed red xylem phenotype, a 10% reduction in lignin amount and a 25 to 41% increase in saccharification efficiency on a plant basis, depending on the applied pretreatment. Analysis of another CCR2 (−/*) line shows that, whereas multiple amino-acid changes in CCR2 can result in lower lignin content (to different extents), they will not all allow normal growth. We propose that in planta screening for combinations of a knockout and a haploinsufficient allele is a promising strategy to fine-tune the desired level of residual gene function. Results CCR2 (−/*) line 12 grows normally while having red xylem To evaluate the effect of fully knocking out CCR2 on the phenotype of poplar, we generated 21 biallelically edited CCR2 mutants in Populus tremula × P. alba by CRISPR/Cas9 using a gRNA (gRNA1) targeting the third exon of the CCR2 gene (Supplementary Fig. 1a ). The twenty lines that contained biallelic frameshift mutations in CCR2 , CCR2 (−/−) lines, were all severely dwarfed (Fig. 1 ; Supplementary Fig. 1b ). Interestingly, one biallelic mutant line did not display observable growth perturbations (Fig. 1 ). CCR2 (−/*) line 12 had a frameshift mutation (1-bp insertion) in the P. tremula CCR2 allele, and a deletion of 3 bp in the P. alba CCR2 allele, which resulted into a substitution of Ile114 and Ala115 for a Thr114 in the corresponding P. alba CCR2 protein sequence (Supplementary Figs. 1b and 2 ). The amino-acid change occurred in α4 of the CCR2 protein, but not in the active site, NAPD-binding domain, or substrate-binding pocket residues 9 , 10 , 11 . Fig. 1: Phenotype of poplar containing biallelic CCR2 mutations. Plants were grown for 11 weeks in the greenhouse. CCR2 (−/−) and CCR2 (−/*) line 12 poplars ( P. tremula × P. alba) were generated via CRISPR/Cas9 using gRNA1 (targeting the third exon of both CCR2 alleles). The status of the CCR2 alleles present in P. tremula × P. alba is denoted between the parentheses; the first one represents that of the P. tremula allele, the second one that of the P. alba allele; −, knockout; *, protein-modified. The plants shown are representative of twenty biologically independent samples for wild type and CCR2 (−/−). One plant was available for CCR2 (−/*) line 12. Scale bar = 10 cm. Full size image To evaluate the growth and xylem phenotypes of CCR2 (−/*) line 12, wild type, and CCR2 (−/*) line 12 were clonally propagated to generate multiple biological replicates. The replicates were grown in the greenhouse for a period of 20 weeks. Plant height was followed weekly, and by the end of the growth period, the trees were harvested, and biomass parameters were determined. The stem height of CCR2 (−/*) line 12 was equal to that of the wild type during the entire growth period (Fig. 2a , Supplementary Table 1 ). At harvest time, the CCR2 (−/*) line 12 plants were morphologically indistinguishable from the wild type (Fig. 2b ), and no differences in either stem diameter or fresh and dry stem weight were observed (Table 1 ). Fig. 2: Phenotype of CCR2 (−/*) line 12 poplar. Plants were grown for 20 weeks in the greenhouse. a Growth curve of wild type and CCR2 (−/*) line 12. No significant differences in height were found between the wild type (individual values shown in white dots) and CCR2 (−/*) line 12 (individual values shown in gray dots) at the 0.01 significance level (two-tailed Student’s t -test). For mean values and exact P values, see Supplementary Table 1 . b Phenotype of representative wild type and CCR2 (−/*) line 12 after growing for 20 weeks in the greenhouse. Scale bar = 20 cm. c Phenotype of debarked wild-type and CCR2 (−/*) line 12 stems grown in the greenhouse for 20 weeks. CCR2 (−/*) line 12 stems display the red xylem phenotype. Scale bar = 10 mm. Wild type, n = 10 biologically independent samples; CCR2 (−/*) line 12, n = 11 biologically independent samples. The source data underlying ( a ) are provided as a Source data file. Full size image Table 1 Biomass and cell wall composition of CCR2 (−/*) line 12 stems. Full size table After debarking the harvested stems, the coloration of the xylem could be judged; whereas wild-type xylem had a white-to-beige color, CCR2 (−/*) line 12 xylem displayed a uniform pink-to-red coloration (Fig. 2c ). This xylem coloration is often observed in lignin-modified plants 3 , 4 , 12 , 13 and suggested a modified lignin in CCR2 (−/*) line 12 as compared to the wild type. To examine the morphology of the xylem cells in the stem, cross-sections of wild type, CCR2 (−/*) line 12, and CCR2 (−/−) were observed via light microscopy after Wiesner and Mäule staining, and via fluorescence microscopy (Fig. 3 ). Wild-type and CCR2 (−/*) line 12 stem sections were morphologically indistinguishable. They both had fiber and vessel cells with lignified cell walls, and the vessel cells were open. By contrast, CCR2 (−/−) stems showed an overall reduction in lignin deposition in the cell walls of both fiber and vessel cells, and the vessel cells had an irregular shape. Fig. 3: Xylem morphology in stems of CCR2 (−/*) line 12 and CCR2 (−/−) poplars. Plants were grown for 20 weeks in the greenhouse. Lignin was visualized using Mäule and Wiesner staining and via lignin autofluorescence ( λ ex = 365 nm, λ em = 420–470 nm). Pictures were taken of the xylem tissue, with the pith localized to the left and the epidermis to the right. Images show representative pictures from five sections per sample and 3 biologically independent samples per line. Scale bar = 100 μm. Full size image Altered lignin in CCR2 (−/*) line 12 To evaluate the lignocellulosic biomass composition of CCR2 (−/*) line 12 stems, the cell wall residue (CWR), the cellulose content, and the lignin content and composition of dried debarked stem material were determined (Table 1 ). CWR was prepared by applying a sequential extraction to remove soluble compounds from the stems. The fraction of CWR (as % of the dry weight) of CCR2 (−/*) line 12 did not differ from that of the wild type. The cellulose content of the prepared CWRs was analyzed via the spectrophotometric Updegraff assay which showed that the crystalline cellulose content of CCR2 (−/*) line 12 did not differ significantly from that of the wild type. The combined amount of matrix polysaccharides and amorphous cellulose—determined as the mass loss upon trifluoroacetic acid treatment—in the CWRs of CCR2 (−/*) line 12 was increased by about 10% when compared to that of the wild type. Next, the fraction of lignin in the prepared CWRs was determined via the Klason and the acetyl bromide methods. Both methods showed that the total lignin amount of CCR2 (−/*) line 12 was decreased by about 10% when compared to that of the wild type. Lignin composition in the CWRs was evaluated via thioacidolysis, an analytical method that quantifies lignin units solely linked by β– O– 4 bonds (Table 1 ). CCR2 (−/*) line 12 lignins showed a trend towards releasing more monomers (H + G + S) when compared to wild-type lignins ( P = 0.092), indicative for a slightly higher frequency of β– O– 4 interunit bonds. In the wild type, H monomers constituted 1.6% of the total identified thioacidolysis-released units. In CCR2 (−/*) line 12, the fraction of thioacidolysis-released H units was only 0.3%. The S/G ratio based on thioacidolysis-released monomers was equal for the wild type and CCR2 (−/*) line 12. Incorporation of ferulic acid (FA), which is a known minor constituent of lignin, results in the release of three different units after thioacidolysis: β– O– 4-FA-I and β– O– 4-FA-II are derived from ferulic acid (or ferulate ester) starting units coupled via their O– 4 position in β– O– 4 interunit bonds, whereas bis -β– O– 4-FA is derived from ferulic acid that has undergone a β– O– 4 coupling twice at its β-position 14 . In agreement with previously reported results for plants deficient in CCR 3 , 5 , 14 , 15 , 16 , the relative abundance of all three thioacidolysis-released ferulic acid units was increased in CCR2 (−/*) line 12 when compared to the wild type. The lignin composition was also analyzed via two-dimensional 1 H– 13 C heteronuclear single-quantum coherence (2D HSQC) nuclear magnetic resonance (NMR) on ball-milled whole-cell-wall material prepared from the stem (Table 1 , Supplementary Fig. 3 ). By analyzing the aromatic regions of the 2D HSQC spectra, it is possible to profile differences in lignin monomeric composition irrespective of the interunit linkage distribution. Using NMR, the estimated relative frequency of H units showed no significant differences between wild type and CCR2 (−/*) line 12. However, the determination of H units via NMR is neither sensitive nor accurate because the peaks used for their estimation are contaminated by those from phenylalanine 17 . In contrast to the results obtained by thioacidolysis, the S/G ratio was decreased by 14% in CCR2 (−/*) line 12 when compared to the wild type. Similar to the results obtained from thioacidolysis, the ferulic acid marker peak was clearly detected in spectra from CCR2 (−/*) line 12 wood, while being absent in spectra from the wild type. In addition, the NMR data enabled the relative measurement of p -hydroxybenzoates that acylate the sidechain γ-OH of G and, predominantly, S units in poplar. The relative frequency of these moieties was increased by 75% in CCR2 (−/*) line 12 when compared to the wild type. This observation is in line with the fact that the biosynthesis of p -hydroxybenzoates, in contrast to that of H, G, and S units, is independent of CCR2 activity. The interunit linkage-type distributions were also deduced from the NMR spectra. The lignin of CCR2 (−/*) line 12 showed an increase in the relative proportion of β-aryl ether (β– O– 4) units, at the expense of resinol (β–β) units. The fraction of phenylcoumarans (β–5) did not differ between CCR2 (−/*) line 12 and the wild type. Finally, analysis by gel-permeation chromatography (GPC) showed that the molecular weight of CCR2 (−/*) line 12 lignin tended to be lower than that of wild-type lignin (Supplementary Table 2 ). Together, these data show that, although the growth of the CCR2 (−/*) line 12 trees is similar to that of wild-type trees, their wood composition still reflects a deficiency in CCR2. Increased saccharification efficiency in CCR2 (−/*) line 12 Because lignin amount, composition, and polymerization degree greatly influence saccharification yield, we further investigated the saccharification potential of CCR2 (−/*) line 12 under conditions of limited saccharification. The cellulose-to-glucose conversion was calculated based on the amount of glucose released upon saccharification of dried debarked stem material after either acidic (1 M HCl, 80 °C, 2 h), alkaline (62.5 mM NaOH, 90 °C, 3 h), or no pretreatment (Supplementary Table 3 ) and the original cellulose content that was measured for each sample (Table 1 ). In all three cases, cellulose-to-glucose conversion of biomass from CCR2 (−/*) line 12 was significantly higher than that of the wild type (Fig. 4 ); the cellulose-to-glucose conversion of the non-pretreated samples increased from 23.9% in the wild type to 32.4% in CCR2 (−/*) line 12 (i.e., a relative increase of 36%), after acidic pretreatment from 30.3 to 46.4% (i.e., a relative increase of 53%), and after alkaline pretreatment from 70.9 to 95.9% (i.e., a relative increase of 35%). Fig. 4: Saccharification assays of CCR2 (−/*) line 12 stems. Plants were grown for 20 weeks in the greenhouse. Cellulose-to-glucose conversion efficiencies of wild type and CCR2 (−/*) line 12 were calculated based on the quantified amounts of cellulose and the amount of released glucose after 72 h of saccharification following no pretreatment, acidic pretreatment (1 M HCl), or alkaline pretreatment (62.5 mM NaOH) (Supplementary Table 3 , Table 1 ). Individual values (dots) and means (bars) of 10 biologically independent samples for wild type (white bars) and eleven biologically independent samples for CCR2 (−/*) line 12 (gray bars). ** P < 0.01 (two-tailed Student’s t -test); P = 5.1 × 10 −4 , 3.4 × 10 −8 , and 6.6 × 10 −8 in the case of no pretreatment, acid pretreatment, and alkaline pretreatment, respectively (see also Supplementary Table 3 ). Error bars indicate standard deviation. Source data are provided as a Source data file. Full size image In many plant species, including Arabidopsis, tobacco, and poplar, lowering CCR activity leads to a substantial yield penalty 3 , 5 , 18 , 19 . As CCR2 (−/*) line 12 had a biomass yield and cellulose amount that was comparable to that of the wild type, the saccharification yield expressed on a per plant basis was also increased in CCR2 (−/*) line 12 under the used pretreatment conditions (Supplementary Table 3 ); using no pretreatment, acidic pretreatment, and alkaline pretreatment, the glucose yield per plant of CCR2 (−/*) line 12 was increased by 25.3%, 41.3%, and 24.9%, respectively. Haplosufficient wild-type P. tremula × P. alba CCR2 alleles In CCR2 (−/*) line 12, the P. tremula CCR2 allele contained a frameshift mutation, thereby fully knocking out this allele. However, the effect of the 114I115A-to-114T amino-acid change in the protein encoded by the mutant P. alba CCR2 allele on total CCR activity was not known. Either the 114I115A-to-114T amino acid change in the P. alba CCR2 protein did not have any effect on the CCR2 activity, in which case the red xylem phenotype and the reduced lignin amount of CCR2 (−/*) line 12 can be explained by the haploinsufficiency of the wild-type P. alba CCR2 allele. In this case, the CCR activity of the protein encoded solely by the wild-type P. alba CCR2 allele does not suffice to secure normal lignin biosynthesis. Alternatively, the wild-type P. alba CCR2 allele is haplosufficient, and the amino-acid change in the mutant P. alba CCR2 protein encoded by CCR2 (−/*) line 12 reduces CCR2 activity resulting in the reduced lignin amount and the consequent red xylem phenotype. To test whether a single CCR2 allele, be it that encoded by the P. tremula or that by the P. alba genome, is sufficient to secure wild-type lignin amount, monoallelic null mutants in CCR2 were generated by CRISPR/Cas9. gRNA2 and gRNA3 were used, targetting the fourth exon of the CCR2 P. alba and P. tremula allele, respectively (Supplementary Fig. 4a , 5a ). Six lines contained monoallelic frameshift mutations in P. alba CCR2 ( CCR2 (+/−); Supplementary Fig. 4b ), while six out of eight lines in which the P. tremula CCR2 allele was targeted contained monoallelic frameshift mutations in P. tremula CCR2 ( CCR2 (−/+); Supplementary Fig. 5b ). The other two lines contained biallelic mutations in CCR2 . CCR2 (−/−) line 202 contained biallelic frameshift mutations in CCR2 and was severely dwarfed, in agreement with the biallelic frameshift mutants generated previously using gRNA1 (Supplementary Fig. 5b , Figs. 1 and 5 ). CCR2 (−/*) line 206 contained a frameshift mutation (5-bp deletion) in the P. tremula CCR2 allele, and a deletion of 3 bp in the P. alba CCR2 allele (Supplementary Fig. 5b ) and will be discussed further below. Fig. 5: Phenotype of poplar containing mono- and biallelic CCR2 mutations. Plants were grown for 11 weeks in the greenhouse. CCR2 (+/−) generated via CRISPR/Cas9 using gRNA2, CCR2 (−/+) and CCR2 (−/−) generated via CRISPR/Cas9 using gRNA3, and CCR2 (−/−) and CCR2 (−/*) line 12 generated via CRISPR/Cas9 using gRNA1. gRNA1 targets the third exon of both CCR2 alleles whereas gRNA2 and gRNA3 target the fourth exon of the P. alba or P. tremula CCR2 allele, respectively. The status of the CCR2 alleles present in P. tremula × P. alba is denoted between the parentheses; the first one represents that of the P. tremula allele, the second one that of the P. alba allele; +, wild type; −, knockout; *, protein-modified. The plants shown are representative for seven biologically independent samples for wild type, CCR2 (+/−), CCR2 (−/+), CCR2 (−/−) (gRNA1), and CCR2 (−/*) line 12. One plant was available for CCR2 (−/−) (gRNA3). Scale bar = 10 cm. Full size image The twelve individual plants containing monoallelic frameshift mutations in CCR2 (as shown in Supplementary Figs. 4b and 5b ) were grown along with wild type and CCR2 (−/*) line 12 in the greenhouse. After a growth period of 11 weeks, both the CCR2 (+/−) and CCR2 (−/+) monoallelic knock-outs and CCR2 (−/*) line 12 plants were equal to the wild type in stem height and diameter, as well as fresh and dry weight (Table 2 , Fig. 5 ). After debarking the harvested stems, the typical red coloration of the xylem was seen to be present in CCR2 (−/*) line 12, but absent in the wild-type and the CCR2 monoallelic knockout plants (Fig. 6 ). These observations indicate that the lignin of the CCR2 monoallelic knockout plants resembles that of the wild type and not that of CCR2 (−/*) line 12. To validate this, the lignin content and composition of dried debarked stem material were determined (Table 2 ). The acetyl bromide lignin content in the CWR from CCR2 monoallelic knockout plants was equal to that of the wild type, whereas that of the CCR2 (−/*) line 12 was reduced by ~15%. In agreement with the normal xylem coloration, no significant differences in thioacidolysis-released aromatic units were observed between the CCR2 monoallelic knockout plants and the wild type. Similar to the results of 20-week-old stems, 11-week-old CCR2 (−/*) line 12 stems had a decreased frequency of H monomers and an increased frequency of all three thioacidolysis-released ferulic acid units when compared to the wild-type and CCR2 monoallelic knockout plants. Table 2 Biomass and cell wall composition of CCR2 (−/+), CCR2 (+/−), and CCR2 (−/*) line 12 stems. Full size table Fig. 6: Phenotype of debarked stems of CCR2 (+/−), CCR2 (−/+), and CCR2 (−/*) line 12 poplars. Plants were grown for 11 weeks in the greenhouse. The red xylem phenotype was only present in CCR2 (−/*) line 12. The status of the CCR2 alleles present in P. tremula × P. alba is denoted between the parentheses; the first one represents that of the P. tremula allele, the second one that of the P. alba allele; +, wild type; −, knockout; *, protein-modified. The stems shown are representative of seven biologically independent samples per line. Scale bar = 1 cm. Full size image Judged by the normal growth, lignin amount and lignin composition of the CCR2 monoallelic knockout plants, the P. tremula and the P. alba CCR2 alleles appear to both be haplosufficient in P. tremula × P. alba . Phenolic profiling of the different CCR2 mutants To further examine the haplo(in)sufficient status of the CCR2 alleles in P. tremula × P. alba , and investigate to what extent the metabolic changes of CCR2 (−/*) line 12 reflect those expected from CCR2 deficiency, comparative phenolic profiling was performed via ultra-high-pressure liquid chromatography-mass spectrometry (UHPLC-MS) on debarked stems of wild type, CCR2 (+/−), CCR2 (−/+), CCR2 (−/*) line 12, and CCR2 (−/−). Principal component analysis (PCA) was performed on a total of 6182 peaks (mass-to-charge ratio [ m/z ] features) (Supplementary Data 1 ). PCA showed that, according to the first principal component (which explains 53% of the variation), the metabolic profiles of wild type, CCR2 (+/−), and CCR2 (−/+) were indistinguishable, whereas those of CCR2 (−/*) line 12 were situated between those of CCR2 (−/−) and wild type (Fig. 7a ). The second principal component, which explains 10.1% of the variation, reflects variation within the genotypes (and not between the genotypes) and can be attributed to biological and/or technical variation. Next, univariate statistical analysis was applied to the selected peaks to screen for peaks with significantly different intensities in CCR2 (+/−), CCR2 (−/+), CCR2 (−/*) line 12, or CCR2 (−/−) mutants compared with their levels in wild-type plants. After applying specific filters (see “Methods”), no significant differences were found between either CCR2 (+/−) or CCR2 (−/+) and the wild type (Fig. 7b ), again showing that the P. tremula and the P. alba CCR2 alleles are both haplosufficient in P. tremula × P. alba . The reduction in lignin amount in CCR2 (−/*) line 12 is therefore not solely a consequence of the P. tremula CCR2 null allele. Fig. 7: Phenolic profiling of CCR2 (+/−), CCR2 (−/+), CCR 2(−/*) line 12, and CCR2 (−/−) stems. Plants were grown for 11 weeks in the greenhouse. a Plot of principal component 1 (PC1) and PC2 of principal component analysis (PCA) on 6182 peaks. Black dots, wild type; blue dots, CCR2 (+/−); cyan dots, CCR2 (−/+); gray dots, CCR2 (−/*) line 12; red dots, CCR2 (−/−). b , c Volcano plots visualizing the differences between b wild type and CCR2 (+/−) or CCR2 (−/+), and c wild type and CCR2 (−/*) line 12 or CCR2 (−/−). Magenta and black dots represent peaks that are different and not different in intensity, respectively, based on the filters: fold change (FC) > 2 and P value <0.01 (one - way ANOVA with Dunnett’s post hoc test). d Venn diagrams of the number of peaks with significantly differential intensities (magenta dots from the volcano plots in ( c )) between wild type and either CCR2 (−/*) line 12 or CCR2 (−/−). Wild type, n = 6 biologically independent samples; CCR2 (−/*) line 12, n = 6 biologically independent samples; CCR2 (−/−), n = 5 biologically independent samples. Full size image By contrast, using the same specific filters, 960 and 1815 peaks were found to have a higher intensity in CCR2 (−/*) line 12 and CCR2 (−/−), respectively, when compared to the wild type, of which 800 were in common (Fig. 7c, d ). In addition, 258 and 1854 peaks had a lower intensity in CCR2 (−/*) line 12 and CCR2 (−/−), respectively, when compared to the wild type, of which 199 were in common. Next, the top 10 highest peaks with significantly higher and lower intensities in CCR2 (−/*) line 12 were structurally characterized based on their mass-to-charge ratio ( m/z ), retention time, and tandem mass spectrometry (MS/MS) (Supplementary Figs. 6 – 11 , Supplementary Table 4 ). The ten highest peaks with significantly higher intensities in CCR2 (−/*) line 12 could be assigned to ten compounds, of which seven could be (partially) structurally characterized as conjugates of ferulic acid, vanillic acid, sinapic acid, and caffeic acid, and three remained unknown. The ten highest peaks with significantly lower intensities in CCR2 (−/*) line 12 could be assigned to nine compounds, of which five could be structurally characterized as oligolignols, and four remained unknown. The relatively high fraction of differential peaks that CCR2 (−/*) line 12 shares with CCR2 (−/−) (Fig. 7d , Supplementary Table 4 ), and the identities of the differential compounds in CCR2 (−/*) line 12, which are in line with the position of CCR2 in the lignin biosynthetic pathway (Supplementary Table 4 ), again underline the haploinsufficiency of the mutant P. alba CCR2 allele in CCR2 (−/*) line 12. A haploinsufficient P. alba CCR2 allele in CCR2 (−/*) line 12 In CCR2 (−/*) line 12, the P. alba CCR2 protein sequence differs in only two amino acids from the wild-type P. alba CCR2 protein sequence (Supplementary Fig. 2 ). Because the analysis of monoallelic CCR2 mutants suggested that the wild-type CCR2 alleles are haplosufficient in P. tremula × P. alba (Table 2 , Fig. 7 ), the reduced lignin content of CCR2 (−/*) line 12 is probably the consequence of the mutation in the P. alba CCR2 allele (in combination with the null P. tremula CCR2 allele), which might encode an enzyme with a lower CCR activity or a lower protein stability as compared to the wild-type P. alba CCR2 -encoded enzyme. To validate this, yeast assays were performed in which the activities of the wild-type and the mutan t P. alba CCR2 proteins were investigated based on the production of coniferaldehyde from feruloyl-CoA, the substrate of CCR2. As it was not possible to feed feruloyl-CoA to the yeast cultures, whereas it was possible to feed ferulic acid, we created yeast strains that express 4-coumarate:CoA ligase ( 4CL ). The 4CL enzyme converts ferulic acid to feruloyl-CoA, allowing the CCR2 enzyme activities to be tested (Fig. 8a ). All yeast cultures were fed with ferulic acid and extracts were analyzed by GC-MS. Yeast cultures expressing only 4CL produced no coniferaldehyde upon feeding with ferulic acid (Fig. 8b ). However, upon co-expression of 4CL and wild-type P. alba CCR2 , coniferaldehyde was formed (peak 1, Fig. 8b ; Supplementary Fig. 12a, b ). In addition to coniferaldehyde, two other metabolites, which were identified based on their EI-MS spectra as dihydroconiferyl alcohol and coniferyl alcohol (peak 2 and peak 3, respectively), accumulated in this strain (Fig. 8b ; Supplementary Fig. 12c, d ). As these two metabolites also accumulated in yeast cultures expressing only 4CL and additionally fed with coniferaldehyde (Fig. 8b ), we concluded that yeast cells further metabolize coniferaldehyde to dihydroconiferyl alcohol and coniferyl alcohol. Therefore, these two metabolites can be used as additional diagnostic markers for the production of coniferaldehyde in yeast cultures. Fig. 8: CCR2 activity assays in yeast. The relative activity of the mutant P. alba CCR2 protein (as present in CCR2 (−/*) line 12) was determined in yeast. a Principle of the yeast feeding assay. Yeast cultures were engineered to express 4CL and the wild-type P. alba CCR2 protein or mutated P. alba CCR2 protein (as present in CCR2 (−/*) line 12). After feeding the yeast cultures with ferulic acid, the activity of the respective CCR2 protein was judged based on the production of coniferaldehyde (the product of CCR2, peak 1), coniferyl alcohol (peak 2), and dihydroconiferyl alcohol (peak 3). See Supplementary Fig. 12 for the spectra of peaks 1–3. b GC-MS chromatograms of an authentic coniferaldehyde standard and extracts from ferulic acid-fed yeast cells expressing 4CL in combination with either an empty vector (EV), mutant P. alba CCR2 or wild-type P. alba CCR2 . The results shown are representative of five biologically independent samples. Full size image In contrast to yeast cultures expressing both 4CL and the wild-type CCR2 , yeast cultures expressing 4CL and the mutated P. alba CCR2 failed to produce coniferaldehyde, dihydroconiferyl alcohol, and coniferyl alcohol (Fig. 8b ). Based on these data, we concluded that, in yeast, no detectable enzymatic activity was present for the mutant P. alba CCR2 protein. The normal growth but reduced lignin content of CCR2 (−/*) line 12 suggests that, in planta , the mutant P. alba CCR2 protein had a reduced CCR activity and/or reduced protein stability as compared to the wild-type P. alba CCR2 protein, but did not fully lose its activity, as this would lead to dwarfism as observed in the CCR2 (−/−) knockout mutants. The apparent null-activity in yeast might be explained by the lack of the proper cellular context of a normal lignifying cell, such as pH, interaction, and/or stabilization partners, etc., which might influence CCR2 enzymatic activity. Not all CCR2 (−/*) lines display a normal growth phenotype Using gRNA3 targeting the fourth exon of CCR2 , another CCR2 (−/*) line was obtained having an indel pattern in the CCR2 alleles that was similar to that present in the CCR2 alleles of CCR2 (−/*) line 12 (albeit in a different position of the CCR2 gene). CCR2 (−/*) line 206 contained a frameshift mutation (5-bp deletion) in the P. tremula CCR2 allele, and a deletion of 3 bp in the P. alba CCR2 allele (Supplementary Fig. 5b ). The latter resulted in a substitution of Trp175 and Asp176 for a Tyr175 in the corresponding P. alba CCR2 protein sequence (Supplementary Fig. 13 ). The amino-acid change occurred in α6 of the CCR2 protein, but not in the active site, NAPD-binding domain, or substrate-binding pocket residues 9 , 10 , 11 . CCR2 (−/*) line 206 had a lignin amount and growth that was increased when compared to that of CCR2 (−/−), but still severely reduced when compared to that of the wild type (Supplementary Fig. 14a, b, d ). However, similar to CCR2 (−/−) stems, CCR2 (−/*) line 206 stems also displayed a red xylem coloration and collapsed vessels (Supplementary Fig. 14c, d). Discussion In previously generated CCR2 -downregulated poplars, the red xylem phenotype appeared in patches, even among clones originating from the same plant, as a consequence of unstable downregulation 3 . A similar observation was made for poplars with RNAi-mediated downregulation of 4CL1 4 . In contrast, CCR2 (−/*) line 12 poplars had a stable reduction in lignin amount along the stem and in all biological replicates, as judged from the uniformly distributed red xylem phenotype, which is a big step forward in achieving stably modified lignocellulosic biomass. After alkaline pretreatment, the cellulose-to-glucose conversion increased from 70.9% in wild type to an almost complete conversion (95.9%) in CCR2 (−/*) line 12. Also without pretreatment, CCR2 (−/*) line 12 had a 36% higher cellulose-to-glucose conversion than the wild type. Since no biomass penalty was observed for CCR2 (−/*) line 12, even on a plant basis, this line yielded substantially more sugar than the wild type upon limited-saccharification experiments. However, before further translation of this knowledge to generate feedstock for the biorefinery, CCR2 (−/*) line 12 poplars remain to be evaluated for growth and lignin amount when cultivated in the field, where they are exposed to different biotic and abiotic stresses. Indeed, it has been shown that genetically modified trees that grow normally in the greenhouse do not always grow as well as wild type when grown in the field 20 . But even if CCR2 (−/*) line 12 plants would have a yield penalty in the field, the specific mutation present in the P. alba CCR2 allele of CCR2 (−/*) line 12 might still be valuable to engineer lignin in another cultivar or crop (such as Eucalyptus), because the effect of a mutation on the phenotype depends not only on the environment, but also on the specific genetic background in which it resides 21 . Typically, using CRISPR/Cas9, knockout lines are generated. However, knockout mutants in lignin biosynthesis frequently suffer from undesired phenotypes such as dwarfism 6 , 7 , 8 , 18 . Lignin levels in plants can be reduced without affecting plant growth, but if lignin drops below a critical level, effects on vessel architecture and growth become apparent 7 , 8 , 18 , 22 . For example, for CCR2 -downregulated (RNAi-mediated) poplars, a range of plants with different levels of CCR2 downregulation and growth was obtained (even amongst biological replicates originating from the same plant) 3 . Approximately 5% of all CCR2 -downregulated lines that have been generated displayed severe dwarfism and could only survive in tissue culture 5 . These lines were most likely the plants with the highest reduction in CCR2 activity. Of the CCR2 -downregulated plants that survived on soil, the plants with the highest amount of red coloration (and thus the lowest amount of CCR expression and lignin amount) had the highest increase in saccharification efficiency, but also showed collapsed vessels and suffered from a biomass yield penalty 3 , 5 . Stems with lower amounts of red coloration did not suffer from a yield penalty, but only had a marginal reduction in lignin and increase in saccharification efficiency. Here, a similar observation was made, but in individual, stably CRISPR/Cas9-edited poplars; CCR2 (−/−) knock-outs had the lowest amounts of lignin and displayed collapsed vessels and severe dwarfism. CCR2 (−/*) line 206 had a slightly increased amount of lignin and biomass yield when compared to CCR2 (−/−) but still displayed collapsed vessels, whereas CCR2 (−/*) line 12, with its mild reduction in lignin amount compared to the wild type, displayed normal vessels and growth, even in clonally replicated trees. As the wild-type CCR2 alleles are both haplosufficient in P. tremula × P. alba , the differences in lignin amount and growth between CCR2 (−/*) line 206 and line 12 could be attributed to the 3 bp deletions occurring in the fourth and third exon of P. alba CCR2 of CCR2 (−/*) line 206 and line 12, respectively (see Supplementary Note 1 for an explanation of the impossibility of the off-target effect). Our research shows that generating a range of allelic variants in the gene of interest by CRISPR/Cas9 is a useful strategy to pick up rare alleles that allow obtaining the desired (intermediate) phenotype (e.g., plants with reduced amounts of lignin without displaying a yield penalty, such as CCR2 (−/*) line 12). For haplosufficient genes in outbreeding diploid species (such as most tree species), an interesting strategy is to knock out one allele, while the other can be screened for mutations leading to (small) amino-acid modifications, as described in this paper. Alternatively, screening for lines in which both alleles contain mutations leading to (small) amino-acid modifications might also be valuable. The yeast assay shows that, although it seems to be a fast, cheap, and easy way to evaluate the activity of specific allelic variants—or even to screen for specific allelic variants with modified activity—the results obtained are not necessarily good predictors for the activity of such variants in planta . Therefore, a direct in planta screening for such variants is a better strategy. Methods Plant material and vector construction To introduce biallelic mutations in CCR2 , a list of 30 protospacers with the N20-NGG motif specific for the P. tremula × P. alba CCR2 alleles (Potri.003G181400) was extracted from the Aspen database ( ) 23 , 24 . Next, the possible protospacers were analyzed based on their position in the CCR2 alleles and the possible off-targets via the Aspen database 23 , 24 . In addition, GC-content and absence of a TTTTT sequence were considered. Based on these parameters, the most suitable gRNA sequence was chosen: GACCAAAAATGTGATCATTG (gRNA1). gRNA1 targets the third exon of both CCR2 alleles. Using the pUC gRNA Shuttle (Addgene plasmid #47024) plasmid, gRNA1 was cloned into the p201N:Cas9 plasmid (Addgene plasmid #59175) by Gibson assembly 25 . The p201NCas9:gRNA1_CCR2 vector was transferred into Agrobacterium tumefaciens strain C58C1 pMP90 by electroporation. Agrobacterium-mediated transformation of P. tremula × P. alba 717-1B4 was performed via co-cultivation 26 . For this, 40 explants (which were preincubated on solidified M1 medium (see below) for 48 h at 24 °C in the dark) were dipped into 25 mL of Agrobacterium solution (grown on M liquid medium (see below) until reaching a concentration of 5 × 10 8 cfu per mL) and slowly stirred for 16 h. After blotting on sterile paper, the explants were placed on solidified M1 medium for 48 h. Subsequently, the explants were washed with tetracycline solution (25 mg per liter) and sterile water, and transferred onto M2 medium (see below) for 10 days at 24 °C. Transferring the explants to solid M3 medium (see below) with 100 mg per liter kanamycin in standard light conditions yielded regenerated shoots that could be excised and transferred to M1/2 medium (see below) with 50 mg per liter kanamycin. After growing for approximately 20 days in standard light conditions, the plants (that developed roots and elongated shoots) were micropropagated on M1/2 medium (see below) with 50 mg per liter kanamycin. To introduce monoallelic mutations in CCR2 , specific gRNAs (targeting either the P. tremula or the P. alba CCR2 allele) were selected based on the criteria described above. The best suitable gRNA sequences were GTGGTATTGCTATGGAAAGG (gRNA2) and GGAACAAGCTGCATGGGATA (gRNA3). gRNA2 and gRNA3 target the fourth exon of the P. alba and P. tremula CCR2 allele, respectively. Cloning of the gRNAs in the p201N-Cas9 vector, subsequent transformation into A. tumefaciens and poplar transformation were performed as described above. For genotyping the regenerated, transformed shoots, the following primers pairs (designed using the Primer3 0.4.0 software) were used for PCR for identifying the mutations present in the CCR2 alleles. For the PCR using the DNA extracted from plants transformed with the construct containing gRNA1: forward 5′-TACAYGGTAATTAATGGTGG-3′, reverse 5′-GATACCTTGGTGTTCTTGC-3′; for gRNA2: forward 5′-AGCTTGCCCGTTCTGTGTT-3′, reverse 5′-CGGTGAGGTACTTGAGGATG-3′; for gRNA3: forward 5′-ACCCCGTTCTGGTAGCTG-3′, reverse 5′-GGAAGGCGTCTCAAAGACT-3′. The PCR products were sequenced by Eurofins (Eurofins Genomics) and analyzed via CLC Main Workbench 8. The wild-type controls ( P. tremula × P. alba 717-1B4) originate from the same batch of plants that was used to generate the transgenic lines. However, instead of generating callus-tissue and performing Agrobacterium-mediated transformation, the wild-type controls were maintained and clonally propagated to serve as a control for the regenerated, transformed shoots. For clonal propagation, poplars that were grown for 3 to 4 months in tissue culture (on M1/2 medium as described below) under long-day conditions (16-h light and 8-h dark photoperiod, 24 °C) were used. More specifically, the stems were cut into pieces of ±2 cm and put on fresh M1/2 medium to allow the development of roots and new shoots. Media used for plant cultivation M: Murashige and Skoog basic medium (Duchefa) 4.4 g per liter, Morel and Wetmore vitamins 10 mL per liter, L -cystein 1 mg per liter, L -glutamine 200 mg per liter, sucrose 30 g per liter, plant agar (Duchefa) 6.2 g per liter. M1/2: as M but with half-strength macro-nutrients, sucrose 20 g per liter, IAA 3 µM. M1: as M but with NAA 10 µM and 2iP 5 µM. M2: as M1 but with carbenicillin 500 mg per liter, cefotaxime 250 mg per liter. M3: as M but with carbenicillin 500 mg per liter, cefotaxime 250 mg per liter, thidiazuron 0.1 µM. Plant growth and harvest After growing for four months in tissue culture under long-day conditions, the transgenic poplars and their wild-type controls were transferred to soil. More specifically, the poplars were transferred to pots of 5.5-cm diameter filled with Saniflor commercial soil (Van Israel nv), placed in a tray filled with water, and covered with a cage liner (Tecniplast APET disposable cage liner for cage body 1291H) for acclimatization. After 2 weeks, one side of the cage liner was lifted 1 cm above water level and kept accordingly for 1 day, after which the other side also was lifted. The next day, the cage liner was removed and the acclimatized plants were transferred to bigger pots filled with a Saniflor commercial soil (Van Israel nv) (10 liter). Note that CCR2 (−/−) plants were found to be extremely sensitive to the adjustment to greenhouse conditions. Therefore, here, the cage liner was kept closed for a period of 5 weeks and slowly opened over a period of 3 weeks to allow a much slower transition to the less-humid greenhouse conditions. Even with this precautionary measure, only a small number of CCR2 (−/−) poplars recovered, whereas all plants of the other genotypes survived. In total, six batches of plants were grown for 11 or 20 weeks in the greenhouse under a 16-h light and 8-h dark photoperiod at ~21 °C. The first batch, containing wild type, the 20 individual CCR2 (−/−) lines, and CCR2 (−/*) line 12, was grown for 11 weeks in the greenhouse and used for a picture of the growth phenotype (Fig. 1 ). The second batch, containing wild type and CCR2 (−/*) line 12, was grown for 20 weeks in the greenhouse. The height of the trees was determined weekly (Fig. 2a , b). After that, the diameter of the stems was determined (3 cm above soil level) (Table 1 ). Next, the stems were harvested (40 cm above soil level), the leaves were removed from the stem and the fresh weight of the stems was determined before and after removing the bark (Table 1 ). The dry weight of the debarked stems was determined after air-drying the stems for 2 weeks at ambient temperature (Table 1 ). The dried, debarked stem was ground in a ball mill for cell wall analysis and saccharification assays (Table 1 , Fig. 4 , Supplementary Tables 2 and 3 ). The third batch consisted of the same lines present in batch 2 (wild type and CCR2 (−/*) line 12), but now also accompanied by CCR2 (−/−) mutants. These plants were also grown for 20 weeks in the greenhouse. For wild type and CCR2 (−/*) line 12, 5-cm-long stem parts between 15 and 20 cm (relative to the soil) were harvested, debarked, and stored in tap water for 1 day. For CCR2 (−/−), the 3-cm-long stem parts between 3 and 6 cm (relative to the soil) were harvested, debarked, and stored in tap water for 1 day. These stem pieces were used for microscopy (Fig. 3 ). The fourth batch, containing wild type, CCR2 (+/−), CCR2 (−/+), and CCR2 (−/*) line 12, was grown for 11 weeks in the greenhouse. After that, the diameter of the stems was determined (3 cm above soil level). Next, the stems were harvested (5 cm above soil level), and the fresh weight was determined after removing the leaves and bark. The dry weight was determined after drying the stems for 5 days at 50 °C (Table 2 ). For acetyl bromide and thioacidolysis (Table 2 ), the harvested stems were debarked, air-dried, and ground in a ball mill (Fig. 6 ). The fifth batch consisted of the same lines present in batch 4 (wild type, CCR2 (+/−), CCR2 (−/+), and CCR2 (−/*) line 12), but now also accompanied by CCR2 (−/−) mutants (Fig. 5 ). After growing for 11 weeks in the greenhouse, the stems were harvested (5 cm above soil level) and the bottom 10 cm of the harvested stem piece was debarked, snap-frozen in liquid nitrogen, and stored at −70 °C for metabolomics (Fig. 7 , Supplementary Table 4 , Supplementary Figs. 6 – 11 ). The sixth batch, containing wild type, CCR2 (−/*) line 206, and CCR2 (−/−) was grown for 20 weeks in the greenhouse. The height of the trees was determined weekly (Supplementary Fig. 14a–c ). For the wild type, CCR2 (−/*) line 206, and CCR2 (−/−), the stems were harvested 10, 3, and 3 cm above soil level, respectively. Of each harvested wild-type, CCR2 (−/*) line 206, and CCR2 (−/−) stem pieces, the basal 3, 1, and 1 cm, respectively, was debarked and stored for 3 days in tap water for microscopy (Supplementary Fig. 14d ). Microscopy Fifteen-micrometer-thick stem slices were made using a Reichert-Jung 2040 Autocut Microtome (Leica). Sections were stained with Wiesner and Mäule reagents; Wiesner staining was performed by adding a drop of phloroglucinol-HCl solution (consisting of one volume of concentrated HCl (37 M) and two volumes of 3% phloroglucinol in ethanol) to the sections. Mäule staining was performed by incubating the sections in 0.5% (w/v) potassium permanganate for 5 min, followed by washing with distilled water. Next, the sections were incubated in concentrated HCl (37 M), followed by the addition of concentrated ammonium hydroxide solution. Images were acquired using an Olympus BX51 microscope (Olympus) with an Olympus PlanC N 10x (0.25 NA) objective or Olympus UplanFl N 20x (0.50 NA) objective and the Toupview x64,3.7.10121 software. Sections were also imaged via autofluorescence using a Zeiss Axio Imager.M1 microscope (Carl Zeiss) with a Plan-Apochromat 10X (0.45 NA) objective and the AxioVs40 V4.8.2.0 software. A 365-nm excitation filter was used together with a 395-nm beamsplitter and a 420- to 470-nm bandpass emission filter. Cell wall characterization To determine the matrix polysaccharides, cellulose, and lignin characteristics, ground wood powder was used for preparing cell wall residue by sequentially washing for 30 min each with milliQ water at 98 °C, ethanol at 76 °C, chloroform at 59 °C, and acetone at 54 °C. The remaining CWR was dried under vacuum. To determine the crystalline cellulose amount, the Updegraff method 27 was used on 10 mg of CWR. First, the samples were incubated with 1 mL trifluoroacetic acid (2 M) for 2 h at 99 °C while shaking at 750 rpm, followed by centrifugation for 3 min at 19,757 × g . Next, the pellet was washed three times with 1 mL milliQ water, two times with 1 mL acetone, dried under vacuum, and weighed. The weight loss upon trifluoroacetic acid digestion was used to determine the matrix polysaccharide content (including hemicelluloses, pectins, and amorphous cellulose). Subsequently, 1 mL of Updegraff reagent (consisting of eight volumes of acetic acid, one volume of nitric acid, and 2 volumes of milliQ water) was added to the pellet, followed by vortexing and heating at 99 °C for 30 min. After centrifuging the samples for 15 min at 10,080 × g , the supernatant was discarded without disturbing the pellet. Next, the pellet was washed one time with 1 mL water and two times with 1 mL acetone. After drying under vacuum, the pellet was incubated with 175 µL of 72% (v/v) sulfuric acid for 30 min at room temperature. Next, the samples were vortexed and incubated for another 15 min at room temperature. After adding 825 µL of water, the samples were centrifuged for 5 min at 10,080 × g . To wells of a 96-well plate, 5 µL of supernatant, 95 µL of water and 200 µL of freshly prepared anthrone reagent (2 mg anthrone per mL pure sulfuric acid) was added. The plate was sealed and incubated for 30 min at 80 °C. After cooling down to room temperature, the absorption was measured at 625 nm with a NanoDrop® ND-1000 spectrophotometer (Thermo Scientific). The Nanodrop 1000 3.8.1 software was used for collecting the absorption data. Lignin content was determined by the acetyl bromide protocol 18 , 28 on 5 mg of CWR, and the Klason protocol 7 , 29 on 100 mg of CWR. In the acetyl bromide protocol, the samples were incubated in 250 µL freshly made acetyl bromide solution (25% acetyl bromide in glacial acetic acid) for 2 h at 50 °C. Next, the samples were incubated for another hour at 50 °C with vortexing every 15 min. After cooling the samples on ice and centrifuging for 15 min at 10,080 × g , 100 µL supernatant, 75 µL freshly made hydroxylamine hydrochloride (0.5 M), 400 µL NaOH (2 M), and 1.425 mL glacial acetic acid were added to an empty 2-mL Eppendorf. The lignin concentrations were determined by measuring the absorbance at 280 nm with a NanoDrop® ND-1000 spectrophotometer (Thermo Scientific) and applying the Bouguer–Lambert–Beer law. The Nanodrop 1000 3.8.1 software was used for collecting the absorption data. In the Klason protocol, the samples were incubated with 1 mL sulfuric acid (72%) in 15-mL glass vials for 2 h at room temperature while stirring. After transferring the samples to a 100-mL flask, 13.4 mL of milliQ water was added. The flasks were autoclaved for 1 h (1 bar, 121 °C). Next, the samples were incubated for 16 h at 4 °C. To measure the acid-soluble lignin, 1 mL of supernatant was collected (see further). The remaining samples were filtered and washed extensively with 200 mL of milliQ water through a preweighed filter paper (Sartorius AG) using a Büchner filter system (Merck Millipore). The filter papers were dried for 16 h at 105 °C and cooled down for 1 h at room temperature. The lignin amount was determined gravimetrically. For determining the acid-soluble lignin, the previously collected 1 mL of supernatant was centrifuged and diluted 20 times in milliQ water. The absorbance at 205 nm was measured using a spectrophotometer (Genesys 10 S UV-Vis, Thermo Scientific). The acid-soluble lignin concentrations were calculated by means of the Bouguer–Lambert–Beer law. For the lignin composition determination via thioacidolysis 14 , 30 , 15 mg of CWR was weighed into a 5‐mL glass Wheaton vial with Teflon‐lined screw‐cap. First, the samples were incubated with 1 mL reaction mixture (consisting of 2.5% boron trifluoride etherate and 10% ethanethiol in dioxane) for 4 h at 98 °C, while shaking the vials manually every half an hour. Next, the samples were incubated for 5 min at −20 °C. To the samples, 0.2 mL tetracosane in dichloromethane (5 mg per mL) and 0.3 mL of 0.4 M sodium bicarbonate was added. Next, the samples were extracted by adding 2 mL milliQ water and 1 mL dichloromethane. By using a Pasteur pipette packed with a small cotton plug and a spatula-point of anhydrous sodium sulfate, 0.5 mL of the organic phase was filtered and added to a new Eppendorf. The samples were dried under vacuum and resuspended in 0.2 mL dichloromethane. Derivatisation occurred by adding 20 μL pyridine and 100 μL N , O ‐bis(trimethylsilyl) acetamide to 20 μL of resuspended sample and incubating the samples for 2 h at 25 °C while shaking at 750 rpm. The reaction product was analyzed by GC-MS. GC-MS analysis was carried out using a 7890B GC system equipped with a 7693A Automatic Liquid Sampler and a 7250 Accurate-Mass Quadrupole Time-of-Flight MS system (Agilent Technologies). One microliter of the reaction product was injected in splitless mode with the injector port set to 280 °C. Separation was realized with a VF-5ms column (30 m × 0.25 mm, 0.25 μm; Varian CP9013; Agilent Technologies) with helium carrier gas at a constant flow of 1.2 mL per min. The oven was held at 130 °C for 3 min post-injection, ramped to 200 °C at 10 °C per min, ramped to 250 °C at 3 °C per min, held at 250 °C for 5 min, ramped to 320 °C at 20 °C per min, held at 320 °C for 5 min, and finally cooled to 130 °C at 50 °C per min at the end of the run. The MSD transfer line was set to 280 °C and the electron ionization energy was 70 eV. Full EI-MS spectra were recorded between m / z 50 and 800 at a resolution of >25,000 and with a solvent delay of 10.0 min. Peak integrations for quantification of the lignin monomers were carried out using the MassHunter Quantitative Analysis (for QTOF) software package (Agilent Technologies). For the lignin composition determination via NMR 31 , 32 , 33 , 40–50 mg of ground stem powder was suspended in 0.6 mL DMSO-d 6 :pyridine-d 5 (4:1, v/v). The samples were sonicated, with occasional mixing by vortexing, until a uniform gel was formed. NMR experiments were performed on a Bruker Biospin (Billerica) Avance 700 MHz spectrometer equipped with a 5-mm QCI 1 H/ 31 P/ 13 C/ 15 N cryoprobe with inverse geometry (proton coils closest to the sample). As in internal reference, the central DMSO solvent peak was used (δ C 39.5, δ H 2.49 ppm). The 1 H– 13 C correlation experiment was an adiabatic HSQC experiment (Bruker standard pulse sequence ‘hsqcetgpsisp2.2’; phase-sensitive gradient-edited-2D HSQC using adiabatic pulses for inversion and refocusing). The parameters used for the HSQC experiments were acquired from 10 to 0 ppm in F2 ( 1 H) with 1398 data points (acquisition time, 100 ms) and 200 to 0 ppm in F1 ( 13 C) with 570 increments (F1 acquisition time, 8 ms) of 32 scans with a 1 s interscan delay. The d 24 delay was set to 0.86 ms (1/8 J, J = 145 Hz). The total acquisition time for a sample was ~5 h. Processing used typical matched Gaussian apodization (GB = 0.001, LB = −0.5) in F2 and Gaussian apodization (GB = 0.001, LB = −0.3) also in F1 (without using linear prediction). Volume integration of contours in HSQC plots used TopSpin 4.0.8 software, and no correction factors were used. Analytical gel-permeation chromatography Ground powder (~150 mg) was treated with cellulase (5 wt% to biomass) in sodium acetate buffer (pH 5.0, 40 mL) at 37 °C for 72 h. After centrifugation, the precipitate was collected and washed five times with deionized water to obtain enzyme lignin for gel-permeation chromatography (GPC) analysis. Enzyme lignin (~10 mg) was dissolved in dimethylformamide (DMF)/lithium bromide (LiBr) (700 μL, 0.1 M LiBr) solution and filtered through a syringe filter (0.45 μm, PVDF). GPC analysis was performed utilizing a Shimadzu LC20-AD LC pump equipped with a Shimadzu SPD-M20A UV-vis detector set at 280 nm and a Shimadzu RID-10A refractive index detector. The GPC column set consists of 4 TOSOH (TOSOH Bioscience, LLC) GPC columns and a guard column (TSKgel Guard Alpha 6.0 mm ID × 4.0 cm, 13 μm → TSKgel Alpha-M 7.8 mm ID × 30 cm, 13 μm → TSKgel Alpha-M 7.8 mm ID × 30 cm, 13 μm → TSKgel Alpha-2500 7.8 mm ID × 30 cm, 7 μm → TSKgel Alpha-2500 7.8 mm ID × 30 cm, 7 μm). The column oven was held at 50 °C during analysis. The mobile phase was DMF with 0.1 M LiBr, the flow rate was 0.5 mL per min, and the oven temperature was 40 °C on an injection volume of 10 μL. Molecular weight distributions were determined using Shimadzu GPC postrun software via a conventional calibration curve using a ReadyCal polystyrene Kit (Sigma-Aldrich, Aldrich # 76552, M(p) 250-70000). Saccharification assay Saccharification was performed on 10 mg of dried, ground stem material 34 . The samples were saccharified for 72 h using no pretreatment, acidic pretreatment (1 M HCl, 80 °C for 2 h while shaking at 750 rpm), or alkaline pretreatment (62.5 mM NaOH, 90 °C for 3 h while shaking at 750 rpm). After the pretreatment, the samples were centrifuged for 5 min at 10,080 × g . The pellet was washed three times with 1 mL milliQ water, and incubated in 1 mL of 70% ethanol for 16 h at 55 °C. After another centrifugation step (5 min at 10,080 × g ), the pellet was washed three times with 1 mL 70% ethanol and one time with 1 mL acetone, centrifuged for 5 min at 10,080 × g , 10,000 rpm, dried under vacuum, and weighed. The enzyme mix consisted of a 5:3 ratio of cellulase from Trichoderma reseei ATCC 26921 and β-glucosidase (Novozyme) which were first desalted over an EconoPac 10DG column (Bio-Rad), stacked with Bio-gel® P-6DG gel (Bio-Rad) according to the manufacturer’s guidelines. The activity of the enzyme mix was measured with a filter paper assay and was 0.18 FPU (filter paper units) per mL. Subsequently, the samples were dissolved in 1 mL acetic acid buffer solution (pH 4.8) and incubated at 50 °C at 750 rpm. After 5 min of incubation, 100 μL freshly prepared enzyme mix was added. After spinning down the samples in a benchtop microcentrifuge, 20 μL of the supernatant was taken after 72 h of incubation at 50 °C and 30-fold diluted with acetic acid buffer (pH 4.8). To determine the concentration of glucose in these samples, a spectrophotometric color reaction was used (glucose oxidase-peroxidase (GOD-POD); 100 mL of GOD-POD reaction mixture contained 50 mg 2,2′-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid), 44.83 mg GOD, and 173 μL POD (4% (w/v)) in acetic acid buffer (pH 4.5)). To this end, 50 μL of 30-fold diluted sample was added to 150 μL GOD-POD mixture and incubated for 30 min at 37 °C. The absorbance was measured spectrophotometrically at a wavelength of 405 nm (Microplate-reader SpectraMax 250 (Sopachem), SoftMax Pro version 5 was used for collecting data). The concentration in the original sample was calculated with a standard curve based on known D-glucose concentrations. Phenolic profiling For phenolic profiling, the frozen debarked stem parts were cut into little pieces using scissors. Subsequently, the stem pieces were extracted by adding 1 mL methanol and incubating for 15 min at 70 °C while shaking at 1000 rpm. After centrifugation at 19,757 × g , 800 μL of the supernatant was transferred to a new Eppendorf and dried under vacuum. After dissolving the pellet in 100 μL cyclohexane, 100 μL milliQ water was added and the cyclohexane/water mixture was vortexed. After centrifugation at 19,757 × g , 70 μL of the water phase was subjected to UHPLC-MS on an ACQUITY UPLC I-Class system (Waters) consisting of a binary pump, a vacuum degasser, an autosampler, and a column oven. Chromatographic separation was performed on an ACQUITY UPLC BEH C18 (150 × 2.1 mm, 1.7 μm) column (Waters), while maintaining the temperature at 40 °C. A gradient of two buffers (A and B) was utilized: buffer A (99:1:0.1 water:acetonitrile:formic acid, pH 3) and buffer B (99:1:0.1 acetonitrile:water:formic acid, pH 3), as follows: 99% A for 0.1 min decreased to 50% A in 30 min, decreased to 0% from 30 to 40 min. The flow rate was 0.35 mL per min, and the injection volume was 10 μL. This UHPLC system was connected to a Vion IMS QTOF hybrid mass spectrometer (Waters). The LockSpray ion source was used in negative electrospray ionization mode under the following specific conditions: capillary voltage, 3 kV; reference capillary voltage, 2.5 kV; cone voltage, 30 V; source offset, 50 V; source temperature, 120 °C; desolvation gas temperature, 550 °C; desolvation gas flow, 800 liter per h; and cone gas flow, 50 liter per h. The collision energy for full MSe was set at 6 eV (low energy) and ramped from 20 to 70 eV (high energy), intelligent data capture intensity threshold was set at 5. For DDA-MSMS, the low mass ramp was ramped between 15 and 30 eV. The high mass ramp was ramped between 30 and 70 eV. Nitrogen (greater than 99.5%) was used as desolvation and cone gas. Leucin-enkephalin (250 pg per μL solubilized in water:acetonitrile 1:1 (v/v), with 0.1% formic acid) was utilized for the lock mass calibration, with scanning every 2 min at a scan time of 0.1 s. Profile data were recorded through a UNIFI Scientific Information System (Waters). Data processing was performed with Progenesis QI software version 2.4 (Waters). All 30,662 peaks (mass-to-charge ratio [ m/z ] features) were integrated in the chromatograms of wild type, CCR2 (+/−), CCR2 (−/+), CCR2 (−/*) line 12, and CCR2 (−/−). The 6182 peaks that had an abundance of at least 0.01% of the highest average in the group with the highest peak abundance were selected for further analysis. After ANOVA, peaks with a P value < 0.01 (false discovery rate adjusted) and a twofold difference in abundance between mutant and the wild type were considered as being significantly different. PCA and volcano plots were generated using MetaboAnalyst 4.0. Chemdraw 16 was used for calculating exact m/z values. CCR2 activity assays in yeast Malus domestica 4CL ( Md4CL ; GenBank Accession number XM_008365460) and yeast codon-optimized versions of the wild-type and mutant P. alba CCR2 alleles (Supplementary Fig. 15 ) were Gateway-cloned into the donor vector pDONR207 and sequence verified. For expression in yeast, the Md4CL entry clone was Gateway-recombined into the high-copy number yeast destination vector pAG426GAL-ccdB (AddGene plasmid 1415535 35 ). The wild-type and mutant CCR2 alleles were also Gateway-recombined into the high-copy number yeast destination vector pAG424GAL-ccdB (AddGene plasmid 1415135 35 ). The resulting expression clones were used to transform yeast strain W303-1A ( MATa ; leu2-3,112 , trp1-1 , can1-100 , ura3-1 , ade2-1 , his3-11,15 ) using the lithium acetate/single-stranded carrier DNA/polyethylene glycol method 36 . To this end, 50 mL YPD medium was inoculated with an overnight grown yeast pre-culture to an absorbance of 0.25 at 600 nm. Subsequently, the culture was grown until the absorbance reached 1.0 by incubating at 30 °C. After harvesting the cells by centrifugation, the cells were washed with 1 mL lithium acetate (0.1 M) and dissolved in 350 μL lithium acetate (0.1 M). For transformation, 50 μL-aliquots of the cells were transferred to an Eppendorf. To the aliquots, 200 μL PLI solution (for 50 mL PLI solution, mix 40 mL of 50% PEG with 5 mL lithium acetate (1 M) and 5 mL milliQ water), 10 μL salmon sperm ssDNA and plasmid DNA (1 μg of each plasmid) was added, followed by incubation for 30 min at 42 °C. After harvesting the cells by centrifugation, they were washed with 800 μL milliQ water and plated on selective plates with synthetic-defined (SD) medium (Clontech) supplemented with –Trp/–Ura amino-acid dropout supplement (Clontech). After incubating the plates for 2 days at 30 °C, five independent transformed colonies were selected. In total, five independent biological cultures were made for strains containing (1) pAG426GAL- Md4CL + pAG424GAL-ccdB (empty vector control), (2) pAG426GAL- Md4CL + pAG424GAL-wild type_ CCR2 , and (3) pAG426GAL- Md4CL + pAG424GAL-mutant_ CCR2 . Yeast precultures of each of the five biological replicates were grown in SD medium with –Trp/–Ura dropout supplement (Clontech) for 24 h at 30 °C with shaking at 300 rpm. Gene expression was induced by washing the precultures with milliQ water and inoculating them in 10 mL SD Gal/Raf medium with –Trp/–Ura dropout supplement. After incubating the cultures for 24 h, 500 μL ferulic acid (20 mM, in 50:50 ethanol:water) or 500 μL coniferaldehyde (20 mM, in 50:50 ethanol:water) was added. The cultures were incubated further for 48 h. Finally, 1 mL of each yeast culture was harvested. After extracting the medium three times with 500 µL ethyl acetate, the solvent of the combined organic fractions was dried under vacuum. The remaining pellet was derivatized by using 10 µL pyridine and 50 µL N -methyl- N -trimethylsilyl-trifluoroacetamide (MSTFA) for GC-MS analysis on a GC model 6890 and MS model 5973 (Agilent Technologies). One microliter of the sample was injected in splitless mode with the injector port set to 280 °C. Separation was realized with a VF-5ms column (30 m × 0.25 mm, 0.25 μm; Varian CP9013; Agilent Technologies) with helium carrier gas at a constant flow of 1 mL per min. The oven was held at 80 °C for 1 min post-injection, ramped to 280 °C at 20 °C per min, held at 280 °C for 45 min, ramped to 320 °C at 20 °C per min, held at 320 °C for 1 min, and finally cooled to 80 °C at 50 °C per min at the end of the run. The MSD transfer line was set to 250 °C and the electron ionization energy was 70 eV. Full EI-MS spectra were recorded between m / z 60 and 800 with a solvent delay of 7.8 min. GC-MS data were recorded and visualized with Agilent firmware and visualized via the AMDIS software (Version 2.6, NIST). Statistical analyses MS Excel 2016 and SAS 9.4 were used for statistical analysis. The specific method used is mentioned in the respective Table and Figure legends. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability Data supporting the findings of this work are available within the paper and its Supplementary Information files. A reporting summary for this Article is available as a Supplementary Information file. The datasets generated and analyzed during the current study are available from the corresponding author upon request. Sequences data that support the findings of this study were obtained from the Aspen database (for the CCR2 sequences from P. tremula × P. alba; ) and NCBI (for the Malus domestica 4CL sequence; ), or reported in Supplementary Information file. Source data are provided with this paper.

Researchers led by prof. Wout Boerjan (VIB-UGent Center for Plant Systems Biology) have discovered a way to stably finetune the amount of lignin in poplar by applying CRISPR/Cas9 technology. Lignin is one of the main structural substances in plants and it makes processing wood into, for example, paper difficult. This study is an important breakthrough in the development of wood resources for the production of paper with a lower carbon footprint, biofuels, and other bio-based materials. Their work, in collaboration with VIVES University College (Roeselare, Belgium) and University of Wisconsin (U.S.) appears in Nature Communications. Today's fossil-based economy results in a net increase of CO2 in the Earth's atmosphere and is a major cause of global climate change. To counter this, a shift toward a circular and bio-based economy is essential. Woody biomass can play a crucial role in such a bio-based economy by serving as a renewable and carbon-neutral resource for the production of many chemicals. Unfortunately, the presence of lignin hinders the processing of wood into bio-based products. Prof. Wout Boerjan (VIB-UGent) said, "A few years ago, we performed a field trial with poplars that were engineered to make wood containing less lignin. Most plants showed large improvements in processing efficiency for many possible applications. The downside, however, was that the reduction in lignin accomplished with the technology we used then—RNA interference—was unstable and the trees grew less tall." New tools Undeterred, the researchers went looking for a solution. They employed the recent CRISPR/Cas9 technology in poplar to lower the lignin amount in a stable way, without causing a biomass yield penalty. In other words, the trees grew just as well and as tall as those without genetic changes. Dr. Barbara De Meester (VIB-UGent) commented, "Poplar is a diploid species, meaning every gene is present in two copies. Using CRISPR/Cas9, we introduced specific changes in both copies of a gene that is crucial for the biosynthesis of lignin. We inactivated one copy of the gene, and only partially inactivated the other. The resulting poplar line had a stable 10% reduction in lignin amount while it grew normally in the greenhouse. Wood from the engineered trees had an up to 41% increase in processing efficiency". Dr. Ruben Vanholme (VIB-UGent) noted, "The mutations that we have introduced through CRISPR/Cas9 are similar to those that spontaneously arise in nature. The advantage of the CRISPR/Cas9 method is that the beneficial mutations can be directly introduced into the DNA of highly productive tree varieties in only a fraction of the time it would take by a classical breeding strategy." The applications of this method are not only restricted to lignin but might also be useful to engineer other traits in crops, providing a versatile new breeding tool to improve agricultural productivity.

10.1038/s41467-020-18822-w

Biology

Beware of sleeping queens underfoot this spring

'Harmonic radar tracking reveals random dispersal pattern of bumblebee (Bombus terrestris) queens after hibernation'. James C. Makinson, Joseph L. Woodgate, Andy Reynolds, Elizabeth A. Capaldi, Clint J. Perry, Lars Chittka. Scientific Reports. DOI: 10.1038/s41598-019-40355-6 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-019-40355-6

https://phys.org/news/2019-03-beware-queens-underfoot.html

Abstract The dispersal of animals from their birth place has profound effects on the immediate survival and longer-term persistence of populations. Molecular studies have estimated that bumblebee colonies can be established many kilometers from their queens’ natal nest site. However, little is known about when and how queens disperse during their lifespan. One possible life stage when dispersal may occur, is directly after emerging from hibernation. Here, harmonic radar tracking of artificially over-wintered Bombus terrestris queens shows that they spend most of their time resting on the ground with intermittent very short flights (duration and distance). We corroborate these behaviors with observations of wild queen bees, which show similar prolonged resting periods between short flights, indicating that the behavior of our radar-monitored bees was not due to the attachment of transponders nor an artifact of the bees being commercially reared. Radar-monitored flights were not continuously directed away from the origin, suggesting that bees were not intentionally trying to disperse from their artificial emergence site. Flights did not loop back to the origin suggesting bees were not trying to remember or get back to the original release site. Most individuals dispersed from the range of the harmonic radar within less than two days and did not return. Flight directions were not different from a uniform distribution and flight lengths followed an exponential distribution, both suggesting random dispersal. A random walk model based on our observed data estimates a positive net dispersal from the origin over many flights, indicating a biased random dispersal, and estimates the net displacement of queens to be within the range of those estimated in genetic studies. We suggest that a distinct post-hibernation life history stage consisting mostly of rest with intermittent short flights and infrequent foraging fulfils the dual purpose of ovary development and dispersal prior to nest searching. Introduction Wild pollinators are under threat by anthropogenic changes in landscape, including agricultural development and urban sprawl 1 . Knowledge of a species’ dispersal patterns can be crucial in predicting how animals respond to environmental change 2 . Bumblebees are important for the pollination of many native and wild plants throughout temperate ecosystems 3 , 4 , 5 , 6 , but the dispersal behaviors of bumblebee queens after hibernation is relatively unknown. Bumblebee colonies are comprised of a queen and up to a few hundred daughter workers. At the end of the colony’s annual life cycle, new queens and males are produced which then leave the hive to mate. The males die before the winter, whereas the new queens hibernate and go on to set up colonies in the spring. Dispersal of new queen bumblebees from their natal nest sites could occur at various stages of life, including before mating, between mating and hibernation, after emergence from hibernation but before nest searching, or while nest searching. To our knowledge, only two studies have explicitly examined the dispersal of queen bumblebees. By looking at genetic relatedness of bees from nearby colonies, Lepais and colleagues 7 estimated that new colonies were founded between three and five kilometers from their queen’s natal nest. Another study using genetic analyses and geographical sampling investigated the invasion of imported bumblebee species into Chile and suggested that Bombus terrestris queens spread up to about 200 km each year 8 . However, this surprisingly long distance was likely, as the authors point out, aided by the prevailing strong winds across the Andes, and perhaps due to anthropogenic causes, e.g. unintentional automotive transportation of queens within plants or soil. In temperate regions, bumblebees take an estimated two to three weeks to locate a new nest site and begin to forage for pollen 9 , but there is very limited information on the behavior of queen bumblebees immediately after emergence from hibernation. Several accounts from almost 60 years ago claim that when queen bumblebees first appear in the early days of spring they spend most of their time actively flying and foraging for nectar and pollen, which helps replenish their fat reserves and develop their ovaries, and within a short time they begin searching for nest sites 10 , 11 , 12 , 13 . However, there are no empirical data specifically on this life stage. The behavior of newly emerged queens and the contribution of this time period to colony dispersal is unknown. Using harmonic radar tracking of artificially over-wintered queen bees, we describe their behavior upon emergence and use a random walk model to estimate their dispersal pattern and displacement. Results Radar tracking of artificially over-wintered bumblebee queens We monitored 20 artificially over-wintered queen bumblebees with harmonic radar (Fig. 1a ). Queen bumblebees were kept at 4 °C until the morning of the start of experiments (Methods), at which point a transponder was superglued to the thorax of each bee (Fig. 1a inset). Subsequently, each bee was placed in a small depression in the center of a mound of earth (approx. 5 cm high and 15 cm in diameter), where they were allowed to warm up naturally. The activity of all bees was monitored with the harmonic radar for between 4 and 15 hours each day (depending on weather and therefore likelihood of any flights occurring) and until all the bees had permanently dispersed outside the range of the radar (Fig. 1a ). Three bees were monitored individually for up to three days each. The other 17 bees were monitored simultaneously for a period of five days. Figure 1 Queen bumblebees ( Bombus terrestris ) that have recently emerged from artificial or natural hibernation spend more time resting than flying. ( a ) Aerial view of field site where harmonic radar tracking took place (an orthomosaic created from drone photography; flight tracks overlaid using MATLAB 2015b). Yellow triangle designates position where queen bumblebees were placed prior to the start of experiment (release site). Orange square and blue circle indicate positions of the harmonic radar for the 17 group- and three individually-monitored bees, respectively. Pink lines indicate flight paths of radar tracked bees. Inset photo: queen bumblebee with a transponder fixed to her thorax. ( b ) Photos show observed behavior of recently emerged queen bumblebees. Most bees we observed landed on the ground after a short flight and proceeded to walk around within a few centimeter radius for several seconds before becoming motionless or, more often, burrowing their body or head into or under vegetation. Full size image The harmonic radar tracks of all 20 bumblebees show that queens tended to make very short (duration and distance) flights (means ± s.e. given throughout the text; 14.2 ± 1.0 s, 34.2 ± 2.3 m, n = 264) separated by long periods of inactivity (29.0 ± 13.8 min, n = 11). This suggests that newly emerged queen bumblebees, prior to nest searching, may spend less than two percent of their day flying, while spending a large majority of their time resting on the ground. Post-hibernation behavior of wild bumblebee queens To ensure that the behavior observed in harmonic radar monitored queen bumblebees was not due to the mass of the transponders attached to them, nor due to the fact that they came from commercially reared colonies, we also observed the behavior of recently emerged wild queen bumblebees (Methods). We found that wild queen bumblebees, recently emerged from hibernation (Methods), also spent long periods of time on the ground (13.3 ± 1.9 min; n = 74) between short (duration and distance) flights (8.7 ± 1.1 s, 4.1 ± 0.7 m; n = 34). Wild queen bumblebees usually sat very still shortly after landing. Immediately after landing and occasionally between periods of stillness, bees would take a very short, usually circuitous walk (radius approx. 2–5 cm), move under vegetation or groom themselves, and then become motionless. Besides these very short and infrequent movements, the time spent on the ground did not seem to involve any activity other than resting. No feeding or interaction with other queens was observed while on the ground. On only three of our 140 observations, occurring only on the last day of observations, the focal queen was seen to fly close to the ground in a zig-zagging pattern characteristic of nest searching, although we did not observe any of these queens enter or examine a potential nest site. On the first day of observations we noticed large numbers of queen bees feeding on the catkins of a nearby willow tree ( Salix caprea ). Of our focal queens, we observed one bee switch from resting to feeding and two bees switch from feeding on the willow tree to resting on the ground. However, most of the bees we observed resting on the ground did not feed from the nearby tree during our observations. On subsequent days we observed no feeding behavior. Resting behavior of queen bumblebees consisted of, in 61% of the bumblebees observed, positioning their heads under a leaf or within the folds of leaves, and many times crawling under leaves (Fig. 1b ). At no point did we observe bees during these between flight periods examining potential nest sites in the ground or burrowing underground. Dispersal behavior of artificially over-wintered queens Monitoring the activity of single queen bees allowed us to examine individuals’ behavior over time. The three individually-monitored queens varied in their flight behavior, but all showed multiple changes in direction during the day (Fig. 2a–c ). Despite not maintaining a consistent flight direction, two bees flew beyond the range of the radar (~600 m within the local topography) on the day of release and the other left the field site within two days, after rain stopped activity on the first day (Fig. 2d–f ). None were observed again on subsequent days. Figure 2 Individually-monitored queen bumblebees disperse slowly from their hibernation site. ( a – c ) Aerial view of field site for radar tracking with the flights (pink lines) of three artificially over-wintered queen bumblebees (an orthomosaic created from drone photography; flight tracks overlaid using MATLAB 2015b). Yellow triangle indicates release site. Blue circle indicates position of the harmonic radar. (e,f) Euclidean distance from the release point to the last positional fix in each flight of each individually-monitored bee. Full size image To get a better understanding of the overall dispersal behavior of queen bees, we monitored a group of 17 simultaneously-released queen bumblebees for five days. By the fifth day, all bees had dispersed beyond the range of the harmonic radar (~600 m; Fig. 3 ). Because all 17 bees were active during the same period and because we were unable to locate and observe individuals once they had left the release site, it was impossible to identify which bee produced which individual flight. However, the density of positional fixes moved from near the origin to spread across the observable field area within one day (Fig. 3 ). Further, the number of flights observed decreased dramatically after the second day (Fig. 3 ). These results indicate that most queens dispersed from the range of the radar within less than two days and did not return. Figure 3 Queen bumblebees disperse from their hibernation site with many short flights between long rest periods. ( a ) Data density plots of bee positional fixes overlaid onto aerial photos (an orthomosaic created from drone photography) of radar tracking field site, monitoring 17 queen bumblebees simultaneously over five days (flight tracks overlaid using MATLAB 2015b). During the first few hours of the first day, most bees remain near the release point, indicated by the high density of positional fixes (yellow). Bees disperse slowly over a day and a half. Near the end of day two, many bees go past the range of the radar indicated by the dramatic drop in positional fixes. Open black triangle indicates position where queen bumblebees were placed into the ground to emerge from prior to experiment. Open black circle indicates position of the harmonic radar. Full size image The distribution of flight headings was not significantly different from a uniform distribution (Kolmogorov-Smirnov Test for uniformity: p = 0.5925, k = 0.0672, n = 264; Fig. 4a ) and this did not change throughout the day, i.e. according to the position of the sun (Kolmogorov-Smirnov Test for uniformity - Before solar noon: p = 0.5757, k = 0.0866, n = 120; After solar noon: p = 0.5651, k = 0.0822, n = 144; Fig. 4b ). There was no correlation between flight headings and wind direction (circular correlation coefficient: p = 0.749, r = −0.019, n = 264) and there was no correlation between flight distance and wind speed (circular correlation coefficient: p = 0.786, r = 0.017, n = 264). In addition, the flight distances followed an exponential distribution (R 2 = 0.9978, n = 264; Fig. 4c ), suggesting a random dispersal. Figure 4 Queen bumblebees slowly disperse from their hibernation site in a random manner. ( a ) Radial histogram of flight directions with cardinal directions indicated. Bees showed no overall directional preference, as the distribution of directions was no different statistically from a uniform distribution. (b) Radial histograms show that the position of the sun had no effect on the directional preferences of bees as the distribution of flight directions before and after solar noon was no different from a uniform distribution. (c) The cumulative probability of flight distances, calculated using the line connecting the first and last positional fix of each flight, followed an exponential distribution, indicative of a random dispersal (black points). Pink line: cumulative probabilities of an exponential distribution for reference. (d) Our random walk model, using the observed flight distances and directions of the harmonic radar monitored bees’ activity combined with the time between flights spent on the ground by our individually radar monitored queen bees, estimates over time an average net dispersal distance per flight from the origin of about five and a half meters. (e) Our model estimates an average total net displacement from the origin in three weeks to be approximately three kilometers (pink line; light grey line = maximum estimated distance, dark grey line = minimum estimated distance over 1000 iterations of the model). ( f ) Our model’s estimation of overall displacement of bees from the origin is biased towards northerly directions, but varies a lot around an average displacement (pink dot = origin, black dots = 30 example displacement positions from our model). Full size image Modelling post-hibernation dispersal To assess the general dispersal of queen bumblebees, we created a random walk model which utilized the observed flight distances and directions within the group-monitored data combined with the data on time between flights spent on the ground from our individually-monitored queens (Methods). Because individually monitored, radar-tracked queen bumblebees took, on average, 27.8 ± 24.5 min sojourns between flights of, on average, 13.3 ± 4.1 s, and daylight during the early Spring (March) in England is 12 hours and increases towards the summer, a conservative estimate of flight frequency would be 25 flights per day. Note that here we used only those sequential individual flights that ended and began within 2.5 m (the accuracy of the radar) of each other. For each flight, our random walk model chose a flight heading based on the distribution of the data from our harmonic radar monitored bees (Methods). Flight distance was chosen randomly from within the minimum and maximum flight distances observed near the chosen heading (Methods). Although our bees’ flight directions were no different from a uniform distribution, the very slight bias of flights towards northerly directions (144 of 264 total flights; Fig. 4a ) resulted, over many flights, in an overall average net displacement away from the origin. Our model estimates a net displacement from the release point over the first day of flights to be 336.2 ± 5.5 meters (25 flights; averaged over 10,000 iterations; Fig. 4e ). The majority of the distance gained on the first day results from the first several flights, as the sign of displacement about the origin becomes more random once the net displacement from the origin is much greater than the maximum flight distance observed. As a result, our model estimates that bees start out displacing themselves on average of about 60 meters from the origin on the first flight, but their net displacement per flight quickly asymptotes towards 5.42 ± 0.08 m per flight (within two days; 50 flights; averaged over 10,000 iterations; Fig. 4d ). The net displacement from the origin in three weeks (525 flights) is estimated to be 3043.8 ± 31.7 m (averaged over 10,000 iterations; Fig. 4e ). Although the slight bias in the observed flights’ directions manifested itself in our model through a net displacement in northerly directions, there was a large amount of variation around the average displacement from the origin (Fig. 4f ). Discussion Characterization of animal movement is vital for efforts dealing with conservation and biological invasions. Here, we used harmonic radar to track the movement of bumblebee queens immediately after emergence from (artificial) hibernation. Strikingly, and contrary to the few observations in prior literature 10 , 11 , 12 , 13 , bees spent the majority of their time resting on the ground with infrequent, short (time and distance) flights between. Observations of wild queens very early in the spring corroborate these results and suggest that the behavior of our radar-monitored bees was not due to the transponders or an artifact of being commercially reared. All the wild bees we observed spent most of their time motionless on the ground, sometimes under foliage and only occasionally taking flights for short durations and distances. The differences between these and those from radar-monitored bees is undoubtedly due in part to the fact that we were unable to visually track wild queens for more than about 15 meters (Methods). The purpose of the wild queen observations was to verify that the transponders were not the cause of the long resting/short flight behavior we observed in the radar tracks. Indeed, the wild queen bees spent long periods of time on the ground between short flights, indicating that the similar behavior observed in our radar tracked bees is a common phenomenon amongst bumblebee queens after hibernation. Wild queen bumblebees would have had previous experience where they emerge, inviting the question of whether the lack of experience with the area played a role in the behavior of the commercially reared radar tracked bees. However, wild queen bees showed similar long resting durations and short flight durations and distances, suggesting the behavior observed in radar tracked bees was normal. Bumblebees, like many other insects, have an internal compass 14 that allow them to fly long distances in relatively straight lines. Our bees’ flights were certainly not straight and were considerably shorter than the kilometers found to be travelled by bumblebee workers (Fig. 1a ) 15 , 16 . Worker bees make characteristic orientation flights when they first leave the hive, in which they loop near their nest gathering information about their immediate surrounding, enabling them to encode vital information so they can find their way back to the nest later on 17 . However, none of our queen bees’ flights showed the characteristic loop features of an orientation flight (Figs 1a and 5a ). Bumblebees are also able to learn and memorize features of their environment (landmarks) in such a manner that they can find their way home after long distance flights or even when being displaced artificially 18 . Our bees did not show any attempt at returning to their place of origin nor did they show any indication that they were consistently avoiding or heading towards local landmarks or prominent features of the horizon profile (Figs 1a and 5b ). Together, queen bees’ flight behaviors in this early phase after hibernation resemble a random walk model, similar to Brownian motion 19 . Figure 5 Comparison of bumblebee flight patterns suggest post-hibernation queens disperse in a random manner. ( a ) Flights of four individual radar tracked queen bumblebees in our study. ( b ) Flights of two individual bumblebees on their first flights from the hive. ( c ) Flight of one experienced individual bumblebee on a foraging flight. Closed diamonds indicated positional fixes. Origin is the location of the colony. Axes show distance from the colony in meters. Note that the two orientation flights display loop patterns, while all flights in b and c show returning to the colony, in contrast to flights of post-hibernation queen bees in a and shown in Fig. 1a . Taken from Osborne et al . 20 within the terms of the Creative Commons Attribution License. Full size image Lévy flights, a type of random walk suggested to produce optimal search strategies in absence of memory, have been found to characterize many animal movements, including the flight patterns of bumblebees 20 . However, the flight distance distribution of our bees did not show the heavy tail distribution described by a power law, characteristic of Lévy flights, but rather followed an exponential distribution indicative of a random walk 19 (Fig. 4b ). Vogt and colleagues observed flying, foraging and nest-searching queens and queens with established colonies, in the spring of North Eastern America and Alaska 9 . Their findings suggest that queen bumblebees in temperate regions take on average about three weeks post-emergence (first sighting of queen bees in the early spring) before starting to look for nest sites (zig-zag flight patterns close to the ground and/or investigating holes in the ground and tussocks of vegetation). Once actively searching for a nest, young queens are thought to spend several days or more locating a suitable nest site 12 . Our random walk model, using the distribution of flight directions and distances observed with harmonic radar, estimates that three weeks of the described behavior could displace queen bumblebees, on average, by a distance of approximately three kilometers from their point of emergence. This result is very similar to distances estimated by Lepais and colleagues, measuring genetic relatedness of colonies over multiple seasons, for Bombus pascuorum and Bombus lapidarius in their natural environment 7 . Our model is limited by the fact that we are only using five days of harmonic radar data. However, our wild bee observations spanned nine days, and nest searching was seen only on the last day from only three bees. Vogt et al . 9 have shown that bumblebee queens take about three weeks to begin nest searching. This suggests that the observed long resting periods and short flight behavior is sustained over more days and importantly corroborates our harmonic radar observations. Further, our findings corroborate those of Dreier et al . 21 who found that bumblebees exhibit low levels of fine scale spatial genetic structure, i.e. a very low relatedness amongst bee colonies near each other. Their findings would suggest a necessary random and widespread dispersal as seen in the results of Lepais et al . 7 as well as our observations and model. A slight bias of the observed flights towards northerly directions (144 of 264 total flights) manifested itself within our model as a net displacement from the origin over time, indicating a biased random dispersal. Further experiments will be required to separate potential causes of the slight directional bias, such as the position of the sun within the southern sky, landscape features (e.g. large patches of trees), and wind direction. Harmonic radar needs a line-of-sight to the transponder in order to detect a signal and cannot reliably detect bees on the ground due to vegetation or undulations of the landscape occluding the transponder. Thus, we could not be sure that between flights, queen bumblebees were not walking, rather than resting. However, wild queen bumblebees spent prolonged periods on the ground and often displayed behavior suggesting that they were resting. During their time on the ground, queens were often observed pushing their heads and body between or under dead leaves or grass (Fig. 1b ). Many of the bees we observed displayed a body posture indicating sleep as described in honeybees 22 where the head sunk downwards and the antennae dropped low and became immobile. Our observations may be indicative of post-hibernation queens having low energy reserves and needing to conserve their fat reserves 10 . Contrary to textbook accounts based on casual observations 10 , 12 , our results show that newly emerged queen bumblebees of Bombus terrestris spend a large majority of their time resting on the ground between very short flights (tens of seconds) slowly dispersing themselves from their hibernation site. We suggest that the observed behavior may serve the dual purpose of overall dispersal and ovary development. Our results indicate that queen bees are not dispersing immediately and directly away from where they emerged from hibernation, and they are not orienting towards and remaining near their place of origin. The ability to disperse and the resultant gene flow are important biological factors that help species cope with changes in their environment, like habitat fragmentation and climate change 23 . Land cover changes and habitat fragmentation strongly affect pollinator community composition 24 . Our results and observations suggest practices that may be valuable for conservation efforts concerning bumblebees. For example, if dispersal occurs as a series of very short flights, pollinator friendly corridors between conserved landscape patches could be beneficial. It may also be helpful to leave vegetation, such as leaf litter and long grass undisturbed until late in the spring, giving queen bumblebees safe places to rest, protect themselves from predators and for shelter at night and colder days. Methods Study Area Harmonic radar field work took place between April 27, 2015 and May 15, 2015 at Rothamsted Research (Hertfordshire, UK, 51′48″13 N 0′22″8 W, Fig. 1a ), with approval from the Rothamsted field experiment committee. The average daily minimum temperature over the period of radar tracking was 9.9 °C and the average daily maximum was 14.8 °C, and an average low of 6.3 °C with all periods of radar tracking occurring during part clouds/sun. Temperature data were collected from worldweatheronline.com. Wind speed and direction was obtained from COSMOS-UK 25 . Queen bumblebees were released and observed in the center of a hay meadow. This area was covered in short grass and normally grows wild flowers during the summer, but over the period of the study no wild flowers were observed. The closest tree cover was approximately 80 m away from the release site. There was an 8.6 ha patch of woodland 180 m away from the release site and residential gardens were within 150 m. The harmonic radar field site had no available sources of forage on the ground within the hay meadow surrounding the release site during our experiments so bees were not foraging during the times that no radar signals were detected within the hay meadow. Wild queen bumblebee observations were made between March 7, 2017 and March 15, 2017 in a woodland clearing (roughly 0.25 ha) in Epping Forest near South Woodford, UK (51′36″1 N 0′0″29 W), that was similar to the woodland close to the release site of radar monitored bees. Wild bumblebees were tracked in similar weather conditions as those for the harmonic radar observations – partly cloudy/sunny each day with an average minimum daily temperature of 8.7 °C and an average maximum daily temperature of 13.3 °C and an average low of 5.7 °C. Animals Fertilized queens of Bombus terrestris audax were obtained directly from Biobest NV in Westerlo, Belgium, and transported at 4 °C to Rothamsted Research. Queens were kept at 4 °C in the dark until the start of the experiment. At the start of the radar monitoring experiments, queen bees were placed into small depressions in mounds of earth (5 cm high and 15 cm diameter) on the ground. Tracking individual bees allowed us to see their patterns of flight and potential for dispersal over a reasonably lengthy period. However, individual tracking took two days per bee, both to track and to make sure that the bee had actually dispersed from the area. The simultaneous release of bees allowed us to gather much more data on the timing and direction of individual flights than could be obtained from individual releases, but at the cost of being unable to identify which individual produced each flight. Our intent was to gather data from queen bees early in the spring, during the time wild bees would be emerging from hibernation. The more bees we released individually, the greater the danger would be that one would remain in the tracking area and prevent us releasing further bees. Thus, it would have been difficult or impossible to collect individual tracking data from so many bees within the period that wild bees were emerging from hibernation. Judicious use of individual and mass tracking allowed us to maximize the data obtained during this period. Three bees were released individually, one on April 27, 2018 at 1:52 pm, a second on April 30 at 8:25 am, and a third on May 1, 2015 at 8:18 am. A group of 17 queen bumblebees were simultaneously released on May 11, 2015 at 10:14 am. These bees were placed in similar mounds arranged in a 4 × 4 grid whose sides were aligned with the principal compass directions with a single extra mound to the East. Mounds were separated from their nearest neighbors by 115 cm. Bumblebees hibernate below the soil-leaf litter interface, usually close to trees 26 . We could not exactly recreate these conditions since the radar transponders risked getting caught in the soil or leaf litter, so we placed the bees on mounds of loose earth. These also raised the bees above the level of the grass to ensure they would not become entangled while emerging from their hibernation state. The bees were placed in small depressions to ensure they would not accidentally fall off their mounds. The mound also gave the bee a recognizable site that she could learn and return to if she wanted. Radar tracking The harmonic radar and tracking procedures are described in detail elsewhere 17 , 27 . Briefly, movements of queen bees were tracked using 32 mm harmonic radar. The radar was located at the Southeast edge of the experimental field where it rotated once every 3 s. Bees we wished to track were fitted with a transponder (consisting of a 16 mm vertical dipole) on their thorax. When activated by the radar beam, these transponders return the radar signal at half the original wavelength and twice the frequency. The radar unit has a second parabolic dish with a receiver tuned to this harmonic frequency allowing us to distinguish the transponder signal from reflections from other nearby objects. The radar returned distance and direction coordinates of the queen bees’ position every 3 s as long as the bees remained within a line-of-sight radius of approximately 800 m (accuracy ~±2.5 m). Transponders were attached to the thorax of each queen bee (Fig. 1a inset) using superglue (Loctite Power Flex Gel, Henkel Ltd., Hemel Hempstead, UK). The harmonic radar requires a line-of-sight between the radar and transponder, so can seldom detect bees on the ground and sometimes does not detect very low flight (below ~0.25 m) when there are obstacles between the bee and the radar. The trackable area of the field was defined by the trees and hedgerows at the edges. We monitored the movements of the three individually released bees for 5–11 hours per day and the 17 simultaneously released bees for 10–15 hours per day over 5 days apart from day 4 on which heavy rain forced us to turn off the radar after 4 hours. Analysis of radar data The harmonic radar returned range and azimuth coordinates for every time a transponder was detected. Flight data were visualized by converting the radar coordinates of the bees’ positions to GPS coordinates using custom scripts written by JLW in MATLAB (Mathworks, Natick, MA, USA). The following variables were extracted from the radar tracks. Time in flight was the time during which the tracks showed the bee to be in motion. This excluded time that the bee was known to be stopped but also necessarily excludes time in which the bee was not detected by the radar but may have been in motion. A flight of a bumblebee was considered to be all sequential positional fixes of a bee that had no more than 6 seconds between each sequential pair of positional fixes. Time between flights was the time from the last radar coordinate of an individual bee’s flight and the first radar coordinate of that individual bee’s following flight. Of the 17 bees monitored simultaneously, data were used for time between flights only if the end of a flight and the beginning of another flight were within 5 meters of each other. Because our observations show that bees, likely by chance, did not tend to land very close to each other, 5 m is a distance that we could be sure included only the subsequent flights of the same bee and rejected any possible flight of a different nearby bee. Wind speed and direction were obtained every 30 mins from a weather station approximately 1250 m from the release site via COSMOS-UK 25 . Each bumblebee flight was paired with the weather readings taken closest in time to the start of that flight. All data processing, analyses and data figures were created using MATLAB (Mathworks, Natick, MA, USA). Observations in the wild On March 7, 9 and 15 of 2017, 140 observations of queen bumblebees ( Bombus terretris ) were made within a grassy clearing in woodland in Epping Forest near South Woodford, UK. Although we could not be certain how many of these observations were of the same individual, given the large number of bees we saw (both focal bees and untracked individuals) and those disturbed simply by walking through the meadow, and the fact that many of the focal bees we observed flew well outside of the area we were monitoring, we are confident that a large majority of the bees were only observed once. We chose focal bees to observe at random from those we saw flying through the meadow. We observed each focal bumblebee until we lost track of her visually, which generally occurred when she flew out of the area through the woods or high up into the air too far and/or too fast for us to maintain visual tracking. We monitored the time each focal bee spent on the ground and in flight, and how far she flew if she landed again before we lost track of her. We measured the distance flown by placing an item next to the landing positions to be measured once we lost track of the bumblebee. Although we did not witness any queen bumblebee emerge from her hibernation, the timing of our observations was early in the year, less than a week after our first sighting of queens within the greater London area in 2017, indicating that these queens had very recently emerged. The fact that we saw no nest searching on the first days of observation and only three instances of a queen nest searching over a week later indicate that the majority of queens had not entered the nest-searching phase, so had likely recently emerged from hibernation. On 50 occasions, the ‘resting’ behavior of bees while on the ground was disturbed by an observer. Because we were unsure whether our actions caused these bees to stop resting and fly away, these observations were not included in our analysis of time between flights. Dispersal model To investigate the dispersal rate of our queen bumblebees, we modelled dispersal distance over time using a random walk model. An individual’s overall flights are represented by a sequence of distinct, independent segments (line segments joining two sequential positional fixes), which lengths and orientations are drawn at random from distributions that were parameterized using the harmonic radar data. Flight directions were chosen at random from the distributions of observed flight directions (Fig. 4a ). Flight lengths were taken to be the mean length of flights observed to occur in the selected flight direction (Fig. 4c ). The number of flights taken per day was estimated from the average amount of time between flights by the three individually-monitored bees, since time between flights was hard to estimate from the simultaneously monitored bee data. Foraging flights of worker bees with transponders have been shown to be slightly longer without transponders 28 , potentially affecting the accuracy of our flight length model parameters. Our transponders weigh approximately 15 mg which is less than 3% of the mass of a normal Bombus terrestris queen bumblebee (~800 mg 29 ; n.b. workers can carry up to 90% of their body mass in nectar 30 ). However, the durations of queen bee flights observed in the wild were similar to those found in bees with transponders (Fig. 1b ), suggesting that the transponders did not have a large effect on the flight behavior of queens and demonstrating that the behaviors we observed – short periods of flight separated by long periods of inactivity – are not an artifact of the tracking technique.

Scientists at Queen Mary University of London have discovered a never before reported behaviour of queen bumblebees. It was long thought that queen bumblebees, after hibernating in the ground over winter, emerged, began feeding and dispersed quite quickly to found their new colony. But new research shows that directly after hibernation, queen bumblebees spend the majority of their time hiding and resting amongst dead leaves and grass. The study, published in the journal Scientific Reports, suggests that this behaviour of long rests with short intermittent flights explains how queen bumblebees find themselves far away from their natal nest. Dr. James Makinson, who co-led the study at Queen Mary University of London but is now based at Hawkesbury Institute for the Environment at Western Sydney University, said: "We wanted to see what queens actually do right after they emerge. By combining state-of-the-art tracking technology with wild bee observations, we were able to uncover a never before seen behavior of queen bumblebees." The researchers placed small antenna on the backs of queens that had just emerged from artificially induced hibernation. At an outdoor field site, radar was used to track the bees via the antennae as they woke up and left the area. Queen bumblebee with antenna Credit: James Makinson The data showed that the queens were spending most of their time on the ground (between 10-20 minutes on average) and making short flights (10-20 seconds on average) in nearly random directions. Observations of wild queen bumblebees verified this was not due to the antennas but rather natural behaviour of recently emerged queens. Computer modelling also showed that this behaviour can explain how bees end up many kilometres from the hibernation spots. Dr. Joe Woodgate, a co-lead author of the study from Queen Mary University of London, said: "Our study suggests that a few weeks of this type of behaviour would carry queen bees several kilometers away from their hibernation site and might explain how queens disperse from the nest in which they were born to the place they choose to found a new colony." Dr. Makinson added: "Better understanding the behavior of queens during this crucial period of their lives can suggest practices to improve their chances of successfully founding new colonies and help their survival. Queen bumblebee resting among leaves Credit: Clint Perry "Our findings suggest that creating pollinator friendly corridors between conserved landscape patches would be helpful. It would also be beneficial to plant pollinator friendly flowers and trees all year round, giving bumblebee queens ample access to food during their early spring emergence. And leaving vegetation, such as leaf litter and long grass, undisturbed until late in the spring would give queen bumblebees safe places to rest." If you see an exhausted bumblebee queen around this time of the year, researchers suggest you can rescue her by giving her sugar solution (half water, half sugar, thoroughly stirred). Put the solution on a teaspoon and move the spoon gently to near her antennae or mouthparts. Drinking the solution will allow the bee to warm up its flight motor and have sufficient energy to find flowers on its own.

10.1038/s41598-019-40355-6

Physics

Cell stiffness may indicate whether tumors will invade

Cell swelling, softening and invasion in a three-dimensional breast cancer model, Nature Physics (2019). DOI: 10.1038/s41567-019-0680-8 , nature.com/articles/s41567-019-0680-8 Journal information: Nature Physics

http://dx.doi.org/10.1038/s41567-019-0680-8

https://phys.org/news/2019-10-cell-stiffness-tumors-invade.html

Abstract Control of the structure and function of three-dimensional multicellular tissues depends critically on the spatial and temporal coordination of cellular physical properties, yet the organizational principles that govern these events and their disruption in disease remain poorly understood. Using a multicellular mammary cancer organoid model, we map here the spatial and temporal evolution of positions, motions and physical characteristics of individual cells in three dimensions. Compared with cells in the organoid core, cells at the organoid periphery and the invasive front are found to be systematically softer, larger and more dynamic. These mechanical changes are shown to arise from supracellular fluid flow through gap junctions, the suppression of which delays the transition to an invasive phenotype. These findings highlight the role of spatiotemporal coordination of cellular physical properties in tissue organization and disease progression. Main Living cells are dynamic systems undergoing processes ranging from gene expression, intracellular dynamics and forces at the molecular level 1 , 2 , 3 to cell contraction, deformation and migration at the cellular level 4 , 5 , 6 , 7 , 8 , 9 , 10 . Within a multicellular tissue, the precise control of these physical characteristics in space and time is critical for the maintenance of mechanical integrity and biological function. Deviation from mechanical haemostasis is associated with diseases, including aberrant wound repair, developmental abnormalities and cancer 11 , 12 , 13 , 14 , 15 , 16 , 17 . In isolated cells in vitro, for example, increased deformability 18 , 19 , 20 , 21 , intracellular dynamics 22 , contractility 23 and mobility have each been identified as a physical hallmark of different types of cancer 24 and have been proposed as potential targets for cancer treatment 25 . Despite the promise of in vitro approaches, extensive studies have shown that the mechanical properties of cells are regulated by their microenvironment, including but not limited to stiffness of the surrounding matrix, cell densities and dimensions 26 , 27 . Hence, in the context of a multicellular system in the three-dimensional (3D) microenvironment, it remains unknown how the physical characteristics of individual cells regulate and coordinate tumour development and invasion. By integrating confocal microscopy with optical tweezers, we have developed a platform to measure morphological characteristics, physical properties and migratory dynamics of individual cells throughout a growing multicellular 3D breast cancer model 11 , 12 . By tracking the spatiotemporal evolution of individual cells during the growing process, we find heterogeneous patterns of cellular physical characteristics that facilitate tumour cell invasion. Compared with cells in the organoid core, those at the invasive leading edge are faster, softer and larger. The volumes of both the cell body and the nucleus are larger at the invasive leading edge, and the temporal fluctuations characterizing cytoplasmic dynamics become stronger. Blocking gap junctions (GJs) significantly suppresses these changes, suggesting that supracellular fluid flow may drive the evolution of the observed patterns of cellular properties. Furthermore, the elimination of the softer subpopulation in the cancer organoid strongly delays the transition to an invasive phenotype. These findings highlight a causal role of spatiotemporal coordination of cellular physical properties—especially cell swelling and softening—in tumour development and invasive dynamics. Epithelial cancer organoids have been widely used to model glandular epithelial cancers in 3D culture systems 28 ; these models recapitulate distinguishing physiological features of epithelial tissues and capture the pathological features of epithelial tumours. For example, the well-ordered epithelial architecture becomes disrupted, the lumen fills with cells and invasive branches then form 28 . To reveal the mechanical underpinnings of this process, we transfected MCF 10A human breast epithelial cells with a green fluorescent protein (GFP) tagged with a nuclear-localization signal (NLS). We then seeded these cells into a 3D interpenetrating network hydrogel composed of 5 mg ml −1 alginate and 4 mg ml −1 Matrigel 12 , with a shear modulus close to 300 Pa to mimic the mechanical microenvironment of a breast carcinoma in vivo 12 , 29 . Starting from a single cell, a multicellular cluster grows, and invasive branches develop over approximately 10 days (Fig. 1a–c ); in accordance with recent definitions, we call this cluster a cancer organoid 30 . During the early stage, an individual MCF 10A cell proliferates to form a spherical cluster (early stage; Fig. 1a , day 3). This cell cluster grows into a larger spheroid with cells both in the core and on the periphery (middle stage; Fig. 1b , day 5). As this spheroid develops further, invasive branches extend from the main body and invade the surrounding extracellular matrix (ECM) (later stage; Fig. 1c , day 10). The phenotype observed in this 3D breast cancer model shows uncontrolled cellular proliferation, lack of cellular polarization and the initiation of matrix invasion, much like those observed in vivo in invasive ductal carcinomas (Fig. 1d ). This contrasts with a normal acinar structure that develops when cells are seeded in a soft matrix (Supplementary Fig. 1 ). Fig. 1: Evolution of heterogeneity and subpopulations of cell stiffness in the growing cancer organoid. a – c , Cross-section images of epithelial cancer organoid developed from GFP-NLS-labelled MCF 10A cells at different stages: early stage (day 3; a ), middle stage (day 5; b ) and later stage (day 10; c ). d , Haematoxylin and eosin stains from grade-2 ER+ invasive ductal carcinoma human breast cancer tissue samples. Tumour glands are indicated using red arrows. e , Schematic of the cytoplasmic mechanics and dynamics measurements in a growing cancer organoid using optical tweezers. f , E A of individual cells in the core, periphery and branch regions of the cancer organoid, quantified from slopes of the normalized force–displacement curves (inset). F and x are the resistant force and displacement of the trapped particle. S and a are the cross-section area and diameter of the particle. The boxes represent the interquartile range between the first and third quartiles, whereas the whiskers represent the 95% and 5% values, and the squares represent the average. g , Mechanical heterogeneity of individual cells within the cancer organoid at different stages ( y axis shared with f ). Measurements are taken in more than three independent cancer organoids for f and g . * P < 0.05; ** P < 0.01, *** P < 0.001. Scale bars in a – d , 50 μm. Source data Source data Full size image The molecular pathways that regulate the invasion process in this model system have been well studied 11 , 12 , but the associated changes in the physical properties of cells remain largely unknown. To characterize cell mechanics in this growing epithelial cancer organoid, we used optical tweezers to perform active microrheology and thus measure the mechanical properties of the cytoplasm within individual cells. To do this, we mixed small particles (0.5 μm diameter) into the ECM that became endocytosed by individual constituent cells. These particles were dragged unidirectionally by optical tweezers at 0.5 μm s −1 to obtain a force–displacement relationship in the cytoplasm (illustrated in Fig. 1e ), which thus characterized the cytoplasmic stiffness within each cell 31 . This force–displacement relationship was found to vary spatially within the growing cancer organoid (Fig. 1f , inset). To compare different regions, we spatially segmented the cancer organoid into three different populations: core, periphery and branch. At early stages, only one population was evident. At middle stages, however, cells were classified as being in the core if they were within the inner 40% of the cancer organoid radius. After branches formed, a central spheroidal mass was defined and separated into core and periphery cells, with all other cells assigned to the branches (Supplementary Fig. 2a ). We found that the force required to deform the cytoplasm of cells in the core of the cancer organoid was appreciably greater than that required for cells in the periphery or in a branch (Fig. 1f , inset). To further quantify this difference, we measured an apparent modulus E A of the cytoplasm by taking the average slope of the linear regime of the normalized force–displacement curve. We found that the cells in the core of the cancer organoid were the stiffest, while the cells in the branches were the softest (Fig. 1f ). Moreover, the range of stiffness and the population heterogeneity increased as the cancer organoid developed (Fig. 1g ), consistent with findings from in vivo samples, where increasing mechanical heterogeneity during disease progression has been observed 32 . This evolution from a homogeneous to heterogeneous population with distinct mechanical properties is reminiscent of the epigenetic heterogeneity that has been observed in breast tumour progression 33 . Given the spatial variation observed in cell stiffness, we wondered whether cytoplasmic dynamics also vary in different regions of the cancer organoid. To quantify cytoplasmic dynamics, we used confocal microscopy to image spontaneous fluctuations of endocytosed fluorescent particles (0.5 μm diameter), which are larger than the typical cytoskeletal mesh size ( ∼ 50 nm); particle motions reflect the active non-equilibrium force fluctuations that spontaneously arise within the cytoplasm and thus also influence the dynamics of proteins and organelles in the cytoplasm 22 . By tracking trajectories of these particles, we calculated the time- and ensemble-averaged mean-squared displacement (MSD) <Δ r 2 ( τ )>, where Δ r ( τ ) = r ( t + τ ) − r ( t ), r and τ represent the particle displacement and the lag time. When comparing the MSD for different regions of the cancer organoid, we found that the dynamics in the core were smallest, followed by those in the periphery cells, with branch cells showing the greatest cytoplasmic dynamics (Fig. 2a ). Beyond cytoplasmic dynamics, we found similar differences in intranuclear dynamics; the diffusion of GFP in the cell nucleus, probed by fluorescent recovery after photobleaching, was also substantially slower in the core region compared with that in the periphery of the cancer organoid (Supplementary Fig. 2b,c ). Fig. 2: Different cell subpopulations in a cancer organoid show distinct dynamic behaviours. a , 2D time- and ensemble-averaged MSD of 0.5-μm-diameter particles are plotted against lag time on log–log axes, in the cytoplasm of cells in the core, periphery and branch regions of the cancer organoid. The data are averaged from 15 independent measurements, and the error bars represent ±1 s.d. b , Cytoplasmic force spectrum < f 2 > calculated from spontaneous fluctuations of tracer particles and the active microrheology measurements through < f 2 ( ω )> = | K ( ω )| 2 < r 2 ( ω )> inside cells at different locations of the cancer organoid. Data are shown as mean ± standard deviation ( n > 10). c , Cell migratory trajectories over 4 h reveal a highly dynamic scenario of cell migration within the central 20 μm cross-section of the cancer organoid. The colour scale indicates the average migratory speed of each cell. d , The migratory speed of cells in each subpopulation is plotted. The boxes represent the interquartile range between the first and third quartiles, whereas the whiskers represent the 95% and 5% values, and the squares represent the average. * P <0.05; ** P < 0.01; *** P < 0.001. Source data Full size image While the change in intracellular dynamics was consistent with the changes in cytoplasmic stiffness, the dynamics are influenced not only by the passive mechanical properties of the cytoplasm but also by active ATP-dependent force fluctuations within the cell 22 . To measure the latter, we used force spectrum microscopy (FSM), which combines measurements of the spontaneous fluctuations of probe particles with the micromechanical measurement in the cytoplasm to quantify the spectrum of the aggregate force fluctuations < f 2 ( ω )> in the cell, where f is the aggregate force in the cytoplasm and ω is the frequency; this force fluctuation spectrum reflects the total enzymatic activity 22 . Using FSM, we found that cells in the branch have the highest cytoplasmic force fluctuations, with magnitudes approximately five times greater at 1 Hz than those of cells in the spheroid core (Fig. 2b ). We also assessed cell motility by imaging the cancer organoid at a high spatial and temporal resolution. Full 3D images of the organoid were recorded by confocal microscopy every 10 min over a period of 24 h subsequent to a branch forming (Supplementary Videos 1 and 2 ). By tracking the position of every cell nucleus, cell tracks were constructed (Fig. 2c ). From the cell tracks, the migration speed of every cell in this 3D organoid was calculated. All cells were constantly migrating, but cells in the branches or in the periphery migrated faster than cells in the core (Fig. 2d ). This spatial dependence in migration speed mirrored the spatial dependence of cytoplasmic dynamics. Cell volume is another important physiological property that is known to correlate with cell stiffness and dynamics 26 , 34 . To compare cell and nuclear volumes, we fluorescently labelled the whole cell using cytoplasmic staining (in addition to the already GFP-NLS-labelled cell nuclei) and measured volumes using 3D confocal microscopy. We found that over a wide range of organoid sizes, the ratio of nuclear to cell volume in individual cells remained constant (13 ± 1%, Supplementary Fig. 3 ), in agreement with previous findings 26 , 35 , allowing us to measure nuclear volume in lieu of cell volume. At early stages of cancer organoid development, when no clear subregions existed, cells had similar nuclear volumes (Fig. 3a , day 3). As the organoid developed, nuclear volumes increased in spatial variability (Fig. 3a , days 5 and 10); the nuclear volume of the cells correlated strongly with the relative cell positions within the cancer organoid (Fig. 3b , days 5 and 10), with smaller cells in the core. The transition from the early to middle period of development led to cells in the core becoming systematically smaller (Fig. 3c , day 5). As growth continued, the cancer organoid eventually underwent an invasive transition marked by the formation and growth of branches into the ECM. While the core and periphery still displayed clearly distinct nuclear volumes, the formation of invasive branches resulted in cells with even larger nuclear volumes (Fig. 3c , day 10). The nuclear volume of cells in the same regime also evolved as the cancer organoid grew; cell nuclei in the core became smaller as the size of the cancer organoid increased from days 5 to 10 (Fig. 3c , blue boxes). This spatial pattern of nuclear volume was also observed in other 3D mammary cancer models using different cell lines and ECMs, such as MCF 10A and MCF 10AT in the collagen–Matrigel system 11 (Supplementary Fig. 4 ). Fig. 3: Temporal and spatial evolution of cell volume during the growth of cancer organoids. a , Nuclear volume heat map showing the evolution of cell nuclear volume distribution in the developing cancer organoid. Scale bars, 50 μm. b , The nuclear volume of every cell in the cancer organoid is plotted against the relative distance to the organoid centre at different stages, showing a strong correlation between nuclear volume and spatial position, especially at middle and later stages. c , Nuclear volume of cells in different geometrical regimes of the cancer organoids ( n > 3). d , Nuclear volume of individual cells in GJ-inhibited cancer organoids. e , Stress release changes the distribution of individual nuclear volumes in the core and periphery if GJs are intact. The boxes represent the interquartile range between the first and third quartiles, whereas the whiskers represent the 95% and 5% values, and the squares represent the average. * P < 0.05; ** P < 0.01; *** P < 0.001; ns, not significant. Source data Source data Full size image GJs are specialized intercellular connections between epithelial cells and have been widely observed in normal and malignant breast tissues 36 . They connect the cytoplasm of two neighbouring cells, and thus allow molecules and ions to directly pass through 37 , 38 . Accumulating evidence has shown that there is compressive stress within tumour tissue and 3D cancer models 39 , 40 , 41 , 42 . Therefore, we hypothesize that the observed volume gradient in our system is a result of supracellular fluid flow through cell–cell GJs, driven by the higher intratumour compressive stress. To test whether the volume gradient is GJ-dependent, we added a GJ inhibitor (carbenoxolone 43 , 500 µM) to our system on day 3, when the volume gradient was not yet present (Fig. 3a,b , day 3). As the cancer organoid continued to grow, we did not observe a significant difference in the nuclear volume of cells between those in the core and the periphery (Fig. 3d ). Similar results were observed when we used a different type of GJ inhibitor, connexin mimetic peptides (CMPs) 44 (Supplementary Fig. 5 ). These results suggest that the GJs within the organoid play a role in the development of the observed cellular volume pattern. It has also been consistently shown that externally applied mechanical stress can drive fluid flow between cells through GJs in a two-dimensional (2D) monolayer 45 and in 3D MCF 10A clusters 46 to induce cell volume variations. To investigate this hypothesis further, we tested whether the intratumour mechanical stress 39 , 40 , 41 , 42 is a factor in the observed cell volume gradient. Using organoids cultured in a collagen–Matrigel network, we reduced the mechanical stresses on day 5 by releasing the organoids from the matrix confinement using collagenase. Six hours after stress release, we observed a significant increase in the cellular volume in the core and a decrease in cellular volume in the periphery (Fig. 3e , stress release). The initial cell volume gradient then became much weaker, suggesting an inverse supracellular fluid flow from the periphery to the core following stress release. To further test our hypothesis that GJs play a role in supracellular fluid flow, we blocked the GJs 6 h before releasing the stress, and we did not observe any significant changes in cellular volume or cell volume gradient on stress release (Fig. 3e , ‘GJ inhibitor + stress release’). These results indicate that the imbalance of intratumour stress drives a fluid flow through GJs in the 3D cancer model system, which results in the swelling and shrinking of cells in the periphery and the core, respectively. For both isolated cells and confluent monolayers in 2D, cell volume has been found to vary with cell stiffness and internal dynamics, with a decrease in cell volume corresponding to an increase in cell stiffness and a decrease in internal dynamics 26 , 34 . Similar trends prevailed in the developing 3D cancer organoid. Cells in the core became smaller, stiffer and less dynamic. By contrast, cells in the branches became bigger, softer and more dynamic. When we tracked individual cell positions over time, we found that cells transit from the core to the periphery or vice versa (Supplementary Fig. 6a,b ). Moreover, if we track a cell in a branch it tends to be swollen, but as it moves towards the core, it shrinks. The converse is true for tracking a cell in the core (Supplementary Fig. 6c,d ); as a cell transits the cancer organoid, it adjusts its physical properties to its local microenvironment. The responsiveness of cell volume to the local microenvironment raises the question of whether similar variability in cell volume exists in real tumour samples. To answer this question, we obtained grade-2 ER+ invasive ductal carcinoma human breast cancer tissue samples from a patient, then fixed and sectioned each sample before staining and imaging them with confocal microscopy (Fig. 4a,b ). Within the tumour mass, spheroidal, acinar clusters of cells surrounded by basement membrane were evident (Fig. 4c and Supplementary Video 3 ), which share similar characteristics with our 3D cancer model. In such spheroidal acinar clusters, we found that the nuclear volume increased as the distance from the centre increased (Fig. 4d ), consistent with our model system. Cells in the core had smaller volumes, while invasive cells that appeared to have escaped from the main cluster had larger volumes (Fig. 4e ). Fig. 4: Characterization of cell volume heterogeneity in patient samples. a , Schematic of a tumour biopsy from a breast cancer patient. b , Large-scale fluorescent image showing cell nuclei from the core to the edge area of the biopsy. Scale bar, 50 μm. c , Invasive acinar structures within the biopsy. Scale bar, 20 μm. d , Individual cell nuclear volume is plotted against the cell’s relative distance to the centre of the invasive acinar structure. e , Nuclear volume of cells in different geometrical regimes of the invasive acinar structures ( n = 3). * P < 0.05; ** P < 0.01; *** P < 0.001. Source data Full size image These findings raise the interesting question of whether the swelling and softening of peripheral cells are important factors in invasive dynamics. To investigate this further, we artificially manipulated the stiffness and volume of peripheral cells. For example, we changed the osmotic pressure to either compress (with an additional 2% PEG300) or swell (with an additional 10% water) the cells on day 3, before the formation of various subpopulations (Fig. 5a–c ). We also tried increasing cell stiffness using the chemotherapy medication daunorubicin (0.5 μM), jasplakinolide (5 μM) 47 or overexpression of actin crosslinking protein α-actinin. We found that the volume of cells in the periphery changed, while little change was observed in the core (Fig. 5f ). The decrease or increase in cell volume was accompanied by a corresponding increase or decrease in cell stiffness (Fig. 5f ). Long-term exposure to any of these interventions did not significantly impair cell proliferation; the projected area of individual growing cancer organoids increased over time in a fashion similar to those in the control medium (Fig. 5d ). However, the invasion changed considerably compared with the control case, where invasiveness was quantified as the percentage of cancer organoids that formed invasive branches after 11 days in each sample, with softer peripheral cells leading to greater invasiveness (Fig. 5e , Supplementary Fig. 7 ). Furthermore, we used GJ inhibitors (carbenoxolone 43 , 500 µM, or CMPs 44 , 600 µM) to perturb the spontaneous cell swelling and softening process in our system on day 3 (Fig. 5f ) and found that the invasion of the tumour cells was also delayed (Fig. 5e ). To exclude the possibility that GJ inhibitors directly impact cell mechanics, we measured the cytoplasmic stiffness of isolated MCF 10A cells with GJ inhibitors on a 2D surface and found that the GJ inhibitors did not affect cytoplasmic stiffness (Supplementary Fig. 8 ). By controlling the volume and stiffness of the peripheral cells, we significantly altered the invasive behaviour of our 3D cancer model. Fig. 5: Stiffening the soft cell subpopulation inhibits the invasion of the tumour cells. a – c , Bright-field images showing the time-dependent morphological changes in the developing cancer organoids under different culture conditions, including complete culture medium ( a ), osmotic compression ( b ) and osmotic swelling ( c ). d , Quantification of the projected areas of cancer organoids shows that the growth rates under different culture conditions are comparable. e , Percentage of the invasive cancer organoids over time under different culture conditions. Asterisks indicates the statistically significant difference between each group and the control on day 11. f , Cell nuclear volume and cell stiffness in the core and periphery of the organoids under the different culture conditions in d . The boxes represent the interquartile range between the first and third quartiles, whereas the whiskers represent the 95% and 5% values, and the squares represent the average. g , Working hypothesis: the intratumour stress gradient drives supracellular fluid flow and thus results in cell volume and stiffness gradients, which together facilitate tumour cell invasion. Scale bars in a – c , 50 μm. Error bars in d and e indicate standard deviation. Measurements were taken in three independent experiments. * P < 0.05; ** P < 0.01; *** P < 0.001. Source data Source data Full size image An understanding of how mechanical changes in cells enable disease advancement is critical for determining how cancer progresses. The invasion of MCF 10A clusters has previously been shown to depend on ECM mechanics through PI3K activation 11 , 12 , 48 ; we indeed confirm that this is critical for disease initiation (Supplementary Figs. 9 and 10 ). However, after this initial stage of disease progression, we found that mechanical heterogeneity increased, giving rise to larger, softer cells and enhanced invasiveness. For isolated cells in the 3D culture, cell–matrix interactions can potentially lead to an increase in cell volume 49 ; this cannot be excluded from the factors that play a role in peripheral cells. Nonetheless, the inhibition of GJs interfered with cell volume change during growth and stress release, indicating that fluid flow across cell boundaries plays a role in mechanical pattern formation. The role of GJs in cancer progression is still under debate, with both promotion and suppression of invasiveness having been observed across various types of cancer and GJ 50 . Our results hint at a purely physical mechanism by which GJs can affect cancer progression. A change in the water content of cells and hence the degree of molecular crowding 26 , 34 will affect a wide range of downstream cell functions and properties. As such, our emerging physical picture of tumour progression now includes 3D spatiotemporal evolution of cellular physical properties. Methods Cancer organoid culture and immunofluorescence staining The mammary epithelial cancer organoid with invasive phenotype was cultured and induced following a previously established protocol 11 , 12 . Briefly, MCF 10A cells (ATCC) were cultured in a DMEM/F12 medium (Invitrogen, 11965-118) supplemented with 5% horse serum (Invitrogen, 16050-122), 20 ng ml −1 epidermal growth factor (Peprotech, AF-100-15), 0.5 μg ml −1 hydrocortisone (Sigma, H-0888), 100 ng ml −1 cholera toxin (Sigma, C-8052), 10 μg ml −1 insulin (Sigma, I-1882) and 1% penicillin and streptomycin (Thermo Fisher, 15140122). The cells were collected using a 0.05% trypsin-EDTA solution (Thermo Fisher, 25300054) after they reached confluence within T-25 flasks in a normal cell culture incubator and were mixed with the gel precursor solution at a low cell concentration (10 4 ml −1 ) to avoid interaction between cancer organoids. For alginate–Matrigel interpenetrating network hydrogels, the gel was composed of alginate (FMC Biopolymer) and Matrigel (Corning, 354234) with final concentrations of 5 mg ml −1 and 4 mg ml −1 , respectively. The gel precursor with cells was put in the cell culture incubator for gelation. After 1 h, an additional complete culture medium was added to keep the gel hydrated. Cancer organoids developed in the 3D gels in the following ∼ 10 days. For the collagen–Matrigel system, the gel was composed of 3.5 mg ml −1 collagen and 0.5 mg ml −1 Matrigel. The MCF 10AT cell line was a gift from J. Nickerson at the University of Massachusetts Medical School and was cultured under the same conditions as the MCF 10A cell line. For immunohistochemical staining, the cancer organoids were cultured from normal MCF 10A cells (without GFP-NLS). The 3D gel with embedded cancer organoids was fixed with 4% paraformaldehyde at room temperature for 30 min. To increase the permeability of the cell membrane, the gel was immersed in PBS supplied with 0.2% Triton-X100 (2 h, room temperature). The sample was then blocked with 0.5% BSA in PBS for 5 h at room temperature and subsequently incubated overnight at 4 °C with primary antibodies for Cytokeratin 8 (1:300 diluted in PBS; Santa Cruz Biotechnology, sc-8020), Cytokeratin 14 (1:300 diluted in PBS; Santa Cruz Biotechnology, sc-53253), Phospho-Akt (1:200 diluted in PBS; Cell Signaling, 9271T), β-catenin (1:300 diluted in PBS; Santa Cruz Biotechnology, sc-7963), Integrin β4 (1:300 diluted in PBS; Santa Cruz Biotechnology, sc-13543) or laminin-5 (1:300 diluted in PBS; Santa Cruz Biotechnology, sc-13587). Corresponding secondary antibodies were added and incubated for another night at 4 °C (1:800; Thermo Fisher, A-11001 or A-11008). Finally, 4,6-diamidino-2-phenylindole (Thermo Fisher, D1306) was added for another 4 h to stain the cell nuclei. After each step, the samples were washed with PBS for at least 6 h. Optical tweezer measurement The laser beam (10 W, 1,064 nm) was tightly focused through a series of Keplerian beam expanders and a high numerical aperture objective (×100, 1.45, oil, Nikon). A high-resolution quadrant detector was used for position detection. To measure the mechanical properties of the cytoplasm, 0.5-μm-diameter latex particles (Sigma, L3280) were embedded in the gel and were endocytosed by the cells as they grew into cancer organoids. The linear regions of the detector and the trap stiffness were calibrated with the same bead using an active power-spectrum method and the equipartition theorem 31 , 51 , 52 . The endocytosed bead was dragged at a constant velocity of 0.5 μm s −1 by the optical trap, and the force–displacement curve of the local cytoplasm was recorded. To calculate E A , the force and displacement were normalized by the cross-section area and the diameter of the particle, respectively. The slope in the linear range of the normalized force–displacement curve was taken as the E A . FSM FSM combines measurements of the spontaneous fluctuation of probe particles with the micromechanical measurement in the cytoplasm to quantify the spectrum of < f 2 ( ω )> in the cell, which in turn reflects total enzymatic activity 22 . To measure the spontaneous fluctuation of the cytoplasm, 0.5-μm-diameter fluorescent particles (Sigma, L5530) were embedded in the gel and were endocytosed by the cells as they grew into cancer organoids. The motions of the particles within the cytoplasm were recorded with a 10 ms interval using confocal microscopy. Particle centres were determined in each frame, and the trajectories of particles were obtained by minimizing the overall displacement between consecutive frames as demonstrated previously 22 , 53 using a customized particle tracking algorithm. Time- and ensemble-averaged values of MSD were then calculated. The force spectrum was calculated using < f 2 ( ω )> = | K ( ω )| 2 < r 2 ( ω )>, where < r 2 ( ω )> was obtained through the Fourier transform of MSD and the cytoplasmic spring constant K ( ω ) was obtained from the optical tweezer measurement with a power law assumption K ( ω ) = K ( ω = 1 Hz) ω β , with β = 0.15 (refs. 22 , 54 ). Cell dynamics within the growing cancer organoid To track the movement of individual cells, cancer organoids of GFP-NLS-labelled cells at different stages (days 3, 5 and 10) were imaged every 10 min for 24 h in a customized incubator (5% CO 2 , 37 °C, 95% humidity) on a confocal microscope (Leica, TCL SP8). The 3D positions of cell nuclei were determined at each frame, and the trajectory of each cell was determined using the same particle-tracking method and algorithm as described above. The velocity of the cells can be calculated from the adjacent positions on the trajectory of each cell. Cell and nuclear volume measurements To visualize the cell nucleus, the MCF 10A cell line was transfected with GFP-NLS using lentivirus (Essen Bioscience, 4475). Stable cell lines were selected in a puromycin-containing culture medium (0.4 mg ml −1 ; Thermo Fisher, A1113802). To study the volume evolution during the growth of the cancer organoid, cancer organoids were formed in the hydrogel and cultured in a customized incubator (5% CO 2 , 37 °C, 95% humidity) on a confocal microscope. The 3D conformation of the cancer organoid was recorded every 6 h for 2 weeks. The 3D structure of each cell nucleus within a cancer organoid was reconstructed, and the nuclear volume was calculated by counting the voxels contained within the 3D structure using a customized algorithm in MATLAB (2017a). Deconvolution (HUYGENS software) was applied to the image before we calculated the volume, which helps improve the z resolution of traditional confocal microscopy. To estimate the error in the volume calculation from anisotropic resolution of the confocal microscope, we repeated our volume measurements using stimulated emission depletion microscopy with a super-resolution mode in the z direction ( ∼ 250 nm), which is similar to the resolution in the x – y plane, to provide an isotropic volume reconstruction. We found a consistent volume pattern, as shown in Supplementary Fig. 11a . We also imaged the same cancer spheroid with both confocal microscopy and the super-resolution mode in stimulated emission depletion microscopy; we compared the cell volume measurement of each cell with these two methods and found consistent results (Supplementary Fig. 11b ). To calculate the distance from a cell to the centre of the cancer organoid, the centre of each cell was determined by locating the local intensity maximum within the nucleus, and the centre of the cancer organoid was calculated as the geometric centre of the shape formed by all of the cells. Cells were classified as being in the core if they were within the inner 40% of the cancer organoid radius at middle stage. After branches formed, a central spheroidal mass was defined and separated into core and periphery cells, with all other cells assigned to the branches. Using this geometrical categorization, the cell nuclear volume distribution is consistent with the results using molecular signatures as metric 55 , as shown in Supplementary Fig. 12 . To obtain the nuclear volume heat map, we projected each cell from a 20-µm-thick section at the middle plane of the cancer organoid into the x – y plane, and the nuclear volume at each location (6.5 μm × 6.5 μm grids on the x – y plane) was obtained by averaging the nuclear volumes of all cells going through this particular location during a 2 h window. To measure the cell volume, the GFP-NLS-expressing MCF 10A cells were incubated with cell-tracker deep-red dye (Thermo Fisher, C34565) and were mixed with non-fluorescent cells (those without cell-tracker red) to make the individual fluorescent cells distinguishable from their neighbours within the formed cancer organoid; these cancer organoids for cell volume measurement were formed using a poly(ethylene glycol)-based microwell. The 3D morphology of both the nucleus and cell body were then imaged with confocal microscopy, and volumes were calculated with the same algorithm in MATLAB. Human tissue Human tissue samples, including the normal breast tissue and tumour tissue, were obtained at the Xuanwu Hospital in Beijing, China, with the pre-approval of the Institutional Review Board. Invasive ductal carcinoma samples were obtained from patients undergoing surgical removal. The pathologically normal breast tissue was obtained 2 cm away from the tumour lesions in the same patient. Informed consent or assent was obtained from all patients and/or their parent or legal guardian. To measure the nuclear volume within surgical tumour samples, these samples were sectioned and fixed with 4% paraformaldehyde for 1 h. The central section crossing the tumour mass centre was then stained with Hoechst 33342 (Sigma, 14533) for 1 h to visualize cell nuclei and washed three times with PBS. The 3D structure of the tumour slice was imaged with confocal microscopy and the volume was calculated as described above. To exclude the lymphocytes from the analysis, we labelled the pan-cytokeratin (a common epithelial cell maker for humans; Thermo Fisher, MA5-13156) in patient samples using antibodies and found that the tumour tissue indeed contains many pan-cytokeratin-negative cells, which suggests non-cancer cells (Supplementary Fig. 13a ). However, we rarely find lymphocytes near the invasive acinar-like structures, and almost all of the cells within the structure are positive for pan-cytokeratin (Supplementary Fig. 13b ). Mechanical and chemical perturbations For osmotic compression experiments, PEG (300 MW; Sigma, 90878) was added to complete the culture media with a final concentration of 2% (v/v) either on day 3 (long-term) or day 5 (short-term). To record the growth of the cancer organoids in the long-term experiments, we frequently recorded bright-field images of cancer organoids for the 10 days following osmotic compression, and the areas of the cancer organoids were quantified using IMAGEJ (1.52e). To study the effect of osmotic compression on cell nuclear volume and cell stiffness within the cancer organoid, we took high-resolution 3D images before and after a 12 h osmotic compression, and the cell nuclear volume and stiffness were analysed as described above. Alternatively, we added common drugs, including 0.5 μM daunorubicin (Sigma, D8809) or 5 μM jasplakinolide (Santa Cruz Biotechnology, sc-202191), or overexpressed α-actinin (Addgene, 54975) using Lipofectamine 3000 (Thermo Fisher, L3000001) on day 3 to increase the stiffness of the cells within cancer organoids. The growth of cancer organoids was recorded for 10 days. To investigate the effect of osmotic compression on organoid proliferation and apoptosis, we performed immunostaining of Ki67 (Cell Signaling, 9129S) and caspase-3 (Thermo Fisher, C10423) and found that these interventions did not affect cell proliferation or induce apoptosis in our system (Supplementary Fig. 14 ). To investigate the role of the PI3K pathway in the evolution of the cellular physical properties during the growing of the cancer organoid, we added 20 μM LY294002 (Sigma, L9908) to inhibit PI3K on day 3, and the growth of the cancer organoids was recorded for 10 days thereafter. The cell nuclear volume and stiffness were measured on day 10. To block the GJ communication between neighbouring cells within the cancer organoids, either 500 µM carbenoxolone (Sigma, C4790) or a mixture of (37,43)Gap 27 (300 µM; Anaspec Inc., AS-62642) and (40)Gap 27 (300 µM; Anaspec Inc., AS-62642) connexin mimetic peptides was added on day 3. Statistics A two-tailed Student’s t -test was used when comparing the difference between two groups. For comparison between multiple groups, one-way analysis of variance with the Tukey method was used. In all cases, * P < 0.05; ** P < 0.01; *** P < 0.001; ns, not significant. In all box plots, the boxes represent the interquartile range between the first and third quartiles, whereas the whiskers represent the 95% and 5% values, and the squares represent the average. Reporting Summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability Data supporting the findings of this study are available within the article and the Supplementary Information and Source Data and from the corresponding author on reasonable request. Code availability MATLAB scripts used in this work are available from the corresponding author on reasonable request.

Engineers at MIT and elsewhere have tracked the evolution of individual cells within an initially benign tumor, showing how the physical properties of those cells drive the tumor to become invasive, or metastatic. The team carried out experiments with a human breast cancer tumor that developed in the lab. As the tumor grew and amassed more cells over a period of about two weeks, the researchers observed that cells in the interior of the tumor were small and stiff, while the cells on the periphery were soft and more swollen. These softer, peripheral cells were more apt to stretch beyond the tumor body, forming "invasive tips" that eventually broke away to spread elsewhere. The researchers found that the cells at the tumor's edges were softer because they contained more water than those in the center. The cells in the center of a tumor are surrounded by other cells that press inward, squeezing water out of the interior cells and into those cells at the periphery, through nanometer-sized channels between them called gap junctions. "You can think of the tumor like a sponge," says Ming Guo, assistant professor of mechanical engineering at MIT. "When they grow, they build up compressive stresses inside the tumor, and that will squeeze the water from the core out to the cells on the outside, which will slowly swell over time and become softer as well—therefore they are more able to invade." When the team treated the tumor to draw water out of peripheral cells, the cells became stiffer and less likely to form invasive tips. Conversely, when they flooded the tumor with a diluted solution, the same peripheral cells swelled and quickly formed long, branchlike tips that invaded the surrounding environment. The results, which the team reports in the journal Nature Physics, point to a new route for cancer therapy, focused on changing the physical properties of cancer cells to delay or even prevent a tumor from spreading. Guo's co-authors include lead author and MIT postdoc Yu Long Han, along with Guoqiang Xu, Zichen Gu, Jiawei Sun, Yukun Hao, Staish Kumar Gupta, Yiwei Li, and Wenhui Tang, from MIT; Adrian Pegoraro and Yuan Yuan of the Harvard John A. Paulson School of Engineering and Applied Sciences; Hui Li of the Chinese Academy of Sciences; Kaifu Li, Hua Kang, and Lianghong Teng of Capital Medical University in Beijing; and Jeffrey Fredberg of the Harvard T. H. Chan School of Public Health in Boston. Cell tweezing Scientists suspect that cancer cells that migrate from a main tumor are able to do so in part because of their softer, more pliable nature, enabling the cells to squeeze through the body's labrynthine vasculature and proliferate far from the initial tumor. Past experiments have shown this soft, migratory nature in individual cancer cells, but Guo's team is the first to explore the role of cell stiffness in a whole, developing tumor. "People have looked at single cells for a long time, but organisms are multicellular, three-dimensional systems," Guo says. "Each cell is a physical building block, and we're interested in how each single cell is regulating its own physical properties, as the cells develop into a tissue like a tumor or an organ." The researchers used recently developed techniques to grow healthy human epithelial cells in 3-D and transform them into a human breast cancer tumor in the lab. Over the next week, the researchers watched as the cells multiplied and coalesced into a benign primary tumor that comprised several hundred individual cells. Several times throughout the week, the researchers infused the growing number of cells with plastic particles. They then probed each individual cell's stiffness with optical tweezers, a technique in which researchers direct a highly focused laser beam at a cell. In this case, the team trained a laser on a plastic particle within each cell, pinning the particle in place, then applying a slight pulse in a attempt to move the particle within the cell, much like using tweezers to pick an egg shell out from the surrounding yolk. Guo says the degree to which researchers can move a particle gives them an idea for the stiffness of the surrounding cell: The more resistant the particle is to being moved, the stiffer a cell must be. In this way, the researchers found that the hundreds of cells within a single benign tumor exhibit a gradient of stiffness as well as size. The interior cells were smaller and stiffer, and the further the cells were from the core, the softer and larger they became. They also became more likely to stretch out from the spherical primary tumor and form branches, or invasive tips. To see whether altering cells' water content affects their invasive behavior, the team added low-molecular-weight polymers to the tumor solution to draw water out from cells, and found that the cells shrank, became more stiff, and were less likely to migrate away from the tumor—a measure that delayed metastasis. When they added water to dilute the tumor solution, the cells, particularly at the edges, swelled, became softer, and formed invasive tips more quickly. As a last test, the researchers obtained a sample of a patient's breast cancer tumor and measured the size of every cell within the tumor sample. They observed a gradient similar to what they found in their lab-derived tumor: Cells in the tumor's core were smaller than those closer to the periphery. "We found this doesn't just happen in a model system—it's real," Guo says. "This means we may be able to develop some treatment based on the physical picture, to target cell stiffness or size to see if that helps. If you make the cells stiffer, they are less likely to migrate, and that could potentially delay invasion." Perhaps one day, he says, clinicians may be able to look at a tumor and, based on the size and stiffness of cells, from the inside out, be able to say with some confidence whether a tumor will metastasize or not. "If there is an established size or stiffness gradient, you can know this will cause trouble," Guo says. "If there's no gradient, you can maybe safely say it's fine."

10.1038/s41567-019-0680-8

Medicine

Researchers discover system that could reduce neurodegeneration in Huntington's disease

Seda Koyuncu et al, The ubiquitin ligase UBR5 suppresses proteostasis collapse in pluripotent stem cells from Huntington's disease patients, Nature Communications (2018). DOI: 10.1038/s41467-018-05320-3 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-018-05320-3

https://medicalxpress.com/news/2018-07-neurodegeneration-huntington-disease.html

Abstract Induced pluripotent stem cells (iPSCs) undergo unlimited self-renewal while maintaining their potential to differentiate into post-mitotic cells with an intact proteome. As such, iPSCs suppress the aggregation of polyQ-expanded huntingtin (HTT), the mutant protein underlying Huntington’s disease (HD). Here we show that proteasome activity determines HTT levels, preventing polyQ-expanded aggregation in iPSCs from HD patients (HD-iPSCs). iPSCs exhibit high levels of UBR5, a ubiquitin ligase required for proteasomal degradation of both normal and mutant HTT. Conversely, loss of UBR5 increases HTT levels and triggers polyQ-expanded aggregation in HD-iPSCs. Moreover, UBR5 knockdown hastens polyQ-expanded aggregation and neurotoxicity in invertebrate models. Notably, UBR5 overexpression induces polyubiquitination and degradation of mutant HTT, reducing polyQ-expanded aggregates in HD-cell models. Besides HTT levels, intrinsic enhanced UBR5 expression determines global proteostasis of iPSCs preventing the aggregation of misfolded proteins ensued from normal metabolism. Thus, our findings indicate UBR5 as a modulator of super-vigilant proteostasis of iPSCs. Introduction As the origin of multicellular organisms, a series of cellular quality control mechanisms must operate at high fidelity in pluripotent stem cells 1 . In culture, embryonic stem cells (ESCs) derived from blastocysts do not undergo senescence and can replicate indefinitely while maintaining their capacity to differentiate into all cell lineages 2 . Alternatively, somatic cells can be reprogrammed to generate induced pluripotent stem cells (iPSCs), which are similar to ESCs in many respects, such as their gene expression, potential for differentiation and ability to replicate continuously 3 . This unlimited self-renewal capacity requires stringent quality control mechanisms, including increased DNA damage responses and antioxidant defense systems 1 , 4 , 5 , 6 , 7 . Growing evidence indicates that pluripotent stem cells also have intrinsic mechanisms to maintain the integrity of the proteome, a critical process for organismal development and cell function 7 , 8 , 9 . Hence, defining the mechanisms of super-vigilant proteostasis in these cells is of central importance. The proteostasis network is formed by multiple integrated processes that control the concentration, folding, location and interactions of proteins from their synthesis through their degradation 10 . Defects in proteostasis lead to the accumulation of damaged, misfolded and aggregated proteins that may alter the immortality of pluripotent stem cells. During the asymmetric divisions invoked by these cells, the passage of damaged proteins to progenitor cells could compromise organismal development and aging. Thus, pluripotent stem cells have a tightly regulated proteostasis network linked with their intrinsic characteristics and biological function 1 , 7 . While ESC identity requires enhanced global translational rates 11 , these cells also exhibit high levels of distinct chaperones to assure proper protein folding 5 , 9 . For instance, ESCs have increased assembly of the TRiC/CCT (T-complex protein-1 (TCP-1) ring complex)/(chaperonin containing TCP-1) complex 12 , a chaperonin that facilitates the folding of approximately 15% of the proteome and reduces the aggregation of disease-related mutant proteins 13 . To terminate damaged proteins, ESCs possess a powerful proteolytic machinery induced by high levels of PSMD11/RPN6 14 , 15 , 16 , a scaffolding subunit that promotes the assembly of active proteasomes 16 , 17 . Remarkably, pluripotent stem cells are able to maintain enhanced proteostasis while proliferating indefinitely in their undifferentiated state 1 , 7 , 8 . However, the differentiation process triggers a rewiring of the proteostasis network that reduces their ability to sustain proteome integrity 7 , 8 , 9 . In addition, post-mitotic and progenitor cells as well as somatic stem cells undergo a progressive decline in their protein folding and clearance activities with age 8 , 18 , 19 . This demise of proteostasis is linked with the onset of age-related disorders such as Alzheimer’s, Parkinson’s and Huntington’s disease (HD) 10 , 18 . On the other hand, the proteostasis network of somatic cells is rewired during cell reprogramming to generate iPSCs with high assembly of active TRiC/CCT and proteasome complexes, resembling ESCs 9 , 12 , 16 , 20 . HD is a fatal neurodegenerative disorder characterized by cognitive deficits, psychosis and motor dysfunction. The disease is inherited in a dominant manner and caused by mutations in the huntingtin ( HTT ) gene, which translates into an expanded polyglutamine stretch (polyQ) 21 . The wild-type HTT gene encodes a large protein of approximately 350 kDa that contains 6–35 polyQ repeats. In individuals affected by HD, HTT contains greater than 35 polyQ repeats 21 . Although loss of normal HTT function could also be a determinant of HD 22 , the dominant inheritance pattern of the disease and numerous experiments in model organisms indicate that gain of function of mutant HTT is toxic and induces neurodegeneration 21 , 23 , 24 . PolyQ-expanded HTT is prone to aggregation, and the accumulation of mutant HTT fibrils as well as intermediate oligomers formed during the aggregation/disaggregation process contributes to neurodegeneration 21 , 23 , 24 . The longer the polyQ-expanded repeat, the earlier HD symptoms (e.g., neurodegeneration) typically appear 21 . However, the length of the pathological polyQ does not affect survival, self-renewal and pluripotency of iPSCs derived from HD patients (HD-iPSCs), which can proliferate indefinitely as control iPSCs 25 , 26 . Moreover, HD-iPSCs do not accumulate polyQ-expanded inclusions 12 , 25 , 27 . These findings indicate that iPSCs have increased mechanisms to maintain proteostasis of mutant HTT. Once differentiated into neural progenitors and neurons, these cells exhibit HD-associated phenotypes such as altered gene expression, increased vulnerability to excitotoxic stressors and cumulative risk of death over time 25 , 28 . However, HD neurons lack polyQ aggregates and robust neurodegeneration phenotype 12 , 25 , 27 , supporting a proteostasis-rejuvenating process during cell reprogramming that allows for HD-iPSC differentiation into neurons with an intact proteome. Although cumulative evidence indicates a strong link between HD-related changes and proteasomal dysfunction 29 , the mechanisms by which the proteasome recognizes polyQ-expanded HTT are poorly understood. With the high levels of proteasome activity exhibited by iPSCs 16 , we ask whether these cells have an intrinsic E3 ubiquitin ligase network to regulate proteostasis of polyQ-expanded HTT. We find that iPSCs have increased levels of UBR5, a HECT domain E3 enzyme that promotes proteasomal degradation of both mutant and wild-type HTT. Notably, an impairment of mutant HTT levels induced by UBR5 downregulation triggers the accumulation of polyQ-expanded aggregates in HD-iPSCs. Prompted by these findings, we examine whether modulation of UBR5 impinges on polyQ-expanded aggregation in distinct models. Since HD iPSC-derived neurons lack aggregates even upon proteasome inhibition, we assess Caenorhabditis elegans and human cells lines that accumulate polyQ-expanded aggregates. We find that loss of UBR5 worsens polyQ-expanded aggregation and neurotoxicity in C. elegans models. Notably, ectopic expression of UBR5 is sufficient to promote polyubiquitination of HTT, resulting in decreased levels and aggregation of mutant HTT in human cell models. Thus, we identify UBR5 as a potential modulator of HTT proteostasis by studying immortal pluripotent stem cells. Results The proteasome suppresses mutant HTT aggregation in iPSCs iPSCs derived from HD patients do not accumulate polyQ-expanded HTT aggregates 12 , 27 . Since pluripotent stem cells exhibit high proteasome activity 16 , we assessed whether this activity is required to prevent mutant HTT aggregation in iPSCs generated from two individuals with juvenile-onset HD (i.e., Q71 and Q180) 25 (Supplementary Table 1 ). Indeed, downregulation of proteasome activity triggered accumulation of mutant HTT aggregates, which were mostly located in the cytoplasm (Fig. 1a and Supplementary Fig. 1 a–b). Although the size of aggregates was generally smaller in iPSCs that express longer polyQ repeats (Q180), these cultures exhibited a higher percentage of aggregate-containing cells (Fig. 1a ). Moreover, proteasome inhibition also induced a high percentage of mutant HTT aggregation in iPSCs derived from an individual with adult-onset HD (Q57) (Supplementary Fig. 1 c–d). On the contrary, we did not detect accumulation of polyQ aggregates in three distinct control iPSCs upon proteasome inhibition (Fig. 1a and Supplementary Fig. 1e ). Likewise, proteasome inhibition did not induce HTT aggregation in two isogenic counterparts of the Q180-iPSC line (Supplementary Fig. 2 ), in which the 180 CAG expansion was corrected to a nonpathological repeat length 30 . Fig. 1 Proteasome inhibition triggers mutant HTT aggregation in HD-iPSCs. a Immunocytochemistry of control and HD-iPSC lines treated with 5 µM MG-132 for 12 h. We used an antibody against polyQ-expanded protein to detect mutant HTT aggregates. Cell nuclei were stained with Hoechst 33342. Scale bar represents 10 μm. The images are representative of six independent experiments. b Chymotrypsin-like proteasome activity in HD Q180-iPSC line treated with MG-132 for 12 h (relative slope to Q180-iPSCs treated with DMSO). Graph represents the mean ± s.e.m. of three independent experiments. c Filter trap analysis of the indicated control and HD-iPSCs. Proteasome inhibition with MG-132 for 12 h results in increased levels of polyQ aggregates in HD-iPSCs lines (detected by anti-polyQ-expansion diseases marker antibody). However, proteasome inhibition does not induce accumulation of polyQ aggregates in control iPSC lines. The images are representative of five independent experiments. d Graph represents the percentage of polyQ aggregate-positive cells/total nuclei in the indicated iPSC lines (mean ± s.e.m., 3 independent experiments, 300–350 total cells per treatment for each line). e Graph represents the percentage of propidium iodide-positive cells/total nuclei in the indicated iPSC lines (mean ± s.e.m., 3 independent experiments, 500–600 total cells per treatment for each line). For each time point, MG-132-treated lines were statistically compared with their respective DMSO-treated line. All the statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001 Full size image To further examine the link between proteasome activity and polyQ-expanded aggregation in iPSCs, we tested different concentrations of proteasome inhibitor. Lower concentrations reduced approximately 50% of the proteasome activity, which was sufficient to induce mutant HTT aggregation as assessed by both filter trap and immunofluorescence experiments (Fig. 1b–d and Supplementary Fig. 3 ). Since these analyses were performed in iPSCs treated with proteasome inhibitor for 12 h, we examined whether this treatment reduces cell viability, a process that could trigger proteostasis collapse and, in turn, dysregulation of protein aggregation. Whereas different concentrations of proteasome inhibitor resulted in polyQ-expanded aggregation (Fig. 1c, d ), only higher concentrations induced a mild increase ( ∼ 8%) in cell death (Fig. 1e ). In addition, we found accumulation of mutant HTT aggregates at an earlier time point (4 h) of the proteasome inhibition treatment when higher concentrations did not trigger cell death (Fig. 1d, e ). Thus, these results indicate that a decline in proteasome activity promotes mutant HTT aggregation in HD-iPSCs, a process that cannot only be explained by impairment of cell viability. Remarkably, control and HD-iPSCs exhibited similar sensitivity to proteasome inhibition (Fig. 1e ), suggesting that mutant HTT aggregation does not induce cell death in iPSCs. Proteasome dysfunction impairs HTT levels in iPSCs With the link between proteasome inhibition and mutant HTT aggregation, we asked whether HTT levels are regulated by the proteasome in HD-iPSCs. For this purpose, we characterized in our model two antibodies that recognize HTT and polyQ-expanded proteins, respectively 12 , 31 . First, we validated that these antibodies detect endogenous levels of HTT in iPSCs (Supplementary Fig. 4 a–c). The HD-iPSCs used in this study express one mutant allele of HTT but also one normal copy (Supplementary Table 1 ) 12 , 25 , 32 , 33 , 34 . Since the length of the polyQ stretch diminishes the electrophoretic mobility of proteins 31 , we could discriminate normal HTT and mutant HTT in both HD Q71 and Q180-iPSC lines by western blot using anti-HTT antibody (Fig. 2a and Supplementary Fig. 4a ). In HD Q57-iPSCs, the differences in the electrophoretic mobilities of both alleles were marginal and they were not efficiently separated on western blot assays (Fig. 2a ). We also observed that anti-HTT antibody was less immunoreactive to mutant HTT (Fig. 2a and Supplementary Fig. 4d ). These differences were more pronounced in HD-iPSCs that express longer polyQ repeats (Q180), as mutant HTT was only detected after high exposure times in these cells (Fig. 2a ). Thus, we used the antibody to polyQ-expanded proteins to examine the expression of mutant HTT in these cells (Fig. 2a ). We confirmed that this antibody only recognizes mutant HTT on western blots and the intensity of the signal correlates with the length of the polyQ expansion (Fig. 2a and Supplementary Fig. 4d ), as previously reported 31 . Although Q57 expansion was not detected with anti-polyQ-expanded proteins by western blot, this antibody strongly detected mutant HTT in Q71 and Q180 lines (Fig. 2a ). Fig. 2 Loss of proteasome activity increases HTT levels in iPSCs. a Western blot of iPSCs with antibodies to total HTT, polyQ-expanded proteins and β-actin. Arrow indicates mutant HTT detected with total HTT antibody in HD Q180-iPSCs. The images are representative of three independent experiments. b Western blot of control iPSCs #2 treated with MG-132 (12 h). The graph represents the relative percentage values to DMSO-treated iPSCs of normal huntingtin (nHTT) detected with total HTT antibody and corrected for β-actin loading control (mean ± s.e.m., three independent experiments). c Western blot of HD Q71-iPSC line #1 treated with MG-132 (12 h). Graphs represent the relative percentage values to DMSO-treated iPSCs (corrected for β-actin) of nHTT and mutant HTT (mHTT) detected with antibodies to total HTT and polyQ-expanded proteins (mean ± s.e.m., three independent experiments). d Western blot of HD Q180-iPSCs treated with MG-132 (12 h). The graphs represent the relative percentage values to DMSO-treated iPSCs (corrected for β-actin) of nHTT and mHTT detected with antibodies to total HTT and polyQ-expanded proteins, respectively (mean ± s.e.m., four independent experiments). Supplementary Fig. 6 presents a higher exposure time of the same membrane for a better comparison of mHTT levels detected with HTT antibody. e Immunocytochemistry of control and HD-iPSCs treated with 0.5 µM MG-132 or 0.25 µM bafilomycin for 12 h. PolyQ-expanded and Hoechst 33342 staining were used as markers of mutant HTT aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of three independent experiments. f Graph represents the percentage of polyQ aggregate-positive cells/total nuclei in the indicated iPSC lines treated with MG-132 or bafilomycin for 12 h (mean ± s.e.m., four independent experiments, 300–350 total cells per treatment for each line). g Western blot of HD Q71-iPSCs #1 treated with bafilomycin (12 h) using antibodies to total HTT, polyQ-expanded proteins, LC3 and P62. Graphs represent the relative percentage values to DMSO-treated iPSCs (corrected for β-actin) of nHTT and mHTT (mean ± s.e.m., three independent experiments). Statistical comparisons were made by Student’s t- test for unpaired samples. * P < 0.05, ** P < 0.01, *** P < 0.001 Full size image Once we characterized these antibodies in iPSCs, we assessed whether proteasome dysfunction impairs HTT levels. In control iPSCs, proteasome inhibition resulted in upregulated HTT protein levels (Fig. 2b and Supplementary Fig. 5a, b ). In HD-iPSCs, proteasome dysregulation not only increased the amounts of normal HTT but also aggregation-prone HTT (Fig. 2c, d and Supplementary Fig. 5c, d ). Although these results suggest a direct link between proteasome activity with HTT levels and aggregation, another possibility is that HTT dysregulation ensues from a global proteostasis collapse induced by proteasome inhibition. To assess this hypothesis, we inhibited the autophagy–lysosome system, which also modulates proteostasis of HTT through its proteolytic activity 29 . Although autophagy inhibition induced mutant HTT aggregation in HD Q71-iPSCs, these aggregates were less compact and the percentage of aggregate-containing cells were lower when compared to proteasome inhibition treatment (Fig. 2e, f and Supplementary Fig. 7a, b ). In HD Q180-iPSCs, autophagy inhibition only induced aggregation in a low percentage of cells (Fig. 2e, f ). Whereas proteasome inhibition increased HTT levels, autophagy downregulation resulted in decreased amounts of HTT protein (Fig. 2g and Supplementary Fig. 8 a–c). This decrease in HTT levels was not associated with potential changes in cell viability or a compensatory upregulation of the proteasome, as we did not find significant changes in these parameters upon autophagy inhibition (Supplementary Fig. 8 d–e). The differences between autophagy and proteasome inhibition supported a direct role of the proteasome in HTT degradation of iPSCs. In these lines, proteasome dysfunction results in increased levels of both normal HTT and aggregation-prone HTT. Since the impairment of clearance of misfolded proteins is key to their accumulation 19 , the increase in mutant HTT levels upon proteasome inhibition could contribute to diminish the ability of HD-iPSCs to suppress HTT aggregation. UBR5 prevents mutant HTT aggregation in HD-iPSCs To assess whether pluripotent stem cells have an intrinsic proteostasis to regulate HTT levels, we examined their E3 ubiquitin ligase network. For this purpose, we analyzed available quantitative proteomics data 35 and found 26 E3 ligases significantly increased in hESCs compared with their differentiated neuronal counterparts (Supplementary Table 2 ). Notably, UBR5 was one of the most upregulated E3 enzymes (Supplementary Table 2 ). UBR5 shows a striking preference for Lys48 linkages of ubiquitin 36 , which is the primary signal for proteasomal degradation 19 . Under proteotoxic stress, UBR5 cooperates with Lys11-specific ligases to produce K11/K48 heterotypic chains, promoting proteasomal clearance of misfolded nascent polypeptides 36 . Recently, a genome-wide association analysis has identified that genetic variations in the chromosome region containing UBR5 gene hasten the clinical onset of HD 37 . Remarkably, loss of UBR5 reduces the modification of overexpressed HTT protein with K11/K48-linked ubiquitin chains in a human cell line 36 . Thus, increased endogenous expression of UBR5 could provide a link between proteostasis and regulation of HTT levels in pluripotent stem cells. Besides its downregulation during differentiation of hESCs (Supplementary Fig. 9a and Supplementary Table 2 ), we confirmed that UBR5 protein levels are also increased in both control and HD-iPSC compared with their neuronal counterparts (Fig. 3a–c and Supplementary Fig. 10 ). In all the lines tested, we observed that UBR5 downregulation is already significant when iPSCs differentiate into neural progenitor cells (NPCs) (Fig. 3a–c and Supplementary Fig. 10 ). The decrease in UBR5 protein levels correlated with a downregulation of messenger RNA (mRNA) amounts during differentiation (Fig. 3d–f and Supplementary Fig. 9b ). Fig. 3 The levels of UBR5 decrease during differentiation of iPSCs. a – c Western blot of UBR5 levels in control #1 ( a ), HD Q71 #2 ( b ) and HD Q180 ( c ) iPSC lines compared with their neural progenitor cell (NPC) and striatal neuron counterparts. The graphs represent the UBR5 relative percentage values to the respective iPSCs corrected for β-actin loading control (mean ± s.e.m. of three independent experiments for each cell line). d – f Quantitative PCR (qPCR) analysis of UBR5 mRNA levels in control #1 ( d ), HD Q71 #2 ( e ) and HD Q180 ( f ) iPSC lines compared with their differentiated counterparts. Graphs ( UBR5 relative expression to iPSCs) represent the mean ± s.e.m. of three independent experiments for each line. * P < 0.05, ** P < 0.01 Full size image Since UBR5 promotes microRNA-mediated transcript destabilization in mouse ESCs 38 and may be involved in transcriptional repression during development 39 , we examined whether UBR5 downregulation during differentiation correlates with increased HTT mRNA levels. However, the amounts of HTT mRNA were either downregulated or not significantly changed with iPSC differentiation (Supplementary Fig. 11 ). As a more formal test, we assessed whether knockdown of UBR5 alters the transcript levels of HTT in distinct iPSC/hESC lines and found no differences (Fig. 4a and Supplementary Fig. 12 ). On the contrary, knockdown of UBR5 induced an increase in the protein levels of HTT in human pluripotent stem cells (Fig. 4b–e and Supplementary Fig. 13 ). We assessed three independent control iPSCs as well as two hESC lines and obtained similar results (Fig. 4b, c and Supplementary Fig. 13 a–c). In HD Q71 and Q180-iPSC lines, the loss of UBR5 not only impaired the levels of normal HTT but also mutant HTT at a similar extent (Fig. 4d, e and Supplementary Fig. 13d ). Although we could not discriminate normal and polyQ-expanded HTT in HD Q57-iPSCs, we confirmed an upregulation of total HTT protein levels in these cells upon UBR5 knockdown (Supplementary Fig. 13e ). Since loss of UBR5 did not impair HTT mRNA levels (Fig. 4a and Supplementary Fig. 12 ), our data supported a role of this E3 enzyme in post-translational regulation of HTT. To examine whether UBR5 modulates HTT levels in a proteasome-dependent manner, we blocked proteasomal degradation in iPSCs. Notably, UBR5 knockdown and MG-132-treated cells exhibited similar levels of HTT (Fig. 4b–e ). Most importantly, UBR5 downregulation did not further increase the levels of HTT in both control and HD-iPSCs with reduced proteasome activity (Fig. 4b–e ), indicating a role of UBR5 in proteasomal degradation of HTT. Given that pluripotent stem cells exhibit high levels of proteasome activity compared with their differentiated counterparts 16 , we assessed whether UBR5 is required for this activity. However, loss of UBR5 did not affect global proteasome activities in iPSCs (Fig. 4f and Supplementary Fig. 14 ), suggesting a specific link between UBR5 and HTT modulation. Besides UBR5, other E3 ligases are also increased in pluripotent stem cells (Supplementary Table 2 and Supplementary Fig. 15a ). We knocked down four of these upregulated enzymes (i.e., UBE3A, RNF181, UBR7, TRIM71) and found no differences in HTT levels of HD-iPSCs (Supplementary Fig. 15 b–g). Moreover, UBR5 interacted with both normal and polyQ-expanded HTT in HD-iPSCs, whereas we were not able to detect this interaction with a distinct upregulated E3 enzyme (Fig. 4g and Supplementary Fig. 16 ). Fig. 4 Loss of UBR5 impairs HTT protein levels in iPSCs. a qPCR analysis of UBR5 and HTT mRNA levels in control iPSCs #1 ( n = 4 independent experiments), HD Q71-iPSCs #2 ( n = 3) and HD Q180-iPSCs ( n = 3). Graphs (relative expression to non-targeting (NT) shRNA) represent the mean ± s.e.m. b , c Western blot analysis of the indicated control iPSC lines with antibodies to HTT and polyQ-expanded proteins. Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. The graphs represent the HTT relative percentage values to DMSO-treated NT shRNA iPSCs corrected for β-actin loading control (mean ± s.e.m. of three independent experiments). d Western blot analysis of HD Q71-iPSC line #2 upon UBR5 knockdown. The graphs represent the relative percentage values to DMSO-treated NT shRNA iPSCs (corrected for β-actin) of nHTT and mHTT detected with antibodies to total HTT and polyQ-expanded proteins (mean ± s.e.m. of four independent experiments). Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. e Western blot analysis of HD Q180-iPSCs upon UBR5 knockdown. Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. The graphs represent the relative percentage values to DMSO-treated NT shRNA iPSCs (corrected for β-actin) of nHTT and mHTT detected with antibodies to total HTT and polyQ-expanded proteins, respectively (mean ± s.e.m. of three independent experiments). f Proteasome activities in HD Q71-iPSCs #1 upon UBR5 knockdown (relative slope to NT shRNA iPSCs). Graphs represent the mean ± s.e.m. of three independent experiments. g Co-immunoprecipitation with UBR5, UBE3A and FLAG antibodies in HD Q71-iPSC line #1 followed by western blot with HTT, polyQ-expanded HTT, UBR5 and UBE3A antibodies. The images are representative of three independent experiments. All the statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001 Full size image Taken together, these results suggest that intrinsic high expression of UBR5 determines HTT levels in iPSCs, a process that could contribute to the remarkable ability of these cells to maintain proteostasis of mutant HTT. In support of this hypothesis, loss of UBR5 triggered the accumulation of mutant HTT aggregates in all the HD-iPSCs tested as we confirmed by immunocytochemistry and filter trap experiments (Fig. 5a–e and Supplementary Fig. 17a ). In contrast, we did not observe accumulation of polyQ aggregates in control iPSCs as well as corrected isogenic counterparts of HD Q180-iPSCs despite the upregulation of normal HTT levels (Fig. 5a, b, f, g and Supplementary Figs. 17b , 18 ). Fig. 5 UBR5 suppresses mutant HTT aggregation in HD-iPSCs. a Immunocytochemistry of control iPSC line #1 (Q21), control iPSC line #2 (Q33) and HD Q71-iPSC line #1 upon UBR5 knockdown. PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of four independent experiments. b Immunocytochemistry of Q180-iPSCs and two isogenic counterparts (i.e., HD-C#1 and HD-C#2), in which the 180 CAG expansion was corrected to a nonpathological repeat length. Scale bar represents 10 μm. The images are representative of three independent experiments. c – g Filter trap analysis of the indicated control and HD-iPSC lines upon UBR5 knockdown with anti-polyQ-expansion diseases marker antibody. The images are representative of at least three independent experiments for each cell line. h Immunocytochemistry of HD Q71-iPSC line #2 upon knockdown of UBR5. PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. The images are representative of four independent experiments. i Graph represents the percentage of polyQ aggregate-positive cells/total nuclei in HD Q71-iPSC line #2 (mean ± s.e.m., 3 independent experiments, 500–600 total cells per condition). Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. j Graph represents the percentage of polyQ aggregate-positive cells/total nuclei in HD Q180-iPSCs (mean ± s.e.m., 3 independent experiments, 300–400 total cells per condition). Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. k Filter trap analysis of the indicated control and HD-iPSC lines with anti-polyQ-expansion diseases marker antibody. Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. The images are representative of three independent experiments. l Filter trap analysis of HD Q180-iPSCs with anti-polyQ-expansion diseases marker antibody. Proteasome inhibitor treatment: 5 µM MG-132 for 12 h. The images are representative of three independent experiments. All the statistical comparisons were made by Student’s t -test for unpaired samples. **** P < 0.0001 Full size image Notably, proteasome inhibition did not further increase the accumulation of aggregates induced by loss of UBR5 in HD-iPSCs (Fig. 5h–l and Supplementary Fig. 19 ), indicating that UBR5 regulates proteostasis of mutant HTT via its proteasomal degradation. To further determine the impact of UBR5 on HTT levels and aggregation, we examined other components of the ubiquitin–proteasome system (UPS) network previously associated with HTT regulation. In particular, we focused on UBE3A, an E3 enzyme that promotes proteasomal degradation of polyQ-expanded HTT in cell lines 40 . In addition, we examined UBE2K, an E2 enzyme that was found to interact with HTT in a yeast two-hybrid screen and reduce polyQ aggregation in cell lines 41 , 42 . Although both UBE3A and UBE2K are upregulated in pluripotent stem cells 35 , their knockdown did not increase mutant HTT levels and aggregation in HD-iPSCs (Fig. 6a, b and Supplementary Fig. 20 ). Thus, these results further support a specific role of UBR5 as a key determinant of HTT levels in iPSCs. Whereas the UPS regulates HTT degradation facilitating its proteostasis, the chaperone network is essential to modulate different steps of the aggregation process 29 . Remarkably, pluripotent stem cells exhibit an intrinsic chaperone network that could contribute to preventing mutant HTT aggregation 12 . For instance, these cells have increased levels of HTT-interacting protein K (HYPK) 35 , a chaperone that reduces polyQ-expanded HTT aggregates in mouse neuroblastoma cell lines 43 . However, loss of HYPK did not trigger aggregation of mutant HTT in HD-iPSCs (Fig. 6b ). Pluripotent stem cells also exhibit increased assembly of the TRiC/CCT complex 12 , a chaperonin that suppresses mutant HTT aggregation 44 , 45 . In hESCs/iPSCs, increased assembly of the TRiC/CCT complex is induced by high levels of CCT8 subunit 12 . As UBR5 knockdown, loss of CCT8 triggers aggregation of mutant HTT in HD-iPSCs (Fig. 6b ) 12 . In contrast, CCT8 downregulation did not affect HTT protein levels in HD-iPSCs (Fig. 6a ), supporting a direct role of the TRiC/CCT complex in mutant HTT aggregation rather than degradation. Altogether, our results indicate a key role of UBR5 in monitoring HTT levels, a process that could facilitate mutant HTT regulation by other proteostasis nodes such as enhanced TRiC/CCT complex. Fig. 6 UBR5 maintains proteostasis of HTT but not ATXN3 in iPSCs. a Western blot analysis of HD Q71-iPSC line #1 with antibody to HTT. The graph represents the nHTT and mHTT relative percentage values (corrected for β-actin loading control) to NT shRNA iPSCs (mean ± s.e.m. of three independent experiments). b Immunocytochemistry of HD Q71-iPSC line #1 upon knockdown of the indicated proteostasis components. PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of three independent experiments. c Western blot analysis with antibodies to HTT and ATXN3 of iPSCs derived from two individuals with MJD (MJD-iPSC lines #1 and #2) and control iPSCs. Anti-ATXN3 antibody detects both normal (nATXN3) and mutant ATXN3 (mATXN3). Proteasome inhibitor treatment: 5 µM MG-132 for 6 h. The graphs represent the HTT, nATXN3, mATXN3 relative percentage values to the respective DMSO-treated iPSCs corrected for β-actin loading control (mean ± s.e.m. of three independent experiments). d Immunocytochemistry of control iPSCs #2, MJD-iPSC lines #1 and #2 upon proteasome inhibition (5 µM MG-132 for 6 h). PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of three independent experiments. e , f Western blot analysis with antibodies to HTT and ATXN3 of MJD-iPSC lines #1 and #2 upon UBR5 knockdown. The graphs represent the relative percentage values of HTT, nATXN3 and mATXN3 to the respective NT shRNA iPSCs corrected for β-actin loading control (mean ± s.e.m. of three independent experiments for each line). g Immunocytochemistry of MJD-iPSC lines #1 and #2 upon UBR5 knockdown. Proteasome inhibitor: 5 µM MG-132 for 6 h. PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of three independent experiments. All the statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, ** P < 0.01, *** P < 0.001 Full size image Loss of UBR5 triggers aggresome formation in hESCs/iPSCs Besides its role in HTT regulation, we asked whether UBR5 modulates other disease-associated proteins. To assess this hypothesis, we first examined ataxin-3 (ATXN3), a distinct polyQ-containing protein. An abnormal expansion of CAG triplets (>52) in the ATXN3 gene causes Machado–Joseph disease (MJD), a neurodegenerative disorder characterized by neuronal loss in the cerebellum and progressive ataxia 46 , 47 . Here we used iPSCs derived from two individuals with MJD (MJD-iPSCs) (Supplementary Table 1 ), both expressing one normal copy of ATXN3 and one mutant allele with 74 CAG repeats (Fig. 6c ). As previously reported 48 , these MJD-iPSCs did not accumulate polyQ-expanded aggregates (Fig. 6d ). To examine whether the UPS regulates proteostasis of ATXN3 in iPSCs, we downregulated proteasome activity by using MG-132 proteasome inhibitor. In contrast with HD-iPSC lines, the treatment with proteasome inhibitor for 12 h induced acute cell death and detachment of MJD-iPSCs. To reduce these effects, we performed our analysis at an earlier time point of the treatment (6 h) (Supplementary Fig. 21 ). Notably, proteasome inhibition did not change the levels of normal and mutant ATXN3, whereas the amounts of HTT were upregulated in these cells (Fig. 6c ). Thus, our results suggest that the UPS does not modulate the levels of ATXN3 in iPSCs. However, proteasome inhibition triggered polyQ-expanded ATXN3 aggregation (Fig. 6d ), a process that could be linked with the global proteostasis collapse induced by this treatment. Although loss of UBR5 increased HTT levels in MJD-iPSCs, we did not find changes in either normal or mutant ATXN3 (Fig. 6e, f ). Moreover, we found that UBR5 binds HTT but it does not interact with wild-type or mutant ATXN3 in MJD-iPSCs (Supplementary Fig. 22a ). In contrast to global proteasome inhibition, UBR5 downregulation did not induce aggregation of polyQ-expanded ATXN3 (Fig. 6g ). This phenotype differed from HD-iPSCs lines, where upregulation of mHTT levels upon UBR5 knockdown was sufficient to trigger polyQ-expanded HTT aggregation (Fig. 5 ). Thus, these data indicate that not all polyQ-containing proteins are modulated by UBR5. To further assess the role of UBR5 in the proteostasis of aggregation-prone proteins, we used iPSCs carrying mutations in the RNA-binding protein FUS 49 . These mutations are linked with amyotrophic lateral sclerosis (ALS), a fatal neurodegenerative disorder characterized by loss of motor neurons and concomitant progressive muscle atrophy 49 , 50 . Although wild-type FUS shuttles between the nucleus and cytoplasm, it is mostly localized in the nucleus in normal conditions 51 . However, many of the ALS-related FUS mutations disrupt the nuclear import of the protein, resulting in aberrant localization and aggregation of FUS in the cytoplasm 52 , 53 . As previously reported 49 , wild-type FUS was essentially located in the nucleus of iPSCs (Fig. 7a ). Likewise, FUS was predominantly detected in the nucleus of iPSCs (FUS R521C/wt ) derived from a patient affected by ALS in mid-late age 49 , 50 (Fig. 7a ). In iPSCs expressing a FUS variant linked with severe and juvenile ALS (FUS P525L/P525L ), the protein was mostly located in the cytoplasm 49 (Fig. 7a ). Under oxidative stress, the cells form stress granules (SGs) where mutant variants of FUS are recruited 54 . Despite the induction of cytoplasmic SGs, wild-type FUS remains in the nucleus in iPSCs under oxidative stress 49 . On the contrary, mutant variants of FUS colocalize with SGs 49 . Similar to oxidative stress, proteasome inhibition induced the accumulation of SGs in iPSCs (Fig. 7a ). In control iPSCs, FUS remained in the nucleus upon proteasome dysfunction (Fig. 7a ). Although FUS signal was mostly nuclear in FUS R521C -iPSCs upon proteasome inhibition, we also found cytoplasmic FUS speckles co-localizing with SGs as reported for oxidative stress 49 (Fig. 7a ). On the other hand, FUS P525L variant showed a strong co-localization with SGs under proteasome inhibition (Fig. 7a ). Remarkably, loss of UBR5 did not stimulate SG formation or FUS delocalization in control or ALS-iPSC lines (Fig. 7a ). In addition, UBR5 knockdown did not impair wild-type or mutant FUS levels, whereas HTT was upregulated in these cells (Fig. 7b ). Accordingly, UBR5 interacted with HTT in both control and ALS-iPSC lines, but we could not detect interaction with wild-type or mutant FUS in these cells (Supplementary Fig. 22b, c ). Thus, these results indicate a specific role of UBR5 on HTT regulation. Although UBR5 was dispensable for ATXN3 and FUS proteostasis, we cannot discard a role of UBR5 in the control of other aggregation-prone proteins associated with disease. Fig. 7 Loss of UBR5 triggers aggresome formation in pluripotent stem cells. a Immunocytochemistry of control iPSCs #4 (FUS wt/wt ), ALS-iPSCs #1 (FUS R521C/wt ) and ALS-iPSCs #2 (FUS P525L/P525L ) with anti-FUS antibody. G3BP1 and Hoechst 33342 staining were used as markers of stress granules (SGs) and nuclei, respectively. Proteasome inhibition: 5 µM MG-132 for 6 h. Scale bar represents 10 μm. Arrows indicate examples of co-localization of FUS with SGs. The images are representative of three independent experiments. b Western blot analysis with antibodies to HTT and FUS of ALS-iPSC lines upon UBR5 knockdown. The graphs represent the relative percentage values of HTT and FUS to the respective NT shRNA iPSCs corrected for β-actin loading control (mean ± s.e.m. of three independent experiments for each line). All the statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, ** P < 0.01, *** P < 0.001. c Staining of aggresomes in H9 hESCs, control iPSCs #2 and HD Q71-iPSCs #1. Hoechst 33342 staining was used as a marker of nuclei. Heat stress: 42 °C for 4 h. Scale bar represents 10 μm. The images are representative of two independent experiments for each line Full size image Besides its role in HTT modulation, we asked whether UBR5 also determines the global proteostatic ability of pluripotent cells. Under normal conditions, misfolded proteins are refolded by chaperones or terminated via proteolytic systems 10 . Metabolic and environmental conditions (e.g., heat stress) challenge the structure of proteins, increasing the load of misfolded and damaged proteins. When proteolytic systems are overwhelmed, misfolded proteins accumulate into aggresomes. In these lines, we observed that heat stress induces the accumulation of aggresomes in pluripotent stem cells despite their increased proteolytic ability (Fig. 7c ). Notably, loss of UBR5 was sufficient to induce the accumulation of aggresomes in these cells (Fig. 7c ). Thus, our data suggest that UBR5 not only regulates the levels of specific proteins such as HTT, but is also involved in the degradation of misfolded proteins ensued during normal metabolism. Mutant HTT aggregates do not affect neural differentiation Loss of UBR5 did not impair the levels of pluripotency and germ layer markers in control iPSCs (Supplementary Fig. 23 a–c). With the pronounced accumulation of mutant HTT aggregates induced by UBR5 knockdown in HD-iPSCs (Fig. 5 ), we asked whether these inclusions alter iPSC identity. However, the accumulation of aggregates did not result in decreased expression of pluripotency markers in HD-iPSCs (Fig. 8a and Supplementary Fig. 24 ). Likewise, we did not observe increased levels of markers of the distinct germ layers (Fig. 8a ), indicating that polyQ-expanded HTT aggregates do not induce differentiation. Accordingly, we did not find a decrease in the number of cells expressing the pluripotency marker OCT4 (Fig. 8b and Supplementary Fig. 25 ). Most importantly, UBR5 knockdown HD-iPSCs could differentiate into neural cells (Fig. 8c and Supplementary Fig. 26 ). Under neural induction treatment, naive HD-iPSCs differentiated into NPCs with no detectable amounts of polyQ-expanded-HTT aggregates (Fig. 8d and Supplementary Fig. 27 ). However, neural cells derived from HD-iPSCs with impaired expression of UBR5 exhibited increased levels of mutant HTT and accumulation of aggregates (Fig. 8d–h and Supplementary Figs. 27 – 29 ). Overall, our results indicate that intrinsic high levels of UBR5 are essential to suppress mutant HTT aggregation in iPSCs, contributing to their ability to generate neural progenitor cells with no detectable polyQ-expanded aggregates. One step further was to determine whether these cells are able to generate terminally differentiated neurons. Since GABAergic medium spiny neurons (MSNs) undergo the greatest neurodegeneration in HD 55 , we differentiated iPSCs into striatal neurons 56 . Among those cells expressing the neuronal marker microtubule-associated protein-2 (MAP2), ∼ 30–60% also expressed GABA depending on the cell line (Supplementary Fig. 30a, b ). As previously reported 25 , these differences were not associated with the expression of mutant HTT (Supplementary Fig. 30a, b ). Knockdown of UBR5 at the iPSC stage did not reduce their ability to differentiate into striatal neurons (Supplementary Fig. 30 a–g). In contrast to NPCs, terminally differentiated neurons derived from HD-iPSCs with downregulated UBR5 levels did not accumulate polyQ-expanded aggregates (Supplementary Fig. 31 ). Thus, the lack of defects in neurogenesis and aggregates in these cells indicate that other mechanisms activate during neuronal differentiation to facilitate proteostasis of mutant HTT. Fig. 8 UBR5 determines HD-iPSC differentiation into NPCs with no aggregates. a qPCR of pluripotency ( OCT4 , NANOG , SOX2 , DPPA4 ), endodermal ( GATA6 , AFP ), mesodermal ( MSX1 ) and ectodermal markers ( PAX6 , NES ) in HD Q71-iPSC line #1. The graph represents the relative expression to NT shRNA HD-iPSCs (mean ± s.e.m. ( n = four independent experiments)). b Immunocytochemistry of HD Q71-iPSC line #1. OCT4 and Hoechst 33342 staining were used as markers of pluripotency and nuclei, respectively. Scale bar represents 20 μm. The images are representative of four independent experiments. c After neural induction of HD Q71-iPSC line #1 with downregulated levels of UBR5, we observed similar numbers of PAX6-positive cells compared to NT shRNA control. Scale bar represents 20 μm. d Upon neural induction of HD-iPSCs with downregulated levels of UBR5, NPCs accumulate mutant HTT aggregates. PolyQ-expanded and Hoechst 33342 staining were used as markers of aggregates and nuclei, respectively. Scale bar represents 10 μm. The images are representative of three independent experiments. e Filter trap analysis of polyQ-expanded aggregates in NPCs derived from HD Q71-iPSCs #1 with downregulated levels of UBR5. The images are representative of four independent experiments. f Filter trap of control NPCs #2 and HD Q180-NPCs derived from iPSCs with downregulated levels of UBR5. The images are representative of three independent experiments. g HD Q71-iPSCs #1 with downregulated UBR5 levels were differentiated into NPCs and analyzed by western blot. Graphs represent the relative percentage values to NT shRNA NPCs (corrected for β-actin) of nHTT and mHTT detected with antibodies to total HTT and polyQ-expanded proteins (mean ± s.e.m. of three independent experiments). h Western blot analysis of NPCs derived from HD Q180-iPSCs with downregulated levels of UBR5. Graphs represent the relative percentage values to NT shRNA NPCs (corrected for β-actin) of nHTT and mHTT detected with antibodies to total HTT and polyQ-expanded proteins (mean ± s.e.m. of three independent experiments), respectively. In g , h arrow indicates polyQ-expanded HTT detected with total HTT antibody. Statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, ** P < 0.01 Full size image UBR5 loss hastens neurotoxicity in polyQ nematode models Although HD-iPSCs can terminally differentiate into MSNs, these cells do not exhibit mutant HTT aggregates even after the addition of proteasome and autophagy inhibitors or the induction of oxidative stress 12 , 20 , 25 , 27 . Thus, these findings support a rejuvenation process during cell reprogramming that prevents aberrant aggregation in differentiated neurons. In addition, the lack of polyQ-expanded aggregates in these cells could reflect the long period of time before aggregates accumulate in HD 25 . In these lines, HD-MSNs derived from iPSCs do not accumulate detectable polyQ-expanded inclusions at 12 weeks after transplantation into HD rat models. However, they accumulate aggregates after 33 weeks of transplantation 27 . Proteasome inhibition did not induce upregulation of HTT protein levels or aggregation in HD-MSNs differentiated from iPSCs (Supplementary Fig. 32 a–d). Likewise, knockdown of UBR5 in differentiated HD-MSNs was not sufficient to induce mutant HTT aggregation and upregulated HTT levels (Supplementary Fig. 33 , 34 ). In addition, UBR5 knockdown did not reduce cell viability of HD-MSNs (Supplementary Fig. 35 ). Given the challenges presented by MSNs derived from HD-iPSCs to study the role of UBR5 in polyQ-expanded aggregation and toxicity, we used a distinct model. Remarkably, a RNA interference (RNAi) screen against E3 ubiquitin ligases found that knockdown of the worm UBR5 ortholog ( ubr-5 ) accelerates paralysis in a C. elegans model expressing 35 polyQ repeats fused to yellow fluorescent protein (YFP) in body wall muscle cells 57 . To assess the requirement of ubr-5 for resistance to polyQ neurotoxicity, we examined a C. elegans model that expresses polyQ-expanded YFP in the nervous system 58 . In these worms, polyQ aggregation and neurotoxicity correlates with the age and length of the polyQ repeat, with a pathogenic threshold of 40 glutamine repeats 58 . Notably, loss of ubr-5 resulted in increased levels of polyQ67-expanded protein with concomitant aggregation (Fig. 9a, b ). Knockdown of ubr-5 had a strong effect in the aggregation propensity of head neurons, resulting in more aggregates in the circumpharyngeal nerve ring and chemosensory processes (Fig. 9c ). We also observed more propensity aggregation in the neurons of the animal mid-body (Fig. 9c and Supplementary Fig. 36 ). To assess whether the polyQ67-YFP foci were immobile protein aggregates, we performed a quantitative fluorescence recovery after photobleaching analysis of head neurons. In both empty vector and ubr-5 RNAi-treated worms, most of the polyQ67-YFP foci signal could not be recovered, indicating an immobile state (Supplementary Fig. 37a ). However, ubr-5 RNAi induced a faster incorporation of new polyQ67-YFP peptides into the aggregates (Supplementary Fig. 37b ). Fig. 9 UBR5 loss hastens aggregation and neurotoxicity in HD C. elegans models. a Western blot analysis of C. elegans with antibodies to GFP and β-actin loading control. The graph represents the Q67-YFP relative percentage values to Q67, vector RNAi-C. elegans corrected for β-actin loading control (mean ± s.e.m. of three independent experiments). b Filter trap analysis indicates that ubr-5 knockdown increases polyQ aggregates (detected by anti-GFP antibody) in C. elegans . The images are representative of four independent experiments. c Representative images of Q67-YFP aggregation under ubr-5 knockdown in whole adult worms. On the right, a higher magnification of head neurons is presented. For higher magnification of C. elegans mid-body, please see Supplementary Fig. 36 . The images are representative of five independent experiments. Scale bar represents 100 μm. White arrows and red arrows indicate nerve ring and chemosensory processes, respectively. DIC, differential interference contrast. d Filter trap analysis with anti-GFP antibody of polyQ-expressing neuronal models in wild-type and rrf-3 ( pk1426 ) background. The images are representative of five independent experiments. e Loss of ubr-5 hastens the motility defects of polyQ67 worms. Bar graphs represent average ( ± s.e.m.) thrashing movements over a 30 s period on day 3 of adulthood (Q19 fed vector RNAi ( n = 62) versus Q19 fed ubr-5 RNAi ( n = 64), P = 0.11; Q19 fed vector RNAi versus Q67 fed vector RNAi ( n = 65), P < 0.0001; Q67 fed vector RNAi versus Q67 fed ubr-5 RNAi ( n = 65), P < 0.0001). All the statistical comparisons were made by Student’s t- test for unpaired samples. ** P < 0.01, **** P < 0.0001 Full size image Although ubr-5 RNAi was sufficient to induce a pronounced increase of polyQ-expanded aggregates (Fig. 9b ), it is important to note that neurons are less sensitive to RNAi when compared with other tissues 59 . For this reason, we introduced a rrf-3 mutation in the polyQ-YFP neuronal models, which confers hypersensitivity to RNAi in all the tissues, including neurons 60 . Accordingly, we found that ubr-5 RNAi treatment induced a strong increase in polyQ67-YFP aggregation of rrf-3 mutants (Fig. 9d ). On the contrary, ubr-5 knockdown did not induce aggregation of polyQ19-peptides, even in the RNAi-hypersensitive mutant strain (Fig. 9d ). Since the neurotoxic effects of polyQ-expanded aggregation correlate with impairment of coordinated movement 58 , we performed motility assays to quantify the role of ubr-5 in the resistance to polyQ-toxicity. Notably, loss of ubr-5 hastened the detrimental effects on motility induced by polyQ67 repeats (Fig. 9e ), correlating with increased polyQ aggregation. On the contrary, knockdown of ubr-5 did not affect the motility of polyQ19-expressing worms (Fig. 9e ). Altogether, these results provide a direct link between UBR5 function and polyQ-expanded aggregation with age. Ectopic expression of UBR5 suppresses mutant HTT aggregation Since loss of UBR5 hastened polyQ-expanded aggregation, we asked whether ectopic expression of this HECT E3 enzyme is sufficient to ameliorate the accumulation of polyQ aggregates. To assess this hypothesis, we generated human cell models that express either control (Q23) or polyQ-expanded (Q100) HTT protein. In these cells, overexpression of mutant HTT resulted in the accumulation of polyQ-expanded aggregates, whereas control HTT did not form aggregates (Fig. 10a ). Notably, ectopic expression of UBR5 ameliorated polyQ-expanded aggregation in Q100-HTT cells (Fig. 10a ), a process blocked by proteasome inhibition (Fig. 10b ). In contrast, UBR5 overexpression did not reduce the insolubility of aggregation-prone β-amyloid protein (Supplementary Fig. 38 ). Fig. 10 Upregulation of UBR5 suppresses polyQ-expanded HTT aggregation. a Filter trap analysis with anti-polyQ-expansion diseases marker antibody of HEK293 human cells overexpressing (OE) Q23-HTT or Q100-HTT. Ectopic expression of UBR5 reduces the accumulation of aggregates in Q100-HTT(OE) cells. On the contrary, overexpression of a catalytic inactive UBR5 mutant (UBR5 ∆HECT) does not ameliorate polyQ-expanded aggregation. The images are representative of three independent experiments. b Filter trap analysis of Q100-HTT(OE) HEK293 cells. Proteasome inhibition with 0.5 µM MG-132 for 16 h blocks the reduction of polyQ-expanded aggregation induced by ectopic expression of wild-type UBR5. The images are representative of three independent experiments. c Western blot analysis of Q100-HTT(OE) HEK293 cells with antibodies to HTT and UBR5. Overexpression of wild-type UBR5 reduces the levels of Q100-HTT, whereas similar levels of UBR5 ΔHECT do not decrease mutant HTT levels. The graph represents the relative percentage values of Q100-HTT to empty vector cells corrected for β-actin loading control (mean ± s.e.m. of three independent experiments). d Western blot analysis of whole cell lysates from Q100-HTT(OE) HEK293 cultures with antibodies to HTT and β-actin loading control. Proteasome inhibition with 0.5 µM MG-132 for 16 h blocks proteasomal degradation of Q100-HTT levels induced by wild-type UBR5 overexpression. The graph represents the relative percentage values of Q100-HTT to DMSO-empty vector cells corrected for β-actin loading control (mean ± s.e.m. of three independent experiments). e Immunoprecipitation with anti-HTT and anti-FLAG antibodies in Q100-HTT(OE) HEK293. Immunoprecipitation was followed by western blot with antibodies to HTT and polyubiquitinated proteins (polyUb) to detect immunoprecipitated total HTT protein and polyUb-HTT, respectively. Prior to immunoprecipitation, cells were treated with proteasome inhibitor (0.5 µM MG-132, 16 h) to block the degradation of HTT induced by UBR5 so we could immunoprecipitate similar amounts of HTT for direct comparison of polyubiquitination among the distinct conditions. The images are representative of three independent experiments. All the statistical comparisons were made by Student’s t -test for unpaired samples. * P < 0.05, **** P < 0.0001 Full size image In support of a direct role of UBR5 in modulation of HTT, we found that UBR5 overexpression decreases the protein levels of mutant HTT (Fig. 10c ). To further determine the role of UBR5, we overexpressed a UBR5 mutant with two point mutations that causes a change in an amino acid (C2769A) located in the HECT domain, resulting in ubiquitin ligase-dead UBR5 61 . Remarkably, similar overexpression levels of this catalytic inactive UBR5 mutant did not diminish polyQ-expanded HTT protein levels and aggregation (Fig. 10a–c ). Given that our results in iPSCs indicate that UBR5 regulates HTT levels via the proteasome, we tested whether UBR5 overexpression promotes proteasomal degradation of Q100-HTT. Indeed, proteasome inhibition blocked the reduction of Q100-HTT levels induced by enhanced wild-type UBR5 expression (Fig. 10d ). To further assess the link between UBR5 and HTT regulation, we performed immunoprecipitation experiments and examined polyubiquitination of HTT. Prior to immunoprecipitation, we treated the cells with proteasome inhibitor to block the degradation of HTT induced by UBR5. Under these conditions, we immunoprecipitated similar amounts of HTT in cells overexpressing wild-type UBR5 when compared with cells expressing empty vector or catalytic inactive UBR5 (Fig. 10e ). Notably, we found that ectopic expression of UBR5 induces a dramatic increase in polyubiquitination of mutant HTT (Fig. 10e ). In contrast, overexpression of the catalytic inactive UBR5 mutant did not promote polyubiquitination of Q100-HTT protein (Fig. 10e ). In support of our findings, a recent study reported that UBR5 downregulation decreases the modification of overexpressed HTT protein with K11/K48-linked polyubiquitin chains in HeLa human cells 36 . Moreover, this study revealed K11/K48-linked polyubiquitin modifications at Lys337 of HTT by proteomics experiments 36 . We performed proteomics analysis of immunoprecipitated HTT in Q100-HTT-overexpressing cells to determine potential lysine sites modified by UBR5 (Supplementary Fig. 39a, b ). Although we did not detect ubiquitination at Lys337 of HTT in our assays, we identified two other ubiquitinated lysine sites: Lys631 and Lys2097. However, only Lys631 shows a small but significant higher ubiquitination in cells overexpressing wild-type UBR5 when compared to catalytic inactive UBR5 (Supplementary Fig. 39b ), suggesting that this site could be ubiquitinated by UBR5. Taken together, our data indicate that the ubiquitin ligase activity of UBR5 modulates proteasomal degradation of mutant HTT, a process that ameliorates polyQ-expanded aggregation. Discussion While the transcriptional, epigenetic and signaling networks of pluripotency have been a primary focus of research efforts, emerging evidence indicates that pluripotent stem cells also exhibit intrinsic proteostasis mechanisms 8 , 12 , 14 , 17 . Thus, a comprehensive understanding of this proteostasis network could be necessary for pluripotent stem cells to hold a great promise for regenerative medicine. As an invaluable resource to generate terminally differentiated cells, pluripotent stem cells can facilitate the study of human diseases and drug screening. This is particularly fascinating in the context of proteostasis-related disorders. At least 30 different human diseases are directly associated with aberrant protein folding protein and aggregation 62 . Whereas a collapse in proteostasis of somatic cells could underlie these diseases, pluripotent stem cells exhibit a striking ability to correct and suppress proteostatic deficiencies. Thus, investigating pluripotent stem cells from patients could contribute to identifying super-vigilant proteostasis mechanisms that could be mimicked in somatic tissues to ameliorate disease. Despite expressing significant amounts of mutant HTT 25 , several studies have demonstrated that HD-iPSCs do not accumulate polyQ-expanded HTT aggregates even after multiple passages 12 , 25 , 27 . In addition, HD-iPSCs can terminally differentiate into neurons lacking the aggregation phenotype characteristic of HD 12 , 25 , 27 , indicating a rejuvenation process during reprogramming that prevents polyQ-expanded aggregation in neurons prior to proteostasis collapse with age. Here we show that increased proteasome activity of iPSCs regulates the levels of both normal and mutant HTT, contributing to suppressing polyQ-expanded HTT aggregation in HD-iPSCs. Conversely, a dysfunction in proteasome activity results in impaired HTT levels, leading to aggregation of mutant HTT in HD-iPSCs. Importantly, we have uncovered that intrinsic high expression of UBR5 is a key component of the UPS to regulate HTT levels in iPSCs. This E3 enzyme interacts with HTT and promotes proteasomal degradation of both normal and polyQ-expanded HTT. Since the clearance of misfolded and aggregation-prone proteins is key to their aggregation, the impairment of HTT levels could be particularly relevant in iPSCs expressing mutant HTT. Accordingly, UBR5 downregulation and concomitant dysregulation of HTT levels results in polyQ-expanded HTT aggregation in HD-iPSCs. Thus, our results suggest that UBR5 monitors HTT levels, facilitating the regulation of mutant HTT by other proteostasis nodes. For instance, pluripotent stem cells also exhibit increased activity of the TRiC/CCT chaperone complex to tightly suppress the aggregation of mutant HTT 12 . We found that the HECT domain containing the E3 ligase activity of UBR5 is required for HTT degradation, providing a further link between increased proteasome activity and stringent control of mutant HTT. It is important to note that pluripotent stem cells also express high levels of other E3 ligases. Although our results discard a direct role of several of these E3s (i.e., UBR7, UBE3A, RNF181) in modulation of HTT levels in iPSCs, it will be fascinating to examine the impact of the other upregulated E3s. Whereas our data highlight the importance of UBR5 in proteostasis of iPSCs, the intrinsic high levels of this enzyme could also suggest a role in pluripotency via the ubiquitination of endogenous substrates. Interestingly, a study has reported that Ubr5 knockdown results in significant loss of pluripotency markers in mouse ESCs 14 . However, a different study did not find impairment of pluripotency markers in murine ESCs upon Ubr5 RNAi treatment 63 . Likewise, we did not find changes in the expression of pluripotency or germ layer markers in control hESCs/iPSCs upon loss of UBR5. Interestingly, knockdown of Ubr5 reduces the induction of Sonic hedgehog (Shh) in murine pluripotent stem cells upon retinoic-acid treatment 63 . In these lines, a conditional Ubr5 mutant mouse presents decreased hedgehog signaling during embryogenesis 63 . Although this model exhibits shorter limbs, the differences were not significant and the mutants do not have other obvious morphological defects 63 . Since our striatal neuronal differentiation protocol is based on the induction of hedgehog signaling, we tested whether UBR5 is necessary for the generation of MSNs from human iPSCs. However, loss of UBR5 did not affect their differentiation into MSNs, indicating that these cells conserved their ability to induce the hedgehog signaling. The distinct impact of UBR5 in pluripotency and differentiation could be associated with genetic differences between mouse and human species. Moreover, it is important to note the distinct pluripotent states exhibited by murine ESCs and hESCs/iPSCs 3 . Mouse ESCs are cultured in the presence of serum and leukemia inhibitory factor (LIF), and exhibit a naive state resembling the pluripotent state observed in the inner cell mass of the pre-implantation embryo 3 . On the other hand, hESCs as well as human iPSCs do not require LIF signaling and exhibit a more primed state that resembles post-implantation embryonic configurations 3 . Although loss of UBR5 did not impair pluripotency markers in human control iPSCs, it induced the formation of misfolded protein aggregates (i.e., aggresomes). However, we found that the accumulation of polyQ-expanded aggregates in HD-iPSCs does not impair pluripotency markers or induce their differentiation. Most importantly, these cells retain the ability to differentiate into neural cells that also accumulate polyQ-expanded aggregates. With these results, we speculate that increased proteostasis of pluripotent stem cells may be required to avoid the generation of precursor cells that accumulate protein aggregates at early organismal stages, a process that could be detrimental to organismal survival and healthspan. However, it is important to remark that most of our studies were performed in iPSCs, and further evidence in hESCs as well as in vivo experiments using mouse models are necessary to assess this intriguing possibility. On the other hand, HD-iPSCs with decreased levels of UBR5 can be further differentiated into MSNs with no aggregates. These results could imply the activation of additional proteostatic control mechanisms to scavenge aggregates during terminal neuronal differentiation. It will be fascinating to define these mechanisms due to its implications for development and organismal aging. Notably, genetic variations in a region of the chromosome 8 that contains the UBR5 gene, among others (i.e., RRM2B , MIR5680, NCALD), have been associated with an early onset of HD by genome-wide association analysis 37 . Whereas our results demonstrate that UBR5 levels decrease during differentiation, an intriguing possibility is that a further dysfunction of UBR5 activity with age could contribute to the onset of diseases such as HD. However, modeling HD as well as other neurodegenerative disorders using patient-specific neurons is challenging, as neurons differentiated from iPSCs lack aggregates and strong cell death phenotype 12 , 20 , 25 , 27 . Moreover, iPSC-derived neurons do not exhibit HTT mutant aggregates even after the addition of cellular stressors such as proteasome inhibitors. Although the relevance of UBR5 in HD pathology remains unclear due to these limitations, we find that ectopic expression of UBR5 is sufficient to reduce the protein levels of polyQ-expanded HTT and its aggregation in mutant HTT-overexpressing cell models. Moreover, UBR5 knockdown hastens the deleterious changes induced by polyQ-expanded expression in C. elegans models, providing direct evidence of a link between UBR5 activity and polyQ-expanded aggregation with age. It is important to note that these C. elegans models express polyQ-expanded fused to YFP but not HTT protein. Thus, these results suggest that UBR5 could also have a role in the proteostasis of other polyQ-containing proteins related with disease. However, we found that UBR5 knockdown does not impair polyQ-expanded ATXN3 levels and aggregation in MJD-iPSCs. Thus, we conclude that not all polyQ-containing proteins are regulated via UBR5 activity. In these lines, we observed that UBR5 does not impinge upon cellular localization and aggregation of FUS variants linked with ALS. Although these results suggest specificity of UBR5 for HTT regulation, we cannot discard a role in the proteostasis of other aggregation-prone proteins associated with disease. Notably, UBR5 downregulation is sufficient to induce the accumulation of aggresomes in control hESCs/iPSCs. Since aggresomes form from the accumulation of misfolded proteins when proteolytic systems are overwhelmed, UBR5 may also be involved in the degradation of misfolded proteins ensued from normal metabolism. Taken together, our studies in immortal pluripotent stem cells identified UBR5 as a determinant of their super-vigilant proteostasis. Methods hESC/iPSC lines and culture The H9 (WA09) and H1 (WA01) hESC lines were obtained from the WiCell Research Institute. The H9 and H1 hESCs used in our study matches exactly the known short tandem repeat (STR) profile of these cells across the 8 STR loci analyzed (Supplementary Table 3a ). No STR polymorphisms other than those corresponding to H9 and H1 12 were found in the respective cell lines, indicating correct hESC identity and no contamination with any other human cell line. The control iPSCs #3 and HD Q71-iPSC lines #1 were a gift from G.Q. Daley. These cells were generated using retroviral induction of c-Myc, Klf4, Oct4 and Sox2 and fully characterized for pluripotency in refs. 26 , 64 . HD Q71-iPSC line #2 (ND42230) was obtained from NINDS Human Cell and Data Repository (NHCDR) through Coriell Institute. These cells were generated from the same parental fibroblast as HD Q71-iPSC line #1 via episomal expression of l-Myc, Klf4, Oct4, Sox2 and LIN28 reprogramming factors. Likewise, ND42242 (control iPSC line #1, Q21), ND36997 (control iPSC line #2, Q33), ND41656 (HD Q57-iPSC) and ND36999 (HD Q180-iPSC) were also obtained from NHCDR through Coriell Institute. Corrected isogenic counterparts of Q180-iPSCs (i.e., HD-C#1 and HD-C#2) were a gift from M.A. Pouladi 30 . By STR analysis, we confirmed correct genetic identity of the control and HD-iPSCs used in our study with the corresponding parental fibroblast lines when fibroblasts were available (that is, control iPSCs #1, control iPSCs #2, HD Q71-iPSCs, HD Q57-iPSCs and HD Q180-iPSCs; Supplementary Table 4 ). Parental fibroblasts GM02183 (control #2), GM04281 (HD Q71) and GM09197 (HD Q180) were obtained from Coriell Institute, whereas ND30014 (control #1) and ND33392 (HD Q57) fibroblasts were obtained from NHCDR through RUCDR Infinite Biologics at Rutgers University. UKBi001-B (MJD Q74-iPSC line #1) and UKBi003-A (MJD Q74-iPSC line #2) were obtained from the European Bank for induced pluripotent Stem Cells (EBiSC). These MJD-iPSC lines were generated using retroviral expression of reprogramming factors (i.e., c-Myc, Klf4, Oct4 and Sox2) and fully characterized for pluripotency in ref. 48 . The MJD-iPSCs used in our study matches exactly the STR profile of the parental fibroblasts provided by the depositor of the lines (Supplementary Table 3b ). Control iPSCs #4 (FUS wt/wt ), ALS-iPSCs #1 (FUS R521C/wt ) and ALS-iPSCs #2 (FUS P525L/P525L ) were kindly provided by A. Rosa and I. Bozzoni. All of them were produced and characterized for pluripotency in ref. 49 . Briefly, control iPSCs #4 were derived from a control donor and checked for absence of mutation in FUS 49 whereas ALS-iPSCs #1 were generated from a patient affected by ALS in mid-late age 50 . Both iPSCs lines were established via lentiviral expression of c-Myc, Klf4, Oct4 and Sox2. ALS-iPSCs #2 were raised from control iPSCs #4 by TALEN (transcription activator-like effector nucleases)-directed mutagenesis and are homozygote for a FUS mutation (P525L) linked with severe and juvenile ALS 49 . We confirmed that the STR profile of the ALS-iPSCs #2 used in our experiments matches with the profile of control iPSCs #4 (Supplementary Table 4 ). The hESCs and iPSCs were maintained on Geltrex (ThermoFisher Scientific) using mTeSR1 (Stem Cell Technologies). Undifferentiated hESC/iPSC colonies were passaged using a solution of dispase (2 mg ml −1 ), and scraping the colonies with a glass pipette. Alternatively, cells were individualized with Accutase (1 unit ml −1 , Invitrogen) and seeded into Geltrex-coated plates for experiments with proteasome and autophagy inhibitors to facilitate homogenous treatment of cultures. All the hESC and iPSC lines used in our experiments had a normal diploid karyotype as assessed by single nucleotide polymorphism (SNP) genotyping (Supplementary Fig. 40 ). Moreover, all the cell lines used in this study were tested for mycoplasma contamination at least once every 3 weeks. No mycoplasma contamination was detected. Research involving hESC lines was performed with approval of the German Federal competent authority (Robert Koch Institute). SNP genotyping The molecular karyotype was analyzed by SNP genotyping with Illumina’s HumanOmniExpressExome-8-v1.2 BeadArray (Illumina, Inc., San Diego, USA) at the Institute for Human Genetics (Department of Genomics, Life & Brain Center, University of Bonn, Germany). Processing was performed on genomic DNA following the manufacturer’s procedures. Copy number regions were detected using the cnvPartition version 3.1.6. STR analysis STR analysis of H9 and H1 hESCs was conducted using the Promega PowerPlex 21 system (Promega Corporation) by Eurofins Genomics (Germany). We analyzed loci D5S818, D13S317, D7S820, D16S539, vWA, TH01, TP0X and CSF1P0 to compare with the known STR profile of these hESC lines 12 . The STR analysis of MJD-iPSC lines was also performed by Eurofins Genomics using the Promega PowerPlex 21 system. Genotype comparison of control and HD-iPSCs lines with their parental fibroblasts was performed using the following microsatellite markers: D17S1303, D16S539, vWA, THO1, CSF1PO and TPOX. Fluorescently labeled PCR products were electrophoresed and detected on an automated 3730 DNA Analyzer and data were analyzed using Genemapper software version 3.0 to compare allele sizes between iPSCs and their parental fibroblasts (Applied Biosystems). Neural differentiation To obtain NPC cultures, we induced neural differentiation of iPSCs with STEMdiff Neural Induction Medium (Stem Cell Technologies) following the monolayer culture method 65 . Undifferentiated iPSCs were rinsed once with phosphate-buffered saline (PBS) and then we added 1 ml of Gentle Dissociation Reagent (Stem Cell Technologies) for 10 min. After the incubation period, we gently dislodged pluripotent cells and added 2 ml of Dulbecco’s Modified Eagle Medium (DMEM)/F12 (ThermoFisher Scientific)+10 μM ROCK inhibitor (Abcam). Then, we centrifuged cells at 300 × g for 10 min. Cells were resuspended on STEMdiff Neural Induction Medium+10 μM ROCK inhibitor and plated on poly-ornithine (15 μg ml −1 )/laminin (10 μg ml −1 )-coated plates at a density of 200,000 cells cm −2 . Striatal neuron differentiation iPSCs were differentiated into striatal neurons by induction of hedgehog signaling pathway 56 . iPSCs were detached by incubating with dispase (1 mg ml −1 ) for 20 min. The detached colonies were cultured in suspension as free-floating embryoid bodies (EBs) in the differentiation medium consisting of DMEM/F12, 20% knockout serum replacement, 100 μM β-Mercaptoethanol (Sigma), 1× minimum essential medium (MEM) non-essential amino acids and 2 mM l -glutamine. On day 4, the medium was replaced with a neural induction medium consisting of DMEM/F12, N2 supplement (ThermoFisher Scientific), 1× MEM non-essential amino acids, 2 mM glutamine and 2 μg ml −1 heparin. On day 7, the EBs were attached to the laminin (ThermoFisher Scientific)-coated substrate in a 35 mm culture Petri dish and cultured in the neural induction medium. In the next week, the EBs flattened and columnar neuroepithelia organized into rosette appeared in the center of individual colonies. On day 12, 0.65 μM purmorphamine (Stem Cell Technologies) was added for 14 days (until day 25). From day 26, neuroepithelial spheres were dissociated with Accutase (1 unit ml −1 , Invitrogen) at 37 °C for 5 min and placed onto poly-ornithine/laminin-coated cover slips in Neurobasal medium containing a set of trophic factors, including brain-derived neurotrophic factor (20 ng ml −1 ), glial-derived neurotrophic factor (10 ng ml −1 ), insulin-like growth factor-1 (10 ng ml −1 ), and cAMP (1 μM) (all from R&D Systems). DARPP32-expressing neurons appeared by day 32 as assessed by immunohistochemistry using Rabbit anti-DARPP32 (Abcam, ab40801, 1:50). We performed the experiments between days 32 and 35 of the differentiation protocol. Lentiviral infection of iPSCs Lentivirus (LV)-non-targeting small hairpin RNA (shRNA) control, LV-HTT shRNA (TRCN0000322961), LV-UBR5 shRNA #1 (TRCN0000003411), LV-UBR5 shRNA #2 (TRCN0000226458), LV-UBE3A shRNA (TRCN0000419838), LV-UBR7 shRNA (TRCN0000037025), LV-RNF181 shRNA #1 (TRCN0000364405), LV-RNF181 shRNA #2 (TRCN0000022389), LV-TRIM71 shRNA #1 (TRCN0000245956) and LV-TRIM71 shRNA #2 (TRCN0000245959) in pLKO.1-puro vector were obtained from Mission shRNA (Sigma). Transient infection experiments were performed as follows. iPSCs colonies growing on Geltrex were individualized using Accutase. Hundred thousand cells were plated on Geltrex plates and incubated with mTesR1 medium containing 10 μM ROCK inhibitor for 1 day. Then, cells were infected with 5 µl of concentrated lentivirus. Plates were centrifuged at 800 × g for 1 h at 30 °C. Cells were fed with fresh media the day after to remove the virus. After 1 day, cells were selected for lentiviral integration using 2 µg ml −1 puromycin (ThermoFisher Scientific). Cells were split for further experiments and collected after 5–7 days of infection. Transfection of HEK293T cells HEK293T cells (ATCC) were plated on 0.1% gelatin-coated plates and grown in DMEM supplemented with 10% fetal bovine serum and 1% MEM non-essential amino acids (ThermoFisher Scientific) at 37 °C, 5% CO 2 conditions. Cells were transfected once they reached 80–90% confluency. Then, 1 μg GFP-UBR5 wild-type or GFP-UBR5 ∆HECT overexpression plasmid and 0.5 μg pARIS-mCherry-httQ23-GFP or pARIS-mCherry-httQ100-GFP were used for transfection, using Fugene HD (Promega) following the manufacturer’s instructions. After 36 h of incubation in normal medium, the cells were harvested for further experiments. GFP-UBR5 wild-type and GFP-UBR5 ∆HECT overexpression plasmids (Addgene plasmids #52050 and # 52051, respectively) were a gift from D. Saunders and were first published in ref. 61 . The pARIS-mCherry-httQ23-GFP and pARIS-mCherry-httQ100-GFP 66 plasmids were a gift from F. Saudou. Analysis of β-amyloid aggregation N-terminal c-Myc-tagged β23 protein construct was a gift from F.U. Hartl 67 . For mammalian expression, c-Myc-tagged β23 sequence was cloned into pCDH-CMV-MCS-EF1 from pcDNA3.1 using Xba I and Bam HI enzymes. Cell fractionation was performed following the protocol described in ref. 67 . Briefly, cells were collected and resuspended in lysis buffer (50 mM Tris (pH 7.8), 150 mM NaCl, 1% (v/v) NP40, 0.25% sodium deoxycholate, 1 mM EDTA and 1 tablet protease inhibitor cocktail (Roche) per 10 ml). The cell lysate was then centrifuged at 20,000 × g for 30 min at 4 °C. The supernatant was applied for analyzing soluble fraction and pellets were dissolved in SDT buffer (4% (w/v) sodium dodecyl sulfate (SDS), 100 mM Tris-HCl (pH 7.6) and 0.1 M dithiothreitol (DTT)). Anti-Myc antibody (Proteintech, #66004-1-Ig, 1:2000) was used for western blot to detect c-Myc-tagged β23 protein. Protein immunoprecipitation for interaction analysis iPSCs were lysed in RIPA buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride (PMSF)) supplemented with protease inhibitor cocktail (Roche). Lysates were homogenized by passing 10 times through a 27-gauge (27 G) needle attached to a 1 ml syringe and centrifuged at 13,000 × g for 15 min at 4 °C. After pre-clearing the supernatant with Protein A agarose beads (Pierce), the samples were incubated overnight with UBR5 antibody (Cell Signaling, #8755, 1:50) on the overhead shaker at 4 °C. As a control, the same amount of protein was incubated with anti-FLAG (SIGMA, F7425, 1:50) or anti-UBE3A antibody (Cell Signaling, #7526, 1:50) in parallel. Subsequently, samples were incubated with 30 µl of Protein A beads for 1 h at room temperature. After this incubation, samples were centrifuged for 5 min at 5,000 × g and the pellet was washed three times with RIPA buffer. For elution of the proteins, the pellet was incubated with 2× Laemmli Buffer, boiled for 5 min and centrifuged for 5 min at maximum speed. The supernatant was taken and loaded onto a sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) gel for western blot analysis. Protein immunoprecipitation for assessing polyubiquitination HEK293 cells were lysed in protein lysis buffer (50 mM Tris-HCl (pH 6.7), 150 mM NaCl, 1% NP40, 0.25% sodium deoxycholate, 1 mM EDTA, 1 mM PMSF, 1 mM Na 3 VO, 1 mM NaF) supplemented with protease inhibitor cocktail. Lysates were homogenized through a syringe needle (27 G) and centrifuged at 13,000 × g for 15 min at 4 °C. The samples were incubated for 30 min with HTT antibody (Cell Signaling, ab#5656, 1:1000) on ice. As a negative control, the same amount of protein was incubated with anti-FLAG antibody (SIGMA, F7425) in parallel. Subsequently, samples were incubated with 50 µl of µMACS Micro Beads for 1 h at 4 °C with overhead shaking. After this incubation, samples were loaded to pre-cleared µMACS column (#130-042-701). Beads were washed three times with 50 mM Tris (pH 7.5) buffer containing 150 mM NaCl, 5% glycerol and 0.05% Triton and then washed five times with 50 mM Tris (pH 7.5) and 150 mM NaCl. For protein elution, the beads were incubated with 1× Laemmli Buffer for 5 min and collected into tubes. The samples were boiled for 5 min at 95 °C and loaded in a SDS–PAGE gel for western blot analysis. Proteomics analysis of HTT ubiquitination sites HEK293 cells were lysed in protein lysis buffer (50 mM Tris-HCl (pH 6.7), 150 mM NaCl, 1% NP40, 0.25% sodium deoxycholate, 1 mM EDTA, 1 mM NaF) supplemented with protease inhibitor cocktail. Lysates were homogenized through syringe needle (27 G) and centrifuged at 13,000 × g for 15 min at 4 °C. The samples were incubated for 30 min with HTT antibody (Cell Signaling, ab#5656, 1:1000) on ice. Subsequently, samples were incubated with 50 μl of μMACS Micro Beads for 1 h at 4 °C with overhead shaking. After this incubation, samples were loaded to pre-cleared μMACS column (#130-042-701). Beads were washed three times with 50 mM Tris (pH 7.5) buffer containing 150 mM NaCl, 5% glycerol and 0.05% Triton and then washed five times with 50 mM Tris (pH 7.5) and 150 mM NaCl. Then, columns were subjected to in-column tryptic digestion containing 7.5 mM ammonium bicarbonate, 2 M urea, 1 mM DTT and 5 ng ml −1 trypsin. Digested peptides were eluted using two times 50 μl of elution buffer 1 containing 2 M urea, 7.5 mM Ambic, and 5 mM iodoacetamide. Digests were incubated overnight at room temperature with mild shaking in the dark. Samples were stage-tipped the next day for label-free quantitative proteomics. All samples were analyzed on a Q-Exactive Plus (Thermo Scientific) mass spectrometer that was coupled to an EASY nLC 1200 UPLC (Thermo Scientific) 68 . Peptides were loaded with solvent A (0.1% formic acid in water) onto an in-house packed analytical column (50 cm × 75 µm I.D., filled with 2.7 µm Poroshell EC120 C18, Agilent). Peptides were chromatographically separated at a constant flow rate of 250 nl min −1 using 150 min methods: 5–30% solvent B (0.1% formic acid in 80% acetonitrile) within 119 min, 30–50% solvent B within 19 min, followed by washing and column equilibration. The mass spectrometer was operated in data-dependent acquisition mode. The MS1 survey scan was acquired from 300 to 1750 m/z at a resolution of 70,000. The top 10 most abundant peptides were subjected to higher collisional dissociation fragmentation at a normalized collision energy of 27%. The AGC (automatic gain control) target was set to 5e5 charges. Product ions were detected in the Orbitrap at a resolution of 17,500. All mass spectrometric raw data were processed with Maxquant (version 1.5.3.8) using default parameters. Briefly, MS2 spectra were searched against the human Uniprot database, including a list of common contaminants. False discovery rates (FDRs) on protein and peptide–spectrum match (PSM) level were estimated by the target-decoy approach to 0.01% (Protein FDR) and 0.01% (PSM FDR) respectively. The minimal peptide length was set to 7 amino acids and carbamidomethylation at cysteine residues was considered as a fixed modification. Oxidation (M), GlyGly (K) and Acetyl (Protein N-term) were included as variable modifications. The match-between runs option was enabled. Label-free quantification (LFQ) was enabled using default settings. The resulting output was processed using Perseus as follows: protein groups flagged as “reverse”, “potential contaminant” or “only identified by site” were removed from the proteinGroups.txt. LFQ values were log2 transformed. Missing values were replaced using an imputation-based approach (random sampling from a normal distribution using a down shift of 1.8 and a width of 0.3). Significant differences between the groups were assessed using Student’s t -test. A permutation-based FDR approach was applied to correct for multiple testing. Western blot Cells were scraped from tissue culture plates by cell scraping and lysed in protein cell lysis buffer (50 Mm Hepes pH 7.4, 150 Mm NaCl, 1 mM EDTA, 1% Triton X-100) supplemented with 2 mM sodium orthovanadate, 1 mM PMSF and protease inhibitor mix). Lysates were homogenized through syringe needle (27 G) followed by centrifugation at 8000 × g for 5 min at 4 °C and then supernatants were collected (with the exception of the western blots showed in Fig. 6d , where we loaded whole cell lysates without centrifugation because proteasome inhibition induced high concentration of Q100-HTT (OE) in the pellet fraction). Protein concentrations were determined with a standard BCA protein assay (ThermoFisher Scientific). Approximately 30 μg of total protein was separated by SDS–PAGE, transferred to polyvinylidene difluoride membranes (Millipore) and subjected to immunoblotting. Western blot analysis was performed with anti-UBR5 (Stem Cell Technologies, #60094, 1:1000), anti-UBE3A (Cell Signaling, ab#7526, 1:1000), anti-RNF181 (ThermoFisher Scientific, PA5-31008, 1:2000), anti-UBR7 (ThermoFisher Scientific, PA5-31559, 1:1000), anti-ATXN3 (Merck, MAB5360, 1:500), anti-FUS (Abcam, #154141, 1:1000), anti-GFP (Immunokontakt, 210-PS-1GF, 1:5000), anti-polyubiquitinylated conjugates (Enzo, PW8805-0500, 1:1000), anti-p62 (Progen, GP62-C, 1:1000), anti-LC3 (Sigma, L7543, 1:1000), anti-DARPP32 (Abcam, #40801, 1:1000) and anti-β-actin (Abcam, #8226, 1:5000). To detect HTT protein, we used anti-HTT (Cell Signaling, ab#5656, 1:1000), a monoclonal antibody produced by immunizing animals with a synthetic peptide corresponding to residues surrounding Pro1220 of human HTT protein. To detect mutant HTT, we used anti-polyQ-expansion diseases marker (Millipore, MAB1574, 1:1000), a monoclonal antibody raised against TATA-binding protein that recognizes peptides overlapping the polyQ stretch of this protein. This antibody also recognizes polyQ-containing proteins such as HTT and ATXN3 with the remarkable property of detecting much better the polyQ-expanded pathological proteins than the wild-type proteins 12 , 31 . Uncropped versions of all important western blots are presented in Supplementary Fig. 41 . Immunocytochemistry Cells were fixed with paraformaldehyde (4% in PBS) for 20 min, followed by permeabilization (0.2% Triton X-100 in PBS for 10 min) and blocking (3% bovine serum albumin in 0.2% Triton X-100 in PBS for 10 min). Human iPSCs/NPCs were incubated in primary antibody for 2 h at room temperature (Mouse anti-polyQ (Millipore, MAB1574, 1:50), Mouse anti-OCT4 (Stem Cell Technologies, #60093, 1:200), Rabbit anti-PAX6 (Stem Cell Technologies, #60094, 1:300), Mouse anti-FUS (Abcam, #154141, 1:500), Rabbit anti-G3BP1 (MBL, #RN048PW, 1:500), Rabbit anti-DARPP32 (Abcam, #40801, 1:50), Rabbit anti-GABA (Sigma, #A2052, 1:100) and Chicken anti-MAP2 (Abcam, #5392, 1:500). Then, cells were washed with 0.2% Triton-X/PBS and incubated with secondary antibody (Alexa Fluor 488 Goat anti-Mouse (ThermoFisher Scientific, #A-11029, 1:500), Alexa Fluor 568F(ab’)2 Fragment of Goat Anti-Rabbit IgG (H+L) (ThermoFisher Scientific, #A-21069, 1:500)), Alexa Fluor 647 Donkey anti-Chicken (Jackson ImmunoResearch, #A-703-605-155, 1:500) and Hoechst 33342 (Life Technologies, #1656104) for 1 h at room temperature. PBS and distilled water wash were followed before the cover slips were mounted on Mowiol (Sigma, #324590). Aggresome detection Pluripotent stem cells were cultured at 37 °C or under heat stress conditions at 42 °C for 4 h. Aggresome formation was detected by PROTEOSTAT Aggresome Detection Kit (ENZO, #ENZ-51035), following the manufacturer’s instructions. Briefly, the cells were fixed in 4% formaldehyde solution and permeabilized with 0.5% Triton X-100, 3 mM EDTA, pH 8.0, in 1× assay buffer from the detection kit. Then, aggresomes were stained with PROTEOSTAT Aggresome Detection reagent, including Hoechst 33342 for nuclear staining. Imaging of aggresomes was performed using a standard rhodamine filter set. Propidium iodide staining Cells were incubated with 2 µg ml −1 propidium iodide (Sigma, #81845) and Hoechst 33342 (Life Technologies, #1656104) for 1.5 h. Then, cover slips were mounted using FluorSave Reagent (Calbiochem, #345789). TUNEL staining TUNEL (terminal deoxynucleotidyl transferase dUTP nick end labeling) measurement was performed with the In situ BrdU-Red DNA Fragmentation (TUNEL) Assay Kit (Abcam, #66110) according to the manufacturer’s protocol. Briefly, cells were washed twice and stained with TdT enzyme and BrdUTP for 1 h at 37 °C. A second incubation with an Anti-BrdU-Red antibody was performed at room temperature for 30 min. Nuclei were stained with Hoechst 33342 (Life Technologies, #1656104). RNA isolation and quantitative reverse transcription-PCR Total RNA was extracted using RNAbee (Tel-Test Inc.). Complementary DNA (cDNA) was generated using qScript Flex cDNA synthesis kit (Quantabio). SybrGreen real-time quantitative PCR (qPCR) experiments were performed with a 1:20 dilution of cDNA using a CFC384 Real-Time System (Bio-Rad) following the manufacturer’s instructions. Data were analyzed with the comparative 2ΔΔ C t method using the geometric mean of ACTB and GAPDH as housekeeping genes. See Supplementary Table 5 for details about the primers used for this assay. 26S proteasome fluorogenic peptidase assays Cells were collected in proteasome activity assay buffer (50 mM Tris-HCl, pH 7.5, 250 mM sucrose, 5 mM MgCl 2 , 0.5 mM EDTA, 2 mM ATP and 1 mM DTT) and lysed by passing 10 times through a 27 G needle attached to a 1 ml syringe needle. Lysates were centrifuged at 10,000 × g for 10 min at 4 °C. Then, 25 μg of total protein of cell lysates were transferred to a 96-well microtiter plate (BD Falcon) and incubated with the fluorogenic proteasome substrate. To measure the chymotrypsin-like activity of the proteasome we used either Z-Gly-Gly-Leu-AMC (Enzo) or Suc-Leu-Leu-Val-Tyr-AMC (Enzo). We used Z-Leu-Leu-Glu-AMC (Enzo) to measure the caspase-like activity of the proteasome, and Ac-Arg-Leu-Arg-AMC (Enzo) for the proteasome trypsin-like activity. Fluorescence (380 nm excitation, 460 nm emission) was monitored on a microplate fluorometer (EnSpire, Perkin Elmer) every 5 min for 1 h at 37 °C. Quantitative proteomics analysis of E3 enzymes For the characterization of protein expression differences in E3 enzymes comparing H9 hESCs with their neuronal counterparts, we analyzed quantitative proteomics data 35 available via ProteomeXchange with identifier PXD007738. Then, we intersected the annotated human E3 network from KEGG (Kyoto Encyclopedia of Genes and Genomes) database 69 with this proteomics dataset. Statistical comparisons were made by Student’s t -test. FDR-adjusted P value ( q value) was calculated using the Benjamini–Hochberg procedure. C. elegans strains and maintenance C. elegans strains were maintained at 20 °C using standard methods 17 . AM23 ( rmIs298[pF25B3.3::Q19::CFP] ) and AM716 ( rmIs284[pF25B3.3::Q67::YFP] ) strains were a gift from R. I. Morimoto. For the generation of the strains DVG144 ( rmIs298[F25B3.3p::Q19::CFP ]; rrf-3(pk1426) ) and DVG145 ( rmIs284 [ pF25B3.3::Q67::YFP ]; rrf-3( pk1426 )), NL2099 strain ( rrf-3(pk1426) ) was crossed to AM23 and AM716, respectively. Screening of rrf-3 ( pk1426 ) worms was done by PCR using the forward primer F: GTTTTGACGCCAAACGGTGA and two reverse primers: TGCAGCATGTCCAGACACAA, which outflanks the deleted region in rrf-3(pk1426) , and CCATTCTGTGCACGTTTCCA, which binds inside the deletion. RNAi experiments in C. elegans RNAi-treated strains were fed Escherichia coli (HT115) containing an empty control vector (L4440) or expressing double-stranded ubr-5 RNAi. ubr-5 RNAi construct was obtained from the Ahringer RNAi library and sequence verified. Imaging of polyQ aggregates in C. elegans AM716 and AM23 strains were grown at 20 °C until L4 stage and then grown at 25 °C on E. coli (HT115) containing either empty control vector or ubr-5 RNAi until day 3. Day 3 adult worms were immobilized following the protocol described in ref. 70 . Briefly, worms were placed on 5% agarose-containing pads on a suspension of polystyrene beads (Polyscience, 2.5% by volume). For imaging, we used a Meta 710 confocal microscope (Zeiss) at the CECAD Imaging Facility. Quantitative fluorescence recovery after photobleaching Q67-YFP day 3 adult worms were immobilized on 5% agarose pads using 0.1% sodium azide. After three prebleaching scans, a constant region of interest (ROI) (44.29 × 30.09 µm) was bleached for 20 scans (860 ms per iteration) in a SP8 Confocal Microscope (Leica). Directly after bleaching, the fluorescence recovery was sampled once every 2 s for 90 times. Average fluorescence intensities within ROIs were measured under the same condition for empty vector and ubr-5 RNAi-treated worms using ImageJ software. The half-life of fluorescence recovery ( t 1/2) was determined by curve fitting of experimental data using the following exponential equation: I ( t ) = a (1 − e^(tau × t ). Motility assay Animals were grown on E. coli (OP50) bacteria at 20 °C until L4 stage and then transferred to 25 °C and fed with E. coli (HT115) bacteria containing empty control vector or ubr-5 RNAi for the rest of the experiment. At day 3 of adulthood, worms were transferred to a drop of M9 buffer and after 30 s of adaptation the number of body bends was counted for 30 s. A body bend was defined as change in direction of the bend at the mid-body 58 . Filter trap AM716, AM23, DVG144 and DVG145 C. elegans strains were grown at 20 °C until L4 stage and then grown at 25 °C on E. coli (HT115) bacteria containing either empty control vector or ubr-5 RNAi for the rest of the experiment. Day 3 adult worms were collected with M9 buffer and worm pellets were frozen with liquid N2. Frozen worm pellets were thawed on ice and worm extracts were generated by glass bead disruption on ice in non-denaturing lysis buffer (50 mM Hepes pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100) supplemented with EDTA-free protease inhibitor cocktail (Roche). Worm and cellular debris was removed with 8000 × g spin for 5 min. Approximately 100 µg of protein extract was supplemented with SDS at a final concentration of 0.5% and loaded onto a cellulose acetate membrane assembled in a slot blot apparatus (Bio-Rad). The membrane was washed with 0.2% SDS and retained Q67-GFP was assessed by immunoblotting for green fluorescent protein (GFP; ImmunoKontakt, 210-PS-1GFP, 1:5000). Extracts were also analyzed by SDS–PAGE with GFP antibody to determine protein expression levels. iPSCs were collected in non-denaturing lysis buffer supplemented with EDTA-free protease inhibitor cocktail and lysed by passing 10 times through a 27 G needle attached to a 1 ml syringe. Then, we followed the filter trap protocol described above. Cell pellet lysates were loaded after solubilization with 2% SDS. The membrane was washed with 0.2% SDS and retained polyQ proteins were assessed by immunoblotting for anti-polyQ-expansion diseases marker antibody (Millipore, MAB1574, 1:5000). Extracts were also analyzed by SDS–PAGE to determine HTT protein expression levels. Data availability Proteomics data of HTT ubiquitination sites have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD009803 . To define E3 enzymes significantly increased in hESC, we analyzed available quantitative proteomics data from PRIDE: PXD007738 35 . All the other data are also available from the corresponding author upon reasonable request. Change history 30 August 2018 This Article was originally published without the accompanying Peer Review File. This file is now available in the HTML version of the Article; the PDF was correct from the time of publication. 17 February 2020 An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Neuroscientist Dr. David Vilchez and his team at CECAD, the University of Cologne's Cluster of Excellence for Aging Research, have made an important step toward understanding the mechanisms that cause the neurodegenerative disorder Huntington's disease. Specifically, they identified a system blocking the accumulation of toxin protein aggregates, which are responsible for neurodegeneration. The results have now been published in the journal Nature Communications. Huntington's disease is a neurodegenerative disorder that results in the death of brain cells, leading to uncontrolled body movement, loss of speech, and psychosis. Mutations in the huntingtin gene cause the disease, resulting in the toxic aggregation of the huntingtin protein. The accumulation of these aggregates causes neurodegeneration and usually leads to the patient's death within 20 years after the onset of the disease. To examine the mechanisms underlying Huntington's disease, Vilchez and his team used so-called induced pluripotent stem cells (iPSC) from Huntington's disease patients, which are able to differentiate into any cell type, such as neurons. Induced pluripotent stem cells derived from patients with Huntington's disease exhibit a striking ability to avoid the accumulation of toxic protein aggregates, a hallmark of the disease. Even though iPSCs express the mutant gene responsible for Huntington's disease, no aggregates were found. The researchers identified a protein called UBR5 as a protective mechanism for the cells, promoting the degradation of mutant huntingtin. These findings can contribute to a better understanding of Huntington's disease and could be a stepping stone to developing further treatment in patients. The researchers screened immortal iPSCs from patients and derived neurons for differences in their ability to avoid mutant huntingtin aggregation. They found that huntingtin can be degraded by the cellular disposal system known as the proteasome. However, this system is defective in the neurons, which leads to the aberrant aggregation of the mutant huntingtin protein. Vilchez and his team found that UBR5 is increased in pluripotent stem cells to accelerate the degradation of huntingtin in the cells. To examine the role of UBR5 in the regulation of the mutant huntingtin gene (HTT), they reduced the levels of UBR5 and could immediately see an accumulation of aggregated proteins in iPSCs. "This was striking to see," says Vilchez. "From nothing, the cells went to huge amounts of aggregates." The authors went a step further and examined whether UBR5 also controls mutant huntingtin aggregation in Huntington's disease organismal models. They found that dysregulation of UBR5 results in a massive increase in the aggregation and neurotoxic effects in neurons. On the other hand, promoting UBR5 activity blocks mutant huntingtin aggregation in the Huntington's disease models. To test the specificity of the results, the researchers also kept an eye on other illnesses. "We also checked the mechanism in other neurodegenerative diseases like amyotrophic lateral sclerosis," says Seda Koyuncu, a doctoral student working in Vilchez's lab and a main author of the publication. "Our result is very specific to Huntington's disease," adds Dr. Isabel Saez, another main author working with Vilchez at CECAD. Even though the results could be important for treatment and drug development, there is no therapy yet. "It's not like you discover something new and then there is a cure, it's more difficult—but in some years, there might be a therapy," Saez says. Until then, more research needs to be done.

10.1038/s41467-018-05320-3

Medicine

Curbing Candida: The cells that keep fungal infections at bay

Jan Dobeš et al, Extrathymic expression of Aire controls the induction of effective TH17 cell-mediated immune response to Candida albicans, Nature Immunology (2022). DOI: 10.1038/s41590-022-01247-6 Journal information: Nature Immunology

https://dx.doi.org/10.1038/s41590-022-01247-6

https://medicalxpress.com/news/2022-07-curbing-candida-cells-fungal-infections.html

Abstract Patients with loss of function in the gene encoding the master regulator of central tolerance AIRE suffer from a devastating disorder called autoimmune polyendocrine syndrome type 1 (APS-1), characterized by a spectrum of autoimmune diseases and severe mucocutaneous candidiasis. Although the key mechanisms underlying the development of autoimmunity in patients with APS-1 are well established, the underlying cause of the increased susceptibility to Candida albicans infection remains less understood. Here, we show that Aire + MHCII + type 3 innate lymphoid cells (ILC3s) could sense, internalize and present C. albicans and had a critical role in the induction of Candida -specific T helper 17 (T H 17) cell clones. Extrathymic Rorc -Cre-mediated deletion of Aire resulted in impaired generation of Candida -specific T H 17 cells and subsequent overgrowth of C. albicans in the mucosal tissues. Collectively, our observations identify a previously unrecognized regulatory mechanism for effective defense responses against fungal infections. Main The transcriptional regulator Aire has an essential role in the induction of self-tolerant T cells during thymic development by controlling the expression of thousands of self-antigen genes in medullary thymic epithelial cells (TECs) 1 , 2 . Presentation of self-antigens by medullary TECs is essential for the deletion of self-reactive T cell clones 3 or their conversion into regulatory T cells (T reg cells) 4 , 5 . Aire deficiency results in impaired T reg cell generation and escape of self-reactive T cells into the periphery, leading to breakdown of immunological tolerance to various parenchymal tissues 1 . Patients with AIRE deficiency develop a rare genetic disorder called autoimmune polyendocrine syndrome type 1 (APS-1; also known as autoimmune polyendocrinopathy candidiasis ectodermal dystrophy; OMIM: 240300 ), which is characterized by autoimmune pathologies such as hypoparathyroidism and primary adrenocortical insufficiency (Addison’s disease), with additional autoimmune disorders such as hypothyroidism, type 1 diabetes, premature ovarian failure, pernicious anemia, vitiligo, alopecia, keratitis or intestinal malabsorption occurring with lower frequency 6 , 7 . In addition, the vast majority (75–100%) of patients with APS-1 develop chronic mucocutaneous candidiasis, mainly characterized by C. albicans overgrowth in the oral cavity, esophagus and nails as early as 1 year of age (median 5 years) 8 , 9 . Because candidiasis is a common complication in people born with a loss-of-function mutation in various genes linked to the CD4 + T H 17 cell-mediated response (e.g., RORC 10 , IL17F 11 , STAT3 (ref. 12 ), CLEC7A 13 , CARD9 (ref. 14 ) or gain-of-function in S TAT 1 (ref. 15 )), T H 17 cells are assumed to have an indispensable role in long-term protection against C. albicans infection 16 . Patients with APS-1 have been reported to develop autoantibodies against the T H 17 effector cytokines interleukin-17A (IL-17A), IL17-F and/or IL-22 (refs. 17 , 18 ), suggesting that the increased susceptibility to C. albicans might also be caused by an autoimmune-mediated mechanism. However, a considerable fraction of patients with APS-1 with very low or no IL-17 or IL-22 autoantibodies still develop chronic mucocutaneous candidiasis 17 , 18 , indicating the correlation between the IL-17- and/or IL-22-specific autoantibodies and candidiasis is incomplete and that additional mechanisms may be involved. In addition to its well-established expression pattern in the thymus, Aire was reported to be expressed in a rare population of cells residing in the lymph nodes (LNs) and bearing the hallmarks of a subset of MHCII + ILC3s 19 . Considering that these Aire + MHCII + ILC3s (hereafter Aire + ILC3s) also express the molecular machinery for antigen presentation and T cell activation (MHCII, CD80, CD86 and ICOSL) 19 , we investigated whether extrathymic expression of Aire in this ILC3 subset may contribute to the adaptive immune response to C. albicans . Here, we show that Aire + ILC3s sensed and internalized C. albicans and effectively presented C. albicans epitopes on their MHCII. Moreover, extrathymic (but not thymus-specific) ablation of Aire impaired the expansion of the Candida -specific T H 17 cell pool and resulted in overgrowth of C. albicans at various mucosal surfaces. Results Aire + ILC3s express receptors involved in C. albicans sensing To test whether peripheral expression of Aire in ILC3s was required for the induction or modulation of an effective immune response to C. albicans infection, we determined whether ILC3s in general and Aire + ILC3s in particular expressed pattern recognition receptors (PRRs) for C. albicans . We used Rorc -Cre + flox-STOP-flox-tdTomato reporter mice that were crossed with Aire -GFP + transgenic reporter mice (hereafter Rorc Tomato Aire GFP mice), in which the tdTomato reporter is expressed in all cells with a history of Rorc expression and the green fluorescent protein (GFP) is expressed in cells with an active Aire locus. ILC3s isolated from the popliteal LN (pLN) by flow cytometry-based sorting as lineage − (CD3, T cell antigen receptor β (TCR-β), CD45RB, CD19, Gr1, F4/80, CD11b and CD11c) Rorc -tdTomato + cells were divided according to their MHCII and Aire GFP expression into MHCII − Aire GFP − conventional ILC3s (hereafter cILC3s), MHCII + Aire GFP − ILC3s (hereafter MHCII + ILC3s) and MHCII + Aire GFP + ILC3s (hereafter Aire + ILC3s) (Fig. 1a ). In addition, to compare the molecular characteristics of these three ILC3 subsets with conventional antigen-presenting cells (APCs) that can sense C. albicans 20 , we isolated CD11c + CD11b + MHCII + dendritic cells (DCs) from the Rorc Tomato Aire GFP mice. All sorted populations were analyzed by bulk RNA sequencing (RNA-seq). Clustering analysis highlighted relatively large transcriptional similarity between all cell subsets expressing MHCII (that is MHCII + ILC3, Aire + ILC3s and DCs), and in particular between the Aire + ILC3s and MHCII + ILC3s (Fig. 1b ). All three ILC3 subsets were, as expected, characterized by low expression of genes specific for hematopoietic stem cells ( Cd34 and Slamf1 ), T cells ( Cd4 , Cd8 and Foxp3 ), B cells ( Cd19 and Cd79a ), granulocytes ( Ly6c1, Ly6g5b and Siglecf ), macrophages and monocytes ( Itgam and Siglech ) and DCs ( Itgax, Csf1r and Cd207 ) and had high expression of genes associated with ILC3s ( Rorc , Il1r1 , Il7r , Kit and Ccr6 ) (Fig. 1c ). Unlike cILC3s, MHCII + ILC3 and the Aire + ILC3s had high expression of genes encoding MHCII ( H2-Aa and H2-Ab1 ) (Fig. 1d ), whereas Aire + ILC3s also had high expression of Aire and genes encoding costimulatory molecules ( Cd80 and Cd86 ) at levels comparable to DCs (Fig. 1d ). The Aire + ILC3s also expressed transcripts for several key PRRs implicated in sensing C. albicans , such as T l r 2 (refs. 21 , 22 ), dectin-1 ( Clec7a ) 23 , 24 or galectin-3 ( Lgals3 ) 25 , 26 (Fig. 1e ) and signaling molecules downstream of these receptors ( Myd88 and Syk ) (Fig. 1f ). The protein expression of the C. albicans -sensing receptors including dectin-1, galectin-3 and Tlr2 on Aire + ILC3s was confirmed by flow cytometry in steady-state conditions and was comparable to the expression on these receptors on DCs (Fig. 1g ). These results collectively suggested that Aire + ILC3s expressed genes implicated in antigen presentation, T cell activation 19 and C. albicans sensing. Fig. 1: Aire + ILC3s express C. albicans -sensing receptors. a , Fluorescence-activated cell sorting (FACS) strategy of cILC3s, MHCII + ILC3s and Aire + ILC3s subsets isolated from the pLNs of Rorc tdTomato Aire GFP reporter mice. Representative plots are shown. b , Heatmap of Pearson correlation according to gene expression values between individual samples as in panel a . c – f , RNA-seq-based heatmap of expression of lineage-specific genes ( c ), genes associated with MHCII presentation or costimulation ( d ), genes encoding receptors implicated in sensing and/or phagocytosis of C. albicans ( e ) and genes encoding signaling molecules downstream of C. albicans -sensing receptors ( f ) in cILC3s, MHCII + ILC3s, Aire + ILC3s and MHCII + CD11c + DCs isolated from pLNs of Rorc tdTomato Aire GFP at steady state. g , Flow cytometry of Rorγt + cILC3s, Rorγt + MHCII + ILC3s, Rorγt + MHCII + Aire + ILC3s and MHCII + CD11c + DCs isolated from pLNs of wild-type animals under steady-state conditions. Gray-filled histogram indicates staining in control dectin-1 or galectin-3-deficient mice or isotype control for CD80, CD86 and TLR2 (representative figure; n ≥ 3). Lin − , lineage negative; MACS, magnetic-activated cell sorting; SSC-A, side scatter area. Full size image C. albicans induces transcriptional changes in ILC3 subsets To test whether C. albicans could induce the activation of ILC3s in general, and Aire + ILC3s in particular, we performed bulk RNA-seq of Aire + ILC3s, MHCII + ILC3s and cILC3s sorted from pLNs of Rorc Tomato Aire GFP mice that were intravenously injected at 1-day intervals for 3 days with heat-killed C. albicans (HKCA) or PBS and analyzed 24 h after the last injection. All three ILC3 subsets, and particularly the Aire + ILC3s subset isolated from HKCA-challenged mice, showed transcriptional changes compared to PBS-challenged controls (Fig. 2a–c ). cILC3s from HKCA-challenged mice significantly upregulated 149 genes (fold change (FC) ≥ 2.0, adjusted P value≤0.05) compared to PBS-treated controls (Fig. 2a ), with genes encoding several PRRs implicated in C. albicans -sensing ( Lgals3 , Clec7a , Clec4d , Clec4e and Cd209b ) or molecules involved in antigen presentation ( H2-Aa , H2-Ab1 , H2-Eb1 and Cd74 ), costimulation ( Cd86 ) or induction of proinflammatory response ( Il1a , Il1b and 1l18 ) among the most upregulated (Fig. 2a and Extended Data Fig. 1a ). This finding suggested that cILC3s may upregulate Candida -sensing receptors and/or MHCII molecules in response to the inflammatory conditions engendered by C. albicans stimulation. MHCII + ILC3s from the HKCA-challenged mice significantly induced the expression of 641 genes, including Aire , which was the most significantly upregulated gene ( P ≤ 1 × 10 −23 ; FC > 7) compared to their PBS-treated counterparts and further upregulated the expression of Clec7a (dectin-1), H2-Ab1 (MHCII) and Cd86 while downregulating the expression of several key ILC3-specific genes, such as Il17a , Il17f and Il12rb1 (Fig. 2b and Extended Data Fig. 1b ). These observations suggested that the HKCA challenge potentiated the antigen-presentation capacity of the MHCII + ILC3 subset while limiting their effector functions as IL-17-producing cells (Fig. 2b and Extended Data Fig. 1b ). Moreover, FC/FC analysis comparing MHCII + ILC3s from HKCA-stimulated versus PBS-stimulated mice and Aire + ILC3 versus MHCII + ILC3s from PBS-treated mice showed that MHCII + ILC3s from HKCA-treated mice acquired a transcriptional signature similar to that of Aire + ILC3s (Fig. 2d ). This finding suggested that exposure of MHCII + ILC3s to HKCA could likely initiate the acquisition of a transcriptional signature characteristic of Aire + ILC3s (Fig. 2d,e and Extended Data Fig. 1d ). Fig. 2: C. albicans induces transcriptional changes in ILC3 subsets. a – c , Volcano plots of bulk RNA-seq analyses showing differential gene expression in nonstimulated versus HKCA-stimulated cILC3 ( a ), MHCII + ILC3 ( b ) and Aire + ILC3 ( c ) subpopulations isolated from the pLNs of Rorc tdTomato Aire GFP mice intravenously injected with HKCA or PBS for three consecutive days, using the same gating strategy shown in Fig. 1a . Dashed lines indicate the cutoff for FC = 2.0 and P = 0.05. Wald test was used to calculate the P value. Based on their function, the selected genes are highlighted in blue (antigen presentation and costimulation), red (cytokine and chemokine signaling), green ( Candida -sensing receptors) or purple ( Aire ). d , FC:FC graph of RNA-seq data showing differential gene expression in Aire + ILC3s in contrast to stimulated or nonstimulated MHCII + ILC3s. Comparison of FC of Aire + ILC3s versus MHCII + ILC3s ( x axis) and nonstimulated versus HKCA-stimulated MHCII + ILC3s ( y axis) isolated by FACS, using the same gating strategy shown in Fig. 1a . Dashed lines indicate the FC = 2.0 cutoff. e , RNA-seq-based heatmap showing the level of expression of selected effector genes in ILC3s, MHCII + ILC3s and Aire + ILC3s with or without HKCA stimulation. Data are plotted as the z -score calculated for a particular row. Data were derived from three independent biological replicates for each population ( a – e ). Full size image Finally, Aire + ILC3s from HKCA-stimulated mice significantly upregulated 777 genes compared to PBS-treated mice, including cytokines or cytokine receptors ( Il6 , Bmp2 , Il7r , Il23r and Il2rb ), chemokines ( Ccl2 ), C. albicans -sensing receptors ( Clec7a ), cell adhesion molecules ( Vcam1 and Cadm1 ), costimulatory molecules ( Cd86 ) and enzymes involved in proinflammatory response ( Ptgs2 ) (Fig. 2c ). Moreover, the HKCA challenge led to a significant upregulation of Il6 , which, along with TGF-β, induces the polarization of naive T cells to the T H 17 subset and their clonal expansion 27 , 28 , 29 (Fig. 2c and Extended Data Fig. 1c ). These results indicated that HKCA induced distinct transcriptional programs in the three ILC3 subsets analyzed, including the upregulation of genes involved in antigen presentation, costimulation and Candida -sensing in cILC3s and MHCII + ILC3s, the upregulation of Aire in MHCII + ILC3s and the upregulation of Il6 in the Aire + ILC3s, suggesting that Aire + ILC3s could play a role in the induction or regulation of the adaptive immune response to C. albicans . Aire + ILC3s internalize C. albicans for antigen presentation Next, we investigated whether Aire + ILC3s could uptake C. albicans by endocytosis. To this end, we performed imaging flow cytometry (Imagestream) analysis of MACS-enriched Lin − (CD3 − , TCR-β − , CD45RB − , CD19 − , Gr1 − , F4/80 − , CD11b − , CD11c − ) cell fraction or CD11c + DCs isolated from pLNs and incubated ex vivo with cell proliferation dye (CPD)-stained HKCA (HKCA-CPD). After 30-min coincubation, most Aire + ILC3s were physically associated with HKCA-CPD (Fig. 3a ), whether through cell–cell interactions or internalization of the HKCA-CPD into their cytoplasm. After 90-min coincubation, most Aire + ILC3s had intracellular CPD signals (Fig. 3b ), suggesting ingestion of HKCA-CPD. Vesicles containing internalized fungi fused with the lysosomes in both Aire + ILC3s and DCs (Extended Data Fig. 2a ), suggesting that Aire + ILC3s had endocytic capacity for C. albicans . Moreover, while the capacity of Aire + ILC3s to endocytose HKCA-CPD was comparable to CD11c + or CD11b + mononuclear phagocytes (MNPs) (Extended Data Fig. 2b ), the endocytic capacity of both MHCII + ILC3s and cILC3s was substantially lower than that of Aire + ILC3s (Extended Data Fig. 2b ). Moreover, similarly to DCs, the endocytic capacity of Aire + ILC3s cells was partially dependent on the expression of dectin-1 and galectin-3 and was reduced at 4 °C (Extended Data Fig. 2c,d ). Fig. 3: Aire + ILC3s internalize C. albicans for antigen presentation. a , b , Imaging flow cytometry showing the physical interaction between HKCA and Aire + ILC3s isolated from pLNs of wild-type mice and incubated ex vivo with CPD-stained HKCA for 30 min ( a ) or 90 min ( b ). Samples were stained for Aire and DAPI ( a , b ) and MHCII ( b ). Shown are representative images of one out of five independent experiments. Scale bars, 8 µm. c – f , In vitro reporter assay measuring the capacity of B cells, Ly6C + MNPs, CD11b + MNPs, CD11c + MNPs and Aire + ILC3s to endocytose HKCA transgenically expressing ovalbumin (HKCA-OVA) either in vitro ( c ) or in vivo ( d – f ) and subsequently to present HKCA-derived OVA antigens on their MHCII molecules to DO11.10 TCR NFAT-GFP reporter cell line that was coincubated with the sorted populations for 17 h. The corresponding APC populations were FACS sorted according to gating strategy shown in Extended Data Fig. 3a,b . Antigen-presentation capacity was measured as a frequency of NFAT-GFP + reporters in the specific sample. Representative data set out of three independent experiments are shown ( n = 5 for each experiment), mean ± standard deviation (s.d.) depicted. d – f , Dot plot of B cells, Ly6C + MNPs, CD11b + MNPs, CD11c + MNPs and Aire + ILC3s depicting APC capacity at 12 h ( d ), 24 h ( e ) or 72 h ( f ) after HKCA-OVA i.v. injection into wild-type mice. g , Two-photon microcopy of explanted pLNs showing interaction between Aire- GFP + cells and Candida -specific T cells from Aire GFP reporter mice adoptively transferred with OT-II tdTomato T cells stimulated by either HKCA or HKCA-OVA. Representative figures from two experiments are shown. Full size image To assess the capacity of Aire + ILC3s to present Candida -derived antigens under in vitro conditions, we FACS-sorted pLN-resident Aire + ILC3s and other types of APCs (B cells, Ly6C + MNPs, CD11b + MNPs, CD11c + MNPs) from BALB/c wild-type mice using a gating strategy described in Extended Data Fig. 2e ), stimulated them with HKCA or transgenic HKCA expressing OVA (HKCA-OVA) and incubated them for 17 h with an NFAT-GFP DO11.10 TCR reporter cell line 4 , which turns on GFP expression in response to TCR stimulation by OVA 323–339 peptide presented in the context MHCII molecules. Based on this in vitro assay, CD11b + MNPs were the most efficient APCs (Fig. 3c ), whereas CD19 + B cells and Ly6C + MNPs showed negligible antigen-presentation capacity (Fig. 3c ). The Aire + ILC3s showed substantial capacity to present HKCA-derived antigens, comparable to that of CD11c + MNPs and approximately fivefold lower than that of CD11b + MNPs (Fig. 3c and Extended Data Fig. 2f ). To assess the capacity of Aire + ILC3s to present Candida -derived antigens under more physiological conditions, we sorted Aire + ILC3s and other types of APCs (B cells, Ly6C + MNPs, CD11b + MNPs and CD11c + MNPs) from the pLNs of BALB/c wild-type mice injected with either HKCA-OVA or HKCA at 12, 24 or 72 h before sorting (Extended Data Fig. 2g ). The sorted cells were then coincubated with the NFAT-GFP DO11.10 TCR reporter cells line for 17 h, and their antigen-presentation capacity was determined by the frequency of cells expressing the NFAT-GFP reporter. CD11b + MNPs isolated 12 h after the HKCA-OVA challenge showed the highest antigen-presentation capacity, as measured by the frequency of NFAT-GFP + cells (~1.7%) (Fig. 3d and Extended Data Fig. 2h ). Their capacity to induce the NFAT-GFP signal declined at 24 h (Fig. 3e and Extended Data Fig. 3d ) and largely diminished at 72 h after the HKCA challenge (Fig. 3f and Extended Data Fig. 2h ). Both Aire + ILC3s and CD11c + MNPs induced the highest frequency of NFAT-GFP + cells (1.2% and 0.7%, respectively) at 72 h after the HKCA-OVA challenge (Fig. 3f and Extended Data Fig. 2h ). These data suggested that although CD11b + MNPs dominated the early antigen-presentation phase, Aire + ILC3s were more efficient at presenting Candida -derived antigens at later time points. To examine the capacity of Aire + ILC3s to present C. albicans -derived antigens to T cells in the pLNs, we performed ex vivo two-photon imaging in Aire GFP reporter mice that were adoptively transferred with OT-II T cells bearing OVA-specific TCR and endogenously expressing tdTomato fluorescence protein (OT-II tdTomato ). The mice were intravenously stimulated either with HKCA-OVA or HKCA as control, and two-photon excitation microscopy on explanted pLNs was used to visualize the interactions between Aire GFP + cells and OVA-specific OT-II tdTomato T cells (Fig. 3g ). Although we did not observe any physical interaction between Aire GFP + cells and OT-II tdTomato T cells in pLNs isolated from mice stimulated with HKCA, we observed numerous such interactions in pLNs isolated from mice stimulated by HKCA-OVA (Fig. 3g ). These observations suggested that Aire + ILC3s might present Candida -derived antigens to T cells and that they could exert this function at later time points compared to conventional APCs. Aire + ILC3s are required for induction of Candida -specific T H 17 cells To determine whether Aire + ILC3s, and, in particular, the expression of Aire in these cells regulated the adaptive immune response to C. albicans , we aimed to conditionally inactivate Aire in Aire + ILC3s. To this end, we generated Rorc -Cre + Aire fl/fl (hereafter ILC3 ΔAire ) 30 , 31 , in which Aire is ablated in all cells that either actively express Rorc or have Rorc expression history. Because the expression of Aire in the pLNs of wild-type mice was restricted to Lin − Rorγt + MHCII + cells 19 (Extended Data Fig. 3a,b ), this strategy allowed us to specifically inactivate Aire expression in Aire + ILC3s. As controls, we generated Foxn1 -Cre Aire fl/fl mice (hereafter TEC ΔAire ), in which Aire is deleted in TECs. Using flow cytometry, we could not detect any Aire protein expression in samples prepared from the pLN of ILC3 ΔAire mice, whereas Aire protein was detected in the ILC3 ΔAire thymus samples (Fig. 4a ). Conversely, Aire protein was detected in the pLNs, but not the thymus, of TEC ΔAire mice (Fig. 4a ). We next intravenously transferred naive OT-II T cells from CD45.1-expressing GFP under the Rorc -promoter (OT-II- Rorc GFP ) equally mixed with control polyclonal T cells from CD45.1/CD45.2 wild-type mice into CD45.2 + wild-type or ILC3 ΔAire mice, followed by stimulation of recipient mice with HKCA-OVA or HKCA as control in consecutive 2-day intervals and analyzed 2 or 14 days after the first injection. Although no significant differences in OT-II- Rorc GFP T cell cellularity and/or proliferation were observed at day 2 after transfer between the HKCA-OVA or HKCA-stimulated wild-type versus ILC3 ΔAire mice (Extended Data Fig. 3c,d ), the OT-II- Rorc GFP T cells transferred in the HKCA-OVA-stimulated wild-type mice showed a ~2.5-fold proliferative expansion at day 14 after transfer compared to the HKCA-OVA-stimulated ILC3 ΔAire mice (Fig. 4b,c ). The frequency of CD45.1 + OT-II T cells in the HKCA-OVA-treated ILC3 ΔAire mice was decreased by more than eightfold compared to HKCA-OVA-treated wild-type controls (Fig. 4b,c ). Moreover, approximately 10% of the CD45.1 + OT-II T cells transferred in the HKCA-OVA-stimulated wild-type mice differentiated into Rorc GFP + T H 17 cells (Fig. 4d ). In contrast, the frequency (Fig. 4d ) and number (Fig. 4e ) of CD45.1 + Rorc GFP + OT-II T H 17 cells in HKCA-OVA-treated ILC3 ΔAire was reduced by ~20 fold and ~100-fold, respectively, compared to HKCA-OVA-treated wild-type controls at day 14 after transfer, suggesting that Aire expression in ILC3s was required for the induction of effector T H 17 cells in response to C. albicans stimulation. Fig. 4: Aire + ILC3s are essential for the generation of C. albicans -specific T H 17 response. a , Flow cytometry analysis showing disruption of Aire -expression Rorc -Cre – Aire fl/fl (WT) and Rorc -Cre + Aire fl/fl (ILC3 ΔAire ) mice. Shown are representative FACS plots ( n ≥ 6) of intracellular Aire staining of thymic stroma (upper panel) or enriched Lin − cells from pLNs (lower panel); b – e , Flow cytometric analysis of OT-II T cell proliferation and differentiation in WT or ILC3 ΔAire mice transferred with naive OT-II and control CD4 + T cells in 1:1 ratio and subsequently injected with HKCA or HKCA-OVA every second day for 2 weeks showing representative flow cytometry dot plot highlighting the frequencies of transferred CD45.1 + OT-II T cell versus CD45.1/CD45.2 double-positive control T cell populations on day 14 after transfer ( b ), the corresponding statistical analysis showing ratios (mean ± s.d., two-tailed Student’s t -test) of OT-II versus control T cells ( n = 5 per group) ( c ), the frequencies of OT-II Rorc -GFP + cells ( d ) and the corresponding statistical analysis of the experiment showing the total counts (mean ± s.d., two-tailed Student’s t -test) of HKCA-induced OT-II Rorc -GFP + cells ( e ). f , g , Flow cytometric analysis assessing counts of Als1-Tet + CD4 T cells in WT, ILC3 ΔAire and TEC ΔAire mice that were injected intravenously with HKCA every second day for 2 weeks, showing a representative FACS plot ( f ) and the corresponding statistical analysis depicting the total counts (mean ± s.d., two-tailed Student’s t -test) in all mice ( n = 5) ( g ). h , i , Flow cytometric analysis assessing the proliferation of human RorγT + T H 17 cells in PBMCs isolated from patients with APS-1 or healthy controls showing representative dot plots of the frequency of proliferating RorγT + T H 17 cells measured by CPD dilution ( h ) and the corresponding statistical analysis of the experiment depicting average frequency (mean ± s.d., two-tailed Student’s t -test, n = 10 per group) ( i ). j , Enzyme-linked immunosorbent assay (ELISA) assessing amounts of IL-17A in the PBMC supernatants from proliferation assay described in panel h . Data are shown as mean of IL-17A concentration ± s.d., n = 10 for each group, two-tailed Student’s t -test. P value indicators: *** P < 0.0001; ** P < 0.001; NS, not significant. Full size image Next, we performed MHCII tetramer staining for C. albicans -specific epitopes derived from the agglutinin-like protein 1 (Als1). Specifically, pooled spleen or pLN-resident CD4 + T cells isolated from ILC3 ΔAire , TEC ΔAire or their wild-type littermates that had been stimulated with HKCA for two weeks prior to isolation were stained with Als1-specific tetramers (Als1-Tet) and analyzed by flow cytometry. We observed a significant reduction in the number of Als1-Tet + CD44 + CD4 + activated T cells isolated from the spleen and pLNs of ILC3 ΔAire mice compared to wild-type or TEC ΔAire mice (Fig. 4f,g ), suggesting that extrathymic, but not thymic expression of Aire was critical for the induction of Candida -specific T cells. To investigate whether a similar defect of T cell response to Candida was found in patients with APS-1, we stimulated the peripheral blood mononuclear cell (PBMC) fraction from patients with APS-1 or healthy individuals with HKCA and monitored T cell proliferation four days later. We observed a significant decrease (~3-fold) in both the frequency of HKCA-induced RorγT + T H 17 cells (Fig. 4h,i ), as well as in the amount of IL-17A released in supernatants from PBMCs isolated from APS-1 versus healthy controls (Fig. 4j ). Collectively, these data suggest that expression of Aire in ILC3s was required for the expansion of Candida -specific CD4 + Rorγt + T cell clones. Extrathymic Aire is critical for an effective response to C. albicans Next, we sought to determine whether loss of Aire expression in ILC3s could impair the clearance of live C. albicans in vivo. Because C. albicans is generally completely absent in mice housed under specific pathogen-free (SPF) conditions 32 , implying that SPF mice would have limited adaptive immune response to this pathogen, we pretreated wild-type and ILC3 ΔAire mice for 3 weeks with intravenous (i.v.) injections of HKCA, before i.v. administration of live C. albicans . ILC3 ΔAire mice were significantly more susceptible to C. albicans challenge, with only ~10% survival at 14 days after challenge, compared to wild-type littermates, which had a ~50% survival rate at this time point (Fig. 5a ). Moreover, the ILC3 ΔAire mice showed a significantly higher burden of C. albicans in their kidneys in comparison to wild-type littermate controls (Fig. 5b,c and Extended Data Fig. 3g,h ). Similarly, Aire −/− mice on either C57Bl/6 or nonobese diabetic (NOD) genetic backgrounds had significantly lower survival rate following i.v. injection of live C. albicans (with 20% and 0% survival, respectively) compared to their Aire +/+ littermates (with 60% and 45% survival, respectively) (Extended Data Fig. 3e,f ). Although the Aire −/− mice on the C57Bl/6 background had significantly poorer survival rate than their wild-type littermates, they did not develop detectable levels of IL-17- and/or IL-22-specific autoantibodies (Extended Data Fig. 3i,j ), further suggesting that the impaired anti- Candida response was primarily T cell dependent. Fig. 5: Mice lacking extrathymic expression of Aire have reduced survival after systemic challenge with live C. albicans . a , Survival curves of WT (Cre – Aire fl/fl ) and ILC3 ΔAire ( Rorc -Cre + Aire fl/fl ) mice ( n ≥ 10 per genotype group) injected i.v. with HKCA every 2 days for 3 weeks prior the infection by live C. albicans . Log-rank (Mantel-Cox) test was used to calculate the P value (0.0414). b, Quantitative PCR analysis of C. albicans- specific DNA in the kidney from WT and ILC3 ΔAire mice ( n = 6) infected with C. albicans as in a (mean ± SD, two-tailed Student’s t -test); c, Colony-forming unit (c.f.u.)-based assay determining the overgrowth of C albicans in kidneys of mice described in panel a , ( n = 6, mean ± s.d., two-tailed Student’s t -test). **** P < 0.0001, *** P < 0.001. Full size image Next, we assessed the adaptive immune response to live C. albicans in a mucosal model of infection. We used a modified version of an existing protocol 33 , which is based on the oral administration of live C. albicans to mice three consecutive times with 1-day intervals, resulting in a long-lasting colonization of their gastrointestinal tract. First, we assessed whether oral Candida colonization of wild-type mice resulted in the expansion of Candida -specific T cells in the secondary lymphoid organs (SLOs) (Fig. 6a,b and Extended Data Fig. 4a,b ). Flow cytometry indicated that while wild-type mice not colonized with C. albicans harbored relatively low numbers of Als1-Tet + T cells (three to five naive CD44 − T cells per SLO in each mouse analyzed), the number of Als1-Tet + T cells increased by ~50- to 100-fold at day 14 after C. albicans colonization (Fig. 6a,b ), and virtually all of them had an activated memory phenotype, as evidenced by high expression of CD44 (Fig. 6a ). In contrast, C. albicans colonization of ILC3 ΔAire mice (Fig. 6c,d ), Aire −/− mice (Extended Data Fig. 4c–e ) or bone marrow chimeric mice in which Aire deficiency occurred in the hematopoietic, but not in the stromal compartment (Extended Data Fig. 4f–h ), resulted in significantly decreased numbers of Als1-Tet + T cells in the SLOs than did C. albicans colonization of their corresponding wild-type littermate controls. Moreover, low numbers of Als1-Tet + T cells were also observed in CD90.2 Rag1 −/− mice that were adoptively transferred with CD90.1 + T cells and B cells and treated with a CD90.2-specific antibody to deplete the endogenous CD90.2 lymphoid compartment, thus rendering them ILC deficient 34 compared to isotype control-treated mice (Extended Data Fig. 4j,k ). Fig. 6: Extrathymic expression of Aire is critical for an effective T H 17 response to C. albicans at mucosal sites. a , b , Flow cytometry analysis assessing frequencies of C. albicans -specific Als1-Tet + , CD4 + T cells in pLNs of WT mice orally colonized by C. albicans and analyzed at different time points after colonization. Representative FACS plot showing counts of Als1-Tet + , C. albicans -specific CD4 + T cells (highlighted in red rectangles in the upper panel and red dots in the lower panel showing T cell activation markers CD44 versus CD69) ( a ) with a corresponding statistical analysis of the same experiment showing the total counts (mean + s.d., two-tailed Student’s t -test ( b ). c – j , Flow cytometry analysis assessing frequencies of C. albicans -specific Als1-Tet + , CD4 + T cells in pLNs ( c – f ) or intestinal lamina propria ( g – j ) of WT versus ILC3 ΔAire mice orally colonized by C. albicans and analyzed 2 weeks after colonization. Representative FACS dot plots showing counts of tetramer-positive CD4 T cells ( c , g ) and RorγT + CD4 T cells ( e , i ) are highlighted in black and red rectangles, respectively. Statistical analyses showing the total counts (mean + s.d., two-tailed Student’s t -test, n = 6 per group) corresponding to data shown in panels c , e , g and i are shown in panels d , f , h and j , respectively. Representative experiment out of three independent biological replicates is shown. k , Quantitative PCR analysis assessing the presence of C. albicans- specific DNA in the ileal part of small intestine from WT and ILC3 ΔAire mice (mean + s.d., two-tailed Student’s t -test, n = 6 per group). l , c.f.u.-based assay determining the overgrowth of C albicans in small intestinal tissues 14 days after oral colonization (mean + s.d., two-tailed Student’s t -test, n = 6 per group). m , c.f.u.-based assay determining the overgrowth of C albicans in oral mucosa 6 days after OPC challenge (mean + s.d., two-tailed Student’s t -test, n = 6 per group). *** P < 0.0001; APC, allophycocyanin; PE, phycoerythrin; ICS, intracellular staining. Full size image We next investigated the presence of Candida -specific T cells at different mucosal sites of the gastrointestinal tract at day 1 or day 14 after C. albicans administration. Although Als1-Tet + T cells could not be detected in the oral cavity, esophagus or small intestine mucosa from wild-type, Aire −/− or ILC3 ΔAire mice 24 h after C. albicans challenge (Extended Data Fig. 5 ), they were present in these locations in wild-type mice at day 14 after challenge (Fig. 6g,h and Extended Data Fig. 5 ). In contrast, ILC3 ΔAire or Aire −/− mice had markedly reduced numbers of Als1-Tet + T cells at these sites along the gastrointestinal tract, in particular in the lamina propria, at day 14 after challenge compared to their corresponding wild-type controls (Fig. 6g,h and Extended Data Fig. 5 ). In addition, the number of the bulk Rorγt + T H 17 population in pLNs and mesenteric LNs isolated was higher in wild-type mice compared to ILC3 ΔAire mice at day 14 after C. albicans colonization (Extended Data Fig. 6 ). In contrast, the number of Rorγt + T H 17 cells within lamina propria were similar in wild-type mice and ILC3 ΔAire mice at this time point (Fig. 6i,k ), suggesting that although ILC3-specific Aire deficiency impaired the expansion of Candida -specific T H 17 cells in the lymphoid organs and lamina propria, the effect on the general T H 17 population was smaller and more variable. In line with the decreased number of Candida -specific T cells at mucosal surfaces, the Aire −/− and ILC3 ΔAire mice had a significantly higher burden of C. albicans in their gastrointestinal mucosal sites compared to their wild-type littermates at day 14 after challenge (Fig. 6k,l and Extended Data Fig. 7 ). Finally, we also utilized a more conventional model for oropharyngeal candidiasis (OPC) 35 , in which the oral mucosal tissue in mice was exposed to live C. albicans for 90 min. In this setting, we pretreated wild-type and ILC3 ΔAire mice with i.v. injection of HKCA every 2 days for 14 days before OPC challenge with live C. albicans , and the oral mucosa was analyzed at day 5 after OPC challenge (Extended Data Fig. 8a ). At this time point, we observed significantly fewer Als1-Tet + T cells (Extended Data Fig. 8b,c ) and a higher burden of C. albicans (Fig. 6m ) in the oral mucosa of ILC3 ΔAire compared to wild-type mice (Extended Data Fig. 7b,c ). Collectively, these data demonstrated that Aire expression in Aire + ILC3s was required for the induction of adaptive immune responses to C. albicans at the mucosal surfaces. Aire + ILC3s induce survival of Candida -specific T H 17 clones To explore how expression of Aire in Aire + ILC3s promoted the expansion of C. albicans -specific T cells, we assessed the impact of Aire deficiency on the gene expression profiles of Aire + ILC3s and Candida -specific T cells in response to C. albicans challenge. For this purpose, we crossed Aire −/− mice with Aire GFP reporter mice to generate Aire GFP Aire −/− mice. Aire + ILC3s sorted (based on Aire GFP expression) from Aire GFP Aire −/− or Aire GFP Aire +/+ littermates that had received HKCA or PBS i.v. every day for 3 days (Extended Data Fig. 9 ) were analyzed by bulk RNA-seq analysis on day 4. Analysis of these transcriptomes indicated that Aire regulated the expression of hundreds of genes, which strongly overlapped with the genes upregulated in response to HKCA (Fig. 7a,b ). Specifically, the Aire + ILC3s isolated from HKCA-stimulated Aire GFP - Aire −/− mice had impaired induction of genes encoding cytokines ( Il6, Il18 and Bmp2) , C. albicans -sensing receptors ( Clec7a ), cell adhesion molecules ( Vcam1 and Cadm1 ), costimulatory molecules ( Cd86 ) and enzymes involved in proinflammatory response ( Ptgs2 ) compared to their HKCA-stimulated Aire GFP Aire +/+ counterparts (Fig. 7a,b ). These data suggested that Aire regulated the expression of several key molecules that may be critical for T H 17 differentiation ( Il6 ), T cell clonal expansion ( Cd80/Cd86 and Vcam1 ) and proinflammatory signaling ( Il18 and Ptgs2 ). Fig. 7: Aire + ILC3s induce a prosurvival program in Candida -specific T H 17 clones. a , Volcano plot of bulk RNA-seq analysis showing differential gene expression in Aire + ILC3s that were isolated from pLN of HKCA-stimulated Aire GFP Aire +/+ versus Aire GFP Aire −/− mice. Dashed lines indicate the FC = 2.0 and P = 0.05 cutoff. Selected genes are highlighted in blue (antigen cell adhesion and costimulation), red (cytokine and chemokine signaling) or green ( Candida -sensing receptors). b , FC:FC graph of RNA-seq data showing differential gene expression in Aire + ILC3s promoted by HKCA stimulation. Comparison of FC of HKCA-stimulated versus nonstimulated Aire + ILC3s isolated from Aire GFP Aire +/+ versus Aire GFP Aire −/− ( x axis) and Aire + ILC3s from stimulated Aire GFP Aire +/+ versus stimulated Aire GFP Aire −/− ( y axis). c , Volcano plot of RNA-seq data showing differential gene expression of Rorc GFP+ versus nonproliferating OT-II T cells derived from WT ( Rorc -Cre – Aire fl/fl ) versus ILC3 ΔAire ( Rorc -Cre + Aire fl/fl ) mice. The mice were transferred with naive OT-II CD4 + T cells and subsequently injected with HKCA-OVA four times during a single week. d , Volcano plot of RNA-seq data showing differential gene expression of Rorc GFP + OT-II T cells derived from WT versus ILC3 ΔAire mice treated as in panel c . Dashed lines indicate the FC = 2.0 and P = 0.05 cutoff. Data are derived from three independent replicates. e , RNA-seq-based heatmap showing the level of expression of selected genes in Rorc GFP + OT-II T cells isolated from WT and ILC3 ΔAire mice. Data are plotted as the z -score calculated for particular row. f , Gene ontology enrichment for upregulated differentially expressed genes from Rorc GFP + OT-II T cells derived from WT versus ILC3 ΔAire mice. Full size image To investigate how Aire deficiency impacted the transcriptional program of C. albicans -specific T cells, we adoptively transferred Rorc GFP + OT-II T cells into wild-type or ILC3 ΔAire mice that were stimulated with HKCA-OVA or HKCA every second day. Rorc GFP + OT-II T cells were then isolated from pLN on day 7 post-transfer and analyzed by bulk RNA-seq (Extended Data Fig. 10a ). At this time point, the frequencies of Rorc GFP+ OT-II T cells in ILC3 ΔAire mice were decreased by approximately threefold compared to wild-type littermates (Extended Data Fig. 10b ). We observed only mild increase in the frequency of CPD + proliferating Rorc GFP – OT-II T cells in wild-type versus ILC3 ΔAire mice (Extended Data Fig. 10b ), suggesting that Aire was not involved in the regulation of Rorc -GFP – T cell priming. The transcriptional analysis indicated that the Rorc - GFP+ OT-II T cells had substantially different transcriptomes compared to the nonproliferating CPD – OT-II T cells. Specifically, the Rorc GFP+ OT-II T cells isolated from wild-type mice upregulated 551 genes, including Rorc, Cd44 , chemokine receptor Ccr5 and proliferation marker Mki67 , and downregulated 220 genes, including Sell and Cd69 , compared to the nonproliferating CPD – OT-II T cells from wild-type mice (Fig. 7c and Extended Data Fig. 10b ), highlighting their activated/memory phenotype and readiness to exit the LN and move to the effector site. We detected more than 1,000 differentially expressed genes between the Rorc GFP + OT-II T cells derived from the wild-type compared to ILC3 ΔAire mice (Fig. 7c and Extended Data Fig. 10b ). Specifically, Rorc GFP + OT-II T cells isolated from ILC3 ΔAire mice had significantly reduced expression of anti-apoptotic factors such as survivin ( Birc5 ) or Bcl2l12 (ref. 36 ) and a less activated/memory phenotype, as suggested by the decreased expression of the chemokine receptor Cxcr3, Il2rb (encoding the IL-2 receptor beta subunit), Igf2r 37 , Il12rb1 (the key subunit of IL-23 receptor; Fig. 7d,e ) and Cd28 , encoding the CD80/86 receptor essential for T cell costimulation (Fig. 7e ). The gene ontology annotation of cell processes indicated that the differentially regulated genes were highly enriched for factors involved in the regulation of cell cycle and/or mitosis (Fig. 7f ). Collectively, these data suggested that Aire controlled the expression of genes in Aire + ILC3s (e.g., Il6, Cd86 and Ptgs2 ) that were critical for the subsequent induction of a prosurvival transcriptional program in T H 17 cells (Extended Data Fig. 10d,e ). Discussion Here, we show that extrathymic expression of Aire in Aire + ILC3s was required for the expansion of Candida -specific CD4 + T cells, in particular T H 17 clones, in response to C. albicans infection and for consequently limiting the pathogenicity of this opportunistic pathogen at mucosal tissues. We found that Aire + ILC3s expressed receptors implicated in C. albicans detection (e.g., dectin-1 and galectin-3) at levels comparable to DCs and effectively endocytosed C. albicans . Moreover, upon C. albicans uptake, Aire + ILC3s presented Candida -derived antigens to CD4 + T cells through MHCII. Therefore, our study suggests that effective response to C. albicans infection involves division of labor between different types of phagocytes and APCs, with Aire + ILC3s playing a nonredundant role in promoting the survival and subsequent expansion of Candida -specific T cell clones in the LN. Although ILCs are generally viewed as the innate analogs of T cells, a growing body of evidence suggests that some ILC subsets are, similarly to Aire + ILC3s, equipped with potent endocytic and/or antigen-presentation capacity 38 , 39 , 40 , 41 , 42 , 43 . For instance, spleen-derived NCR – CCR6 + MHCII + ILC3s, were reported to internalize latex beads and present model protein antigen to CD4 + T cells in vitro 43 . Similarly, MHCII + ILC2s can endocytose and present OVA protein and induce antigen-specific T cell proliferation 42 , whereas intestinal MHCII + ILC3s were reported to regulate T cell responses to bacterial antigens in an antigen-dependent manner 38 , 40 , 41 , arguing that some ILC3 subsets may act as potent APCs, with the capacity to modulate antigen-specific T cell responses in different contexts and different anatomical niches 38 . Moreover, our data also suggested that LN-resident MHCII – ILC3s could upregulate the expression of MHCII and CD86 coding genes in response to C. albicans challenge. This finding is in line with reports showing that the proinflammatory cytokine IL-1β promotes the expression of MHCII and costimulatory molecules on spleen-derived, but not gut-derived, MHCII – ILC3s 43 . Moreover, our data indicated that the impaired expansion of Candida -specific CD4 + T cells in SLO of ILC3 ΔAire mice was associated with their decreased accumulation at different mucosal sites, including the oral cavity, esophagus and intestine, and with increased burden of C. albicans at these mucosal tissues. It is likely that the impaired surveillance of the mucosal tissues due to a loss of Candida -specific CD4 + T cells resulted in reduced ability to control the fungal burden at the mucosal surfaces and loss of barrier integrity. In AIRE -deficient patients, in which both the extrathymic and the thymic expression of AIRE are defective, additional mechanisms (e.g., autoantibodies against IL-17A/F, IL-22 (refs. 17 , 18 , 44 ), defensins 45 and mucins 46 or loss of Paneth cells 45 ) may further impair the integrity of mucocutaneous surfaces and thereby enhance the invasiveness of C. albicans . Therefore, the increased susceptibility to C. albicans in Aire deficiency may combine aspects of the role of Aire in central tolerance with its role in shaping the C. albicans -specific T H 17 response in the periphery. Although our data are in line with the interpretation that the increased susceptibility to Candida infection in both humans and mice is due to defects in the T H 17 response 12 , 14 , 15 , 47 , 48 , it was also suggested that in the oral mucosa, Aire -deficient mice and humans have intact T H 17 responses to C. albicans and that these phenotypes are due to overproduction of interferon-γ in CD4 + and CD8 + T cells at the epithelial barriers 49 . In this specific study, however, the T H 17 response was measured only 24 h after C. albicans challenge 49 , which might be too early a timepoint to assess an adaptive immune response, because C. albicans is virtually absent in mice housed under SPF conditions 32 , 33 . Indeed, we found that the induction of Candida -specific T cells in SPF-housed mice peaked only 2 weeks after C. albicans colonization. In addition, the report assessed bulk T H 17 cell responses 49 and not Candida -specific T cell clones, as was done in this study. The molecular features of Aire + ILC3s described here seem to largely overlap with a subset of Lin − Aire + Rorc + MHCII + Janus cells recently identified using single-cell RNA profiling of mouse LN 50 , suggesting they might represent the same cell population. Although the study suggested that the Lin − Aire + Rorc + MHCII + Janus cells are a subset of tolerogenic DCs 50 , they lacked expression of key DC canonical markers such as Cd11c, Cd11b, Dec205, Clec9a or Cd4/Cd8 (ref. 50 ), as well as of other key myeloid or lymphoid markers. Moreover, similarly to Aire + ILC3s, the Janus cells highly expressed ILC3 markers, including Rorc, Il1r1, c-Kit, il7r, Id2, Ccr6, Ccr7 and Il18r1 , as well as genes linked to antigen presentation and costimulation. Therefore, although both Aire + ILC3s and Janus cells lacked the expression of key ILC3 effector molecules such as Il17a , Il17f or Il22 , they bore molecular features more characteristic of MHCII + ILCs rather than of myeloid APCs. This notion is also supported by a recent study showing that extrathymic Lin − Aire + Rorc + MHCII + cells have an interconverting potential with ILC3s in a fate-mapping analysis based on Aire -Cre reporter mice 51 . Irrespective of nomenclature, our study provides experimental evidence that the expression of Aire in LN-resident Lin − Aire + Rorc + MHCII + cells has a critical and a nonredundant role in the induction of Candida -specific T cells and the control of C. albicans colonization at mucosal tissues. Collectively, our data help to not only shed more light on the mechanisms underlying chronic mucocutaneous candidiasis in AIRE-deficient individuals but also identify an additional functional role for Aire beyond its well-established role in central tolerance induction in the thymus. Methods Mice The Aire fl/fl (Jax: 031409, ref. 30 ), Aire −/− (004743 C57BL/6 J and 006360 NOD genetic background 1 , 52 ), Aire -GFP ( Aire -IGRP-GFP 53 , a kind gift of M.S. Anderson (University of California, San Francisco)), CD45.1 congenic strain (002014), CD90.1 congenic strain (000406), Clec7a −/− (012337, ref. 54 ), Foxn1 -Cre (018448, ref. 55 ), Lgals3 −/− (006338, ref. 56 ), OT-II (004194, ref. 57 ), Rag1 −/− (002216, ref. 58 ) Rorc -Cre (022791, ref. 31 ), Rorc -GFP (007572, ref. 31 ) and Rosa -tdTomato (007914, ref. 59 ) strains were used. Unless indicated otherwise, all mouse strains were of C57Bl6/J genetic background and were purchased from Jackson Laboratories. Mice were housed at the Weizmann Institute of Science in SPF conditions. All experiments were approved by the local ethics committee (9661117-2, 01420218-2, 04690718-2 and 14850619-2). Usually, 6- to 8-week-old mice were used for the experiments, with the exception of bone marrow chimeras, where mice were 12–15 weeks old. For the generation of bone marrow chimeras, recipients were irradiated by single dose of 900 rad and transplanted by 1.10 7 bone marrow cells. Only mice with reconstitution level higher than 95% were used for experiments. For the generation of CD90-disperate chimeras, a protocol described elsewhere was used 34 . Briefly, 6-week-old Rag1 −/− mice were intravenously adoptively transferred by 8.10 7 MACS-enriched T cells and B cells from CD90.1 mice. These cells were allowed to homeostatically proliferate for 2 months. After this period, mice were intraperitoneally injected with 250 µg CD90.2 depleting antibody (BioXcell) every 3 days. Whenever possible, littermates were used as the controls. Human samples Patients were included from Norwegian National Registry of Organ Specific Autoimmune Diseases and fulfilled the APS-1 diagnostic criteria. All patients with APS-1 have C. albicans infection. Gender-matched controls were recruited from the local blood bank at Haukeland University Hospital. All participants gave informed and written consent, and the study was approved by The Regional Committee for Medical and Health Research Ethics for Western Norway. PBMCs were obtained from the whole blood by centrifugation in Ficoll-Pague (GE Healthcare), frozen and stored in liquid nitrogen. The proliferation assay was done for all samples together; the whole PBMC fraction (5.10 6 cells) was stained with cell proliferation dye and stimulated by 1.10 5 HKCA particles (Invivogen). Proliferative response was measured 4 and 6 days later. Materials All materials and reagents used in this study are described in the particular relevant section and specified in detail in Supplementary Tables 1 and 2 . Infection by live C. albicans Experimental mice were intravenously injected every 2 days with 10 6 particles heat-killed C. albicans in PBS for 3 weeks. After this period, mice were infected by single dose of 10 5 particles of live C. albicans . Mice were then monitored daily and killed when they lost 20% of their initial weight or showed signs of distress. Mucosal colonization by C. albicans To establish the gastrointestinal colonization by C. albicans , mice were supplemented ad libitum by 1 mg. ml −1 ampicillin in drinking water and were kept on it during the experiment. After 2 days, mice were colonized with 50 μl 10 6 particles C. albicans in PBS. The inoculation was performed dropwise into the mouth of mice. Experimental mice were monitored daily. ELISA For the detection of autoantibodies, the ELISA microtiter plate was precoated with 5 μl ml −1 recombinant IL-17 or IL-22 in bicarbonate buffer overnight at 4 °C. The plate was washed and blocked with 5% milk. Detection of autoantibodies was performed using anti-IgG specific antibody conjugated to horseradish peroxidase (Jackson ImmunoResearch). For the detection of human IL-17A cytokine, the human IL-17A ELISA kit (BioLegend) was used according to the manufacturer’s instructions. C. albicans strains and preparation of HKCA The wild-type C. albicans strain used in the study is of SC5314 origin 60 . GFP and OVA coding sequences were inserted in the coding frame after the C-terminal end of the Eno1 gene, resulting in generation of an OVA-expressing strain derived from the wild-type 61 . Both strains were a kind gift of J. Berman (Tel Aviv University, Israel). C. albicans was grown at 30 °C using YPD agar. HKCA variant was prepared by heat inactivation of the yeast at 60 °C for 1 h in a thermoshaker. Heat inactivation was tested by seeding the HKCA on YPD plates. Cell isolation for flow cytometry and cell sorting Aire + ILC3s, MHCII + ILC3s, cILC3s and DCs were isolated as described previously 19 . Although all systemic LNs were found to contain Aire + ILC3s 19 , for consistency, most of the experiments were done by analyzing cells and cellular responses in pLNs, unless stated otherwise. Briefly, pLNs were collected and subjected to several rounds of enzymatic digestion by Dispase I. (Roche). Single-cell suspension was depleted of Lin + cells using LS-column-based MACS enrichment with a cocktail of biotinylated antibodies (TCR-β, CD3, CD19, B220, CD11b, F4/80, Gr1 and CD11c; BioLegend) and anti-biotin microbeads (Miltenyi Biotec). T cells were isolated by meshing the skin-draining LNs and spleens through 40-μm nylon mesh. For surface staining, cells were incubated with antibodies for 25 min on ice. DAPI (Sigma-Aldrich) and viability dye eF506 (eBioscience) were used for live/dead cell discrimination. For intracellular staining fixation, the Foxp3/transcription factor staining buffer set (eBioscience) was used according to the manufacturer’s recommendation. Subsequently, intracellular targets were stained by antibodies for 1 h at room temperature. Cells from the small intestinal lamina propria, oral cavity or esophagus were collected by enzymatic digestion. Briefly, small intestinal tissue was subjected to two rounds of epithelial cells removal by incubation with 2 mM EDTA in HBSS for 20 min at 37 °C. Tissues were digested in 1 mg ml −1 Collagenase D and (Roche) for 1 h, and immune cells were enriched by Percoll gradient (Sigma-Aldrich). For details concerning antibodies, please refer to Supplementary Table 2 . Flow cytometry analysis and cell sorting were performed using BD CantoII, LSRII and AriaIII machines (BD). FlowJo (V10; Tristar) software was used for flow cytometry data analysis. RNA-seq Single-cell suspensions were directly FACS sorted to Lysis/Binding buffer (Invitrogen) and frozen on dry ice. RNA was isolated using Dynabeads (Invitrogen) according to manufactures protocol. The MARS-seq protocol described elsewhere was followed to generate the sequencing libraries 62 . The sequencing of the library was performed using the NextSeq high-output kit and NextSeq 500 sequencer (Illumina). Obtained data were analyzed for differential gene expression using the UTAP pipeline 63 . Imagestream analysis of endocytosis Lin − cells or enriched DCs were coincubated in-test described time period together with Cell proliferation dye eF660 (Thermo Fisher Scientific) stained 10 5 ml −1 HKCA particles in 37 °C. Cells were fixed by Foxp3 / Transcription factor staining buffer set (eBioscience) and stained immediately after the end of incubation period and subjected to Imagestream analysis (Amnis). Data were analyzed using Ideas (v6.2) software (Amnis). Image acquisition by two-photon laser scanning microscopy MACS-isolated OT-II tdTomato + cells were adoptively transferred to Aire GFP hosts and stimulated by heat-killed HKCA or HKCA-OVA. Zeiss LSM 880 upright microscope fitted with Coherent Chameleon Vision laser was used for LN imaging experiments. Images were acquired with a femtosecond-pulsed two-photon laser tuned to 930 nm. The microscope was fitted with a filter cube containing 565 LPXR to split the emission to a photomultiplier tube detector (with a 579- to 631-nm filter for tdTomato fluorescence) and an additional 505 LPXR mirror to further split the emission to two GaAsp detectors (with a 500- to 550-nm filter for GFP fluorescence). Pictures were acquired at 512 × 512 x - y resolution, and the zoom was set to 1.5. Analysis of endocytosis and antigen-presentation capacity by FACS Experimental mice were intravenously injected with 10 6 cell proliferation dye eF660 (Thermo Fisher Scientific) stained HKCA. Mice were analyzed in indicated described time periods. For intravascular staining, mice were injected five minutes prior the analysis intravenously by 5 µg anti-mouse CD45 BV605 monoclonal antibody (30-F11, BioLegend). For the antigen-presentation assays, experimental mice were intravenously injected with 10 6 HKCA or HKCA-OVA particles. Cells with antigen-presentation capacity were isolated using FACS and incubated with DO11.10 TCR NFAT-GFP cell line 4 for 17 h at a ratio of 1:5. GFP fluorescence was measured using FACS. Adoptive T cell transfer and stimulation of mice by HKCA Naive OVA-specific TCR + OT-II cells from CD45.1 + mice and wild-type-derived CD4 + T cells (CD45.1/CD45.2) were isolated using a naive CD4 + T cell isolation kit (Miltenyi Biotec), mixed in 1:1 ratio, stained using cell proliferation dye eF660 (Thermo Fisher Scientific) and transferred via tail vain to recipient mice. Once in 2 days, mice were injected with 10 6 HKCA or HKCA-OVA particles via the tail vein. Tetramer staining of C. albicans-specific T cells Als1 tetramers conjugated with PE and APC were used to detect C. albicans-specific T cells from HKCA-stimulated mice or mice after C. albicans colonization. The staining by tetramers and pulldown of tetramer-positive cells by anti-PE and anti-APC conjugated microbeads (Miltenyi Biotec) was performed as described elsewhere 64 . Each batch of tetramer reagent was titrated to determine the optimal staining concentration. We thank the NIH Tetramer Core Facility for providing tetramer reagents. Isolation of DNA from tissue and intestinal content and quantification of C. albicans burden Approximately 3 mm ileum or kidney or liver tissue was surgically resected including its content. DNA was extracted using Quick-DNA kit (Zymo Research) according to manufactures instructions. 10 ng isolated DNA was used for downstream quantitative PCR reaction using SYBR green (Roche) and following set of primers for detection of C. albicans (primer forward (pF): 5′-TTTATCAACTTGTCACACCAGA-3′, primer reverse (pR): 5′-ATCCCGCCTTACCACTACCG-3′) and bacterial ribosomal subunit 16S (pF: 5′-ACTCCTACGGGAGGCAGCAGT-3′, pR: 5′-ATTACCGCGGCTGCTGGC-3′) as the calibrator. The relative C. albicans DNA content in the samples was calculated using a method described elsewhere 65 . Isolation of RNA from tissues and quantification of gene expression Approximately 0.1 g tongue tissue, esophagus, ileal part of small intestine and kidney was collected and RNA was extracted using Nucleospin RNA Mini kit (Macherey Nagel) according to the manufacturer’s instructions. Isolated RNA was subjected to a reverse transcription reaction using RevertAid RT Reverse Transcription Kit (Thermo Fisher Scientific). Quantitative PCR reaction using SYBR green (Roche) and following set of primers was used; Il17a (pF: 5′-TGACCCCTAAGAAACCCCCA-3′, pR: 5′-TCATTGTGGAGGGCAGACAA-3′), Il17f (pF: 5′-GAAGGCTGGGAACTGTCCTC-3′, pR: 5′-CGGAGTTCATGGTGCTGTCT-3′), Il22 (pF: 5′-TTGACACTTGTGCGATCTCTGA-3′, pR: 5′-AAAGGTGCGGTTGACGATGT-3′), and Casc3 as housekeeping gene (pF: 5′-TTCGAGGTGTGCCTAACCA-3′, pR: 5′-GCTTAGCTCGACCACTCTGG-3′). The relative gene expression was calculated using method described elsewhere 65 . Mouse model of OPC Mice were first primed by repeated injection of 10 6 HKCA particles every second day for 2 weeks. Then, a previously established protocol was followed 35 . Briefly, mice were sedated and exposed to C. albicans orally for 1.5 h using cotton swabs soaked with C. albicans diluted in PBS (10 7 particles ml −1 ). Mice were analyzed 5 days after oral inoculation. Determination of C. albicans colony-forming units A total of 0.2 g tongue, esophagus, ileal part of small intestine and kidney tissue was mechanically disrupted in PBS and plated on Sabouraud Dextrose Agar (Merck) in two dilutions. The number of colonies was calculated after 24 and 48 h. Colony-forming units were recalculated per gram of original tissue. Statistical analysis Unless indicated otherwise, statistical significance was assessed using two-tailed Student’s t- test calculated in GraphPad Prism program. To summarize P values, the following marks were used: *** P < 0.0001, ** P < 0.001, * P < 0.05. Access codes for all transcriptomics data All RNA-seq data have been deposited to the Gene Expression Omnibus under accession number GSE203158 . Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Of all the fungi that live in the human body, the most infamous is probably the yeast Candida. This distant cousin of baker's yeast is notorious for causing various types of thrush that can be a major nuisance, but it can also lead to an invasive infection that may, on occasion, prove fatal. In a study published today in Nature Immunology, a Weizmann Institute of Science research team headed by Prof. Jakub Abramson uncovered a previously unknown defense mechanism employed by the immune system in fighting Candida infections. Candida is present at low levels in the bodies of most healthy people, forming part of the microbiome—a diverse spectrum of microbes that reside peacefully in our gut and on our skin. Under normal circumstances, Candida is held in check by the immune system, but it can occasionally grow excessively, invading the lining of the mouth, the vagina, the skin or other parts of the body. In severe cases, it can spread to the bloodstream and from there to the kidneys. Such life-threating infections may occur when a person's immune system has been weakened, for example, by AIDS or by immunosuppressive drugs such as cancer chemotherapy or steroids. Antibiotics, which wipe out many of the beneficial bacteria within our microbiome, can also unleash local or invasive Candida eruptions by providing this yeast with an unfair advantage vis-à-vis other microorganisms. That's why, for instance, women sometimes develop a vaginal yeast infection after taking antibiotics. Until now, the immune cells that got most of the credit for defending the body against Candida were the small, round lymphocytes of the T cell type, called TH17. These cells were also the ones to take the blame when this defense failed. In the new study, postdoctoral fellow Dr. Jan Dobeš, working together with colleagues in Abramson's lab in Weizmann's Immunology and Regenerative Biology Department, discovered that a powerful commando unit of TH17 cells capable of fighting Candida cannot be generated without crucial early support from an entirely different contingent: a subset of rare lymphoid cells known as type-3 innate lymphoid cells, or ILC3, that express a gene called the autoimmune regulator, or Aire The two groups of cells belong to the two different arms of the immune system, which, like foot patrols and specialized units, join forces against a common enemy. The Aire-ILC3s—part of the more ancient, innate arm—spring into action almost immediately upon encountering a threat—in this case, a Candida infection. The TH17s belong to the immune system's more recent, adaptive arm, which takes several days or even weeks to respond, but which launches a much more targeted and potent attack than the innate one. Credit: Weizmann Institute of Science The scientists found that as soon as Candida starts infecting tissues, the Aire-ILC3s engulf the yeast whole, chop them up and display some of the yeast pieces on their surfaces. That's how these bits are presented to the TH17s, a few of which are generally on call in the lymph nodes, ready for an infection alert. This kind of presentation instructs the specialized T cells to start dividing rapidly, soaring in number from a few lone commandos to several hundred or even thousands of Candida-specific fighters, capable of destroying the yeast at the sites of infection. "We have identified a previously unrecognized immune system weapon that is indispensable for orchestrating an effective response against the fungal infection," Abramson says. Abramson became intrigued by Candida because it commonly leads to severe, chronic infections in people with a rare autoimmune syndrome caused by defects in the Aire gene. Abramson's lab had conducted extensive studies of this gene, helping to clarify its role in preventing autoimmune disorders. That research, as well as studies by other scientists, had shown that Aire-expressing cells in the thymus instruct developing T cells to refrain from attacking the body's own tissues. When Aire is defective, T cells fail to receive proper instructions, consequently causing widespread autoimmunity that wreaks havoc in multiple body organs. But one puzzle remained: Why would Aire-deficient patients suffering from a devastating autoimmune syndrome also develop chronic Candida infections? While trying to complete the Aire puzzle, Dobeš and colleagues found that outside the thymus, Aire is also expressed in a small subset of ILC3s in the lymph nodes. The researchers then genetically engineered two groups of mice: One lacked Aire in the thymus, and the other group lacked it in the ILC3s in the lymph nodes. The first group developed autoimmunity but was able to successfully fight off Candida. In contrast, those in the second group, the ones lacking Aire in ILC3s, did not suffer from autoimmunity, but were unable to generate numerous Candida-specific TH17s. Consequently, they failed to effectively eliminate Candida infections. In other words, without Aire-expressing ILC3s, the specialized T cells needed for fighting Candida were not produced in sufficient numbers. "We found an entirely new role for Aire, one that it plays in the lymph nodes—turning on a mechanism that increases the numbers of Candida-fighting T cells," Dobeš explains. These findings open up new directions of research that in the future may help develop new treatments for severe Candida, and possibly for other fungal infections. The newly discovered mechanism might, for example, help produce large numbers of Candida-fighting T cells to be used in cell therapy. And if scientists one day identify the signals by which Aire-ILC3s boost T cell proliferation, these signals themselves might provide the basis for new therapies.

10.1038/s41590-022-01247-6

Medicine

Researchers closer to understanding how a drug could induce health benefits of exercise

Baptiste Rode et al, Piezo1 channels sense whole body physical activity to reset cardiovascular homeostasis and enhance performance, Nature Communications (2017). DOI: 10.1038/s41467-017-00429-3 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-017-00429-3

https://medicalxpress.com/news/2017-08-closer-drug-health-benefits.html

Abstract Mammalian biology adapts to physical activity but the molecular mechanisms sensing the activity remain enigmatic. Recent studies have revealed how Piezo1 protein senses mechanical force to enable vascular development. Here, we address Piezo1 in adult endothelium, the major control site in physical activity. Mice without endothelial Piezo1 lack obvious phenotype but close inspection reveals a specific effect on endothelium-dependent relaxation in mesenteric resistance artery. Strikingly, the Piezo1 is required for elevated blood pressure during whole body physical activity but not blood pressure during inactivity. Piezo1 is responsible for flow-sensitive non-inactivating non-selective cationic channels which depolarize the membrane potential. As fluid flow increases, depolarization increases to activate voltage-gated Ca 2+ channels in the adjacent vascular smooth muscle cells, causing vasoconstriction. Physical performance is compromised in mice which lack endothelial Piezo1 and there is weight loss after sustained activity. The data suggest that Piezo1 channels sense physical activity to advantageously reset vascular control. Introduction The health value of exercise has been described since the time of Hippocrates, but it was not until 1953 that it was demonstrated scientifically 1 . It is now known that whole body physical activity and other forms of physical exercise afford major protection against chronic disease 2 , 3 . Such protection seems likely to have evolved because animals survived by tuning their biology to regular physical activity to avoid predators and source prey and other food. Without this activity many humans today are in suboptimal environments, increasing the risk of dysregulation and disease. Therefore there has been intense research into exercise 2 . However, the existence and identity of molecular sensors of exercise has remained unclear. If we could identify such sensors, we might find ways to better tune human biology to advanced societies. Because the likely site of exercise sensors is the endothelium 2 , we were interested in whether Piezo1 might act as an exercise sensor. Piezo1 is a relatively recently discovered membrane protein which assembles as a trimer to form Ca 2+ -permeable non-selective cationic channels activated by physical force 4 , 5 , 6 . It is highly expressed in endothelial cells and known to be important for integrating vascular architecture with physical force during embryonic development 7 , 8 . In this study, we conditionally disrupted Piezo1 in the endothelium to investigate its relevance to adult mice. We found that elevated blood pressure of whole body physical exercise depended on endothelial Piezo1. The mechanism was a vascular bed-specific effect of Piezo1 which opposed endothelium-dependent relaxation mediated by endothelium-derived hyperpolarization (factor) (EDH(F)) to cause vasoconstriction when fluid flow was elevated. We conclude that endothelial Piezo1 is an exercise sensor which enables optimized redistribution of blood flow to enhance physical performance. Results Mice with disrupted endothelial Piezo1 are superficially normal To investigate the relevance of Piezo1 in the adult endothelium, we engineered mice with conditional Cre-Lox-mediated disruption of Piezo1 in the endothelium (Piezo1 ΔEC mice) (Supplementary Figs. 1 and 2 ). The mice appeared normal and had normal body weights, weight gains and organ weights and serum urea, K + and Na + ; gross anatomies and functions of the heart and aorta were also normal (Fig. 1a–m ) (Supplementary Fig. 3 ). Retinal vasculature and endothelial response to injury were normal (Fig. 1n–q ). Therefore endothelial Piezo1 in the adult appeared to be without consequence. Fig. 1 Mice with disrupted endothelial Piezo1 are superficially normal. a Physical appearance of control and endothelial Piezo1-deleted (Piezo1 ΔEC ) mice. b Body weight and percentage change in body weight of control ( n = 13) and Piezo1 ΔEC ( n = 14) mice before and 10–14 days after tamoxifen (TAM) treatment. c As percentages of total body weight, weights of heart, lung, kidney and liver in control ( n = 13) and Piezo1 ΔEC ( n = 14) mice. d Histological examples of control ( top row ) and Piezo1 ΔEC ( bottom row ) sections of aorta, heart, lung, kidney and liver stained with H&E. Scale bars 100 μm. e – g Serum concentrations of urea, K + and Na + in control ( n = 7) and Piezo1 ΔEC ( n = 7) mice. h – m Ultrasound study of the heart h – j and aorta k – m of control ( n = 5) and Piezo1 ΔEC ( n = 5) mice under anaesthesia. h Example of left ventricle images of control and Piezo1 ΔEC at diastole and systole. The left ventricle chamber is circled with a white dashed line . Scale bar 1 mm. i , j Cardiac parameters measured by ultrasound. k Example of aorta images of control and Piezo1 ΔEC at diastole and systole. The left ventricle chamber is circled with a white dashed line . Scale bar 1 mm. l Aorta anteroposterior diameter at systole. m Aortic distensibility. n Retinal vasculatures stained with isolectin ( green ) from control and Piezo1 ΔEC mice. Entire retinas ( full view ) and close up views ( zoom ). o Quantification of retina distal artery diameter and capillary plexus area from control ( n = 5) and Piezo1 ΔEC ( n = 4) mice. p Endothelial regeneration 5 days after femoral artery injury. Images of the arteries in which the blue colour shows Evans blue staining of areas which were not re-endothelialized after injury. Scale bars 0.5 mm. q Quantification of endothelial regeneration in control ( n = 9) and Piezo1 ΔEC ( n = 7) mice. Independent data points are displayed with superimposed bars indicating mean ± s.e.m. Data sets are compared by t -test. No significant differences were detected Full size image To investigate the mice in more detail, we made isometric tension recordings from second-order mesenteric arteries, looking for relevance of Piezo1 to endothelium-dependent tone. As expected, arteries from control genotype mice contracted in response to the α 1 -adrenoceptor agonist phenylephrine and then relaxed in response to the endothelium-dependent vasodilator acetylcholine (Fig. 2a, b ). Arteries from Piezo1 ΔEC mice behaved similarly (Fig. 2c, d ). These data also suggested that endothelial Piezo1 in the adult was of no consequence. Fig. 2 Endothelial Piezo1 channels have an anti-EDH(F) effect. Isometric tension recordings from mouse second-order mesenteric artery. a Example recordings from control genotype artery before (+ EC) and after endothelium-denudation (−EC). Upward deflection is increasing tension. Phenylephrine (PE, 0.3 μM). Acetylcholine (ACh) and the nitric oxide donor amino-3-morpholinyl-1,2,3-oxadiazolium (SIN-1) were applied at increasing concentrations as indicated by the dots (0.01, 0.03, 0.1, 0.3 and 1 μM). b As for a but mean data ( n = 14 mice). c , d As for a , b but Piezo1 ΔEC mice ( n = 10 mice). e Schematic illustration of the dichotomy Piezo1 presents for endothelial biology and vascular function. EC, endothelial cell. VSMC, vascular smooth muscle cell. eNOS, endothelial nitric oxide synthase. NO, nitric oxide. EDH(F), endothelium-derived hyperpolarization (factor). f Example recordings from control genotype artery before and after application of apamin (Apa, 0.5 μM) and charybdotoxin (Ch, 0.1 μM). Phenylephrine (PE, 0.3 μM). Acetylcholine (ACh) was applied at increasing concentrations as indicated by the dots (0.01, 0.03, 0.1, 0.3 and 1 μM). g As for f but mean data ( n = 8 mice). h , i As for f , g but for Piezo1 ΔEC mice ( n = 6 mice). Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by * P < 0.05, ** P < 0.01, *** P < 0.001 Full size image Endothelial Piezo1 channels have an anti-EDH(F) effect The Ca 2+ -permeable non-selective cationic pathway generated by Piezo1 channels 4 , 5 , 6 poses an intriguing dichotomy for endothelial biology (Fig. 2e ). Ca 2+ entry theoretically drives Ca 2+ dependent mechanisms such as activation of endothelial nitric oxide synthase 7 , 9 . But cation entry as a whole (and especially entry of the monovalent ion Na + ) theoretically drives membrane potential depolarization which could be important as a vasoconstrictor mechanism because EDH(F) is well established as a mechanism for endothelium-dependent vasodilatation 10 . Therefore, we investigated the effect of inhibiting EDH(F) by the established method of combining apamin and charybdotoxin; toxin inhibitors of two K + channels which are critical for EDH(F) 10 . In mesenteric arteries from control genotype mice the toxins caused slight inhibition of the ACh response but they strongly inhibited the ACh response in arteries from Piezo1 ΔEC mice (Fig. 2f–i ). Other properties of the arteries were unchanged by Piezo1 ΔEC (Supplementary Figs. 4 , 5 ). The data suggest that endothelial Piezo1 channels oppose EDH(F) and may have specific roles under certain circumstances. Importance for elevated blood pressure in exercise Because of the relevance of mesenteric arteries to peripheral resistance and thus blood pressure, we inserted telemetry probes for continuous recording of blood pressure. Mice were provided with free access to a running wheel. Blood pressure was not different in Piezo1 ΔEC mice during periods of physical inactivity (Fig. 3a ). In contrast, the increase in blood pressure seen during physical activity was reduced (Fig. 3b ). The data suggest that endothelial Piezo1 has specific importance in blood pressure regulation during whole body physical activity. Fig. 3 Importance for elevated blood pressure during whole body physical activity. Telemetry measurements of mean, systolic and diastolic blood pressures in conscious freely moving control ( n = 6) and Piezo1 ΔEC ( n = 7) mice. Data were analysed when the mice were inactive during the day a and voluntarily active on a running wheel during the night b . Time zero is when the mouse was introduced to the running wheel cage. Measurements were not made during the first 2 days of acclimatization. Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by * P < 0.05, ** P < 0.01, *** P < 0.001 Full size image Constitutive and flow-induced Piezo1 in mesenteric artery To investigate how Piezo1 might mediate an anti-EDH(F) effect and elevation of blood pressure, we first developed a unidirectional monovalent cation flux assay for Piezo1 channels which showed strong monovalent cation permeability when the channels were over-expressed in HEK293 cells or natively expressed in cultured endothelial cells (Supplementary Fig. 6 ). Constitutive monovalent cation flux was evident, suggesting non-inactivating Piezo1 channels capable of mediating sustained depolarization (Supplementary Fig. 6 ). To determine the relevance to physiological cells, we acutely isolated endothelial cells from second-order mesenteric arteries. Cell-detached outside-out membrane patches were used to enable identification of Piezo1 channels by their unitary current size (and therefore unitary conductance) and avoid contaminating effects from the cytosol and intracellular organelles. Outside-out patches are outwardly convex 11 and so the patch pipette did not protect the membrane from fluid flow. Dominant constitutive channel activity was observed with the expected 25 pS unitary conductance of Piezo1 channels (Fig. 4a, b ) 4 , 5 . Fluid flow enhanced the activity (Fig. 4a, c ). Piezo1 channel identity was confirmed by sensitivity to inhibition by Gd 3+ (Fig. 4a, c ) which blocks Piezo1 channels 4 , and absence of the channels in patches from Piezo1 ΔEC mice (Fig. 4c, d ). There was no response to fluid flow in the absence of Piezo1 (Fig. 4c, d ). Moreover the fluid flow effect was mimicked by Yoda1 (Supplementary Fig. 7 ), a small-molecule activator of Piezo1 channels 12 , 13 . The data suggest that constitutive and fluid flow-enhanced Piezo1 channel activity is common in mesenteric artery endothelial cells and that these channels do not inactivate or depend on intracellular factors. Fig. 4 Piezo1 channels are flow sensors in endothelium of mesenteric resistance artery. Data are for ionic current recordings from outside-out patches excised from freshly isolated endothelium of second-order mesenteric arteries. a Example recording at −70 mV. Two sections are expanded to clarify unitary current events (C: channel closed) (O: 1, 2 or 3 simultaneous channel openings). The patch was placed at the outlet of a capillary from which flowed ionic solution at 20 μl s −1 . Gadolinium ion (Gd 3+ , 10 μM). b Mean unitary current amplitudes for channels activated by flow as in a ( n = 10). c Mean channel activity (NP o : number × probability of opening) for experiments of the type exemplified in a for no flow and flow conditions and the two genotypes (Control and Piezo1 ΔEC ). Individual data points for each independent experiment are shown as symbols , superimposed on which are the mean ± s.e.m. values ( n = 10 for each group). d Example original trace for a patch from Piezo1 ΔEC endothelium exposed to 20 μl s −1 flow. Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by *** P < 0.001 Full size image To explore the significance in membrane potential control, freshly isolated sheets of mesenteric endothelium were used for measurements without contamination from other cell types. The sheets were syncitia of endothelial cells 14 to which a patch-clamp pipette was attached in whole-cell recording mode. The resting membrane potential averaged about −46 mV in static resting conditions and application of fluid flow caused reversible depolarization (Fig. 5a, b ). Importantly, application of the Piezo1 channel inhibitor Gd 3+ or deletion of Piezo1 (Piezo1 ΔEC ) caused hyperpolarization of the resting potential to an average of about −60 mV, and fluid flow and Yoda1 now had no depolarizing effect (Fig. 5a, b ) (Supplementary Fig. 7 ). Incremental increases in fluid flow up to and above fluid flow rates reported in anaesthetised mice 15 incrementally depolarized the membrane potential in a Piezo1-dependent manner (Fig. 5c ). Fig. 5 Ca 2+ channel activation in vascular smooth muscle cells. a , b Membrane potential measurements from freshly isolated endothelium of second-order mesenteric arteries. a Example traces from the control and Piezo1 ΔEC genotypes in the absence of Gd 3+ and the control genotype in the presence of 10 μM Gd 3+ . Endothelium was exposed to flow at 20 μl s −1 . b As for a but individual data points for the three independent experiments shown as symbols. Control genotype: no flow −46.3 ± 2.4 mV vs flow −38.4 ± 2.1 mV (***). Control genotype in Gd 3+ : no flow −60.1 ± 3.1 mV vs flow −56.5 ± 3.3 mV (***). Piezo1 ΔEC genotype: no flow −59.5 ± 2.3 mV vs flow −60.7 ± 2.5 mV (*). c As for a , b but a separate series of experiments in which endothelium was exposed to increasing flow in the absence of Gd 3+ for control ( n = 10) and Piezo1 ΔEC ( n = 6) genotypes. d Example current recordings from vascular smooth muscle cells (VSMC) freshly isolated from second-order mesenteric artery. Square-step depolarizations were applied at the time of the arrow from a holding voltage of −80 mV to the test voltage indicated. Linear leak and capacitance currents were subtracted. e Mean data for peak VSMC inward currents of the type exemplified in d (current, n = 8). Superimposed in grey are endothelial shear stress values calculated from the control genotype data in c . Membrane potential for the highest shear stress was obtained by extrapolation based on a least-squares fit of the Hill equation. Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by *** P < 0.001 Full size image The observations suggested a type of Piezo1 channel which is sufficiently abundant and active to strongly regulate the resting membrane potential. It is a specialised non-inactivating form of Piezo1 channel but otherwise it has the expected unitary conductance and pharmacology. The channel is strongly constitutively active but can further activate in proportion to fluid flow. The commonly reported inactivation of exogenously over-expressed Piezo1 channels has previously been found to be lost after pulses of mechanical force and suppressed by application of force to extracellular domains without effect on channel unitary conductance 16 , 17 . There is therefore precedence for Piezo1 channels adopting a non-inactivating state. Coupling to voltage-gated Ca 2+ entry in smooth muscle cells To understand how the Piezo1 mechanism could become relevant at times of whole body physical activity, we hypothesized that the increased blood flow in exercise drives depolarization which has been shown previously to be efficiently coupled to the adjacent vascular smooth muscle cells 10 , 18 , 19 and could be sufficient to activate pro-contractile voltage-gated Ca 2+ channels 20 only when the depolarization reaches a particular range of values. To test this hypothesis, we recorded from vascular smooth muscle cells freshly isolated from second-order mesenteric artery. Increasing depolarizations were applied by voltage-clamp to activate voltage-gated Ca 2+ currents. As expected these currents were small, close to the technical limits of detection (Fig. 5d ). Such channels have no distinct threshold for activation but show exponential increases in opening probability with depolarization 20 . To take account of this behaviour, we fitted the data with the Boltzmann equation, leading to the suggestion that increases in Ca 2+ entry occurred progressively from −40 mV to more positive (Fig. 5e ). In the whole body, the cell-rich fluid of blood causes shear stress at the endothelium and whole body physical activity increases shear stress 21 . We demonstrated the principle that increasing fluid flow, and thus shear stress, caused depolarization which reached the activation range for voltage-gated Ca 2+ channels (Fig. 5e ). To test this idea in the intact artery, we cannulated segments of second-order mesenteric artery to apply pressure and flow to the lumen. Importantly, increased flow caused vasoconstriction which was inhibited by nicardipine, a blocker of voltage-gated Ca 2+ channels (Fig. 6a–c ). In Piezo1 ΔEC mice, flow-induced vasoconstriction was absent (Figs. 6d, e ). These data support a hypothesis whereby precipitous increases in voltage-gated Ca 2+ entry occur in smooth muscle cells, leading to vasoconstriction when fluid flow along the endothelial surface is sufficiently high. Fig. 6 Flow-evoked vasoconstriction in mesenteric artery. Isobaric external diameter recordings from second-order mesenteric artery. a Example images of a cannulated artery before and after luminal pressure difference (Δ P ) and then after 10 μM nicardipine was added to the recording chamber. Control genotype mouse. Scale bar , 200 μm. b Example diameter recording for a control genotype mouse during incremental increases in Δ P as indicated by the black dots (20, 40, 60, 70, 80, 90 and 100 mm Hg). Nicardipine (Nic, 10 μM), phenylephrine (PE, 1 μM) and acetylcholine (ACh, 10 μM) were applied as indicated. The first arrow indicates addition of 100 μM N(ω)-nitro-L-arginine methyl ester (L-NAME) to the recording chamber and the second arrow multiple washes of the chamber to remove nicardipine and L-NAME. c Mean data for the type of experiment shown in b presented as the constriction to Δ P as a percentage of the PE response (9 arteries from n = 6 mice). Nicardipine significantly (***) reduced the 100 mm Hg Δ P response to 30.7 ± 3.9% ( n = 6 mice). d , e The same as for b , c but using Piezo1 ΔEC genotype mice (8 arteries from n = 3 mice). Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by ** P < 0.01, *** P < 0.001 Full size image There is absence of specific information on the absolute shear stress experienced by endothelial cells in second-order mouse mesenteric arteries at rest or during whole body physical activity. In vivo, shear stress is complicated by the pulsating cardiac rhythm, vascular architecture and arterial calibre, viscosity and cellular content of the blood, and glycoprotein-polysaccharide structures between blood and membrane proteins of the endothelial cells. Nevertheless, shear stress in mice is considered to range from about 3 to 60 Pa 22 , 23 , 24 , consistent with our studies (Fig. 5e ). Values in humans are usually lower 22 , 23 . Endothelial Piezo1 is important for physical performance Based on the above findings, we hypothesised a Piezo1 mechanism which senses whole body physical activity in order to constrict mesenteric resistance arteries with the purpose of directing mesenteric blood flow away from the gastrointestinal tract 25 , 26 to other organs—in particular skeletal muscle—to improve physical performance. To test this idea, we quantitatively investigated running wheel performance. Although the Piezo1 ΔEC mice were superficially normal (Fig. 1 ), their performance was compromised (Fig. 7a–d ) (Supplementary Fig. 8 ). The impact of Piezo1 ΔEC declined with continued exposure to the wheel, suggesting compensated performance due to physical training (Fig. 7a–c ) despite the blood pressure lowering effect being sustained (Fig. 3b ). Although Piezo1 ΔEC mice gained weight normally in the absence of the wheel (Figs 1b and 7e ), they lost more weight than controls once they had access to the wheel, suggesting that they were working harder to achieve their expectations (Fig. 7e ). Fig. 7 Physical performance depends on endothelial Piezo1. Voluntary running wheel data for control and Piezo1 ΔEC genotypes from the dark cycle (active period) showing distance run on the wheel a , percentage time for which mice were active on the wheel b and the number of active bouts of exercise c ( n = 12 mice per group). d As for c but summary analysis for all bouts of activity. e For the same mice analysed in a – d , changes in body weight during the 12 days prior to and 7 days after access to the wheel ( n = 12 mice per group). Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by * P < 0.05 Full size image Therefore, we suggest a molecular mechanism for sensing whole body physical activity: specialised Piezo1 channels in the endothelium which beneficially impact on overall physical performance. We outline a mechanistic principle by which the channels can sense fluid flow and transduce it into constriction of mesenteric arteries and thus increase blood pressure during physical activity because these arteries contribute a major component of total peripheral resistance. Contribution of endothelial Piezo1 is vascular bed specific It is important to recognise that different vascular beds respond differently to whole body physical activity. While blood flow to the intestines decreases in physical activity 25 , 26 , blood flow to skeletal muscle increases 2 , so a vasodilator rather than vasoconstrictor mechanism must dominate in resistance arteries of skeletal muscle 2 . Similarly, blood flow to the brain is maintained or slightly increased during physical activity to avoid syncope 2 . Therefore, we investigated the significance of endothelial Piezo1 in saphenous artery, which supplies the leg, and carotid artery, which supplies the brain. Saphenous artery was similar in calibre to second-order mesenteric artery: its external diameter when pressurized to 60 mm Hg was 284.2 ± 13.4 μm ( n = 15) compared with 279.8 ± 12.7 μm in mesenteric artery ( n = 17). Carotid artery was 428.8 ± 17.4 μm under similar conditions ( n = 8). In contrast to mesenteric artery (Fig. 2 ) the EDH(F) effect was not amplified by Piezo1 disruption in saphenous or carotid artery (Fig. 8a–h ) (Supplementary Fig. 9 ). The data suggest vascular bed specificity of Piezo1 in a way which enables it to enhance physical performance. Fig. 8 Contribution of endothelial Piezo1 is vascular bed specific. Isometric tension recordings from mouse saphenous or carotid artery. a Example recordings from control genotype saphenous artery before and after application of apamin (Apa, 0.5 μM) and charybdotoxin (Ch, 0.1 μM). Phenylephrine (PE, 0.3 μM). Acetylcholine (ACh) was applied at increasing concentrations as indicated by the dots (0.01, 0.03, 0.1, 0.3 and 1 μM). b As for a but mean data ( n = 5 mice). c , d As for a , b but for Piezo1 ΔEC mice ( n = 5 mice). e – h As for a – d but carotid artery ( n = 5 control mice, n = 5 Piezo1 ΔEC mice). Averaged data are displayed as mean ± s.e.m. Data sets are compared by t -test. Statistical significance is indicated by * P < 0.05, ** P < 0.01 Full size image Discussion We have described the challenge presented by the Ca 2+ and depolarizing effects of Piezo1 channels in endothelial cells and suggested a role for the depolarising effect in whole body physical activity. Here, however, we have not addressed the relationship between Piezo1 and nitric oxide synthase and nitric oxide suggested previously 7 , 13 . When we studied flow-induced vasoconstriction, an inhibitor of endothelial nitric oxide synthase was present in order to focus on the Piezo1 depolarization mechanism (Fig. 6 ). Importantly, the Piezo1-dependence of increased blood pressure in whole body physical activity suggests dominance of the vasoconstrictor mechanism in vivo even without exogenous nitric oxide synthase inhibition (Fig. 3b ). This implies that the nitric oxide mechanism is naturally suppressed in mesenteric arteries and perhaps also in other arteries which constrict during whole body physical activity such as those of the kidneys and liver 2 . A flow-induced vasodilatation mechanism exists in mesenteric arteries 13 , 27 but our data suggest that it is less important than vasoconstriction during whole body physical activity. The vasodilatory effect might of course be important in other circumstances. The role of endothelial Piezo1 has been studied in sedate mice where relevance of the nitric oxide mechanism has been suggested 13 . The effect reported was however on systolic blood pressure, with no diastolic data shown 13 . Because vasodilatation should affect diastolic pressure, it remains to be clarified if endothelial Piezo1 is important for vasodilation in vivo in sedentary mice. Our data suggest no role of endothelial Piezo1 when mice are inactive between periods running on a wheel (Fig. 3a ). The principles we describe here in the mouse may be important in people because Piezo1 is a functional endothelial protein in humans 28 and whole body physical activity increases vascular wall shear stress in humans 21 . In healthy individuals, systolic blood pressure usually increases due to increased cardiac output whereas diastolic blood pressure may remain unchanged depending on the extent of vasodilation in skeletal muscle 2 . Nevertheless, whole body physical activity redistributes blood away from the intestines, suggesting importance of vasoconstriction in the mesenteric bed 25 , 26 . Exercise is commonly used as an approach for protecting against or reducing hypertension in the human population 29 and, conversely, exercise-induced hypertension is suggested as an important predecessor of persistent hypertension, which remains one of the major health concerns of the twenty-first century 30 . Therefore developing further methodologies and tools for studying Piezo1 and its relationships to exercise-induced adaptations should be a valuable area for future research. Methods Piezo1-modified mice All animal use was authorized by the University of Leeds Animal Ethics Committee and The Home Office, UK. All animals were maintained in GM500 individually ventilated cages (Animal Care Systems), except during telemetry recordings, at 21 °C 50–70% humidity, light/dark cycle 12/12 h on RM1 diet (SpecialDiet Services, Witham, UK) ad libitum and bedding of Pure’o Cell (Datesand, Manchester, UK). Genotypes were determined using real-time PCR with specific probes designed for each gene (Transnetyx, Cordova, TN). C57BL/6 J mice with Piezo1 gene flanked with LoxP sites (Piezo1 flox ) were described previously 7 . To generate tamoxifen (TAM) inducible disruption of Piezo1 gene in the endothelium, Piezo1 flox mice were crossed with mice expressing cre recombinase under the Cadherin5 promoter (Tg(Cdh5-cre/ERT2)1Rha and inbred to obtain Piezo1 flox/flox /Cdh5-cre mice. TAM (Sigma-Aldrich) was dissolved in corn oil (Sigma-Aldrich) at 20 mg ml −1 . Mice were injected intra-peritoneal with 75 mg kg −1 TAM for 5 consecutive days and studies were performed 10–14 days later. Piezo1 flox/flox /Cdh5-cre mice that received TAM injections are referred to as Piezo1 ΔEC . Piezo1 flox/flox littermates (lacking Cdh5-cre) that received TAM injections were the controls (control genotype). For experiments, mice were males aged 12–16 weeks, except for telemetry (14–18-week old) and femoral artery injury (18–22-week old). Analysis of Piezo1 deletion Samples (about 10 mm 3 ) of liver, lung, aorta, femoral artery and mesenteric artery were digested overnight at 37 °C in a lysis buffer containing 10 mM Tris pH 7.4, 50 mM EDTA, 1 % SDS; 5 µg ml −1 proteinase K (Sigma-Aldrich). Samples were then vortexed and 400 µl of phenol/chloroform/isoamyl (Sigma-Aldrich) was added. Tubes were mixed by inverting/shaking 20 times every 15 min for 1 h then centrifuged at 13,000× g for 15 min at room temperature. Four-hundred microliter of the top layer was transferred to a new tube followed by an addition of 440 µl of isopropanol and 40 µl of 3 M NaCl. Tubes were mixed by gentle inverting and left to stand for 1 h at room temperature. DNA was pelleted by centrifugation at 13,000× g for 30 min at room temperature. The supernatant was discarded and the pellet was washed with 70% ethanol, briefly air dried and resuspended in 60 µl of TE buffer. DNA was amplified using 12.5 µl Bioline MyTaq Red Mix, 0.5 µl of DNA solution, 1 µM of each primer. Sequences of PCR primers are specified in Supplementary Table 1 . PCR was 95 °C for 5 min; 32 cycles of 95 °C for 30 s, 60 °C for 30 s, 72 °C for 30 s; 72 °C for 5 min. PCR products were electrophoresed on 2 % agarose gels containing SYBR safe (Roche) at 80 V for 45 min. Mouse liver endothelial cells Mouse liver sinusoidal endothelial cells were isolated using an immunomagnetic separation technique. A whole mouse liver was minced using 2 scalpel blades and resuspended in a dissociation solution consisting of 9 ml 0.1 % collagenase II, 1 ml 2.5 U ml −1 dispase, 1 µM CaCl 2 and 1 μM MgCl 2 in Hanks Buffer solution. The tissue-dissociation mix was incubated at 37 °C for 50 min in a MACSMix Tube Rotator (Miltenyi Biotech) to provide continuous agitation. At the end of enzymatic digestion the sample was passed through 100 and 40 μm cell strainers to remove any undigested tissue. Cells were washed twice in PEB buffer consisting of Phosphate Buffered Saline (PBS), EDTA 2 mM and 0.5% Bovine Serum Albumin (BSA), pH 7.2. The washed pellets were resuspended in 1 ml of PEB buffer and 200 µl of dead cell removal paramagnetic microbeads per 1 × 10 7 cells (Miltneyi Biotec) at room temperature for 15 min. After incubation the cells were passed through an LS column prepared with 1 × binding buffer (Miltenyi Biotec) in a magnetic field (MiniMACS Separator, Miltenyi Biotec). The eluate was then incubated with 20 ml red blood cell lysis buffer consisting of 0.206 g Tris base, 0.749 g NH 4 Cl in 100 ml PBS pH to 7.2. Cells were washed again in PEB buffer, and the pellet was resuspended in 1 ml PEB buffer and 30 µl CD146 microbeads (Miltenyi Biotec) at 4 °C for 15 min under continuous agitation. After incubation this solution was passed through an MS column prepared with PEB buffer. CD146 positive cells were retained in the column and CD146 negative cells passed through as eluate. CD146 positive cells were washed through with warm EGM-2 media and the CD146 selection process was repeated a second time. After a second purification cells were plated and grown in a 5% CO 2 incubator at 37 °C. Media were changed at 12 h and then every 24 h until confluent. Piezo1 inducible cell line Piezo1-GFP 7 was used as a PCR template to clone human Piezo1 coding sequence into pcDNA™4/TO between HindIII and EcoRI restriction sites. Piezo1 was amplified as two fragments using the following primers: (HindIII-Piezo1-Fw: AATAAGCTTATGGAGCCGCACGTG and BamHI-Int.Piezo1-Rv: AATGGATCCCCCTGGACTGTCG) and (BamHI-Int.Piezo1-Fw: AATGGATCCTCCCCGCCACGGA and EcoRI-Piezo1-Rv: AATGAATTCTTACTCCTTCTCACGAGT). The two fragments were fused using BamHI restriction site, resulting in the full length Piezo1 coding sequence with the c4182a silent mutation. T-RExTM-293 cells were transfected with pcDNA4/TO-Piezo1 using Lipofectamine 2000 (Thermo Fisher Scientific). Subsequently cells were treated with 10 μg ml −1 blasticidin and 200 μg ml −1 zeocin (Invitrogen, Thermo Fisher Scientific) to select stably transfected cells. Single cell clones were isolated and analysed individually. Expression was induced by treating the cells for 24 h with 10 ng ml −1 tetracycline (Sigma-Aldrich) and analysed by quantitative RT-PCR and western blot. Fura-2 Ca 2+ measurements Intracellular Ca 2+ was measured using the ratiometric Ca 2+ indicator dye fura-2. Experiments were performed on confluent cells in a 96-well plate. Cells in each well were incubated with 50 μl fura-2 AM loading solution for 1 h at 37 °C. The loading solution consisted of 2 µM fura-2 AM and 0.01% pluronic acid in Standard Bath Solution (Ca 2+ -SBS) consisting of 130 mM NaCl, 5 mM KCl, 1.2 mM MgCl 2 , 1.5 mM CaCl 2 , 8 mM d -glucose and 10 mM HEPES (pH 7.4). After 1 h the loading solution was removed and 100 μl of Ca 2+ -SBS was added to each well and left at room temperature for 30 min. A compound plate was prepared at twice the final concentration tested in Ca 2+ -SBS. The FlexStation II 384 was set to add 80 μl of the compound solution to each well on the test plate containing 80 μl of Ca 2+ -SBS. Baseline fluorescence ratios were recorded before addition of the compound solution to the cell plate after 60 s, with regular recordings thereafter for a total of 5 min. Thallium FluxOR measurements Cells were plated at 80–90% confluence in 96-well plates 24 h prior to recordings (5 × 10 4 Human Embryonic Kidney (HEK) T-REx cells; 1.92 × 10 4 Human Umbilical Vein Endothelial Cells (HUVECs)). HEK T-Rex cells were from Thermo Fisher Scientific (catalogue #R71007) and HUVECs were from Lonza (catalogue #CC-2519). HEK T-Rex cells were validated to be tetracycline-responsive (as expected) (Supplementary Fig. 6 ). HUVECs were validated by positive staining with anti-CD31 antibody, response to vascular endothelial growth factor, and alignment of the cells to shear stress; human nucleotide sequences were detected, confirming human origin. To measure thallium (Tl + ) influx, cells were loaded with the FluxOR TM dye for 1 h at room temperature, transferred to assay buffer and stimulated with a Tl + containing K + -free solution as per the manufacturer’s instructions (Molecular Probes). Measurements were made on a fluorescence plate reader (FlexStation II 384 ). FluxOR was excited at 485 nm and emitted light collected at 520 nm, measurements expressed as a ratio increase over baseline ( F / F 0 ), with vehicle (DMSO) values subtracted from Yoda1 values at each time point (∆ F / F 0 ). Rates of increase in fluorescence intensity were determined between 7.5 and 35 s after injection of Yoda1 (∆ F / F 0 /ms). Arterial contraction studies Animals were culled by cervical dislocation according to Schedule 1 procedure approved by the UK Home Office. Mesenteric arcades were dissected out and placed immediately into ice-cold Krebs solution (125 mM NaCl, 3.8 mM KCl, 1.2 mM CaCl 2 , 25 mM NaHCO 3 , 1.2 mM KH 2 PO 4 , 1.5 mM MgSO 4 , 0.02 mM EDTA and 8 mM d -glucose, pH 7.4). Second-order mesenteric, saphenous or carotid arteries were cleaned of fat and connective tissue under a dissection microscope. Segments of 1 mm length were mounted in an isometric wire myograph system (Multi Wire Myograph System, 620 M), bathed with Krebs solution warmed at 37 °C and gassed with 95% O 2 /5% CO 2 then stretched stepwise radially to their optimum resting level to an equivalent transmural pressure of 100 mm Hg and equilibrated for 1 h prior to experiments. For studies of luminal flow in second-order mesenteric artery, vessel segments were mounted on glass cannulas in a pressure myograph (Model 110p, Danish Myo Technology A/S, Denmark). Flow was generated by increasing the pressure difference (Δ P ) between inflow and outflow without change in the absolute intraluminal pressure. The outer arterial diameter was monitored using a CCD camera (DMX41 AU02, Imaging Source Europe, Germany) and recorded with MyoView II software. Arteries were only used for investigation if they constricted in response to phenylephrine (PE) and dilated in response to acetylcholine (ACh). Blood pressure measurements Conscious long-term recordings of arterial blood pressure (mean, systolic and diastolic) were achieved via a radiotelemetry probe (model TA11PA-C10, Data Sciences International). Adult male mice (14–18-week old) were anaesthetised with isoflurane (5% induction 1.5% maintenance) in 95% O 2 and body temperature maintained via a heating pad. The probe catheter was advanced, via the left carotid artery, into the ascending aorta. The body of the transmitter was placed in a subcutaneous pocket along the left flank. A period of at least 14 days was allowed for recovery from surgery before the start of experimental recordings. TAM treatment started 4 days after probe implantation and recordings started 10 days after the last TAM injection. Mice were housed singly in cages and synchronized to a light–dark cycle of 12:12 h with lights on at 06:00 h. Cages were positioned over receivers connected to a computer system for data recording. Blood pressure waveforms and parameters were analysed using DSI analysis package, Dataquest ART 4.1. Continuous 24 h recordings were begun 3 days following singular housing and obtained over a 7 day period. During the recording period animals were allowed free access to a voluntary running wheel. Freshly isolated mesenteric endothelial cells Endothelial cells were freshly isolated from second-order branches of mouse mesenteric arteries as described previously 14 . Briefly, dissected second-order mesenteric arteries were enzymatically digested in dissociation solution (126 mM NaCl, 6 mM KCl, 10 mM Glucose, 11 mM HEPES, 1.2 mM MgCl 2 , 0.05 mM CaCl 2 , with pH adjusted to 7.2) containing 1 mg ml −1 collagenase Type IA (Sigma-Aldrich, Dorset, UK) for 14 min at 37 °C and then triturated gently to release the endothelial cells on a glass coverslip. Patch-clamp electrophysiology Membrane potential was measured using the perforated whole-cell configuration of the patch-clamp technique in current clamp mode with an Axopatch-200A amplifier (Axon Instruments, Inc.) equipped with Digidata 1440 A and pCLAMP 10.6 software (Molecular Devices, Sunnyvale, CA, USA) at room temperature. Outside-out membrane patch recordings were made using the same equipment but in voltage-clamp mode. Endothelial cells and endothelium were bathed in a solution consisting of 135 mM NaCl, 4 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM glucose and 10 mM HEPES (pH 7.4). Heat-polished patch pipettes with tip resistances between 3 and 5 MΩ were used. For membrane potential recordings, amphotericin B (Sigma-Aldrich) was used as the perforating agent, added in the pipette solution composed of 145 mM KCl, 1 mM MgCl 2 , 0.5 mM EGTA and 10 mM HEPES (pH 7.2). For application of fluid flow, endothelium or membrane patches were manoeuvred to the exit of a capillary tube with tip diameter of 350 μm, out of which ionic (bath) solution flowed at rates specified in the main text and figure legends. Calculation of shear stress ( τ ω ) was achieved using the Hagen-Poiseuille formula 31 ( τ ω = 4 μ Q / πR 3 ) where μ is dynamic viscosity, Q is flow rate and R is radius of the capillary tube. Echocardiography Animals were maintained under steady-state isofluorane anaesthesia and placed on a heated platform with ECG and respiration monitoring. Core temperature was measured using a rectal probe (Indus Instruments) and maintained at 37.5 °C throughout recording. Echocardiography was performed using a Vevo2100 high resolution, pre-clinical in vivo ultrasound system (VisualSonics) with the MS-550D transducer at 40 MHz frequency and 100% power. Imaging was performed on a layer of aquasonic gel after the pre-cordial skin had been clipped and de-epliated with cream (Veet). Parasternal long-axis view (PLAX) images were obtained in EKV mode (set at 1000 Hz for recording) over the entire cardiac cycle. The left ventricular area was traced in end-distole (LVAd) and end-systole (LVAs) and used to derive the ejection fraction (EF) with the Vevo LAB cardiac package software. The investigator performing sonography was blinded to the genotype of the animals. Transverse EKV recordings were also obtained over the abdominal aorta just below the diaphragm using the same settings as described for the heart. These images were evaluated in the VevoVasc software package to determine vessel distensibility. Maximal anteroposterior aortic diameter (from inner wall to inner wall) was measured in the same images in systole and diastole using Vevo LAB general imaging package software. Retina vasculature staining and analysis Retinas were dissected from eyes after fixation in 4% paraformaldehyde in PBS for 4 h at room temperature, then stored overnight at 4 °C in permeabilisation and blocking buffer (PBS; 0.5% triton; 1% BSA; 0.01% sodium deoxycholate; 0.02% sodium azide; 0.1 mM CaCl 2 ; 0.1 mM MgCl 2 ; 0.1 mM MnCl 2 ). Retinal vasculature was then stained overnight at 4 °C with isolectin B4 Alexa Fluor 488 conjugate (Molecular Probes, Thermo Fisher), diluted 1:100 in PBLEC buffer (PBS; 1% triton; 0.1 mM CaCl 2 ; 0.1 mM MgCl 2 ; 0.1 mM MnCl 2 ). Retinas were washed with 0.25% Triton in PBS, then flat-mounted on slides with ProLong Gold (Molecular Probes, Thermo Fisher). Confocal microscopy (LSM 880, Zeiss) was used to image retinas, with analysis blinded to genotype conducted using ImageJ software (NIH, Bethesda, MD). Distal arterial diameter was analysed 1500 μm from the optic disc in the largest branch of each artery emanating from the disc. Capillary area was determined in regions of interest, which excluded arteries and veins, using the threshold function and fractional area measurement. Femoral injury Experiments were carried out on 18–22-week-old male mice. Femoral injury was performed 12–16 days after the last TAM injection. Mice were anesthetized with isoflurane (1.5–2%) before a small incision was made in the mid-thigh and extended. Having carefully isolated the femoral artery, an arteriotomy was made in the saphenous artery using iris scissors (World-Precision Instruments, Sarasota, FL) and a 0.014-inch-diameter angioplasty guidewire with tapered tip (Hi-Torque Cross-It 200XT, Abbott-Vascular, IL) was introduced. The guidewire was advanced 1.5 cm in to the femoral artery, and three passages performed per mouse, resulting in complete endothelial denudation. The guidewire was removed completely and a suture tightened rapidly immediately distal to the bifurcation of the femoral artery. The skin was closed with a continuous suture. Animals received peri-operative analgesia with buprenorphine (0.25 mg kg −1 s.c.). Mice were anesthetized at 5 days after wire injury and 50 μl of 0.5% Evans blue dye injected into the inferior vena cava. The mice were perfused/fixed with 4% paraformaldehyde in PBS before the femoral arteries were collected. The vessels were opened longitudinally. The areas stained and unstained in blue were measured in a 5 mm injured segment beginning 5 mm distal to the aortic bifurcation, and the percentage areas were calculated using ImagePro Plus 7.0 software (Media Cybernetics, Bethesda, MD). Running wheel analysis Mice were individually housed and had free access to a running wheel. Custom-built hardware and software allowed detailed characteristics of running activity to be recorded for each animal. A mouse was considered to be active when there were ≥ 2 revolutions of the running wheel during each 1 min recording period. This equates to ≈10% of the mean dark cycle velocity of running for control animals (0.34 m/s). Continuous periods of activity (bouts) were defined as activity seen in two or more consecutive minutes. One Piezo1 ΔEC mouse showed complete inactivity for the first 3 days and was excluded from the analysis along with its control genotype pair. Cell and tissue staining Epididymal fat pad and liver tissues were fixed for 48 h in 4% PFA at 4 °C prior to processing on a Leica ASP 200 and embedding in CellWax (Cellpath) on a Leica EG1150H embedding station. Sections of 4 μm were cut on a Leica RM2235 microtome onto Plus Frost slides (Solmedia) and allowed to dry at 37 °C overnight prior to staining. Slides were de-waxed in xylene and rehydrated in ethanol. H&E was performed by staining in Mayer’s Haematoxylin for 2 min and eosin for 2 min. Slides were imaged on an Aperio AT2 (Leica Biosystems) high definition digital pathology slide scanner with a maximal magnification of ×20. Tissue processing and imaging were performed at Section of Pathology and Tumour Biology, Leeds Institute of Cancer and Pathology. For CD31 immunofluorescence, cells were fixed on coverslips in 4% paraformaldehyde and permeablised with 0.1% TritonX-100 at room temperature. Cells were blocked with donkey serum for 30 min to prevent non-specific binding. Cells were then incubated with 1% BSA in PBS containing rabbit anti-mouse CD31 (1:50, Abcam ab28364). Following incubation with primary antibody, cells were washed in PBS and incubated with Alexa Fluor 488-conjugated affinipure donkey anti-rabbit IgG (1:300, Jackson Immuno Research Laboratories) for 45 min at room temperature. Cells were mounted with Prolong Gold Antifade Reagent containing DAPI (Invitrogen) and visualised using a LSM 880 confocal microscope (Zeiss). RNA isolation and quantitative PCR Total RNA was isolated using a standard TriReagent protocol and treated with DNAse (TURBO DNA-free, AM1907M, Ambion). An aliquot was used for cDNA synthesis using a High Capacity RNA-to-cDNA kit (Applied Biosystems, UK) containing Oligo-dT and random primers. Real-time PCR was performed using Roche Fast Start SYBR Green I on a Lightcycler2 with Lightcycler 3.5 software or using Roche 480 SYBR Green I on a Lightcycler480II with Lightcycler 1.5.62 software. DNA amplification was for 35 cycles with an initial 10 min at 95 °C followed by 10 s at 95 °C, 6 s at 55 °C and 14 s at 72 °C. Primers were used at 0.5 μM. Sequences of PCR primers are specified in Supplementary Table 1 . The specificity of PCR was verified by reactions without RT (-RT) and by melt-curve analysis. PCR cycle crossing-points (CP) were determined by fit-points methodology. Relative abundance of target RNA was calculated from (E18sCp)/(EtargetCp). All quantitative PCR reactions were performed in duplicate and the data averaged to generate one value per experiment. Data analysis Genotypes of mice were always blinded to the experimenter and mice were studied in random order determined by the genotype of litters. Data were generated in pairs (control mice and Piezo1 ΔEC mice) and data sets compared statistically by independent t -test without assuming equal variance. Paired t -tests were used when comparing data before and after application of flow or a substance to the same membrane patch or cell. Statistical significance was considered to exist at probability ( P ) < 0.05 (* < 0.05, ** < 0.01, *** < 0.001). Where data comparisons lack an asterisk, they were not significantly different. The number of independent experiments (mice or independent cell cultures) is indicated by n . For multi-well assays or multiple cell on coverslip studies, the number of replicates is indicated by N . Descriptive statistics are shown as mean ± s.e.m. unless indicated as mean ± s.d. (standard deviation). Origin Pro software was used for data analysis and presentation. Data availability All relevant data are available from the authors upon reasonable request.

The research team - led by the University of Leeds - has found that a protein called Piezo1 in the lining of blood vessels is able to detect a change in blood flow during exercise. They have described the protein as an 'exercise sensor'. During physical activity - as the heart pumps more blood around the body - the Piezo1 protein in the endothelium or lining of the arteries taking blood from the heart to the stomach and intestines senses the increased pressure on the wall of the blood vessels. In response, it slightly alters the electrical balance in the endothelium and this results in the blood vessels constricting. In a clever act of plumbing, that narrowing of the blood vessels reduces blood flow to the stomach and intestines, allowing more blood to reach the brain and muscles actively engaged in exercise. The scientists say this is ground-breaking research because it identifies for the first time a key biomolecular mechanism by which exercise is sensed. They believe the health benefit of exercise maybe linked with the fact that blood flow is being controlled to the intestinal area. Professor David Beech, from the Leeds Institute of Cardiovascular and Metabolic Medicine and lead investigator, said: "If we can understand how these systems work, then we may be able to develop techniques that can help tackle some of the biggest diseases afflicting modern societies. "We know that exercise can protect against heart disease, stroke and many other conditions. This study has identified a physiological system that senses when the mammalian body is exercising." The research - which is based on studies using mice - has been published in Nature Communications. The Piezo1 protein is also present in humans - and scientists have recognised that physical activity in humans also increases the pressure on the walls of the endothelium in the stomach and intestinal area, pushing blood towards the brain and muscles. Looking for a drug treatment The researchers also investigated the effect of an experimental compound called Yoda1 - named after the character from Star Wars - on the action of the Piezo1 protein. They found that it mimicked the action of increasing blood flow on the walls of the endothelium which is experienced during physical activity, raising the possibility that a drug could be developed which enhances the health benefits of exercise. Professor Beech said: "One of our ideas is that Piezo1 has a special role in controlling blood flow to the intestines and this is really an important part of the body when we start to think about something called the metabolic syndrome which is associated with cardiovascular disease and type 2 diabetes. "By modifying this protein in the intestines then perhaps we could overcome some of the problems of diabetes and perhaps this Yoda1 compound could target the Piezo1 in the intestinal area to have a functional effect. "It may be that by understanding the working of the Yoda1 experimental molecule on the Piezo1 protein, we can move a step closer to having a drug that can help control some major chronic conditions." The scientists have received funding from the British Heart Foundation to move to the next phase of the project. Working with chemists also based at the University at Leeds, they will modify the Yoda1 molecule so it can be used on further animal studies.

10.1038/s41467-017-00429-3

Biology

New hope for novel therapies has emerged from computational models

Ahmed M. Ibrahim, The conditional defector strategies can violate the most crucial supporting mechanisms of cooperation, Scientific Reports (2022). DOI: 10.1038/s41598-022-18797-2 Journal information: Scientific Reports

https://dx.doi.org/10.1038/s41598-022-18797-2

https://phys.org/news/2022-11-therapies-emerged.html

Abstract Cooperation is essential for all domains of life. Yet, ironically, it is intrinsically vulnerable to exploitation by cheats. Hence, an explanatory necessity spurs many evolutionary biologists to search for mechanisms that could support cooperation. In general, cooperation can emerge and be maintained when cooperators are sufficiently interacting with themselves. This communication provides a kind of assortment and reciprocity. The most crucial and common mechanisms to achieve that task are kin selection, spatial structure, and enforcement (punishment). Here, we used agent-based simulation models to investigate these pivotal mechanisms against conditional defector strategies. We concluded that the latter could easily violate the former and take over the population. This surprising outcome may urge us to rethink the evolution of cooperation, as it illustrates that maintaining cooperation may be more difficult than previously thought. Moreover, empirical applications may support these theoretical findings, such as invading the cooperator population of pathogens by genetically engineered conditional defectors, which could be a potential therapy for many incurable diseases. Introduction A long-standing puzzle in evolution theory is how cooperative behavior can evolve and persist within the selfish natural world. Once cooperation exists, it is always prone to exploitation by defective free riders who adopt selfish strategies for reaping the highest possible profit without paying the share. Thus, they invest most of their energy in reproduction. Therefore, cheaters could outcompete the cooperators and take over the population. Haldane pointed out that there was no general principle to solve this problem 1 . Since then, many partial mechanisms have been suggested, such as kin selection, group selection, reciprocity, policing, spatial structure, sanction, reward, and punishment 2 , 3 , 4 . Darwin himself suggested some core concepts of these mechanisms. This paper introduced conditional defector strategies that violate the kin selection, punishment, and spatial structure mechanisms. What are conditional defector strategies? From a conceptual viewpoint, a conditional defector strategy may be any cheating strategy that could somehow cooperate. In other words, it is a cheater who pays additional costs or wastes a portion of the profit to survive. Therefore, it is not necessarily defective in all behaviors or at all times. However, it may cooperate in some behaviors now and then. Consequently, they are not pure defectors. Some forms of conditional defector strategies Cooperate for the spread Dispersal is beneficial because it decreases Kin competition, sustains resources, and outsets colonization. Without dispersal, the fate of all populations is extinction 5 . Hence, an intermediate dispersal rate of cooperators is essential for cooperation maintenance 6 , 7 . Furthermore, migration can induce an explosive outbreak of cooperation, even in a world of selfish individuals with various sources of randomness, starting with defectors only 8 . Recently 9 showed for the first time that strategy-neutral migration triggered strategy oscillation in a two-strategy game even without the bridging of any transition strategies (states), thus defining a novel oscillating behavior fundamentally different from the conventional cyclic dominance previously found in a game of at least three or more strategies. Reference 10 studied the impact of varying migration preferences in promoting cooperation. They revealed the role of orientation-driven migration, where individuals prefer to move closer to their neighboring cooperators or favor escape away from their neighboring defectors. Nevertheless, dispersal is a costly behavior that increases the mortality of dispersers or decreases their fecundity 11 , 12 . Therefore, some studies have focused on the joint evolution of dispersal and cooperation 13 . Alternatively, there is a correlation between cooperation, dispersal rate, and dispersal cost 14 . Their findings asserted that the low dispersal cost selects against cooperation. Thus, if cheaters reduce their dispersal costs, they may turn the game against cooperators. Usually, cheats are not good migrators because dispersal itself is a cooperative behavior. The migrators leave their suitable habitats to other unknown environments and face dangerous predators to colonize a new patch. Therefore, such behavior is costly for migrators. Nevertheless, its benefits are also gained by non-migrators because it decreases kin competition. Thus, dispersal is considered a cooperative behavior that naturally does not expect to be abundant in cheats. Hence, cheaters go extinct rapidly with the depletion of local patches they dominated without global prevalence like cooperators; this might be the fundamental problem of cheaters. However, the probable solution is adopting a conditional defection strategy wherein free riders would cooperate only for the spread. The actors of this selfish strategy would have a high dispersal rate with the lowest possible cost because they share migration costs. Thus, the exploitation rate of public goods and interactions among defectors and cooperators will increase. In other words, the conditional defectors can exclusively cooperate in all collective behaviors related to migration (coalition dispersal) but defect otherwise. Therefore, these selfish successful migrators can convert the structured meta-population into a well-mixed game and violate the spatial structure mechanism. In fact 15 , is considered empirical evidence to assume that individuals reduce their dispersal costs by sharing it. Thereby, they can achieve successful migrations. Additionally, in metastatic cancer, migrating in groups (coalition dispersal) raises the efficacy to 50-fold more than individual dispersal 16 , 17 . Pay for the escape Conditional defectors can pay some of their wealth or waste some profits to escape punishment by producing substances to mislead punishers. Or possession of the significant tag that marks cooperators. Similarly, by reducing their payoff to be more similar and familiar to cooperators. If cheaters reduced the benefits, it might be hard to have been noticed by a quorum-sensing system or other defense mechanisms. It is considered a kind of imitation or tag-based decision that prevents cooperators from detecting and punishing defectors. Those cheaters pay a cost to escape sanctions or reduce the accuracy of the monitoring/punishment system. Therefore, they can merge with cooperator populations accordingly, violating punishment and kin-selection mechanisms. Methods We used two agent-based simulation models to investigate the concepts of "cooperate for the spread" and "pay for the escape," both were net logo models created by Dr. Susan Hanisch. Afterward, we modified the first model to represent the concept of sharing the dispersal costs. We used the second model without modifications. Instead, we assigned definite values of some parameters that highlight the pay for the escape strategy. First model The original model was entitled "Evolution and patchy resource" 18 . She first developed it for educational purposes. It illustrates the concepts of cooperator-cheater competition, natural selection, spatial structure mechanisms, multilevel selection, and founder effects. Changeable variables Distance-resource-areas: the distance between the centers of the resource areas. Size-resource areas: the size of resource areas as a radius in the number of patches. Living costs: the costs that each agent has to deduct from energy per iteration for basic survival. Mutation rate: The probability that offspring agents have different traits than their parents. Evolution: the ability of agents to produce offspring. Constant variables The number of patches is 112 × 112 patches. Carrying capacity per patch: Resource = 10, Agents = 1 The growth rate of the resource = 0.2 The resources on a patch regrow by a logistic growth function up to the carrying capacity: New resource level = current resource level + (Growth-Rate × current resource level) × (1 - (Current resource level/carrying capacity)). The cost for producing offspring is ten subtracted units of energy. The initial level of energy of agents is set at living costs. Role of randomness Agents are distributed randomly in resource areas at the beginning of a simulation. Sustainable behavior is distributed randomly with a probability of percent sustainables among the initial agent population. The order in which agents move and harvest within one iteration is random. Agents move to a randomly selected patch if several patches fulfill the objectives. The order in which agents produce offspring within one iteration is random. Agents reproduce offspring with a probability of (0.0005 × Energy). Agents place offspring on a randomly selected unoccupied neighboring patch. Offspring mutate with a potential mutation rate. Model processes In each iteration, each agent moves around in random order. There are three likelihoods: If there are no unoccupied patches in a two-patch radius, they stay on the current patch. If there are unoccupied patches with resources amounting to more than living costs, the agents move to them. If the resource amount is less than the living costs, the agents move randomly to other unoccupied patches. The agents harvest the resources from separated patches to gain energy for metabolism and proliferation. If the energy level of any agent falls to zero, it dies. The cooperator type harvests half of the resource, while the greedy type consumes 99%. The living costs are deducted from the energy amount of the agent constantly everywhere all the time. This process occurs whether an agent moves within the patch, between the patches, or even not. Therefore, the model does not consider dispersal cost explicitly. If there is an unoccupied neighbor patch, the agent can reproduce with a probability of 0.0005 of his energy, place the offspring on the unoccupied neighbor patch, and then transfer ten units of the energy to his offspring. Resources regrow only on resource patches. When the resource amount is more than or equal to 0.1, then it regrows. When the resource is less than 0.1, its value is set to 0.1. Output diagrams and monitors The average energy of agents: average energy levels of sustainable and greedy agents, resulting from resource harvest minus living costs and reproduction. Trait frequencies: the relative frequencies of sustainable and greedy agents in the total population, resulting from mutations, different reproduction rates, and death. Agent population: the absolute number of the total population size resulting from reproduction and death. Modifications In the first modification, we added a different type of cost that agents only incur when they disperse from one patch to another (in-between the patches). It is the slider entitled "dispersal costs". In the second modification, we added another sharing dispersal costs tool to reduce them by dividing their value by the number of included agents (flock-mates) in the identified range from the same type. It is the slider entitled "group-dispersal-range." which is the flock mate's areas as a radius in the number of patches. Therefore, changing the value of the group dispersal range will change the area around every agent. Accordingly, the number of its flock mates who share the dispersal costs also adjusts. The group dispersal range is not confined to greedy agents but applies to all agents. Therefore, it represents the case of the wild-type cooperators who can also cooperate for the spread. The group dispersal range also does not only target the agents in between patches. However, it counts the agents inside and outside the patches. For example, once an agent starts its dispersion with a determined range containing ten agents, four from another type, three non-dispersal agents from the same type that existed inside a patch, and three dispersal agents from the same type outside the patches. The dispersal costs for this agent will be divided by 6. Our assumption that non-dispersal agents at the pre-departure stage share dispersion costs with dispersal agents; seems justified because they reap mutual benefits by reducing kin competition inside patches if they promote the migrators. However, can agents remotely pay the dispersion costs? Yes. For instance, some bacterial species can trigger the migration of other species if located in their vicinity, even if the two bacterial colonies are separated by a barrier 19 , 20 or if they are non-motile 21 . On the other hand, dispersion is an extended process with many factors, including escape from predators, suppression of host defense mechanisms, and production of biosurfactants to reduce surface tension to facilitate motility. Therefore, the agent's contribution (inside/outside the patches) to support such factors is considered a shared dispersal cost. Finally, cheaters can arise within cooperator patches by mutation or immigration. Therefore, to investigate the efficacy of migration, the mutation rate value should be 0 to cancel its effect in the meta-population dynamics. Second model The model is entitled "Evolution, resources, monitoring, and punishment." 22 is a simulation of a population with four types of agents competing for the same resource. It demonstrates many concepts, such as kin selection, cooperation, selfishness, public good, monitoring, punishment, sharing the costs, positive/negative frequency-dependent selection, and multilevel selection. The four agent colors and types: (1) Red: greedy, non-punishing. (2) Orange: greedy, punishing. (3) Turquoise: sustainable, non-punishing. (4) Green: sustainable, punishing. Punishing agents can perceive other agents in their environment to some degree (perception accuracy) and react to their behavior. There are three kinds of punishment: Punishers can kill agents with greedy harvesting behavior, stop them from harvesting in the next iteration, or have them pay a penalty fee to their neighbors. Agents have a cost (energy) to pay for, both detection and punishment, so this behavior is altruistic. Punisher agents of one type share punishment costs equally. Changeable variables Death rate: the probability that agents die independent of their energy level. Carrying capacity: the maximum amount of resource units on a patch from 1 to 100. Growth rate: the rate at which resources on patches regrow. The maximum sustainable yield is calculated based on the carrying capacity and growth rate. Harvest-sustainable: the number of resource units harvested by sustainable agents. Harvest-greedy: the number of resource units harvested by sustainable agents. Perception accuracy: the probability with which punishing agents notice greedy agents. Costs-perception: the costs in units of energy, punishing agents have to pay for perceiving other agents. Costs-punishment: the costs as units of energy that punishing agents have to pay in each iteration to punish other agents. All punishing agents of an agent divide the costs of punishment. Punishment: the kinds of punishing behavior that punishing agents perform. Fine: if the kind of punishment is "pay fine", the fine in energy units that punished agents have to pay (shared between all their neighbors). Living costs and mutation rate: see the first model. Constant variables The number of patches: There are 60 × 60 patches in the world. The initial energy level of agents is set at living costs + 1. The initial number of resource units on a patch is set to the carrying capacity. The resources on a patch regrow: see the first model. Role of randomness * In addition to items in the first model. Agents take on their traits (harvest preference and ability to notice and punish) randomly based on the probability of percent-sustainable and percent-punishers. The order in which punishing agents notice greedy agents within one iteration is random. Greedy agents are noticed by punishing agents with a probability of perception accuracy. The order in which detected greedy agents are punished within one iteration is random. Agents produce offspring with a probability of (0.001 × Energy). Agents die with a probability of (death-rate). Model processes In each iteration, each agent attempts to harvest resources from the patches it is on and the eight neighboring patches until the harvest preference level is reached, except for the punished agent with the sanction (suspend harvest once), its harvest amount = 0 in the current iteration. If the amount of resources available is lower than the amount that the unpunished agent attempts to harvest. Then, the agent moves to a neighboring unoccupied patch with the most resources after losing one energy unit as a move cost. Punishers pay the costs of perceiving the greedy agents. The greedy neighbors have been noticed with the probability of perception accuracy. The agent lost an amount of energy as living costs. The agent dies with the likelihood of death rate or if the energy level falls to zero. If there is an unoccupied neighbor patch, the agent can reproduce with a probability of 0.001 of its energy, place the offspring on the unoccupied neighbor patch, and then transfer half of its energy to its offspring that mutate according to the probability of the mutation rate. Resources regrow on all patches. When the resource amount is more than or equal to 0.1, then it regrows. When the resource is less than 0.1, its value is set to 0.1. Output diagrams and monitors Populations (% of carrying capacity): the state of the resource and the agent population in the world as a percentage of total carrying capacity resulting from resource harvesting behavior and resource regrowth, agent reproduction, and death. Average harvest per iteration: the average harvested amounts of agents per iteration by trait, resulting from harvested resource units, minus costs for monitoring and punishing (for punishing agents), minus fines (for punished agents in case of punishment “Pay fine”) The average energy of agents and trait frequencies: see the first model. How does the model represent a conditional defector strategy? The model aims to highlight the role of kin selection and punishment mechanisms in supporting cooperation evolution against cheats. We did not need to modify the model but just thought about what the conditional defector should do to upside down the game. The answer was to pay for the escape. For instance, if the standard Harvest-greedy of a cheater (greedy, non-punishing) was 13 and the Perception-accuracy of its actual punishers was 75%. Now suppose this cheater faces troubles, and it cannot dominate. However, if it gives up some of its profit to become 12, to escape punishment, and to reduce the perception accuracy to 60%, it could dominate and take over the population. The conditional cheater can pay something and reduce its profit to escape punishment by reducing perception accuracy if there is a positive correlation between these two variables. Therefore, this model is appropriate if it can support/deny such a correlation. Results All experiments were carried out under the Net Logo Behavior Space. All data analyses were carried out via a Python library called Glueviz and Excel . The experiments of the first model The default values of the variables: mutation rate = 0 (to investigate only the effect of dispersal). Dispersal costs = 8 (high value). The agent's shape is Bacteria. Size-Resource-Areas = 4 (Relatively small). Living costs = 1. Percent-Sustainables = 90% (most of the population initially consists of cooperators). Number-Agents = 80 (started number). Distance-Resource-Areas = 20, (Relatively far). The evolution switch is true (natural selection is working). Group dispersal range = 0, 30, 50, 70, 100, 150, and 200. Seven experiments were carried out with 63 runs. Fifteen repeated runs for group dispersal range = 0, and 8 repeated runs for each other value. Approximately all runs with group dispersal range = 0 finished in favor of cooperators and the extinction of cheaters; as expected, cheaters cannot sustain their patches and cannot arrange successful migrations to other patches due to the high dispersal costs. This situation significantly changed in the rest of the runs of group dispersal, ranging from 30 to 200, where cheaters could share the dispersal costs. Consequently, all these runs ended in favor of cheaters, and all cooperators were extinct. Figure 1 . Additionally, cheaters in these runs outcompete cooperators quickly with fewer steps, as long as the group dispersal range increases from 30 to 70. Then, the average number of steps is somewhat convergent for the group dispersal range from 70 to 200. Figure 2 . In addition, Fig. 3 . Figure 1 The final numbers of cheaters (red dots) and cooperators (blue dots) at different group dispersal ranges: cheaters could thrive only when they started to share the dispersal costs to some degree. However, when the group dispersal range = 0, each cheater pays the dispersal costs by itself. Therefore, cheaters cannot arrange successful migrations and cannot violate the spatial structure mechanism. Hence, they encounter local extinction at their patches. Full size image Figure 2 The runs that finished in favor of cooperators: (gray dots) most of these runs finished from 9000 to 30,000 steps by the complete extinction of cheaters, except one run reached the stop limit of our experiments at 50,000 steps, as three cheater agents succeeded to persist. The group dispersal range was 0 in all these runs. Therefore, cheaters cannot violate the spatial structure mechanism. The runs that finished in favor of cheaters by the complete extinction of cooperators: (1) (orange dots), runs finished after 8220 steps. (2) (green dots), finished from 5000 to 8220 steps. (3) (pink dots), finished before 5000 steps. Full size image Figure 3 Different group dispersal ranges: (blue dots), group dispersal range = 30. (dark green dots), group dispersal range = 50. (sky blue dots), group dispersal range = 70. (light green dots), group dispersal range = 100. (orange dots), group dispersal range = 150. (red dots), group dispersal range = 200. Cheaters outcompeted cooperators in all of these runs. However, the extinction of cooperators is likely to be done more quickly, with fewer steps in the higher group dispersal ranges. Full size image The results follow the intuitive predictions that cheaters could thrive, violate the spatial structure mechanism, and dominate the whole meta-population as long as they could cooperate to decrease the dispersal costs. The experiments of the second model. The default values of the variables: mutation rate = 1%. The kinds of Punishment are "suspended harvest once", "pay fine", or "kill". The fine if existed = 5. Carrying capacity = 100. Number Agents = 250, (started number). Costs perception = 0.5. Growth rate = 0.3. Costs-punishment = 0.8 Percent-Punishers = 20%, (started ratio). Harvest-sustainable = 7. Percent-Sustainables = 99% (most of the population initially consists of cooperators). Living costs = 4. Death rate = 1. Perception-accuracy%. Harvest-greedy (see Table 1 ). The runs were 15,000 ticks (iterations or steps) for the punishment types (suspended harvest once and pay fine). However, there were 30,000 ticks for the third type (Kill). The experiments began with a 99% frequency of cooperators, ending with greedy, non-punishing agents taking over the population. The final frequency of greedy non-punishing was above 90% in most runs, and the mean frequency of all steps was above 80% (Fig. 4 ). We excluded 100% accuracy, as it seems to us that there is no such perfect monitoring case in nature. In the first type of punishment (suspended harvest once), we began with 99% accuracy and then degraded to reach 30%, parallel to similar degradation in the greedy harvest amount from 15 to 9, Table 1 . In the second type of punishment (pay fine), we used the same values for greedy harvest amount and perception accuracy as the first type of punishment. The results of the two types were similar when the fines were five or less. In the third type of punishment (kill), we used different values for perception accuracy (from 70 to 30%) and greedy harvest amount (from 40 to 12) Table 1 . Figure 4 In the upper right, the frequencies of the greedy non-punishing agents through different experiments on three types of punishment: (1) Suspend harvest once (red dots). (2) Pay fine (green dots). (3) Kill (blue dots). The final frequencies of greedy non-punishing agents were above 90% in most runs (except one run for pay fine type was 88%). The mean frequencies of greedy non-punishing agents were above 80% for all runs. In the lower left, the frequencies of the other agents (brown dots): (1) Sustainable, punishing. (2) Sustainable, non-punishing. (3) Greedy, punishing. Full size image Table 1 Experimental details of the second model. Full size table Our findings demonstrate a strong positive correlation between the two variables (harvest greed and perception accuracy). The correlation coefficient (r) = 0.99 for the first and second punishment types. (r) = 0.95 for the third punishment type (Fig. 5 , Table 1 ). Figure 5 A strong positive correlation between the variables (harvest greed and perception accuracy) through different punishment types: (1) The first type, "suspend harvest once" (small blue squares). The second type, "pay fine", is the same. (2) The third type "kill" (large orange squares). (3) The blue line is the linear relationship of the selected values through the first/second type of punishment. (4) The orange line is the linear relationship of the selected values through the third type of punishment. Full size image The selected correlated values led to the dominance of greedy non-punishing agents through the three types of punishment. The dominance of cheaters here means they can violate kin selection and punishment mechanisms when they pay to escape punishment. Discussion The frequent dominance of conditional defection strategies in the computational experiments that we have conducted declares the failure of several well-established and crucial mechanisms responsible for augmenting cooperation, such as kin reciprocity, punishment, and spatiality. Therefore, it can be said that maintaining cooperation may be far harder than previously thought. The two models 18 , 22 we followed paved the way by encapsulating substantial problems of cooperation as well as the most crucial supporting mechanisms. These models aimed to assert the efficient role of kin selection, punishment, and spatial structure mechanisms for supporting cooperators against cheats. The novelty of the present article is that it takes the results to the reverse side by designing the conditional defection strategies and modifying the program codes of the previous models to involve the concept of sharing the dispersal costs. It also reveals the effective behavior of paying to escape different types of punishment. The two forms we presented in this article are simple and general. However, the concept of the conditional defection strategy is much broader. For example, the zero-determinant (ZD) extortion strategy is also a conditional defector strategy if it has a tag-based decision to cooperate with relatives who adopt the same (ZD) extortion strategy but cheat otherwise. At that time, it could be stable and win the game against the opponent's strategies 23 . In addition, when selfish strategies can modulate benefits and costs, they can outcompete tit-for-tat and generous strategies 24 . On the other hand, cheaters who can increase their dispersal rate without decreasing the dispersal costs often cannot achieve triumph. Therefore, they cannot drive the cooperators (wild type) to go extinct or even harm themselves if the benefits of exploitation do not offset the costs of dispersal. For instance, the social parasite of P. punctatus ants is a wingless cheater queen. Although it has a high dispersal rate, it has costly migration on foot for long distances. Therefore, the colonies persisted for a long time instead of the supposed rapid collapse of the whole population 25 . The findings of the present paper suggest a potentially therapeutic application. Conditional defectors can be used as suicidal agents to drive the population of pathogens into the self-destruction process. From an evolutionary perspective, tumors or microbes are considered populations of cooperating cells that struggle for survival by adopting many collective costly actions to produce the intrinsic common resources 26 , 27 , 28 . However, conditional defectors can violate the crucial mechanisms that support cooperation. Thereby, cheaters take over the population. Then they also go extinct after cooperators because they cannot do the necessary collective actions. Undoubtedly, the production of the common resources or the public good we meant is not independent of cooperators, as in the two models in the present paper. Instead, its production ought to rely on cooperators. For example, the essential excretions of microbes deplete after the extinction of the cooperator. Cheaters can drive the whole population to go extinct; it is a well-established evolutionary prediction. This robust outcome appears in many theoretical and empirical studies and is known as the tragedy of the commons or evolutionary suicide 29 , 30 , 31 . This phenomenon can occur if free riders have a fitness advantage over cooperators (wild-type) in an environment set by the cooperators. Creating evolutionary suicide within pathogen populations would mean the end of infections or even endemics, as cheaters are not static chemical substances but infectious transmissible organisms. It is not the first time someone has suggested using cheaters in attacking pathogens as a cooperator population. For instance, Brown et al. 32 suggested trojan horse therapy to reduce the virulence of pathogens or release beneficial medical substances inside its colonies. Weinberger et al. 33 suggested therapeutic interfering particles (TIPs) or hijacker therapy. It is a therapeutic utilization for defective interfering particles (DIPs) that are molecular parasites of viruses or incomplete RNA particles lacking essential packaging elements. It is believed to defeat HIV and other viruses (such as SARS-CoV-2). Moreover, DIPs are antivirals that can be transferred from one person to another until the end of endemicity in infected areas such as sub-Saharan Africa 34 . Archetti 35 suggested autologous therapy. It aims to increase the diffusion range of the growth factors that the tumor is excreting. Hence, this could increase the tumor's vulnerability to exploitation 36 . Domingo-Calap et al. 37 manipulated a defector strain of vesicular stomatitis virus called Δ51. It does not excrete a costly enzyme that suppresses interferon. So it could defeat the wild type. Then led to the tragedy of the commons. Other treatments and descriptive game-theoretic models of cancer are reviewed here 38 . To date, many previous papers have suggested closely related ideas. However, the defense mechanisms of cooperators are always a huge obstacle. We think now conditional defector strategies can surpass these obstacles. Data availability The datasets generated or analyzed during the current study include the three codes of Net Logo 6.1.1 (both codes of Dr. Susan and my own modified one) are available and accessible at the following CoMSES Computational Model Library: .

In the war between good and evil, I was there seeking to design powerful selfish strategies for investigating the defense mechanisms of cooperators. Surprisingly, the results serve a different domain of science. It gave me the hope to develop novel therapies using selfish strains as traitors betray their original species in favor of us. They commit treason inside cancer and microbial pathogen populations that usually resist traditional drugs, and destroy these harmful populations and drive them to go extinct. In a nutshell, I recruited villains to do something good. Unlikely beginning From an unlikely beginning, I took on the task of finding new treatments. The story began when I attempted to better understand how massive cooperative behavior evolved in populations of the living world under natural conditions in which selfishness was the only way to survive. Since Darwin's time, the problem of cooperation has preoccupied evolutionary biologists. How can cooperative behavior evolve and persist within the selfish natural world? This is "the most important unanswered question in evolutionary biology," according to Lord Robert May. Why does cooperation matter? Cooperation is fundamental not only for social insects, as you may think, but also for humans, bacteria, viruses, cancer cells, and every living organism. However, it is inherently prone to exploitation by cheats or defective free riders, who reap the highest profit without paying their share. Therefore, they invest more in proliferation than cooperators. Consequently, their numbers rise and take over the population. However, the realm of living organisms is still here despite the infinite attacks of cheats. Evolutionary biologists explain this phenomenon by suggesting mechanisms that promote cooperation against cheats as kin selection, punishment, multilevel selection, and spatial structure. Computational models I created agent-based simulation models to examine the substantial promoting mechanisms—the ones that have already been mentioned above—against some specific selfish strategies I designed and coined "conditional defector strategies." The conditional defector strategies were frequently dominant in the computational experiments, as they could outcompete the cooperating majority, which in turn declares the failure of the most crucial promoting mechanisms of cooperation. "Therefore, it can be said that maintaining cooperation may be far harder than previously thought," the study says. The conditional defector (red) significantly dominated the other cooperative and selfish strategies. Credit: Figure 2. Scientific Reports (2022). DOI: 10.1038/s41598-022-18797-2 Conditional defector therapy Although the new insights revive a long-standing puzzle in evolution theory, they also pave the way for a new therapy for incurable diseases. The study said, "Conditional defectors can be used as suicidal agents to drive the population of pathogens into self-destruction." From an evolutionary point of view, we can consider tumors or microbes as cooperating populations. They are clusters of cells wherein each cell incurs costly actions to produce essential substances in favor of all. They resist cheats by the mentioned defense mechanisms. However, conditional defectors can violate these mechanisms and outcompete natural cooperating strains. Ultimately, the entire population would die, as the selfish agents could not last after the extinction of cooperators due to the lack of the necessary collective actions, which is a phenomenon known as the tragedy of the commons. Creating the tragedy of the commons within pathogen populations means healing many uncontrolled incurable diseases. What are conditional defector strategies? The conditional defector strategy may be any cheating strategy that could occasionally cooperate, such as cooperating in some behaviors and cheating in others, or incurring additional costs to reap a group advantage even though still selfish in other behaviors. For instance: 1. Cooperate for spread The fundamental problem of ordinary cheaters is that they cannot arrange successful migrations. Therefore, their patches rapidly go to extinction. However, the conditional defection strategy can avoid this problem, as they cooperate only for the spread. They only share migration costs to decrease the portion of each participant. Hence, the conditional defectors have a high dispersal rate with the lowest possible price. Consequently, they could escape the local extinction threat. (Fig. 1.) 2. Pay for the escape "Conditional defectors can pay some of their wealth or waste some profits to escape punishment," the study says. For example, some individuals produce substances to mislead punishers, or possess a significant tag that marks cooperators to escape punishment. Therefore, conditional defectors can safely merge within the cooperator's populations without consequences. (Fig. 2.) The idea of using cheaters to attack pathogens as a cooperator population is not new; many papers have suggested similar ideas previously. However, cooperators' defense mechanisms are always a major obstacle. Now, conditional defector strategies can surpass these obstacles to provide promising therapeutic applications. This story is part of Science X Dialog, where researchers can report findings from their published research articles. Visit this page for information about ScienceX Dialog and how to participate.

10.1038/s41598-022-18797-2

Biology

New research could make ethanol production more efficient and economic

Deepak Kumar et al. Dry-grind processing using amylase corn and superior yeast to reduce the exogenous enzyme requirements in bioethanol production, Biotechnology for Biofuels (2016). DOI: 10.1186/s13068-016-0648-1 Journal information: Biotechnology for Biofuels

http://dx.doi.org/10.1186/s13068-016-0648-1

https://phys.org/news/2016-11-ethanol-production-efficient-economic.html

Abstract Background Conventional corn dry-grind ethanol production process requires exogenous alpha and glucoamylases enzymes to breakdown starch into glucose, which is fermented to ethanol by yeast. This study evaluates the potential use of new genetically engineered corn and yeast, which can eliminate or minimize the use of these external enzymes, improve the economics and process efficiencies, and simplify the process. An approach of in situ ethanol removal during fermentation was also investigated for its potential to improve the efficiency of high-solid fermentation, which can significantly reduce the downstream ethanol and co-product recovery cost. Results The fermentation of amylase corn (producing endogenous α-amylase) using conventional yeast and no addition of exogenous α-amylase resulted in ethanol concentration of 4.1 % higher compared to control treatment (conventional corn using exogenous α-amylase). Conventional corn processed with exogenous α-amylase and superior yeast (producing glucoamylase or GA) with no exogenous glucoamylase addition resulted in ethanol concentration similar to control treatment (conventional yeast with exogenous glucoamylase addition). Combination of amylase corn and superior yeast required only 25 % of recommended glucoamylase dose to complete fermentation and achieve ethanol concentration and yield similar to control treatment (conventional corn with exogenous α-amylase, conventional yeast with exogenous glucoamylase). Use of superior yeast with 50 % GA addition resulted in similar increases in yield for conventional or amylase corn of approximately 7 % compared to that of control treatment. Combination of amylase corn, superior yeast, and in situ ethanol removal resulted in a process that allowed complete fermentation of 40 % slurry solids with only 50 % of exogenous GA enzyme requirements and 64.6 % higher ethanol yield compared to that of conventional process. Conclusions Use of amylase corn and superior yeast in the dry-grind processing industry can reduce the total external enzyme usage by more than 80 %, and combining their use with in situ removal of ethanol during fermentation allows efficient high-solid fermentation. Background Due to increasing population and industrialization, global energy demand has increased steadily over the last few decades, and currently, about 80 % of this energy is derived from non-renewable fossil fuel supplies [ 1 ]. Transportation sector is one of the major consumers of the fossil fuels in the United States [ 2 ]. The concerns of depleting fossil fuel and the negative environmental impacts from their use necessitate the need to identify and develop renewable and sustainable energy sources. Bioethanol is considered as the most promising renewable transportation fuel, which can be produced in significant quantities from fermentation of sugars obtained from starch, sugary or cellulosic materials. United States is the biggest bioethanol producer in world with about 14.3 billion gallon (54.1 billion liters; 58 % of world production) production in year 2014 [ 3 ]. Most of the ethanol in the United States is produced from corn using dry-grind or wet milling process. Dry-grind is the most common used method for corn ethanol production [ 4 ]. In year 2014, about 5.4 billion bushels (25.4 kg in one bushel) of corn (37.8 % of total production) was processed in dry-grind industry [ 5 ]. Figure 1 illustrates the major steps used during laboratory scale conventional dry-grind process. The ground corn and water slurry is liquefied using α-amylase enzymes at high temperatures to convert starch into dextrins. The dextrins are further converted to glucose using glucoamylase (GA) enzymes during saccharification process, which is fermented to ethanol by yeast. Currently, these alpha and glucoamylases enzymes are added externally in liquid form during the liquefaction and saccharification process respectively. Saccharification and fermentation are performed in single step in the same reactor by process known as simultaneous saccharification and fermentation (SSF). Ethanol is recovered from the fermentation broth using distillation process. Remaining non-carbohydrate fractions in corn (germ, fiber, and protein) are recovered as a co-product called distillers dried grains with soluble (DDGS) at the end of the process. Fig. 1 Schematic of laboratory scale dry-grind corn process for ethanol production. Figure illustrates the steps followed during lab scale dry-grind processing for ethanol production from corn Full size image Over the last few decades, several advances have been made to improve the ethanol yields and profitability of the dry-grind process, including modifications in the production process [ 6 ], recovery of high-value co-products [ 5 , 7 ], use of advanced enzymes [ 8 , 9 ], and use of high-yield corn varieties [ 10 ]. A new corn developed by transgenic technology, known as amylase corn, produces an endogenous α-amylase in endosperm that is activated at high temperature and moisture [ 10 , 11 ]. Due to high expression levels of enzymes, only a small amount of the corn is required to be mixed with the conventional dent corn. Use of the amylase corn mix during the dry-grind process can eliminate the need of external addition of exogenous α-amylase. Similarly, a new engineered yeast, referred as “superior yeast” in this manuscript, is an advanced strain of Saccharomyces cerevisiae which expresses endogenous glucoamylases and provides novel metabolic pathways for high ethanol yields by reducing glycerol production. Use of this yeast can eliminate or alleviate the addition of expensive glucoamylase enzymes during SSF process, potentially improving the process efficiency, and reducing the overall ethanol production cost. Increasing the solid loadings during dry-grind process can be another approach to reduce the overall cost of ethanol production process. Using high-solid slurries in dry-grind process can decrease the overall energy use and process cost by reducing load on downstream processing of ethanol and co-product recovery and lowering the volumes of the processing equipment. However, the solid loadings during the ethanol process are restricted to 30–32 % w/w due to high viscosities, and yeast stress by high glucose and ethanol concentrations [ 12 – 14 ]. High-solid loadings can lead to higher final ethanol concentrations; however, low ethanol yields (liters/metric ton or gallons/bushel) are observed because of strong ethanol inhibition [ 15 ]. Simultaneous stripping off ethanol under vacuum during SSF process is one of the potential approaches to reduce the ethanol inhibition and achieve high-solid loadings [ 16 ]. With application of vacuum, ethanol can be evaporated at the normal fermentation temperature without affecting the yeast activity. Some studies on ethanol and butanol production have concluded that fermentation efficiencies can be improved significantly by applying only few cycles of vacuum [ 12 , 13 , 17 ]. Objectives of this work were to investigate the strategies to reduce external exogenous enzyme requirements during dry-grind process and improve ethanol yields at high-solid loadings. The fermentation characteristics of dent corn and amylase mix corn were evaluated using a superior yeast at various loadings of glucoamylase enzyme (0, 25, and 50 %), and the performance was compared with conventional yeast and glucoamylase used in the dry-grind process. The fermentation behavior of amylase mix corn using superior yeast was investigated using vacuum flashing process to achieve high ethanol yields by reducing ethanol inhibition at high-solid loadings. Methods Materials Conventional yellow dent corn was generously donated by a commercial seed company (DuPont Pioneer). The amylase corn was obtained from another commercial seed company (Syngenta Biotechnology, Inc., Research Triangle Park, NC). Corn samples were hand-cleaned and sieved using a 12/64″ (4.8 mm) sieve to remove broken corn and foreign materials. The cleaned corn was stored in refrigerator at 4 °C till analysis. The moisture content in corn was determined by drying the samples in hot air oven at 135 °C for 2 h (AACC International Approved Method 44-19.01) [ 18 ]. Starch content in the ground corn flour was determined using enzymatic assay (AACC International Approved Method 76-13.01) using the Total Starch Kit (Megazyme, Bray, Co. Wicklow, Ireland) [ 18 ]. The α-amylase and glucoamylase employed in this study were commonly used commercial enzymes. The α-amylase enzyme has an activity of 6400 µmol maltose/min mL. The glucoamylase enzyme activity has been reported 775 AGU/mL. Conventional active dry yeast (ethanol red) was obtained from the Fermentis-Lesaffre Yeast Corporation (Milwaukee, Wisconsin). The superior yeast was provided by the Lallemand Biofuels and Distilled Spirits (Milwaukee, WI). Dry-grind process The cleaned samples were ground in a laboratory scale hammer mill (model MHM4, Glen Mills, Clifton, NJ) at 500 rpm and using a 0.5-mm screen. Conventional dent corn and amylase corn were ground separately and later mixed to form a 15 % (by dry weight) amylase corn mixture, referred as “amylase mix corn” in this manuscript. All dry-grind experiments were performed at 250 mL scale in 500 mL stainless steel reactors in triplicate. Ground corn was mixed with deionized (D.I.) water to make slurry having 30 % solids on dry basis. For the liquefaction of control samples (100 % dent corn), the pH of the slurry was adjusted to 5.1 using 10 N sulfuric acid and 25.7 µL of α-amylase was used per 100 g dry corn, as per the manufacturer’s recommendations. The pH was not adjusted in case of amylase corn mix and no external α-amylase was added. The liquefaction was performed at 85 °C for 90 min using Labomat incubator with continuous agitation (Labomat BFA-12, Werner Mathis AG, Switzerland). It is important to note that heating and cooling time (heating and cooling rate of 3 °C/min) were in addition to liquefaction time (90 min). The pH of the liquefied slurry was adjusted to 4.8 using 10 N sulfuric acid for the SSF process. In control samples, yeast inoculum (2 mL), urea (0.4 mL of 50 % w/v solution), and GA (56.3 µL/100 g dry corn) were added, and the slurry was fermented at 32 °C for 72 h in an automatic incubator (New Brunswick Innova 42R Inc/Ref Shaker, Eppendorf, Connecticut) with continuous agitation at 150 rpm. Yeast inoculum was prepared by mixing 5 g of active dry yeast with 25 mL water and incubated at 32 °C for 20 min. SSF experiments using superior yeast were performed at three GA loadings (0, 25, and 50 % of recommended dosage). The superior yeast was inoculated at the rate of 0.176 g per liter of slurry (~50 µL for 250 mL slurry) as recommended by the manufacturer. Similar to the control experiments, urea solution was used as nitrogen source and slurry was fermented at 32 °C for 72 h in an automatic incubator with continuous agitation at 150 rpm. To monitor the fermentation, about 2 mL of sample was drawn at 0, 4, 8, 12, 24, 36, 48, and 72 h and centrifuged at 10,000 rpm (Eppendorf Centrifuge 5415 D, Eppendorf AG, Hamburg) for 10 min. The liquid was immediately filtered through 0.2 μm Acrodisc nylon syringe filters (Pall Life Sciences, Port Washington, N.Y.) into HPLC vials. The vials were frozen at −20 °C until further analyzed for sugar and ethanol content. Vacuum-assisted fermentation The vacuum-assisted fermentation experiments were performed using a lab scale modified vacuum-reactor system as shown in Fig. 2 . It consists of a 3 L modified jacketed fermenter, modified to accommodate thermocouples, agitating motor with stirring blades, and a sampling port. A dry vacuum pump (DryFast model 2044, Welch, Niles, IL) was used to create the vacuum in the fermenter. The system has the facility to condense the evaporated ethanol and water vapors by passing those through a coiled condenser (5977-19, Ace Glass, Vineland, NJ) with chilled liquid circulated at 1 °C. The condensate was collected in a 250 mL conical flask kept under low temperature using ice. For other constructional and operational details of the system, please refer to Huang et al. (2015) [ 17 ]. Fig. 2 Schematic of lab scale system for the corn fermentation with vacuum stripping system facility. The figure illustrates detail of vacuum-assisted fermentation system used in study Full size image Slurry at 40 % solids was prepared by mixing 500 g (dry basis) of 15 % amylase mix corn with calculated amount of D.I. water. The slurry was liquefied at 85 °C for 90 min in multiple 500 mL stainless steel reactors using Labomat incubator as described in the previous section. The liquefied slurry from multiple reactors was mixed in the 3 L fermenter, and pH was adjusted to 4.8 using 10 N sulfuric acid. The slurry was inoculated with 2 mL urea solution, 0.25 mL superior yeast, and 140.8 µL of glucoamylase (50 % of recommended dose for conventional yeast) and was incubated in water bath set at 32 °C for 72 h. Vacuum pressure at 6.7 kPa (28 in Hg gage) was applied for 1.5 h at 24, 36, 48, and 60 h of the fermentation. The vapors formed due to boiling of slurry were condensed and collected in 250 mL conical flask. A sample was withdrawn from each condensate to determine the ethanol concentration using HPLC. For fermentation profile, about 2 mL of sample was withdrawn at 0, 4, 8, 12, 24, 36, 48, and 72 h of fermentation from the slurry and prepared for HPLC analysis as explained earlier. The samples were also withdrawn after the application of vacuum and analyzed for the sugar and alcohol concentrations. Sample analysis (HPLC analysis) The fermentation samples were analyzed by high-performance liquid chromatography (HPLC; Waters Corporation, Milford, MA) using an ion-exclusion column (Aminex HPX-87H, Bio-Rad, Hercules, CA). The mobile phase was 0.005 M sulfuric acid at 50 °C with a flow rate of 0.6 mL min −1 . For each sample, a 5 μL injection volume was used with a run time of 30 min. The amounts of sugars, alcohols, and organic acids were quantified using a refractive index detector and using multiple standards. Ethanol yields and conversion efficiency Theoretical ethanol yields were estimated using Eqs. 1 and 2 , based on the starch content and free glucose of the corn, assuming complete starch conversion and 100 % fermentation efficiency. V_{{{\text{max\_EtOH}}}} = \frac{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)*\left[ {\left( {S*1.11*0.511} \right) + \left( {G*0.511} \right)} \right]}}{{\rho_{\text{EtOH}} }}, V max\_EtOH = W C ∗ ( 1 − MC C ) ∗ [ ( S ∗ 1.11 ∗ 0.511 ) + ( G ∗ 0.511 ) ] ρ EtOH , V_{{{\text{max\_EtOH}}}} = \frac{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)*\left[ {\left( {S*1.11*0.511} \right) + \left( {G*0.511} \right)} \right]}}{{\rho_{\text{EtOH}} }}, (1) E_{{{\text{Th\_EtOH}}}} = \frac{{V_{{{\text{max\_EtOH}}}} }}{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)}}, E Th\_EtOH = V max\_EtOH W C ∗ ( 1 − MC C ) , E_{{{\text{Th\_EtOH}}}} = \frac{{V_{{{\text{max\_EtOH}}}} }}{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)}}, (2) where V max_EtOH is the maximum possible volume of ethanol, mL; W C is weight of the corn, g; MC C is the moisture content in the corn; S is starch content; G is free glucose in corn; ρ EtOH is density of ethanol, 0.789 g/mL; E Th_EtOH is theoretical ethanol yield, L/kg dry corn; 1.11 is the gains during hydrolysis of starch; 0.511 is glucose to ethanol conversion ratio, kg/kg. Actual ethanol yields were determined by calculating liquid volume in final slurry after 72 h of fermentation. Weight of the final slurry was noted and a sample of the slurry was dried in hot air oven at 105 °C till constant weight achieved (~24 h) to estimate the solid percent in the slurry. The actual ethanol yields were calculated using Eqs. 3 , 4 , 5 . W_{\text{L}} = W_{\text{slurry}} *(1 - {\text{Solids}}_{\text{slurry}} ), W L = W slurry ∗ ( 1 − Solids slurry ) , W_{\text{L}} = W_{\text{slurry}} *(1 - {\text{Solids}}_{\text{slurry}} ), (3) V_{\text{EtOH}} = \frac{{W_{\text{L}} }}{{\rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} }}*C_{\text{EtOH}}, V EtOH = W L ρ H 2 O/EtOH ∗ C EtOH , V_{\text{EtOH}} = \frac{{W_{\text{L}} }}{{\rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} }}*C_{\text{EtOH}}, (4) E_{\text{EtOH}} = \frac{{V_{\text{EtOH}} }}{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)}}, E EtOH = V EtOH W C ∗ ( 1 − MC C ) , E_{\text{EtOH}} = \frac{{V_{\text{EtOH}} }}{{W_{\text{C}} *\left( {1 - {\text{MC}}_{\text{C}} } \right)}}, (5) where W L is the weight of liquid in the fermented slurry, g; W slurry is the weight of fermented slurry, g; Solids slurry is the solid fraction in the slurry; V EtOH is the volume of ethanol produced, mL; \rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} ρ H 2 O/EtOH \rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} is the density of water–ethanol mixture (g/L) at final ethanol concentration; C EtOH is the final ethanol concentration, mL/L; E EtOH is the actual ethanol yield, L/kg. Ethanol conversion efficiencies were calculated by dividing actual ethanol yields with the theoretical ethanol yield (Eq. 6 ). \eta_{\text{EtOH}} = \frac{{E_{\text{EtOH}} }}{{E_{{{\text{Th\_EtOH}}}} }} * 100. η EtOH = E EtOH E Th\_EtOH ∗ 100. \eta_{\text{EtOH}} = \frac{{E_{\text{EtOH}} }}{{E_{{{\text{Th\_EtOH}}}} }} * 100. (6) Statistical analysis The final ethanol concentrations, ethanol yields, starch to ethanol conversion efficiencies, and final glycerol concentrations during various treatments were statistically compared using analysis of variance and Fisher’s least significant difference (SAS version 9.3). The level selected to show the statistical significance in all cases was 5 % ( P < 0.05). Results and discussion Comparison of yellow dent corn and amylase mix corn Ethanol and glucose concentration profiles during fermentation of dent corn and amylase mix corn are illustrated in Fig. 3 . After 72 h of fermentation, average final ethanol concentrations for dent corn and amylase mix corn were 17.62 and 18.05 % (v/v), respectively. The small increase in final ethanol concentration for amylase corn could be due to relatively lower glucose inhibition. The peak glucose concentrations for yellow corn were much higher (13.8 %) compared to that from using amylase corn mix (8.22 %). The ethanol yield from amylase corn mix was calculated 0.444 L/kg dry corn (2.98 gal/bu), which was 4.1 % higher than that of dent corn. Most of the fermentation was complete in 48 h for both cases, observed by the small (<0.25 %) amounts of residual glucose, maltose, and maltotriose (Table 1 ). The results indicated that 15 % addition of amylase corn mixed with conventional corn can eliminate the need of exogenous liquefaction enzyme currently used in the dry-grind process. Fig. 3 Fermentation profile of dent corn and amylase corn mix. The figure provides the comparison of ethanol concentrations (% v/v) and glucose concentrations (% w/v) during SSF of 100 % dent and 15 % amylase corn mix. The data points in the figure are means of triplicate runs, and error bars represent standard deviations Full size image Table 1 Comparison of sugar concentrations during SSF process among yellow corn and amylase mix corn (mean ± standard deviation of triplicate runs) Full size table Performance of superior yeast SSF of conventional corn with superior yeast The ethanol and sugar production profiles during fermentation of conventional corn using conventional yeast at 100 % GA loading and superior yeast with various glucoamylase loadings are illustrated in Fig. 4 . Use of superior yeast even without any addition of glucoamylase (0 %) resulted in similar final ethanol yield as that of control ( P > 0.05), indicating that superior yeast has sufficient GA expression required to achieve similar ethanol profiles as with control (Table 2 ). One important factor for these results could be lower substrate inhibition to yeast. The glucose concentrations were relatively low throughout (1.41–5.24 % w/v) the fermentation process for 0 % GA loading, indicating relatively slow conversion of dextrins to glucose, which was simultaneously converted to ethanol by yeast. During initial 12 h of SSF, fermentation rates were very low for superior yeast for all GA loadings. Ethanol concentrations were observed higher by addition of 25 and 50 % GA along with the superior yeast (Fig. 4 ). Another major reason for high ethanol production using superior yeast was lower levels of glycerol production during fermentation process. The glycerol production was lower in all cases of superior yeast compared to that for conventional yeast (Fig. 5 ). Glycerol production is considered as an indicator of yeast stress, and typically about 1.2–1.5 % glycerol concentrations are observed in dry-grind ethanol fermentations [ 19 , 20 ]. In this study, for the superior yeast, maximum glycerol was observed 0.91 % at 50 % GA loading, which was still about 35 % less than that of control. In case of superior yeast use without any addition of GA, final glycerol was observed only 0.34 %, which was about 75 % less than that of control. The ethanol yields of dent corn fermented using superior yeast were in the range of 0.423–0.461 L/kg of dry corn (2.84–3.1 gal/bu). Maximum starch to ethanol conversion efficiency of 88.5 % was observed in case of 50 % GA addition (Table 2 ). Peak glucose concentration was maximum for superior yeast with 50 % GA addition. In case of superior yeast, it was observed that the peak glucose was observed at 12 h instead of at 8 h as in case of control, indicating relatively slow saccharification initially during SSF. Fig. 4 Concentrations of ethanol and glucose during fermentation of yellow dent corn in the conventional dry-grind process by conventional and superior yeast. Figure illustrates the fermentation profile of dent corn mix during SSF by superior yeast at various GA loadings and conventional yeast. Solid lines refer to ethanol concentrations (% v/v), and dotted lines refer to glucose concentrations (% w/v). The data points in the figure are means of triplicate runs and error bars represent standard deviations Full size image Table 2 Ethanol yields and conversion efficiencies (mean ± standard deviation of triplicate runs) Full size table Fig. 5 Comparison of glycerol concentration (% w/v) during SSF of dent corn among conventional yeast (control) and superior yeast at various GA loadings. The bars in the figure are means of triplicate runs and error bars represent standard deviations Full size image SSF of amylase mix corn with superior yeast The performance of superior yeast with amylase corn mix was similar to that of conventional corn. The peak glucose during amylase corn mix fermentation using superior yeast was observed at 12 h instead of at 8 h as in case of control, indicating relatively slow conversion (Fig. 6 ). Compared to those for conventional corn, overall glucose concentrations were low for all GA loadings for amylase corn mix, as observed with the conventional yeast also. Amylase corn mix fermented using superior yeast was considered as control for these experiments. The final ethanol concentration using superior yeast without any addition of GA was about 7.3 % lower than that of control. Addition of only 25 % GA resulted in high ethanol concentration (18.31 %), similar to that of control (18.05 %, using conventional yeast). These results indicate that combined use of amylase corn and superior yeast in the dry-grind process reduced the total external enzyme (α-amylase and glucoamylase) addition by more than 80 %, which would significantly reduce the processing cost. Ethanol concentration as high as 18.7 % was observed at 50 % GA addition along with superior yeast use. At this GA loading, ethanol yield was estimated 0.454 L/kg dry corn (3.05 gal/bu), about 2.35 % higher than that of control. Ethanol conversion efficiencies for amylase mix corn using superior yeast ranged from about 77.57 to 87.01 %. Similar to the case of dent corn, lower levels of glycerol production could have resulted in higher ethanol yields when using superior yeast (Fig. 7 ). In case of 25 % GA addition with use of superior yeast, final glycerol concentration (0.54 %) was 56.4 % lower than that for conventional yeast (1.24 %). Maximum glycerol concentration of 0.64 % was observed at 50 % GA loading, and was about 49 % less than that of control. The glycerol concentrations in all cases were lower than that of conventional corn. Fig. 6 Ethanol and Glucose concentration during SSF of amylase corn mix by conventional and superior yeast. Figure illustrates the fermentation profile of amylase corn mix during SSF by superior yeast at various GA loadings and conventional yeast. The data points in the figure are means of triplicate runs and error bars represent standard deviations. Solid lines refer to ethanol concentrations (% v/v), and dotted lines refer to glucose concentrations (% w/v) Full size image Fig. 7 Comparison of glycerol concentration (% w/v) during SSF of amylase corn mix among conventional yeast (control) and superior yeast at various GA loadings. The bars in the figure are means of triplicate runs, and error bars represent standard deviations Full size image Effect of solid loadings To examine the performance of superior yeast at high-solid loadings, amylase mix corn was also liquefied at 35 and 40 % solids, and the slurry was fermented using superior yeast with 50 % GA addition. Figure 8 illustrates the glucose and ethanol concentrations during fermentation at these solid loadings compared to those at 30 % solids. Although final ethanol concentrations at 35 % solids (19.28 %) were higher than that at 30 % solids (18.97 %), however, about 3.14 % glucose remained unconverted after 72 h of fermentation compared to complete conversion at 30 % solids. Final ethanol concentrations at 40 % solids were lower (17.1 %) than both 30 and 35 % solids and 10.5 % of glucose remained unconverted. The ethanol yields for 35 and 40 % solids were 0.358 and 0.268 L/kg dry corn (2.40 and 1.76 gal/bu), respectively, which were 21.14 and 42.0 % lower than that at 30 % solids. High viscosities and yeast stress due to high glucose and ethanol concentration reduce the yeast productivity and result in lower ethanol yields. In this study also, the peak glucose concentrations for 35 and 40 % solids were 1.55 and 1.42 times higher than that at 30 % solids. Fig. 8 Fermentation profile of amylase corn mix during SSF at various solid loadings. This figure illustrates the effect of solid loadings on the ethanol concentration (% v/v) and glucose concentrations (% w/v) of amylase corn mix during SSF using superior yeast and 50 % GA. The data points in the figure are means of triplicate runs and error bars represent standard deviations Full size image In situ ethanol removal during high-solid SSF Simultaneous stripping of ethanol during SSF process can reduce the ethanol inhibition and improve yeast activity. Preliminary experiments were performed to identify the suitable vacuum conditions (vacuum cycles and their frequency) for fermentation at 40 % solids. Application of vacuum for 1 h at 24, 36, and 48 h during fermentation resulted in relatively very high ethanol yields; however, still there were about 2.78 % glucose left unconsumed at the end of fermentation (Fig. 9 ). Even after removal of significant amount of ethanol during the fermentation process, the final ethanol concentrations were close to that of conventional fermentation (16.33 vs. 17.05 % v/v). Ethanol yield was calculated 0.38 L/kg (2.55 gal/bu), about 44 % higher than that of conventional fermentation at 40 %. Fig. 9 Fermentation profile of amylase corn mix using superior yeast during conventional and vacuum-assisted fermentation (vacuum for 1 h at 24, 36, and 48 h). Figure illustrates the comparison of glucose concentrations (% w/v) and ethanol concentrations (% v/v) during fermentation of amylase corn mix using superior yeast and 50 % GA among conventional and vacuum-assisted fermentation Full size image To further improve the fermentation efficiency, another vacuum cycle was added at 60 h and the vacuum time was increased to 90 min. Application of vacuum for 1.5 h at 24, 36, 48, and 60 h during SSF process resulted in complete fermentation compared to 10.5 % residual sugars in case of conventional process (Fig. 10 ). After vacuum application for 90 min, the ethanol concentrations dropped in the range of 10.4–41.9 mL/L, depending upon the ethanol concentrations at the start of vacuum application. The ethanol drop was higher than those in previous case with 60 min vacuum application (8.2–32.3 mL/L). The final ethanol yield with 82.89 % to ethanol conversion efficiency was estimated 0.433 L per kg dry corn, which was about 1.65 times that for the conventional fermentation at 40 % solids and only 4.6 % lower than that at 30 % solids. Similar results were observed by Shihadesh et al. for dent corn ethanol production using granular starch hydrolyzing enzymes (GSHE) and conventional dry active yeast [ 13 ]. The ethanol yields at 40 % solid fermentation with vacuum application produced similar ethanol yields as those of 30 % solids during conventional fermentation. Fig. 10 Fermentation profile of amylase corn mix using superior yeast during conventional and vacuum-assisted fermentation (vacuum for 1.5 h at 24, 36, 48, and 60 h). The figure illustrates the comparison of glucose concentrations (% w/v) and ethanol concentrations (% v/v) during fermentation of amylase corn mix using superior yeast and 50 % GA among conventional and vacuum-assisted fermentation Full size image The ethanol concentrations in the collected condensates ranged from 42.23–71.75 % (v/v), with an average of 57.1 % (v/v). This concentrated ethanol solution can potentially be directly guided to the rectification column during the distillation process for ethanol recovery, which can significantly reduce the energy load on the beer column (first stage of the ethanol recovery process) and overall cost of the dry-grind process. Conclusions Conventional dent corn and amylase mix corn were processed in dry-grind process using superior yeast that expresses glucoamylase and reduces the external enzyme addition. Only 15 % mix of amylase corn was sufficient to eliminate the need of α-amylase addition during liquefaction and achieve similar fermentation profiles. For yellow dent corn, no significant differences were observed in the ethanol yields between the control and using superior yeast without any external addition of glucoamylases. Use of superior yeast can significantly reduce the glucoamylase requirement, improve ethanol yields, and reduce the glycerol production. The vacuum flashing process successfully removed ethanol from the fermentation broth and resulted in complete sugar consumption for 40 % solid slurry. The ethanol yield of 2.9 gal/bu of dry corn with more than 80 % ethanol conversion efficiency was about 65 % higher than that at 40 % solids for conventional fermentation. The study provided a valuable insight about using amylase corn and superior yeast in the dry-grind processing industry and application of vacuum-assisted fermentation to improve fermentation at high solids. Abbreviations C EtOH : final ethanol concentration DDGS: distillers dried grains with soluble D.I.: deionized E EtOH : actual ethanol yield E Th_EtOH : theoretical ethanol yield G: free glucose in corn GA: glucoamylase GSHE: granular starch hydrolyzing enzymes MC C : moisture content of the corn S: starch content SSF: simultaneous saccharification and fermentation SY: superior yeast Solids slurry : solid fraction in the slurry V max_EtOH : maximum possible volume of ethanol V EtOH : volume of ethanol produced W C : weight of the corn W slurry : weight of fermented slurry ρ EtOH : density of ethanol \rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} ρ H 2 O/EtOH \rho_{{{\text{H}}_{ 2} {\text{O/EtOH}}}} : density of water–ethanol mixture

New research at the Integrated Bioprocessing Research Laboratory (IBRL) on the University of Illinois Urbana-Champaign campus could significantly change ethanol production by lowering operating costs and simplifying the dry grind process. "There are currently more than 200 dry grind plants that are processing corn to produce ethanol," says Vijay Singh, director of IBRL and a professor in agricultural and biological engineering. "The dry grind process requires two different enzymes to convert corn starch to glucose, which is further fermented to ethanol by yeast." Singh says that process has been simplified by combined use and optimization of three new technologies. "A new corn developed by transgenic technology, known as amylase corn, produces one of these enzymes in the grain itself, and a newly engineered 'superior yeast' provides the second enzyme, as well as fermenting the glucose. "There is a high expression level of the first enzyme, α-amylase, in the new corn, so only a small amount [15 percent was tested in these studies] of this corn is required to be mixed with conventional dent corn," Singh notes. "The superior yeast provides the second enzyme, glucoamylase, and also provides an alternate metabolic pathway to reduce by-product formation during fermentation. Combined use of this corn and superior yeast can reduce the total enzyme addition by more than 80 percent." Another approach to improve the dry grind process is to use high solids in the plant. However, according to Singh, high solid concentrations leads to high ethanol build-up in the tank. "High ethanol affects the yeast viability and inhibits its fermentation performance, so we have added a third technology to the process. We remove the ethanol as it is being produced, using a vacuum flashing process that is patented technology from the University of Illinois. Only a couple of vacuum cycles of 1 to 1.5 hours can bring the ethanol concentration below the inhibitory levels without affecting yeast health and allow complete fermentation of corn solids up to 40 percent," says Singh. Deepak Kumar, a postdoctoral research associate in agricultural and biological engineering, says because the dry grind process uses a significant amount of water, using more solid material in the slurry - 40 percent as opposed to 30-35 percent - means less water going into the process. "When ethanol is produced, it is in a very dilute solution. You have a small amount of ethanol and a large amount of water," says Kumar. "We cut down the water use by pushing high solids. When we reduce the amount of water, we also reduce the amount of energy required to remove the water." Singh believes this new research has the potential to improve the economics and process efficiencies and simplify the dry grind process. "By developing highly optimized technologies, we will benefit the entire dry grind industry," he concludes. Singh and Kumar received the 2016 Bioenergy Society of Singapore (BESS) Achievement Award for their work, in particular their paper "Dry-grind Processing using Amylase Corn and Superior Yeast to Reduce the Exogenous Enzyme Requirements in Bioethanol Production."

10.1186/s13068-016-0648-1

Biology

Cryptic sense of orientation of bats localized: The sixth sense of mammals lies in the eye

Lindecke O, Holland RA, Petersons G, Voigt CC (2021): Corneal sensitivity is required for orientation in free-flying migratory bats. Communications Biology. DOI: 10.1038/s42003-021-02053-w Journal information: Communications Biology

http://dx.doi.org/10.1038/s42003-021-02053-w

https://phys.org/news/2021-05-cryptic-localized-sixth-mammals-lies.html

Abstract The exact anatomical location for an iron particle-based magnetic sense remains enigmatic in vertebrates. For mammals, findings from a cornea anaesthesia experiment in mole rats suggest that it carries the primary sensors for magnetoreception. Yet, this has never been tested in a free-ranging mammal. Here, we investigated whether intact corneal sensation is crucial for navigation in migrating Nathusius’ bats, Pipistrellus nathusii , translocated from their migratory corridor. We found that bats treated with corneal anaesthesia in both eyes flew in random directions after translocation and release, contrasting bats with a single eye treated, and the control group, which both oriented in the seasonally appropriate direction. Using a Y-maze test, we confirmed that light detection remained unaffected by topical anaesthesia. Therefore our results suggest the cornea as a possible site of magnetoreception in bats, although other conceivable effects of the anaesthetic are also explored. Furthermore, we demonstrate that the corneal based sense is of bilateral nature but can function in a single eye if necessary. Introduction While the capacity for magnetoreception among mammals is evident from a number of behavioural experiments 1 , 2 , 3 , 4 , 5 , 6 , 7 , the anatomical location of the involved receptors remains as enigmatic as in any other animal to date 8 , 9 . Interestingly enough, when tested in darkness, mammals 10 , 11 , 12 , 13 , fish 14 , 15 and sea turtles 16 were able to orient by a magnetic polarity compass. The underlying magnetic sense is hypothesized to involve intra-cellular iron oxide, i.e., magnetite nanoparticles (Fe 3 O 4 ), which would be sensitive to the horizontal polarity of a magnetic field, enabling these animals to distinguish between magnetic north and south, independent of light. Intra-cellular iron oxide could also be responsible for magnetic signal transmission through control of ion channels depending on the alignment of animals in relation to the magnetic field 8 , 15 , 17 , 18 . Wegner and colleagues postulated that the cornea may be the location of the primary magnetoreceptors in mammals 19 . Specifically, they showed that in mole rats, Fukomys anselli , bilateral anaesthesia of the cornea resulted in randomly oriented nest-building, contrary to the usually magnetic polarity-dependent nesting behaviour 10 , 19 . According to the innervation of the cornea, the ophthalmic branch of the trigeminal nerve would transmit the magnetic signal to the midbrain where magnetic stimuli could be processed 11 , 13 , 20 , 21 , 22 . Yet, to date, the hypothesis of a corneal magnetic sense has never been challenged nor expanded from laboratory conditions to freely moving animals by performing a true navigation task in the field, e.g., during seasonal migration. Non-migratory bats are known to possess a polarity-sensitive magnetic compass, which they use for homing tasks 5 , 12 , 23 . Furthermore, results from a classic ‘Kalmijn-Blakemore’ pulse re-magnetization experiment in big brown bats ( Eptesicus fuscus ) are consistent with the hypothesis that magneto-sensory cells located somewhere in a bat’s body carry single-domain magnetite 24 . In contrast, the compass cues and sensory structures the migratory bats use for long-range in-flight navigation still remain undetermined. Only recently, it was demonstrated that bats calibrate their compass system to the solar azimuth at sunset and could take up a seasonally appropriate migratory heading after moderate displacement from their migration corridor 25 , 26 . To study the role of the cornea in navigation of vertebrates that are adapted to long-range navigation, we performed translocation experiments with 80 adult Nathusius’ bats ( Pipistrellus nathusii ) caught at the Baltic Sea during the late summer migration season. A geographical displacement of the bats was necessary to study their individual orientation behaviour at an unfamiliar site after astronomical twilight and when remote from the high density of conspecifics along the migration corridor, as well as the landmark cues emanating from the shore over short range. Half of the bats received either unilateral or bilateral topical corneal anaesthesia prior to release, while the other half were treated with a saline solution to create a sham control condition. Importantly, we also conducted tests of photoreception in another 76 bats using a Y-maze choice experiment to validate retinal function; specifically, the capacity of bats to still discriminate between light and dark in their environment, despite corneal anaesthesia, was tested. We hypothesized that migratory bats depend on corneal magnetoreception for navigation. If the cornea plays a role in magnetic orientation, we predicted that translocated bats treated with a topical anaesthetic on both eyes would vanish in random directions after release. However, bats with a single eye treated would be able to navigate because the other eye’s cornea would still be functional, i.e., to transmit sensory stimuli through the ophthalmic branch of the trigeminal nerve, and to enable the released bats to fly in a correct migratory direction similar to bats of the sham treated group. Results Detection of a light source is unimpaired after topical corneal anaesthesia In previous Y-maze experiments with one dark and one lit exit, the bats chose the lit exit instead of the dark one 27 . In contrast, blindfolded bats totally deprived of light perception chose the exits randomly 28 . We performed similar tests, yet without blindfolding, to evaluate our bats’ principal ability of light detection after administering topical corneal anaesthesia by oxybuprocaine eye drops. When we tested the unilateral and bilateral treatment groups that received the topical corneal anaesthetic, and the two sham control groups that received eye drops of saline solution bilaterally, our tests did not indicate a differential effect between these applications on the bats’ phototactic behaviour, i.e., animals of both the treatment groups and the sham control groups preferred the lit exit of the Y-maze (sham control 1: n = 22, 77% (proportion of bats choosing the lit exit in %), χ 2 = 6.55, W = 0.55, P = 0.011; single eye treated: n = 16, 81.3%, χ 2 = 6.25, W = 0.625, P = 0.0124; sham control 2: n = 16, 75%, χ 2 = 4.0, W = 0.5, P = 0.046; both eyes treated: n = 22, 86%, χ 2 = 11.64, W = 0.727, P < 0.001). Further, exit latency did not differ between bats with bilateral corneal anaesthesia and the respective sham control (both eyes treated, mean ± SD: 11.9 s ± 18.9 SD, median: 4.0 s; sham control 1, mean ± SD: 11.1 s ± 10.7 SD, median: 7.5 s; Mann–Whitney U -test: n = 44, U = 199.5, P = 0.321). Cornea sensation is crucial for accurate navigation after translocation Nathusius’ bats with their eyes untreated were previously shown to spontaneously vanish in a southerly, seasonally appropriate direction after experimental translocation during migration 25 , 29 . Here, we tested bats caught during their late summer migration at the Latvian Baltic Sea coast. The vanishing bearings of the two sham control groups were also significantly oriented towards the south (Rayleigh’s test, sham control 1, Fig. 1a : mean vector orientation 183° ± 34° (95% confidence intervals), n = 20, r = 0.495, Z = 4.91, P = 0.006; sham control 2, Fig. 1b : 187° ± 34°, n = 19, r = 0.502, Z = 4.78, P = 0.007) and the circular distributions obtained were best described by unimodal orientation models (Table 1 ). Fig. 1: Migratory Nathusius’ bats vanish in random directions if corneal sensation is inhibited in both eyes. a and b show control bats that received eye drops of saline solution as a sham treatment before release. c Experimental bats that randomly received a topical anaesthetic to the left or right eye’s cornea and sham treatment for the other eye, accordingly. d Bats with bilateral topical corneal anaesthesia. Empty and filled dots indicate animals that were tracked on the same nights. Arrows depict the group mean vectors in non-randomly oriented groups of bats with the magnetic North (0°) always on top of all plots. Grey sectors encompassing the group mean vectors indicate the 95% confidence intervals for the mean. P -values from Rayleigh tests are shown. Total sample size: n = 76. Full size image Table 1 Model-based analysis of bat orientation. Full size table There was no difference between the mean orientations and the variances around the mean vector in the two sham control groups (Mardia–Watson–Wheeler test, W = 0.189, P = 0.91). Bats of the experimental group that received corneal anaesthesia in one eye and sham treatment for the other also vanished in a southerly direction (Rayleigh’s test, single eye treated, Fig. 1c : 199° ± 37°, n = 19, r = 0.469, Z = 4.183, P = 0.013). Hence, the group mean vector did not differ from the mean of the respective sham control group (Mardia–Watson–Wheeler test, W = 1.011, P = 0.603). The variance of individual orientations around the mean also did not differ between the unilateral treatment group and the sham control one from the same migration season (Levene’s test, F 1,37 = 0.224, P = 0.639). In contrast to all other groups, bats released with bilateral topical corneal anaesthesia departed in random directions (Rayleigh’s test, both eyes treated, Fig. 1d : 240°, n = 18, r = 0.061, Z = 0.066, P = 0.937) and their circular distribution was best described by the uniform orientation model (Table 1 ). This lack of a preferred direction was distinguishable from the orientation of the control group ( p < 0.001: the bootstrapped 99.9% confidence interval for the r -value from the bilateral sham control group was 0.09 < r < 0.86, which does not overlap with the r -value of 0.06 in the bilateral anaesthesia group) and also significantly different from the other treatment group ( p < 0.001: the bootstrapped 99.9% confidence interval for the r -value from the unilateral anaesthesia group was 0.14 < r < 0.81, which does not overlap with the r -value of 0.06 in the bilateral anaesthesia group). The variance of individual orientations between bats that received bilateral anaesthesia and the respective sham control differed significantly (Levene’s test, F 1,35 = 5.824, P = 0.021). In addition, the variances around the means of the two groups that received corneal anaesthesia differed (single eye treated vs . both eyes treated: Levene’s test, F 1,35 = 5.310, P = 0.027). Experimental and sham control bats vanished promptly from the release site (mean values ± SD, single eye treated: 19.3 min ± 6, median=20 min; sham control 1: 16 ± 6 min, median=14.5 min; both eyes treated: 17.3 min ± 7, median=19.5 min; sham control 2: 16.1 min ± 6, median=16 min). Groups did not differ in the lengths of vanishing times (analysis of variance[ANOVA], F = 1.203, d.f.= 3, P = 0.135). Discussion To our knowledge, these experiments are the first to elicit a response in the navigation behaviour of a free-ranging mammal migrant without manipulating any sensory cues of the surrounding environment. Further, these data support, for the first time, the hypothesis of an orientation system in bats that relies on corneal sensitivity. Although direct evidence that this is an effect on the magnetic sense in this species is not yet available, it is consistent with previous work from microphthalmic mole rats, which suggests that such a sensory system could be part of a magnetic sense in mammals 19 . Briefly, when bats of our sham treatment groups were released after translocation from their migration corridor, flights were oriented in a seasonally appropriate migratory direction, which is in line with previous data from the same study location 25 . The same was also true when bats were deprived of corneal sensation in only one of their eyes. Yet, with both corneas made temporally insensitive, bats vanished in random directions but at the same speed as other bats. Our Y-maze study shows that under corneal anaesthesia the photoreceptive function of the retina was not neutralized, which meant that the bats could discriminate between light and dark. For take-off, bats would see enough to crawl out of the apparatus and through the preferred lit exit. Thus, upon release, free-ranging bats could have used some visual cues, yet the cornea-anaesthetized bats did not seem to use any visual cues that would enable them to pick their migratory direction. Similar disruption of orientation, independent of retinal impairment, has also been observed in migrating birds and in experiments with homing pigeons, when these encountered magnetic anomalies or fluctuations of the Earth’s magnetic field 30 , 31 , 32 , 33 , 34 . Also, domestic dogs abandoned their directional preferences for magnetic body alignment during excretion when the rate of change in declination of the Earth’s magnetic field changed 35 . Such disorientation responses were associated not only with an impaired magnetic compass but also with a malfunction of the “map sense” in animals, i.e., when they cannot obtain positional information 17 . This is supported by pigeons that were unable to compensate with other intact compass systems, such as a sun compass, when released in magnetic anomalies 30 , 32 . Recent evidence supports a “magnetic map sense” in birds based on magnetic iron particles that transmit magnetic field information through the trigeminal system 36 , 37 , 38 . Interestingly, such magnetic particles (magnetite) have also been found in the heads of different bats 39 , 40 , 41 , yet no physical link to any sensorial neuronal network has been established so far. However, magnetic pulsing, which should trigger re-magnetization of any magnetite-based sensor and, thus, provide directionally reversed magnetic compass or map information, led to deflections in adult homing bats that have established a map of their home range 24 . In migratory songbirds, disruption of the magnetic map sense (but not the magnetic compass 42 ) can be elicited by bilaterally cutting the ophthalmic branch of the trigeminal nerve 37 , 38 , 43 , which is the same branch whose sensation we blocked here through anaesthesia of the corneal nerve endings. Finally, magnetite particles are considered to support a magnetic polarity compass that is independent of light, which was observed in mammals, and also in bats 2 , 12 . Another possibility is that if Nathusius’ bats possess a star compass, the observed disorientation in the both-eyes treatment group would suggest side effects of the topical anaesthesia on their vision, e.g., on their visual acuity and, thus, the capacity to discriminate between stars in the sky. Although this possibility cannot be entirely excluded, two lines of evidence suggest it is a less likely explanation for our results. First, the anaesthetic oxybuprocaine is routinely used in mouse models for vision research where a stable ocular pressure and full retinal function are a prerequisite 44 , 45 , 46 , 47 , 48 , 49 , 50 . Side effects of the anaesthetic therefore are unlikely, but this alternative requires further testing in bats specifically. Second, although we do not have direct evidence for a magnetic compass in this species, no experiment on the compass system in any bats, or indeed any mammal, has yet provided positive evidence for a role for the stars: either as their primary mechanism of orientation, or as a calibration reference 5 , 23 , 26 . Future studies need to clearly establish the role of the magnetic sense in migratory bat navigation, and focus on the location and, particularly, the cellular mechanisms behind any trigeminal magnetosensor 51 . However, as nocturnal animals, bats have relatively large corneal surfaces and the cornea ranks among the most densely innervated tissues in the mammalian body, which renders it a promising organ for the search of biological “compass needles” 19 , 21 , 52 , 53 , 54 . Methods Animal subjects All work was conducted under the permits #10/2015, #31/2016, #33/2017-E and #3.6/85/2017-N-E issued by the Latvian Nature Conservation Agency to the Institute of Biology, University of Latvia. Over the course of three field seasons, we captured 156 adult Nathusius’ bats ( P. nathusii ), using a custom-made directional funnel trap (35 × 50 × 15 m; length × width × height) set up adjacent to the shoreline of the Baltic Sea at Pape Bird Ringing Station (PBRS; 56°09' N 21°03' E, Rucava Municipality, Latvia). Capture effort was most intense during the peak of the late summer migration season (between 14 Aug to 1 Sep 2015, 19 Aug to 23 Aug 2016, and 18 Aug to 4 Sep 2017). Bats were aged based on the closure of the epiphyseal gaps. While bats assigned to the retina function test ( n = 76) were only controlled for seasonally appropriate body mass (≥7.0 g), individuals assigned to the translocation experiment ( n = 80) were also transitionally ringed and measured for body mass and forearm length. Subsequently, animals were transferred to a keeping facility, where they were kept in groups of up to five individuals in wooden boxes (38 × 19 × 13 cm 3 ) in a dark and quiet environment, simulating a natural daytime roost in a tree hollow. Each evening the animals were fed. The duration of animal maintenance ranged from 2 to 5 days to secure, suitable release conditions for experimental nights (relatively high ambient temperature, no rain and low wind conditions). The retinal function experiments were conducted indoors and on the night subsequent to the capture of bats. Animals were housed in small groups and had no access to the natural night sky before release. Captive bats were fed individually with mealworms (larval stages of Tenebrio molitor , Coleoptera) during the evening hours and provided ad libitum water. Prior to feeding on experimental evenings, bats also received three small drops of saline solution (NaCl) into the nostrils, as they served as a control group for another study, which, however, did not require any additional experiments for the individuals of this study. That way, we also guaranteed blind study procedures. We do not expect an effect on visual performance and corneal sensation from this nasal treatment. Topical anaesthesia of the cornea Bats were gently held in an upright position and treated with one drop of oxybuprocaine hydrochloride (0.4%, Novesine®, Novartis, Germany) to the central cornea using a pipette. We chose this topical anaesthetic over lidocaine, which is commonly applied in studies of orientation physiology, because lidocaine is known to occasionally cause ophthalmic side effects in birds and mammals, including visual impairment, when penetrating deep into tissues, potentially crossing the blood–brain barrier 51 , 55 , 56 , 57 . Oxybuprocaine is different to lidocaine as it numbs only the outermost layers of the cornea while leaving the retina unaffected; however, its anaesthetic efficiency decreases after 30 min and thus the sensory impairment is quickly reversible 58 , 59 . For these reasons, oxybuprocaine is therefore routinely used in human and veterinary ophthalmology 44 , 45 , 46 , 47 , 48 , 49 , 50 , 58 , 59 , 60 . As a control, i.e., for a sham treatment, we used eye drops of sterile saline solution (NaCl 0.9%, B. Braun Melsungen AG, Germany), which is a standard in both human and veterinary ophthalmology, and eye care 59 . The bilateral treatment group received corneal anaesthesia to both eyes. Bats from the unilateral, i.e., single-eye treatment group, and control bats (sham control groups 1 and 2) received a drop of the sham treatment to the contralateral or both eyes, respectively. After 20 s of exposure, any supernatant was gently removed from the surface of the eye using sterile tissue, and only then the contralateral eye was treated. It is noteworthy that bats neither blinked during this procedure, nor did they show any signs of discomfort, such as emitting of distress calls or spontaneous movements. The choice for the individual cornea treatment was made in a blinded fashion, with students assisting the experimenter. The experimenter received two identical unlabelled pipettes and a note on lateral allocation for the application of eye drops. The left–right ratio of the unilateral treatment was kept at balance over the course of the study period, yet lateral allocations followed a randomized order each night. Behavioural testing started immediately after the eye drops were applied. Bats assigned to the navigation experiment received the treatment only after translocation, just before individual releases. Testing retinal function and phototactic behaviour To make sure topical corneal anaesthesia did not completely abolish the bats' visual capability, we tested 76 bats for phototactic responses in a Y-maze task. Bats are known to choose lit exits over dark ones for emergence from Y-mazes 27 , 28 . We compared the bilateral anaesthesia treatment ( n = 22; 14/8 males/females) with a sham control ( n = 22; 10/12; “sham control 2”) in 2015, and the unilateral treatment ( n = 16; 5/11) with another sham control ( n = 16, 5/11, “sham control 1”) in 2017. Tests were performed indoors at PBRS, at room temperature, and were performed between 0300 and 0600 h over the course of two nights in both years. Experimental individuals were kept in wooden boxes until tested. The maze apparatus was made out of plywood, and was inclined towards the exits 10° following recommendations of previous works 28 . The Y-maze had an arm length of 200 mm; cross-sectional dimensions of the runway were 80 × 60 mm 2 (width × height). All surfaces were coloured dark-brown to minimize light reflections. For the floor, an easy-to-clean PVC coating with a structured surface was used, which was not slippery for crawling bats. The entrance of the Y-maze had a light level of 0.02 lx. Dim light (120 lx) was provided at the exit of one arm using three commercial white torch LEDs indirectly illuminating the space behind the exit, while the exit of the other arm was kept dark (0.01 lx). The area of the bifurcation inside the Y-maze was illuminated indirectly (0.12 lx) via the lit arm. Each bat was transferred manually to the acclimatization compartment of the Y-maze, directly after corneal anaesthesia or sham treatment, respectively. Besides the Y-maze illumination, the testing room was kept dark. After 20 s for acclimatization, a bat had to crawl a 100 mm runway to reach the bifurcation. We timed the emergence latency. Bats of both groups were tested in alternate order, with the lit arm of the maze changed after the first half of bats had been tested. Clean sheets soaked with ethanol (70%) were used to clean the runways between trials. Individuals were tested only once and released in the nearby coastal forest after 1 h to ensure that anaesthesia had ceased before bats were free again. When dawn was approaching, bats were kept for the next day, fed and watered in the evening and released immediately after that at the site of capture. Emergence latency was compared using the Mann–Whitney U -test since data were not normally distributed ( P < 0.05). Directional choices for exits of each group were analysed for a preference using a test of goodness of fit (chi-squared test; R version 3.2.1, package shiny ). Testing navigational performance after translocation and corneal anaesthesia We used 80 adult P. nathusii (36 males, 44 post-lactating females) for the release experiment. On the day of the translocation, bats were fed and watered from 1800 to 2000 h. Subsequently, they were equipped with VHF radio transmitters (operating frequency wavelengths: 150.00–152.00 MHz; LB-2XT, Holohil Systems Ltd., Ottawa, Canada, 0.31 g; V1 and V3, Telemetrie-Service Dessau, Dessau-Roßlau, Germany, 0.35 g; Pip Ag337 and PicoPip Ag379, BioTrack Ltd., Wareham, UK, 0.35 and 0.43 g). One radio transmitter was glued onto the fur of the lower dorsum of each bat using skin glue (Manfred Sauer GmbH Hautkleber, Lobbach, Germany). Transmitters were selected so that the mass of the tag was lower than 5% of the individual body mass. Until translocation to the release site, bats were kept individually in large cloth bags to allow acclimatization to the tag. Translocation and releases were performed between 2300h and 0400 h of a given night and over the course of 26 nights. The release site was a flat field about 11 km east of the capture site and outside the coastal migration corridor where bats were caught. The location offered a clear line of sight of the horizon for 360°. To increase the motivation to continue migratory transit flights, bats were offered water and mealworms again prior to release but before any cornea treatments. The person who tracked the animals was blind to the treatment conditions. To achieve this, the assisting personnel randomly chose the substances to be applied, i.e., chose the test group, and consequently provided the experimenter with one pipette per eye for applications. Thereby, we ensured unbiased measurements of vanishing bearings. We aimed to release an even number of bats per group and per night. Only the assisting personnel tracked the sequence of experimental and control bat releases on a given night and could balance the number and succession of releases of both groups of treatments. Before treatment and release from the roof of the car, we surveyed the vicinity of the site for the presence of any other bats using a bat detector (Echometer EM3 + , Wildlife Acoustics, Inc., Maynard, MA, USA). If any bat would have been recorded, the experiment would have been paused to avoid confounding via eavesdropping. After the cornea treatment and prior to releasing, the surrounding was surveyed for bats again for 1 min. In the absence of bat activity, test bats were offered to take-off at their own speed while the release direction was chosen randomly. Bats were then tracked at about 4 m above the ground using a handheld three element Yagi antenna attached to an Australis 26k receiver (Titley Scientific). When the signal of the radio transmitter had vanished, the bearing of the fading signal and the time elapsed since the release was noted. After 2 min, we confirmed the absence of bats by monitoring the area for the individual radio signal again. This was also repeated for all individuals of the given night after the last bat had vanished. The next night, a complete scan for all frequencies was repeated before any new bat was released. For statistical comparisons we did not include data from bats that took >30 min for vanishing ( n = 4) because the full efficiency of the corneal anaesthesia lasts for half an hour 44 , 47 , 49 , 58 , 59 . Also, in a previous study, P. nathusii vanished from the tracking range in less than 20 min from the same release site 25 , indicating that significantly longer vanishing times most likely represent outliers. The mean bearings and vector lengths of each group were calculated using the Oriana 4.02 circular statistics software package (Kovach Computing Services). Groups were tested for departure from a uniform circular distribution using the Rayleigh’s test 61 . In order to further evaluate the distribution of bearings of our experimental groups, in particular the pattern of the non-unimodally oriented group with bilaterally topical anaesthesia, we applied a likelihood-based modelling approach (package CircMLE, R version 3.5.2) that has recently been introduced to compare circular data with multiple potential models of orientation behaviour 62 . Beyond the uniform distribution representing random scatter of bearings (M1), these models comprise three unimodal variants (ordinary, M2A; symmetric modified, M2B; modified unimodal, M2C) and six bimodal variants of distributions (homogenous symmetric bimodal, M3A; symmetric bimodal, M3B; homogenous axial bimodal, M4A; axial bimodal, M4B; homogenous bimodal, M5A; and bimodal, M5B) 63 . For each experimental group, resulting models were then compared by means of the corrected Akaike information criterion (AICc) and the corresponding model weights 64 . Tests for significant differences between group orientations were performed using the Mardia–Watson–Wheeler test. For a more sophisticated comparison of the directedness between the bilateral treatment group and the other two significantly unimodally oriented groups (the unilateral treatment group and sham control 2; Fig. 1b, c ), we followed a recently introduced bootstrap technique 65 . For this, the mean resultant vectors ( r -values) of different experimental groups are used to observe whether the r -value of a non-significantly oriented group falls within some confidence intervals of another r -value that derives from a significantly oriented group. To do so, a random subsample of n orientation angles is drawn with replacement from a significantly oriented sample of n orientation angles present in the significantly oriented treatment groups ( n = 19 for both the unilateral treatment group and the sham control 2). Then the corresponding r -value is calculated based on these n = 19 orientation angles. With a new randomization each time, this procedure is repeated 100,000 times. The resulting 100,000 r -values are ranked lowest to highest. The r -values at the ranks 2500 and 97,500, 500 and 99,500, and 50 and 99,950 define the 95%, the 99%, and 99.9% confidence limits for the observed r -value of the significantly oriented group, respectively. If the r -value observed in the actually non-significantly oriented group lies outside these confidence intervals, the significantly oriented group is significantly more directed than the non-significantly oriented group with a significance of p < 0.05, p < 0.01, or p < 0.001, respectively. Navigational accuracy between groups was assessed by testing for homogeneity of variances across groups, i.e., the scatter of the bearings. For this, the original bearings were transformed to absolute residuals from the group-specific orientation mean. With these we computed a Levene’s test 66 , which does not assume underlying normality of the data (R package car version 2, R). Departure flight times were compared using an ANOVA Kolmogorov–Smirnov test for normality passed, P = 0.083). Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request.

Mammals see with their eyes, hear with their ears and smell with their nose. But which sense or organ allows them to orient themselves on their migrations, which sometimes go far beyond their local foraging areas and therefore require an extended ability to navigate? Scientific experiments led by the Leibniz Institute for Zoo and Wildlife Research (Leibniz-IZW), published together with Prof. Richard A. Holland (Bangor University, UK) and Dr. Gunārs Petersons (Latvia University of Life Sciences and Technologies) now show that the cornea of the eyes is the location of this important sense in migrating bats. If the cornea is anesthetized, the otherwise reliable sense of orientation is disturbed while light detection remains unimpaired. The experiment suggests the localisation of a magnetic sense in mammals. The paper is published in the scientific journal Communications Biology. A research team led by Dr. Oliver Lindecke and PD Dr. Christian Voigt from Leibniz-IZW has demonstrated for the first time that environmental signals that are important for navigating over long distances are picked up via the cornea. They conducted experiments with Nathusius' bats (Pipistrellus nathusii) during the late summer migration period. In bats of one test group, the scientists locally anesthetized the cornea with a drop of oxybuprocaine. This surface anesthetic is widely used in ophthalmology to temporarily desensitize the patient's cornea when the eyes are overly irritated. Effects on orientation, however, had not been previously recorded. In another test group of bats, the research team anesthetized the cornea of only one eye. The individuals in the control group were not anesthetized, but instead received an isotonic saline solution as eye drops. All animals in this scientific experiment were captured within a migration corridor at the coastline of the Baltic Sea and released singly in the open field 11 kilometers inland from the capture site immediately after treatment. The scientists first used bat detectors to make sure that there were no other bats above the field at the time of release that the test animals could have followed. The person observing the direction of movement of released bats was unaware of how bats were treated experimentally. "The control group and the group with unilateral corneal anesthesia oriented themselves clearly in the expected southerly directions, whereas the bats with bilateral anesthetized corneas flew off in random directions," explains Dr. Oliver Lindecke, first author of the paper. "This evident difference in behavior suggests that corneal anesthesia disrupted a sense of direction, yet orientation apparently still works well with one eye." As corneal treatment wears off after a short time, the bats were able to resume their journeys south after the experiment. "We observed here for the first time in an experiment how a migrating mammal was literally blown off course—a milestone in behavioral and sensory biology that allows us to study the biological navigation system in a more targeted way." In order to rule out the possibility that the anaesthetisation of the cornea also affects the sense of sight and that the scientists would thus come to the wrong conclusions, they carried out a complementary test. Once again dividing the bats into experimental and control groups, the researchers tested whether the response of bats to light changed after anesthesia of the corneas on one or both sides. "We know from previous research that bats prefer an illuminated exit when leaving a simple Y-shaped labyrinth," explains PD Dr. Christian Voigt, head of the Leibniz-IZW Department of Evolutionary Ecology. "In our experiment, the animals with one-sided or two-sided anesthesia also showed this preference; we therefore can rule out that the ability to see light was altered after corneal treatment. The ability to see light would, of course, also influence long-distance navigation." Many vertebrates such as bats, dolphins, whales, fish and turtles, for example, are able to safely navigate in darkness, whether it is under the open night sky, when it is cloudy at night or in caves and tunnels as well as in the depths of the oceans. For many decades, scientists have been searching for the sense or a sensory organ that enables animals to perform orientation and navigation tasks that seemed difficult to imagine for people. A magnetic sense, so far only demonstrated in a few mammals but poorly understood, is an obvious candidate. Experiments suggest that iron oxide particles within cells may act as microscopic compass needles, as is the case in some species of bacteria. Recent laboratory experiments on Ansell's mole rats, relatives of the well-known naked mole rats that spend their lives in elaborate underground tunnel systems, suggest that the magnetic sense is located in the eye. This magnetic sense of orientation has not been verified in migratory mammals nor has it been possible to identify the specific organ or tissue which could provide the morphological basis for the required sensory receptors. The experiments by the team around Lindecke and Voigt now provide the first reliable data for the localisation of a sense of orientation in free-ranging, migratory mammals. Exactly what the sense in the cornea of the bats looks like, how it works and whether it is the long-sought magnetic sense must be shown in future scientific investigations.

10.1038/s42003-021-02053-w

Medicine

Therapeutic potential of Mozart for medication-resistant epilepsy

Musical components important for the Mozart K448 effect in epilepsy, Scientific Reports (2021). DOI: 10.1038/s41598-021-95922-7 , www.nature.com/articles/s41598-021-95922-7 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-021-95922-7

https://medicalxpress.com/news/2021-09-therapeutic-potential-mozart-medication-resistant-epilepsy.html

Abstract There is growing evidence for the efficacy of music, specifically Mozart’s Sonata for Two Pianos in D Major (K448), at reducing ictal and interictal epileptiform activity. Nonetheless, little is known about the mechanism underlying this beneficial “Mozart K448 effect” for persons with epilepsy. Here, we measured the influence that K448 had on intracranial interictal epileptiform discharges (IEDs) in sixteen subjects undergoing intracranial monitoring for refractory focal epilepsy. We found reduced IEDs during the original version of K448 after at least 30-s of exposure. Nonsignificant IED rate reductions were witnessed in all brain regions apart from the bilateral frontal cortices, where we observed increased frontal theta power during transitions from prolonged musical segments. All other presented musical stimuli were associated with nonsignificant IED alterations. These results suggest that the “Mozart K448 effect” is dependent on the duration of exposure and may preferentially modulate activity in frontal emotional networks, providing insight into the mechanism underlying this response. Our findings encourage the continued evaluation of Mozart’s K448 as a noninvasive, non-pharmacological intervention for refractory epilepsy. Introduction Epilepsy impacts approximately 1% of the global population, and of these people, 1/3 suffer from medication-resistant or refractory epilepsy 1 . Besides seizures and their associated comorbidities, persons with epilepsy experience interictal epileptiform discharges (IEDs). IEDs arise from the brief, synchronous firing of neural populations that are typically involved with epileptic networks 2 . These IEDs are known epileptic biomarkers that are associated with seizure frequency and impaired cognition 3 , 4 , 5 , 6 . Thus, IED-related interventions may provide insight into novel therapies for epilepsy and its related comorbidities. An IED-related intervention with accumulating evidence is the use of music as a noninvasive, non-pharmacologic form of neuromodulation 7 , 8 . Specifically, Mozart’s Sonata for Two Pianos in D Major (K448) has been shown to reduce ictal and interictal epileptiform activity in several scalp-EEG and fMRI studies 9 , 10 , 11 , 12 . While effect sizes varied, a meta-analysis demonstrated that approximately 84% of subjects had significant IED reductions during Mozart’s K448 13 . This reputed “Mozart K448 effect” was first described in 1993 by Rauscher et al. 14 when they demonstrated enhancement on a spatial task during exposure to K448. Later, Hughes et al. (1998) 15 were the first to witness the “Mozart K448 effect” in persons with epilepsy by showing that K448 was associated with reduced epileptiform activity. Following Hughes et al.’s discovery, there has been continued support for the “Mozart K448 effect” in epilepsy research—generally demonstrating that exposure to K448 was associated with some therapeutic reduction in seizures and IEDs 7 , 9 , 12 , 16 , 17 . Apart from one other composition—Mozart’s Piano Sonata in C Major (K545)—the therapeutic properties of K448 could not be replicated with other musical stimuli 18 . Stimuli previously tested were other Mozart compositions 16 , Beethoven’s Fur Elise 19 , and a string version of K448 10 . This led to several theories about the mechanisms underlying Mozart’s therapeutic effects for epilepsy; however, the specific properties driving the “Mozart K448 effect” remain unknown. Consequently, there is a general reluctance to fully accept this effect due to the unknown mechanism of K448 and to heterogeneous past findings that are likely linked with the use of different study protocols and inferior imaging modalities. The latter limitation is noteworthy, as scalp-EEG is much less sensitive for quantifying epilepsy-related outcomes, especially interictal events 20 , 21 . Our previous work demonstrated that 40 Hz auditory stimulation could reduce IEDs in subjects with refractory epilepsy and high baseline IED rates 22 . Historically, the relevance of gamma sensory stimulation emerged from findings of reduced gamma oscillations in humans with Alzheimer’s disease (AD) 23 . This was followed by observations of improved disease states (e.g., AD 24 , 25 , 26 and stroke 27 ) after exposure to exogenous gamma stimulation. Lin et al. (2010) 10 even demonstrated that musical stimuli with more fundamental tones (i.e., higher gamma power) reduced the number of epileptiform discharges. A major pitfall to this noninvasive intervention is that while the 40 Hz tone could effectively reduce IEDs in refractory epilepsy, it was not especially pleasant to listen to for a prolonged time. In this study, we evaluate the use of Mozart’s K448 to see (1) if we can validate previous scalp-EEG findings with intracranial Stereo-EEG in adults with refractory epilepsy, (2) if there is a temporal dependence for eliciting the “Mozart K448 effect”, and (3) if the “Mozart K448 effect” is associated with preferential brain networks. We also examined if preferred music and music with enhanced gamma frequencies (i.e., either gamma-matched to K448 or gamma-boosted) could elicit a therapeutic response; this was motivated by our past findings and the theory that increased fundamental frequencies may be beneficial for epilepsy 10 , 22 . We hypothesized that eliciting the “Mozart K448 effect” would be dependent on a longer stimulus duration and prolonged internal musical segments. This is based on the theory that emotional responses result from positive reward prediction errors 28 . Further, we expected this effect would extend to regions outside of the primary auditory pathways, owing to past observations of music and its involvement with higher order systems (e.g., emotion and mirror neurons) 29 . This research may guide future work in uncovering how Mozart’s K448 elicits therapeutic responses, which may facilitate the development of novel, noninvasive music therapies for refractory epilepsy. Results Detecting interictal activity in subjects with refractory epilepsy An automated template-matching interictal epileptiform discharge detector was utilized to calculate subject-specific IED rates (Fig. 1 ). We recruited 16 neurosurgical subjects undergoing clinical monitoring for refractory epilepsy to participate in sessions of a music task (Fig. 2 a,b). To determine if the duration of exposure was an important factor for eliciting the “Mozart K448 effect”, stimuli were presented for either 15-s (“Group 15”) or 90-s (“Group 90”). Subjects in Group 15 had a mean age of 43.75 (SD 16.46), an average normalized baseline IED rate of 1.23 (SD 1.09), and 50% were male. Group 15 subjects had 32.5 (SD 14.40) electrodes implanted in the left hemisphere and 35.86 (SD 12.78) electrodes in the right hemisphere. Subjects in Group 90 had a mean age of 34.88 (SD 10.02), an average normalized baseline IED rate of 1.43 (SD 0.94), and 75% were male. Group 90 subjects had 38 (SD 22.67) electrodes implanted in the left hemisphere and 33.38 (SD 22.77) electrodes in the right hemisphere. Subjects from both groups performed 1.81 (range 1–2, SD 0.40) 25-min sessions on average. Other subject demographic and clinical characteristics are provided in Table 1 . Figure 1 Automated spike detector pipeline. A template-matching IED detector first cross-correlated a 60-ms triangular template with preprocessed Stereo-EEG, then normalized the cross-correlation by the median standard deviation from 1-s sliding windows. The absolute value of the normalized cross-correlation was then used to mark local peaks above a specified threshold as IEDs. Full size image Figure 2 Task structure and validation method. ( a ) Trials consisted of auditory stimuli selected randomly without replacement, each presented for 15-s, followed by a 15-s rest period (“Group 15”). ( b ) Trials consisted of musical stimuli selected randomly without replacement, where a simultaneous attention task was performed during the final 30-s of each auditory stimulus. This was followed by a 60-s control period and a True/False question assessing whether the subject attended to the auditory stimulus (“Group 90”). ( c ) A marginal model (GEE) revealed a nonsignificant difference in global normalized IED rates between the control periods of each study group ( p = 0.92). RM-ANOVA on z-scored IED rates demonstrated no significant fluctuation in IEDs between control periods for Group 15 ( p = 0.16) ( d ) and Group 90 ( p = 0.40) ( e ); means and standard deviations are depicted. Full size image Table 1 Subject information. Full size table Validation of the control We first confirmed that Group 15 and Group 90 were comparable by verifying a nonsignificant difference in the global normalized IED rates between the average control periods from each group ( p = 0.92) (Fig. 2 c). Our GEE model indicated that ASM status was a significant confounder ( p = 0.027); therefore, all future models controlled for ASM status and session time, as these factors were previously shown to influence IED rates 30 , 31 , 32 . Similar GEE models were used to show that there was no significant difference in the global IED rates between the pre-stimulus baseline period and the control period for Group 15 ( p = 0.82) and Group 90 ( p = 0.88). Our RM-ANOVA of z-scored IED rates showed no significant fluctuation in IEDs between the control periods of all trials for Group 15 ( p = 0.16) (Fig. 2 d) and Group 90 ( p = 0.40) (Fig. 2 e). Together, these findings supported our use of the nested control periods as a reference in subsequent models. Global IED reductions are dependent on the duration of K448 After confirming that the control periods were similar between groups, we could more confidently compare interictal epileptiform responses to auditory stimuli. Our GEE models demonstrated a significant reduction in global IED rates during 90-s of exposure to the original version of K448 both inside and outside of the seizure onset zone (SOZ) (SOZ RR = 0.33, p < 0.001; Non-SOZ RR = 0.34, p = 0.0013) (Fig. 3 a). This effect was only present for the original K448, as we observed a nonsignificant change for the filtered version of K448 with 90-s of exposure (SOZ RR = 0.95, p = 0.48; Non-SOZ RR = 0.82, p = 0.23) (Fig. 3 b). Nonsignificant IED rate reductions were also shown with 15-s of exposure to the original K448 (SOZ RR = 1.05, p = 0.65; Non-SOZ RR = 1.04, p = 0.76) (Fig. 3 a) and the amplitude modulated version of K448 (SOZ RR = 1.03, p = 0.93; Non-SOZ RR = 0.96, p = 0.55) (Fig. 3 b). Figure 3 Reduced global IED rates are dependent on the duration of music exposure. ( a ) GEE models showed that the original version of K448 was the only stimulus effective at reducing IEDs with at least 90-s of exposure. ( b ) Nonsignificant reductions were observed for the altered versions of K448 (top = modulatedK448, bottom = filteredK448). ( c ) Partitioning the 90-s window of Mozart’s original K448 revealed that IED reductions only began after 30-s of exposure. There was a significant IED reduction between the 0–15 and 30–45 windows ( p = 0.004). Control stimuli (musical control = Wagner’s Lohengrin [ Prelude to Act I ], nonmusical control = violet noise) demonstrated nonsignificant IED reductions for each time window. ( d ) All other musical stimuli presented to Group 90 showed nonsignificant IED reductions. “T” or “N” following each song label indicates if the gamma-range auditory modulation spectrum of that song matched (“T”) or did not match (“N”) that of K448. “Altered” indicates signals with secondary gamma modulations. Significance at * p < 0.05, ** p < 0.01, *** p < 0.001. Full size image In evaluating the data from Group 90, where the 90-s window was divided into six 15-s windows, we revealed that IED rate reductions were only present after at least 30-s of exposure (30–45 s RR = 0.31, p < 0.001; 45–60 s RR = 0.34, p < 0.001; 60–75 s RR = 0.33, p < 0.001; 75–90 s RR = 0.31, p < 0.001) (Fig. 3 c). Nonsignificant IED reductions were observed for all times less than 30-s (0–15 s RR = 1.09, p = 0.99; 15–30 s RR = 0.83, p = 0.21) (Fig. 3 c). Our paired-sample comparison of the 0–15 window and the 30–45 window corroborated this finding in showing a significant reduction in IEDs ( p = 0.004) (Fig. 3 c). Applying the same procedure to Wagner’s Lohengrin ( Prelude to Act I ) demonstrated nonsignificant IED reductions for all time windows (0–15 s RR = 1.98, p = 0.41; 15–30 s RR = 2.27, p = 0.13; 30–45 s RR 2.07, p = 0.28; 45–60 s RR = 2.08, p = 0.28; 60–75 s RR = 1.84, p = 0.73; 75–90 s RR = 1.98, p = 0.40) (Fig. 3 c). Similarly, violet noise demonstrated nonsignificant IED reductions for all time windows (0–15 s RR = 0.95, p = 0.30; 15–30 s RR = 1.18, p = 0.98; 30–45 s RR 0.80, p = 0.07; 45–60 s RR = 0.82, p = 0.08; 60–75 s RR = 0.97, p = 0.35; 75–90 s RR = 0.85, p = 0.11) (Fig. 3 c). Our evaluation of all other musical stimuli presented to Group 90 revealed nonsignificant IED reductions for music from the preferred genre (Preferred T RR = 2.83, p = 0.20; Preferred N RR = 1.42, p = 0.78; Preferred N Altered RR = 1.51, p = 0.15) and the classical genre (Classical T RR = 1.16, p = 0.83; Classical N RR = 1.45, p = 0.79) (Fig. 3 d). K448 preferentially reduced IEDs in bilateral frontal regions We next examined region-specific IED rate alterations for regions outside of a subject’s specified SOZ. This was done to see if we could localize the “Mozart K448 effect” in less pathologic brain tissue, identified with implanted intracranial electrodes (Fig. 4 a), while also minimizing the impact that subject-specific SOZs had on IED rates. Linear mixed effects models demonstrated significant IED reductions in the bilateral frontal cortices (right frontal cortex (FC) % reduction = 59.55, p = 0.049; left FC % reduction = 63.25, p = 0.017) (Fig. 4 b). Nonsignificant IED reductions were observed for all other brain regions (right superior temporal cortex [STC] % reduction = 12.69, p = 0.22; right middle temporal cortex [MTC] % reduction = 10.60, p = 0.25; right mesial temporal cortex [Mesial] % reduction = 18.01, p = 0.99; left STC % reduction = 31.02, p = 0.06; left MTC % reduction = 32.96, p = 0.07; left Mesial % reduction = 8.39, p = 0.29) (Fig. 4 b). Figure 4 Bilateral frontal regions responded to K448. ( a ) Stereo-EEG electrodes aggregated across subjects. ( b ) Linear mixed models revealed nonsignificant reductions in all regions other than the bilateral frontal cortices (right frontal cortex (FC) % reduction = 59.55, p = 0.049; left FC % reduction = 63.25, p = 0.017). “n” represents the number of subjects with electrode coverage in the corresponding brain region and “s” represents the number of unique experiment sessions. Significance at * p < 0.05, ** p < 0.01, *** p < 0.001. Full size image Transitions from longer K448 segments increased frontal theta activity An acoustic analysis of the nested structural components of Mozart’s K448 revealed several segment boundaries for repeated sequences (Fig. 5 a). These segment boundaries coincided with a professional musician’s annotations of the musical score (Fig. 5 b). We investigated the association between segment boundaries and frontal activity, as this was the only region with significant IED effects. There was a significant association between increased frontal theta activity and transitions from longer musical boundaries (ß = 0.17, % increase = 19.10, p = 0.002) (Fig. 5 c). All other powerbands and musical segment categories showed nonsignificant associations (Fig. 5 c). Repeating this procedure with the filtered version of K448 revealed a nonsignificant relationship between all powerbands and musical segment categories in the frontal cortex (Supplementary Fig. S1 a); this suggests that the broad structural components of K448 are preserved but altering the frequency structure of the original composition attenuated neural responses. Applying this procedure to Wagner’s composition also revealed nonsignificant associations between all powerbands and musical segment categories in the frontal cortex (Supplementary Fig. S1 b). Figure 5 Enhanced frontal theta activity during shifts from long musical segment boundaries. ( a ) Detected segments for Mozart’s K448 with dashed lines indicating 15-s window boundaries used for our IED-related analyses. ( b ) Constant-Q spectrogram of K448 overlay with annotations from a theoretic analysis of the exposition of the first movement of Mozart’s K448 (i.e., K448’s musical score). ( c ) An assessment of the association between different Stereo-EEG powerbands and musical segment shifts (short = 3-s or less, medium = 3–10 s, long = 10-s or more, control = periods of no segment shift). There was a significant association between increased frontal theta activity and transitions from longer musical boundaries (ß = 0.17, % increase = 19.10, p = 0.002). All other powerbands and musical segment categories showed nonsignificant associations. ß values above zero reflect increased power, and ß values below zero reflect decreased power during a musical segment shift. Significance at * p < 0.05, ** p < 0.01, *** p < 0.001. Full size image Discussion In this study, we observed an association between noninvasive musical stimulation and reductions in intracranial interictal activity in persons with refractory epilepsy. Our study is one of two extant in the literature that examined the “Mozart K448 effect” in an adult population with intracranial recordings 33 . We advance past findings by testing if there was a minimum length of exposure needed to elicit this therapeutic effect and if novel music alteration methods could enhance this phenomenon. To our knowledge, this is the first study to systematically evaluate the relationship between musical segment boundaries and spectral power changes as they relate to the “Mozart K448 effect” in persons with epilepsy. Although previous studies have investigated the role of K448 on interictal and ictal activity 9 , 11 , 12 , 15 , 16 , 18 , 19 , 34 , our study further demonstrates this effect using intracranial Stereo-EEG implants in an adult population with refractory epilepsy. We showed that the original version of K448 could effectively reduce IED rates with exposures as short as 30-s. The 66.5% average global IED reduction observed in our study is consistent with the upper limit of IED responses to K448 reported in the past. For instance, Lin et al.’s most recent work showed a 79.4 ± 20.0% average reduction in IEDs detected with scalp-EEG after one month of K448 exposure 9 . A meta-analysis of other scalp-EEG studies also reported an average IED reduction of approximately 35% during Mozart’s K448 13 . Recently, Štillová et al. (2021) used intracranial recordings to report a median IED reduction of 32% 33 . The enhanced effect observed in our study could be explained by our use of Stereo-EEG instead of scalp-EEG 9 , 10 , 11 , 12 , 18 , as Stereo-EEG is more reliable for detecting “true” intracranial interictal activity 20 , 21 , or by differences attributable to heterogeneous sample populations immanent with intracranial studies (i.e., high variability between subjects with refractory epilepsy). Our observation of no carry-over of the “Mozart K448 effect”, evident by no effect persisting into the post-stimulus periods, showed that 90-s of exposure was likely too short for a lasting impact on neural activity. This contrasts with previous work, which reported significant effects on seizure frequency in the post-treatment follow-up 34 , 35 . For instance, Bodner et al. 34 showed a significant 33% reduction in seizures that persisted into the follow-up phase after K448 exposure, indicating that brief exposures of 90-s or less may invoke a less pronounced neural response than those witnessed with longer stimulus durations 9 , 10 , 11 , 19 . Nonsignificant observations for Group 15 also suggest a weaker neural response with transient exposures. Nonetheless, our null IED findings for the altered music, music matched to K448, and music from subject-preferred genres reinforce the claim that there might be something special about Mozart’s original composition, especially for interacting with the pathology of epilepsy. These findings reveal the importance of stimulus duration and encourage future work to determine the optimal duration of music for generating enduring therapeutic responses. Models investigating less pathologic brain regions (Non-SOZ) highlighted the bilateral frontal cortices as regions important for the “Mozart K448 effect”. This agrees with past observations that listening to music was associated with increased activation of prefrontal cortices 36 , 37 , 38 , 39 . Such as Mansouri et al.’s (2017) 39 observation that high-tempo music activated prefrontal cortical areas, while transcranial direct current stimulation (tDCS) over prefrontal regions negated the influence that music had on executive functions. These current findings also agree with Rauscher et al.’s (1993) 14 original observation that K448 enhanced spatial–temporal working memory, a process directly linked with dorsal frontal activation. We applied a structural decomposition technique to the original version of Mozart’s K448 to identify local and long-range nested structures based on the composition’s harmonic and timbral features. Our investigation revealed enhanced frontal theta power following shifts from longer musical segments (i.e., 10-s or more) that was not present during transitions from shorter musical segment boundaries and during all transitions within the filtered version of K448 and Wagner’s Lohengrin Prelude to Act I. These findings are concordant with past music research, which demonstrated that pleasant music was associated with increased frontal theta power 40 , 41 , 42 , 43 . Previous studies even proposed that frontal theta may represent a gating mechanism for the passage of information to the limbic system 41 and showed that the limbic system’s activity directly correlates with theta oscillations in the frontal cortex, particularly in response to emotionally arousing musical stimuli 40 , 42 , 43 . Further evidence for the relationship between music and frontal emotion networks is provided by Tillmann et al.’s (2003) 44 fMRI study, which showed enhanced activation of the bilateral inferior frontal regions for unexpected targets. More specifically, the structural syntactic relations between musical events led to increased bilateral frontal activation, where greater activation was correlated with processing incoherent, unexpected events 44 . In 2017, Arjmand et al. 45 augmented these findings by showing that unexpected changes in musical features, such as intensity and tempo, activated frontal brain regions linked with positive emotional responses. In conjunction with our current findings, this suggests that the generation of neural predictions about musical features may depend on both the duration of exposure and transitions from prolonged segments within the musical stimulus—as this may be driving enhanced activation of internal emotion networks regulated by frontal cortices. Our theory for the “Mozart K448 effect” raises a critical distinction between subjective emotional responses to music and internal, evoked emotional brain responses. This is supported by our findings of nonsignificant IED changes for musical pieces from the subject preferred genre. Additional support for this theory is provided by Hughes et al.’s (1998) landmark study 15 , which showed that K448 reduced IED activity even in subjects in a comatose state. In this study, Hughes et al. (1998) also showed that theta activity decreased in the central areas, while delta activity increased in frontal areas during K448 15 . Our increase in frontal theta is comparable to their increase in frontal delta, whereby the slight difference in frequency may be associated with their use of scalp-EEG, which is inferior at detecting higher frequency components. Era-related differences in EEG hardware, study protocols, subject populations, and analytical control (i.e., our control for the influence of the SOZ and ASM status) could further explain these discrepancies. In revealing that the musical structure of K448 may be contributing to its therapeutic effect, we shed light on a new theory for the “Mozart K448 effect” in epilepsy: the musical structure defined by the sonata form may elicit positive emotional responses that may be important for anti-epileptic effects. This is further supported by past observations that the only other composition with anti-epileptic properties was Mozart’s Piano Sonata in C Major (K545) 18 . We also confirmed the importance of other musical features, such as the stimulus’s frequency components, by showing that the filtered version of K448 failed to elicit a therapeutic response 10 , 33 . Thus, despite similar broad structural components, the filtered version of K448 may have decreased emotional salience (i.e., frequency distortions made it less acoustically pleasurable), resulting in a reduced likelihood of developing internal musical predictions and engaging emotionally with the piece. A theoretical evaluation of the first 90 s of Mozart’s K448 shows that it is structurally organized by contrasting melodic themes, each with its own underlying harmony. This is contrasted by the first 90 s of Wagner's Prelude to Act I of Lohengrin , which has no recognizable melodies. Called “the first piece of hypnosis by music” 46 , and one of Wagner's most popular musical works, the selection consists solely of static chords that are held for long durations before small shifts in instrumentation and harmony occur. Thus, the structure of Wagner’s selection is organized by subtle and gradual changes instead of contrasting melodic themes, as seen in Mozart’s K448. This work was selected to control for the effect of melody-with-harmony versus harmony alone. It also underscores the importance of selecting proper negative musical controls to systematically uncover components essential for the “Mozart K448 effect” and enhance the validity of experimental findings. Future work will focus on using additional musical controls to further identify components of K448 that are essential for its therapeutic effect. That is, we will focus on analyzing carefully curated musical controls that are specifically matched to certain features of K448 (e.g., frequency, musical structure, musical segment durations) to isolate components essential for beneficial responses. This may enable us to replicate the “Mozart K448 effect” with other musical stimuli through (1) algorithmically searching for stimuli with matching essential components or (2) adding essential components with secondary signal alterations. Several factors limit the implications of this current study. Our automated IED detection could introduce bias; however, it provided a means for objectively marking IEDs in light of the discordance between human reviewers 2 . We are also missing surgical outcome data and ASM blood levels, which may further bias IED-related findings. The relatively small number of Stereo-EEG subjects presents another limitation, which could be responsible for some of the non-significant results presented. However, our sample size was typical for most intracranial studies, which require fewer subjects due to significantly larger effects detectable with intracranial recordings. Our study also provides the foundation and methodology for future multicenter studies that can recruit a larger number of subjects with refractory epilepsy. We did not run the same experiment in all subjects, as we did not consider the importance of stimulus duration in our initial study, which would have been ideal. Thus, Group 15 offers complementary evidence for Group 90’s findings in different persons with refractory epilepsy. Another general limitation is that our study did not specifically measure if positive, subjective emotions were evoked while subjects listened to Mozart’s K448. Nonetheless, we provide evidence for internal representations of emotions through previously reported neural patterns. That is, our findings were concordant with the literature in showing frontal activation following shifts in musical expectations. We also demonstrated that while persons with epilepsy do not generally listen to classical music, it does not preclude them from enjoying and benefitting from Mozart’s K448 at an internal, neural level. In conclusion, the current findings demonstrate that musical stimulation with Mozart’s K448 may reduce IED rates inside and outside the seizure onset zone in persons with refractory epilepsy. We show that the “Mozart K448 effect” has a lower limit of approximately 30-s for evoking therapeutic neural responses and provide evidence for the preferential reduction of IEDs in bilateral frontal regions, with implications for the activation of emotion networks regulated by the frontal cortex. Our data suggest a strategy for the noninvasive modulation of intracranial interictal activity, which may alleviate IED-related comorbidities. They support the future investigation of other sonatas with similar structural characteristics to Mozart’s K448, as they may hold therapeutic potential for epilepsy. Ultimately, our study provides insight into intracranial mechanisms that may be important for the anti-epileptic properties of Mozart’s K448. Material and methods Participants Sixteen subjects undergoing intracranial electroencephalographic monitoring for the clinical treatment of refractory focal epilepsy participated in this study (Table 1 ). All subjects reported little to no previous musical training and limited exposure to classical music. The research protocol for this study was approved by the Committee for the Protection of Human Subjects (CPHS#: 12495) at Dartmouth College, and informed consent was obtained from each subject. All methods were carried out in accordance with the relevant guidelines and regulations of this ethics committee. Electrophysiological data were collected from depth electrodes implanted within the brain parenchyma to best localize epileptogenic regions. Stereo-EEG data Stereo-EEG macroelectrodes recorded electrophysiological data at sampling rates ranging from 500 to 1500 Hz (Natus Medical Inc.). Recording channels were excluded if the raw signal was greater than two standard deviations from the median value across channels to remove non-physiological artifacts or if channels were outside of co-registered brain regions. Stereo-EEG data were band-pass filtered from 1 to 50 Hz, re-referenced to an average referential montage, excluding the channels with artifacts, then resampled at 256 Hz. Anatomical localization For all subjects, pre-implant T1-weighted and T2-weighted MRI images were co-registered with postoperative computed tomography (CT) to obtain the position of small-spacing Stereo-EEG depth electrodes. Freesurfer and the Desikan–Killany atlas were used for hippocampal subfield localization and cortical parcellation, and then final electrode positions were manually reviewed by two neuroradiologists 47 , 48 , 49 , 50 . Electrodes were finally reclassified into the following broader regions: left superior temporal cortex (STC), right STC, left middle temporal cortex (MTC), right MTC, left mesial temporal cortex (Mesial), right Mesial, left frontal cortex (FC), and right FC. Automated spike detection An automated template-matching detector was used for the detection of all IEDs in this study. This detector was previously validated and performed comparably to clinicians at Dartmouth–Hitchcock (DH) and other published detectors 4 , 51 , 52 . The detector used the following pipeline to mark IEDs: (1) cross-correlate a 60-ms triangular template with preprocessed Stereo-EEG, (2) normalize the cross-correlation by the median standard deviation from 1-s sliding windows, (3) calculate the absolute value of the normalized cross-correlation, and (4) mark local peaks above a specified detection threshold as IEDs (Fig. 1 ). We then collapsed temporally overlapping detections into a single marked event and excluded IEDs occurring within 2-s of another IED to account for bursts of spikes. An illustration of this pipeline and sample IED detections are provided in Fig. 1 . Spectral power Due to spectral perturbations associated with IEDs, we first divided task epochs into 1-s segments, then rejected all task epochs within 3-s of an IED. Our goal was to assess spectral power between brief pre- and post-segment boundaries (i.e., musical transitions), so we used the multitaper spectral analysis method, which convolved orthogonal Slepian sequences with the Stereo-EEG signal to provide new periodograms. A final spectrum, obtained by averaging over all the periodograms, was used to calculate the average spectral power within the following canonical frequency bands: delta (2–4 Hz), theta (4–7 Hz), alpha (8–12 Hz), beta (12–30 Hz), gamma (25–40 Hz), and high gamma (40–100 Hz) 53 . We log-transformed and z-scored the power within each experiment session for each electrode for each subject, then averaged power values into non-overlapping 1-s time bins for each trial 54 . Musical boundary detection We defined musical boundaries using a technique developed and detailed by McFee and Ellis (2014) 49 . Specifically, we applied techniques that operate on the graph Laplacian to identify repeated patterns in the musical composition to create block structures in the spectrum based on expanded diagonal bands of a self-similarity matrix 49 . This music information retrieval technique identified a hierarchical structure using (1) harmonic features for long-range repetitions and (2) timbral features for short-range patterns to define structural boundaries in the music 55 , 56 . Spectral clustering was performed using k-means clustering (k = 10) with the normalized eigenvectors of the symmetric normalized Laplacian. The input signal was sampled at 22,050-Hz (mono) and analyzed with a 2048-sample FFT window and 512-sample hop. Music segments were reclassified into the following categories: short (3 s or less), medium (3–10 s), and long (10 s or more). These musical boundaries demarcated shifts in musical themes within the 90-s clips, allowing us to test the hypothesis that frontal activity (i.e., emotion networks) was associated with positive reward prediction errors, observable during transitions out of longer musical segments 28 . That is, we hypothesized that (1) longer segments with a similar musical structure were required for subjects to develop expectations, then (2) violations of those internally generated expectations would be correlated with the preferential activation of emotion networks. Auditory task The first group of subjects (referred to as “Group 15”) was presented with blocks of 15-s clips of distinct auditory stimuli, presented in a random order for each subject session through sampling without replacement (Fig. 2 a). Auditory stimuli were followed by a 15-s control period that consisted of normal ambient room noise (i.e., silence in the acoustic speaker). Although four different auditory stimuli were presented to this group, we only examined Mozart’s Sonata for Two Pianos in D Major (K448) and Mozart’s Sonata for Two Pianos in D Major that was amplitude modulated with a 40 Hz sinusoid (modulatedK448), using the control period as a reference. The second group of subjects (referred to as “Group 90”) was presented with blocks of 120-s clips of musical stimuli, presented in a random order for each subject session through sampling without replacement (Fig. 2 b). During the last 30-s of each auditory stimulus, each subject was presented with a simultaneous attention task. Again, the control period following each auditory stimulus consisted of normal ambient room noise, followed by a Boolean question assessing whether the subject attended to the auditory stimulus. Nine different musical stimuli were presented to this group. However, our primary analysis focused on Mozart’s Sonata for Two Pianos in D Major (K448) and Mozart’s Sonata for Two Pianos in D Major that was band-pass filtered to boost gamma frequencies (filteredK448), using the control period as a reference. We also included an orchestral version of Wagner’s Lohengrin Prelude to Act I and violet noise as two control stimuli. We focused on Wagner’s Lohengrin Prelude to Act I because it had similar general popularity to Mozart’s K448 within the classical genre and included a comparable number of musical boundaries identified by our segmentation method with significantly fewer long musical segments (SFig. 1 ). It also paralleled Lin et al.’s (2010) use of a string version from the classical genre 10 , making it an ideal negative control. Violet noise was selected as a non-musical auditory control because it has a power density that increases per octave, weighting it towards the top of the spectrum. This makes it an ideal negative control for testing our hypothesis that lower frequencies (e.g., gamma boosted) are important for reducing IED rates. Each experiment session had a typical duration of 25-min and was repeated twice per subject on average. The first 15- and 90-s from Mozart’s K448 ( Allegro con spirito ) were used for this study. We utilized a Roland C30 loudspeaker to deliver the auditory stimuli at a comfortable sound level determined by the subject, ranging from 60 to 70 dB. All experiment sessions were performed at least four hours after the most recent seizure and at least 24 h post-implantation. Auditory stimuli A complete list of the auditory stimuli presented to Group 15 may be found in our previous publication 22 . Apart from the original version of K448, altered version of K448, Wagner, and violet noise, other auditory stimuli presented to Group 90 were Frederic Chopin's Bolero in C—Op. 19 for piano, performed by Nikita Magaloff (“Classical T”); Franz Liszt’s Piano Sonata in B Minor, 1st movement: Lento assai—Allegro energico, performed by Leslie Howard (“Classical N”); and three songs from a preferred musical genre. That is, each subject was asked to select a preferred genre after sampling a preselected list of songs from the classical country (Tumbling Tumble Weeds by Sons of the Pioneers [“Preferred T”]; Barbara Allen by Bradley Kincaid [“Preferred N”]), heavy metal (Jugulator by Judas Priest [“Preferred T”]; Just For by Nickelback [“Preferred N”]), and rock and roll (Na Na Hey Hey Kiss Him Goodbye by Steam [“Preferred T”]; Peggy Sue by Buddy Holly [“Preferred N”]) genres. Within this subject-preferred genre, we also altered the “Preferred N” song by boosting lower frequencies (“Preferred N Altered”). The “T” or “N” following each song label indicates whether the gamma-range auditory modulation spectrum of that song matched (“T”) or did not match (“N”) that of K448 using the amplitude-modulation-analysis toolbox [ ]. The modulation-spectrum analysis was run on a large musical corpus to identify maximally matching (“T”) and divergent (“N”) musical pieces to be investigated in our study. All selected songs, both matching and divergent, were additionally tempo-matched to the mean tempo of their respective genre using the librosa library to implement automatic tempo extraction 57 . “Altered” indicates signals with secondary gamma modulations. The first 90 s from each musical selection were used in our study. The order of stimuli presented for each subject session is provided in SFig. 2 . Statistical analysis We used Generalized Estimating Equations (GEE) log-linear regression models with extra-Poisson variance assumptions to determine if there was a difference in the normalized IED rates across all control periods between Group 15 and Group 90 and to compare pre-experiment baseline IED rates with control IED rates. In these models, the within-subject association was specified in terms of an unstructured pairwise correlation pattern, and the variances of the counts were adjusted with sandwich estimators. To ensure that natural transient fluctuations in IEDs did not drive subsequent findings, we z-scored IED rates within each subject, then examined if there was a difference between any of the experiment trials using a repeated measures analysis of variance (RM-ANOVA). An a priori power analysis used the effect sizes reported by Lin et al.’s (2014) most recent work 9 , an alpha of 0.05, and power of 0.90 to determine that a minimum of eight subjects were required to evaluate the main objective of this study. Similar GEE log-linear regression models, retaining the same previous assumptions about the variance and correlation, were used to determine if there was a significant global normalized IED rate reduction during the original K448 relative to the average control period both inside (SOZ) and outside (Non-SOZ) of the seizure onset zone. Analogous models were used to assess the altered versions of K448 relative to the control period. We subsequently partitioned Group 90’s data from the original K448 into six 15-s windows, then evaluated each window independently. Due to the reduced number of IEDs per 15-s time window, we used Generalized Linear Mixed Models (GLMM) that assumed a zero-inflated Poisson distribution with random slopes and intercepts for each subject and an offset term for exposure time. We repeated this procedure for Group 90’s data during the control stimuli, using similar models to evaluate each window independently. A post hoc paired-sample Wilcoxon test was used for pairwise comparisons to determine if there was a significant difference between specific windows of interest. Another post hoc analysis used analogous GEE log-linear regression models to assess all other musical stimuli presented to Group 90. We evaluated subject-specific changes, between control and K448 conditions, in the rate of IEDs for specified brain regions outside of the SOZ. We fit the same mixed effects log-linear regression model separately for each region using region-specific IEDs with the following predictors: ASM status and session time, including random slopes and intercepts for the within-subject factor, and an offset term for exposure time. This model assumed IED rates had a zero-inflated Poisson distribution and controlled for natural heterogeneity between subject IED rates and expected changes in IED rates over time. The presence of prolonged segment boundaries (i.e., transitions out of persistent structural patterns) in the time windows corresponding to significant IED reductions inspired our subsequent analyses, which focused on determining if powerband shifts in the frontal cortices were associated with musical segment boundaries. These analyses were also motivated by (1) our finding that bilateral frontal regions were preferential for the “Mozart K448 effect” and (2) the theory of music and emotion, whereby positive emotional responses are thought to be correlated with the neural processing of acoustic patterns in frontal brain regions 44 . GLMM were used to assess the statistical relationship between shifts in musical boundaries and power in the frontal cortex. These models compared the average frontal power from 1-s before a segment boundary to the average frontal power from 1-s after a segment boundary. We repeated this analysis for frontal brain regions during exposure to the filtered version of K448 and Wagner’s composition as a validation. Of note, past studies relied primarily on paired t-tests 9 , 10 , 11 or its nonparametric alternative, the paired samples Wilcoxon test 12 , 33 to assess changes in interictal activity. For our group-level analyses, we used a semiparametric test, the GEE, which is similar to conventional paired tests but is generally more robust, as it is less affected by departures from parametric assumptions 58 . Differences in statistical power are especially notable when comparing the GEE with the Wilcoxon test, as the latter relies on ranks rather than actual count outcomes. In the setting of our experiment, the GEE and paired t-test are essentially equivalent, except that the GEE uses the asymptotic normal distribution for inference rather than the t distribution, making it the preferred method for a smaller sample size 58 . Relative Risks (RRs) were computed from odds ratios (ORs) for better estimations and were reported with 99.9% confidence intervals to reflect the significance of corrected p values 59 . All models included offset terms to account for different task window durations while also controlling for the influence of anti-seizure medication (ASM) status, unique session time, and subject heterogeneity. The family-wise error rate (FWER) was controlled at 0.05 using Bonferonni correction unless otherwise specified. Data availability Deidentified Stereo-EEG data are available upon reasonable request.

Listening to Mozart's Sonata for Two Pianos in D Major (K448) for at least 30 seconds may be associated with less frequent spikes of epilepsy-associated electrical activity in the brain in people with medication-resistant epilepsy. The findings, which also suggest that positive emotional responses to K448 may contribute to its therapeutic effects, are published in Scientific Reports. Previous research has shown that listening to K448 is associated with less frequent spikes of epilepsy-associated electrical activity in the brain in people with epilepsy. However, the impact of music duration on this association and the reasons for it have been unclear. Robert Quon and colleagues used electroencephalogram (EEG) to measure electrical activity in the brains of 16 adults with medication-resistant epilepsy as they listened to a series of either 15 or 90 second music clips, including K448. The authors found that listening to between 30 and 90 seconds of K448, but not the other music clips, was associated with a 66.5% average reduction in the number of epilepsy-associated electrical activity spikes throughout the brain. These reductions were found to be greatest in the brain's left and right frontal cortices, parts of the brain involved in regulating emotional responses. The researchers also observed that when participants listened to the ending of long, repetitive sections within K448, a type of electrical activity known as theta activity increased in their frontal cortices. Previous research has suggested that theta activity may be associated with positive emotional responses to music. The authors hypothesize that listening to K448 for as little as 30 seconds may activate networks within the brain that are associated with positive emotional responses to music and are regulated by the frontal cortex. The activation of these networks may contribute to reductions in epilepsy-associated electrical activity spikes among those with medication-resistant epilepsy, they suggest.

10.1038/s41598-021-95922-7

Biology

Machines can help wine grape industry survive labor shortage

S. Kaan Kurtural et al, Mechanization of Pruning, Canopy Management, and Harvest in Winegrape Vineyards, Catalyst: Discovery into Practice (2021). DOI: 10.5344/catalyst.2021.20011

http://dx.doi.org/10.5344/catalyst.2021.20011

https://phys.org/news/2021-06-machines-wine-grape-industry-survive.html

Abstracts Collections Free Sample Issue Print on Demand Information For Authors Open Access and Subscription Publishing Submission Subscribers Proprietary Rights Notice for AJEV Online Permissions and Reproductions About Us Feedback Alerts Help Login ASEV MEMBER LOGIN User menu Log in Search Search for this keyword Advanced search Log in Follow ajev on Twitter Follow ajev on Linkedin Search for this keyword Advanced Search Home Content Current Volume AJEV and Catalyst Archive Best Papers ASEV National Conference Technical Abstracts Collections Free Sample Issue Print on Demand Information For Authors Open Access and Subscription Publishing Submission Subscribers Permissions and Reproductions About Us Feedback Alerts Help Login ASEV MEMBER LOGIN Review Mechanization of Pruning, Canopy Management, and Harvest in Winegrape Vineyards S. Kaan Kurtural , Matthew W. Fidelibus Am J Enol Vitic. June 2021 5: 29-44; published ahead of print June 07, 2021 ; DOI: 10.5344/catalyst.2021.20011 S. Kaan Kurtural 1 Department of Viticulture and Enology, University of California Davis, 1 Shields Avenue, Davis, CA 95616. Find this author on Google Scholar Find this author on PubMed Find this author on ADS search Find this author on Agricola Search for this author on this site For correspondence: [email protected] Matthew W. Fidelibus 1 Department of Viticulture and Enology, University of California Davis, 1 Shields Avenue, Davis, CA 95616. Find this author on Google Scholar Find this author on PubMed Find this author on ADS search Find this author on Agricola Search for this author on this site /.panel-row-wrapper Article Figures & Data Supplemental Info & Metrics PDF Summary Aim: In winegrape production, pruning, canopy management, and harvest are essential practices that are increasingly being done by machines. How well these practices are executed can substantially affect fruit yield and quality. Mechanization offers timeliness, uniformity, and cost benefits, but most methods available to date are nonselective and optimal execution requires careful attention to vineyard design, management, and machine settings. This review provides information to help growers make the best use of machines for these tasks. Key Themes: Vineyard design considerations Winter pruning Canopy management Harvesting Impact and Significance: The need to manage large vineyards in a contracting labor market is achievable with mechanization. This review summarizes the best practices in consideration of vineyard design, as well as operation of machines for optimal productivity for the winegrape grower. This review also provides information to help growers effectively incorporate the machines in their vineyards for consistent and economical production of winegrapes including pruning, shoot-thinning, fruit-zone leaf removal, crop load management, and mechanical harvest. canopy management crop removal fruit-zone leaf removal mechanization pruning shoot-thinning viticultural practice Overview Many winegrape cultural practices, including pruning, canopy management, and harvest, are laborious and time sensitive. However, California and many other winegrape production regions are facing rising wage costs and labor shortages. 1 , 2 Ongoing labor shortages were heightened in 2020 because of the novel coronavirus pandemic. 3 Thus, there is an increasing need to farm with fewer people, in a more cost-effective and timely manner. 4 The most laborious vineyard tasks are dormant pruning, canopy management, and harvesting, 5 and mechanization for these methods has received considerable attention from academics, equipment manufacturers, and growers. 6 As growers adapted to mechanical harvest 7 and pruning, they searched for other methods to mechanize other cultural operations such as shoot 6 and fruit-zone leaf removal, 8 , 9 , 10 berry thinning, 11 and shoot positioning. 12 Machinery was developed and commercialized by researchers at the University of Arkansas 13 , 14 and commercialized by partners for these practices. Adoption of such machinery is increasing with economic necessity and as growers develop the knowledge and experience necessary to use these tools effectively. Current vineyard mechanization equipment reduces the need for seasonal manual labor for certain tasks, but does not eliminate it. 15 The degree to which manual labor can be reduced depends on the growing region (coastal versus inland), grapevine cultivar (upright versus downright growth habit), and the number of cultural practices (harvest only or pruning and other practices) the grower is able to mechanize. Furthermore, because vineyard mechanization relies heavily on information-based decision-making, growers need fewer, but more highly skilled, personnel for optimal vineyard management. This review will provide information to help growers effectively incorporate machines in their vineyards for consistent and economical production of winegrapes. Key Themes Vineyard design considerations Preparation for vineyard mechanization ideally starts before grapevines are planted. Some key elements to consider, as in traditionally farmed vineyards, are soil uniformity within blocks, 16 blocking by cultivar or irrigation need, 17 drainage, slope, and consistency of grade. Ideally, mechanization will optimize investment per unit land area and production consistency. This is best achieved by optimizing uniformity within blocks for soil properties such as fertility and soil water drainage, which are helpful in this regard. 18 Sites with inconsistent grade with irregular knolls should be avoided because of safety concerns and variability. Variable grades within a vineyard will often increase vine-to-vine 19 variability because of changes in soil properties. 4 , 19 , 20 Vineyard management in mechanized vineyards is further simplified when vineyard blocks are planted to the same rootstock and cultivar. Well-drained soils are important in mechanized vineyards because equipment access and deployment are hindered in waterlogged vineyards after heavy precipitation. Most equipment will operate safely on slopes up to 7%. 5 , 21 Factors that need consideration when designing a vineyard for mechanization are available equipment, adequate inter- and in-row spacing, headland size, trellising, irrigation, and cultivar selection. Growers must pay attention to certain details to maximize success in future years when designing a vineyard for mechanization. Desired properties are listed in Table 1 . View this table: View inline View popup Download powerpoint Table 1 Checklist of desired vineyard properties for successful vineyard mechanization. Long, straight rows that are adequately spaced are essential. Long rows increase operating efficiency, while straight rows minimize vine injury and damage to line posts ( Figure 1 ). Adequate interrow spacing is desired to allow for lateral adjustment of equipment during operations. Growers need to carefully consider which implements and power units are needed and provide adequate space for their safe operation. Because most growers have full-size equipment, row spacing in mechanized vineyards needs to adhere to the 2.73 to 3.35 m distance for single canopy systems and 3.66 to 4.27 m for double, vertically divided, canopy systems. Download figure Open in new tab Figure 1 Long, continuous 3.2 km rows in a Pinot gris vineyard in Kern County, California. Long rows are preferred to minimize the number of turns for the continuous operation of mechanical equipment. Uniform vine spacing is essential for mechanized vineyards. Currently available equipment is compatible with both single and double curtain trellis systems, and uniform vine spacing permits the use of this equipment. Unlike manual operations where a laborer can adjust density of buds depending on grapevine in-row spacing, equipment is insensitive to these irregularities. Variability in grapevine spacing may also lead to variability in harvest dates, 22 crop load, 23 and canopy density, 24 thereby negatively affecting fruit composition at the farm gate. 2 , 20 Grapevine spacing in the middle to upper range is most desirable in mechanized vineyards. In-row spacing of 1.83 to 2.13 m has proven adequate under most conditions. 4 , 9 , 21 In manually tended vineyards, headland width of 7.6 m to 9.2 m have been adequate. However, because longer equipment with a wider turning radius is used in mechanized vineyards ( Figure 2 , Supplemental Video 1 ), headlands having a minimum width of 10.7 m are needed. In adjacent vineyard blocks that share a headland and have rows that are in line with each other, wider alleyways further facilitate equipment movement between blocks, minimize turns, and align implements with rows, thereby maximizing efficiency. Download figure Open in new tab Figure 2 Proper spacing between alleyways will ensure long-machines can turn into rows easily. Although most trellises can be mechanized to some extent, growers seeking a high level of mechanization in their vineyards should consider using the trellis that is most amenable to mechanization. The currently available vineyard mechanization equipment works best with single curtain trellis systems; however, quadrilateral and vertically separated trellises can also be managed with mechanization. The adaptability of most common trellis and training types are summarized in Table 2 . In selecting a trellis system, consider the growth habit of the cultivar, 21 , 22 the climate, and potential yields of the vineyard. Mechanized vineyards need stout, high-quality trellis materials with strong tie-back assemblies to withstand crop weight without sagging and to bear the torque and force applied by equipment. Row lengths up to two miles are common in the San Joaquin Valley of California, so high-quality materials are needed to set up the trellis and secure it. Commonly, a support-line post is installed at every vine with a cordon wire gauge of 2.3 to 2.6 mm (American Wire Gauge [AWG] 10 to 11) to carry the weight of the crop and withstand cultural operations. Support-line posts are made of galvanized steel with beveled edges to deflect wind, facilitate harvest, and increase durability and longevity ( Figure 3 ). The end posts are constructed from steel and have a narrow profile and spaded ends. Desired dimensions for these materials are presented in Table 3 . View this table: View inline View popup Download powerpoint Table 2 Adaptability of common trellis types to vineyard mechanization found in major production areas of California. VSP, vertical shoot-positioned. Download figure Open in new tab Figure 3 ( A ) High quality steel end posts and line posts are needed to ensure that the load of fruit and the torque from machines can be handled. ( B ) Narrow profile steel end posts with spaded ends are preferred in mechanically managed vineyard installation. View this table: View inline View popup Download powerpoint Table 3 Desired dimensions for end posts and wires used in mechanically managed vineyards. VSP, vertical shoot-positioned. For single high-wire trellis systems, the grapevine trunks are trained against the line post 4 ( Figure 4A ). The line posts should not extend >5 cm above the cordon wire so that they will not obstruct mechanical pruners. As the grapevine’s spurs grow, 1 they will extend above the line posts and can be pruned without the pruning blades impacting the posts. Similarly, used in vertical shoot-positioned trellis systems ( Figure 4B ), the cordon wire is positioned at a height compatible with the implement to aid mechanization. Download figure Open in new tab Figure 4 ( A ) Single high-wire trellis with beveled line post and the cordon wire, and ( B ) vertically shoot-positioned trellis with three grapevine panels with line stakes. For quadrilateral systems, the grapevine is trained with the line post as well. 21 In this system, a 0.6 m-wide cross arm is secured to the line post at 1.4 to 1.6 m above the vineyard floor and two bilateral cordons are trained on the cordon wires ( Figure 5A and 5B ) Download figure Open in new tab Figure 5 ( A ) Quadrilateral trellis as cordons are being trained and ( B ) at full canopy with the sprawling shoots caught by the catch-wires at the t-top. The design of irrigation systems in mechanized vineyards should incorporate the needs of the rootstock and cultivar combination to meet crop evapotranspiration demand. 17 , 25 Drip irrigation lines should be placed high enough to avoid interfering with collector plates of mechanical harvesters. It is advised that inline emitters are used instead of button types, which can be broken off by machinery. Usually, a line height of 55 to 60 cm above the vineyard floor is required. The risers and gate valves of irrigation systems need to be constructed out of flexible tubing to minimize the damage from impact of harvesters or other implements ( Figure 6 ). Download figure Open in new tab Figure 6 Properly installed irrigation gate valve, fertilizer injection port at the correct height of 55 cm above vineyard floor. More vineyard operations can be mechanized if cultivars grown can be trained to have straight trunks and lateral cordons. Cultivars having fruitful basal buds are more easily adapted to mechanization using currently available equipment. Cultivars requiring adjustments to spur length, to retain sufficient numbers of fruitful buds, are less amenable to mechanized pruning. 6 , 26 Winter pruning Dormant pruning is one of the most labor-intensive practices in vineyards. 1 , 8 It is estimated that ~80% of all labor operations costs in vineyards can be attributed to the combination of pruning and harvesting practices. 27 For mechanized vineyards, dormant pruning is best depicted along a continuum of two extremes. 12 On one extreme is hand pruning that produces precisely controlled numbers of short spurs and buds. On the other extreme is minimal pruning to retain numerous longer spurs with less precision. Mechanical pruning types Minimal pruning . The reasoning behind minimal pruning is that the development of high numbers of clusters would be balanced by the early growth of numerous vegetative shoots. 11 The result is production of high yields composed of many very small clusters and berries. However, minimal pruning has all but been abandoned because of the inability to control trunk diseases, the proliferation of dead wood, reduction in primary bud cold-hardiness, and less than ideal fruit composition, which limits marketability. 28 Mechanical box-pruning . As the name suggests, the grapevine’s bearing spurs are pruned from the top, bottom, and sides to resemble a box. 2 , 4 , 17 , 21 Box pruning is the mechanized technique that most closely resembles hand pruning, although box pruning is not selective, leaving all the nodes within the perimeter of the cuts ( Figure 7 , Supplemental Video 1 ). Box height and the width can be manipulated by the pruning-machine operator. A prepruning pass 15 may leave a 0.3 m wide by 0.4 m high box (recommended in frost prone areas ( Figure 8 ), whereas a precision pruning pass may leave a 0.10 to 0.15 m high × 0.10 to 0.15 m wide box. The following approaches are generally used when implementing mechanical box-pruning: Download figure Open in new tab Figure 7 Mechanical box-pruning at a single-high wire Cabernet Sauvignon vineyard during the dormant season. Download figure Open in new tab Figure 8 Mechanical prepruning at a Cabernet Sauvignon vineyard in Napa County, California. Prepruning, hand pruning, and thinning follow-up . Dormant shoots are mechanically prepruned to a 0.30 cm to 0.40 m tall × 0.10 m wide box retaining 120 to 200% of the desired number of buds. After budbreak and the danger of frost has passed, the desired shoot density is achieved by manual pruning and shoot-thinning. 15 Prepruning, mechanical shoot thinning follow-up . Dormant canes are mechanically prepruned to a 0.1 m wide × 0.3 to 0.4 m high box, retaining 120 to 200% predicted bud load. After budbreak and the danger of frost has passed, the desired shoot density is achieved by shoot-thinning. Mechanical box-pruning, mechanical shoot thinning follow-up . Canes are mechanically box-pruned to a 0.10 to 0.15 m tall × 0.10 m wide box. After budbreak and the danger of frost has passed, the desired shoot density is achieved by mechanical shoot-thinning (described later). Early mechanical pruners were adaptations of existing equipment such as vertically mounted hedger bars, which would be used to clean row ends or do summer trimming. As the industry demand for mechanical hedgers increased, the hedger bars became more elaborate and articulated. Hedger bars make the main vertical plane of cut in combination-type pruners. Hedger bar pruners . Hedger bar pruners ( Figure 9 ), when mounted singly, have a single plane of cut and low penetration into the dormant canopy. They have an efficient cutting mechanism consisting of two opposing serrated cutting surfaces, which operate in opposing directions of the plane of travel, matching the ground speed of the tractor. They have low hydraulic flow requirements and can easily be operated with a hydraulic pump having a 38 L/min flow rate. Because they have low penetration into the canopy, they are mostly used in minimal pruning applications in the dormant season, for summer pruning, or as part of combination pruners. Download figure Open in new tab Figure 9 Hedger bars removing the sprawl and undergrowth of a sprawling canopy in Madera County, California. Rotary and drum pruners . Rotary and drum pruners ( Figure 10 ), whether mounted singly or in stacks, have greater penetration into the dormant canopy than do hedger bars. Depending on the cutting surface used (rotating drums or saws), they can be used either for summer pruning or for setting height of the bearing spurs during dormant pruning. When mounted singly and operating on a single row, rotary pruners have a low hydraulic flow requirement and can be operated with a 38 L/min flow rate hydraulic pump. Download figure Open in new tab Figure 10 A rotary and drum mechanical pruner used in mechanical box pruning with hedger bar cutters on the side removing the sprawl. Combination and sprawl pruners . Combination pruners make multiple plane cuts by combining rotary pruners and several hedger bars ( Figure 11 , Supplemental Video 1 ). For example, a combination of rotary pruners can be positioned horizontally and vertically to define the height and width of the bearing surface. In combination with hedger bars, they will prune away the canes that sprawled out of the canopy or above the permitted canopy height. These pruners usually require a toolbar to be installed, with an independently operating power unit to deliver the hydraulic flow to drive these implements. There are units that can be mounted into mechanical harvester frames or trailers with operator stations ( Figure 12 ). Rotary pruners are the most common and efficient pruners, capable of pruning 1 ha in 4.5 hr, saving significant labor costs compared 6 , 15 to hand pruning. Download figure Open in new tab Figure 11 A combination mechanical pruner that is used in California sprawl type trellis in the San Joaquin Valley of California. Download figure Open in new tab Figure 12 A multipurpose carrier frame with combination mechanical pruners installed that can prune two rows at a time in a Chardonnay vineyard in the central coast of California. Effects on yield, fruit composition . Pruning is a rough regulator of yield because of the unpredictable nature of fruitful buds breaking from noncount positions. 15 Therefore other techniques are needed to control yield. Generally, mechanically pruned vineyards are higher yielding 29 than manually pruned vineyards by ~30% in the initial eight years. 8 However, as the vineyard balances and the fruit zone becomes less fruitful because of mutual shading, 22 the yield levels off, and there are negligible differences between manually and mechanically pruned vineyards. 4 There is agreement in literature that berry total soluble solids accumulation in mechanically pruned vineyards is slower than in manually pruned vineyards. 30 , 31 , 32 This is attributed to a combination of higher crop level with shorter shoots, and more leaf layers shading the clusters. 10 In regions with longer growing seasons this is usually not a concern. However, in regions with short growing seasons and early killing frosts, growers should shoot thin. 33 Furthermore, mechanically pruned vineyards produce smaller berries 8 , 34 than do manually pruned vineyards. Smaller berries are purported to have a skin-to-pulp 6 ratio preferred by winemakers because it leads to higher polyphenolic concentrations in wine. Recent research revealed that mechanically pruned vineyards have more exposed leaf area relative to total leaf area compared to hand-pruned grapevines, 35 enabling the flavonoid pathway to be upregulated. 20 Research in interior and coastal California1, 8 , 27 , 36 , 37 has shown that using one of the pruning management strategies described above would save between 60 to 80% of labor operation costs per acre compared to manual pruning alone. 1 , 8 , 27 , 37 In a study comparing traditional Guyot cane pruning, manual spur pruning, and complete mechanization in Madera, CA, researchers found that converting traditionally farmed systems will save 80% of labor operations cost without any differences in berry primary and secondary metabolites. A detailed breakdown of labor operation costs in using this approach, including mechanical pruning, is provided in Table 4 . 1 View this table: View inline View popup Download powerpoint Table 4 Labor operations costs ($/ha), gross revenue, and net income per hectare ($/ha) of Merlot grapevine on Freedom rootstock during transitioning to mechanical pruning in central San Joaquin Valley of California. © American Society for Horticultural Science 2019 HortTechnology 29:128-139. Canopy management mechanization Manufacturers developed mechanized implements to complete all canopy management requirements for the growing season. Below are some mechanized canopy management practices that have verifiable physiological effects with economic benefit. Shoot-thinning . Mechanical shoot thinners use soft silicone “finger-like” protrusions on a rotating drum to brush the cordon ( Figure 13A ) and a rotating shoot thinner with soft paddles that strike the canopy at a known pace (Figure 13B). Growers need to assess the canopy density (number of count shoots, noncount shoots per foot of row) before calibrating and using the mechanical shoot thinner. The mechanical shoot thinner is not selective, that is to say it does not remove specific shoots, but rather strikes the canopy at a known frequency that can be adjusted ( Supplemental Video 2 ). The shoot thinner is ideal for regions that are prone to frost where prepruning passes have retained more nodes than necessary or for cultivars that are prone to over cropping. Growers are advised to apply mechanical shoot-thinning around the time of cluster elongation. The mechanical shoot thinners have four planes of motion and will require an external power unit to provide enough hydraulic flow to power the implements. Download figure Open in new tab Figure 13 A shoot removal implement with a cordon brush with silicone fingers on (A) a rotating drum and (B) the shoot removal paddle in the front of the rotating drum. Mechanical shoot-thinning reduces shoot density 6 , 8 , 35 and is an efficient crop-thinning method. However, its effect may be temporary unless a mild water deficit is imposed to inhibit lateral and secondary shoot growth. 8 It further assists in establishing fruiting positions in the following year. In Table 5 , the effects of shoot-thinning on leaf area, yield, and cumulative yield reduction to Colombard grapevines grown on Freedom rootstock in Fresno County, CA, are shown. Mechanical shoot-thinning applied early reduced leaf layers and increased the exposure of clusters to moderate levels of sunlight, thereby improving berry phenolic content of red wine grape varieties.6 In considering whether or not to shoot thin, the grower has to consider the potential loss of income from reduced yield and the possibility of increased income from better fruit composition. The application cost of mechanical shoot-thinning is $197/ha 27 , 37 compared to $1482/ha for manual shoot-thinning. 27 View this table: View inline View popup Download powerpoint Table 5 Leaf area reduction of Colombard grapevine grown on Freedom rootstock under mechanical shoot-thinning to 21 shoots/m of row and cumulative reductions in yield and gross farm income in three consecutive seasons in Crush District 13 of California. Trunk suckering . During the spring growth flush, shoots may develop from latent buds on the trunk. Traditionally, these unwanted shoots were removed manually 37 with crews. However, as the height of canopies increased with modern trellis systems such as single high-wire or high quadrilateral, the necessity to do this practice by machine has increased. 15 , 27 The machines that can do these tasks are now mounted to the front of the tractor ( Supplemental Video 3 ) and strike the trunk of the grapevine with silicone fingers affixed to a rotating drum. The trunk sucker machines use the hydraulic flow of a common vineyard tractor with a flow rate of 38 L/min and can remove suckers on both sides of the row. In a study conducted in Napa County, CA, the cost to manually sucker trunks was $1493/ha. 27 When done by the machine, the cost was reduced to $154/ha, providing almost an order of magnitude in labor operations cost savings. 37 Leaf removal . Various implements can remove leaves mechanically. The goals of fruit-zone leaf removal can include limiting crop level by reducing the number of berries that set, 38 improving cluster exposure to sunlight 36 depending on timing, and improving air flow to reduce fungal infections, 39 but the outcome must balance crop level with exposed leaf area. 36 Generally, fruit-zone leaf removal equipment operates on the principle that leaves are lighter than clusters and can either be sucked into a baffle and cut off ( Figure 14A ) or rolled over an expanded aluminum drum ( Figure 14B ) that screens out flowers and clusters but “plucks” away leaves ( Supplemental Video 4 ). There are also implements that force leaves off petioles with short bursts of air. Download figure Open in new tab Figure 14 Common fruit-zone leaf removal implements used in vineyards with ( A ) a suck-and-cut type and ( B ) a roll-over type of fruit-zone leaf removal machine with a 50 cm baffle exposed. Timing . Balancing crop level and exposed leaf area requires precise fruit-zone leaf removal timing. If conducted prebloom, 40 the response of the grapevine is to reduce the number of berries set, 41 providing greater exposure with reduced yield. If conducted postbloom, 23 the result for the grapevine is increased solar radiation and temperature in the fruit zone. The cost to apply mechanical fruit-zone leaf removal is $247/ha compared to $1500/ha when conducted manually in coastal California. 27 Effects on yield, fruit composition . Depending on timing and the climate in which the fruit is grown, fruit-zone leaf removal might not affect yield 9 , 10 or it may reduce yield. In warm regions, fruit-zone leaf removal often has a minimal effect on yield, but an improvement on leaf area to fruit ratio should be expected 36 as a result of removing excessive leaf layers, which shade the cluster in the fruiting zone. 10 By removing ~20% of leaf area, a more balanced grapevine was achieved, 20 however, the biggest beneficial effect of fruit-zone leaf removal is increased berry flavonoid content. 36 As depicted in Table 6 , even in warm-climate viticulture regions, anthocyanin accumulation in the berry can be increased by early season fruit-zone leaf removal. 10 View this table: View inline View popup Download powerpoint Table 6 Effect of mechanical removal timing and fractions of crop evapotranspiration replacement on labor operations cost of canopy management and cost of producing total skin anthocyanins per hectare in northern California. © 2015 American Society for Enology and Viticulture, AJEV 66:266-278. Crop load management using vineyard mechanization Crop load is the ratio of fruit to pruning weight, and a commonly recommended desirable range is 5 to 12, 8 , 42 depending on the cultivar and location of the vineyard. The whole-season approach to using machines to regulate crop load could use all or some of the implements (e.g., pruner, shoot thinner, leaf remover) described previously. Possible management options are described below and differ according to grapevine growth habit. Grapevines that have a downward growth habit such as Merlot, 1 Zinfandel, 6 , 10 and Syrah 22 are not as amenable to whole season crop load management using mechanization as are vines with an upward growth habit such as Cabernet Sauvignon, 4 Cabernet franc, or Chardonnay. The issue has to do with increased sunlight received from postmechanical pruning, which results in vegetative compensation by the grapevine. 22 , 28 In the case of downward growth habit grapevines, the canopy is quickly repopulated, thereby negating any effects of higher amounts of solar radiation received, resulting in poor berry development. In cultivars such as Syrah, 23 Merlot, 1 and Zinfandel, 30 mechanical box-pruning followed by mechanical shoot-thinning resulted in less than ideal yield-to-pruning weight ratio and poor anthocyanin accumulation. 23 Crop load management in procumbent grapevines may be limited to a prepruning pass with a mechanical pruner and hand shoot-removal follow up and mechanical harvest. Grapevines with an upward growth habit are more amenable to whole-season crop load management with mechanization. All the steps of canopy management can be performed mechanically and reliably because upward growth habit grapevines respond more beneficially to increased solar radiation exposure to growing shoots, and ideal yield-to-pruning weight ratios can be achieved. Crop-thinning . Crop-thinning can be accomplished with mechanization. 31 Operating a mechanical harvester at a low frequency with some beater bars removed is a simple way to remove a portion of the berries on a vine. In a study conducted in Fresno County, Sauvignon blanc berries were shaken off at BB size (3 to 5 mm) with beater bars positioned 1 ft below the fruiting zone, at a ground speed of 2.4 kph and 470 RPM (rotations per minute). This reduced yield by 50%, and reduced cluster compactness by 60% compared to grapevines that were not crop-thinned. The cost to do manual crop-thinning was $1340/ha for vertically shoot-positioned canopy 27 in Napa County. To achieve a similar cropping level with a mechanical harvester, the cost was reduced to $803/ha. 27 However, the benefit of this practice has to be managed by the grower for loss of yield versus an improvement in berry composition. 36 The effectiveness of crop-thinning depends on timing. A rule of thumb is to thin postfruit-set and preveraison. If thinning is done too soon, the grapevine may set more fruit, whereas thinning too late may not affect fruit quality. 43 , 44 , 45 Most studies have shown that the ideal time to thin is when berries are the size of BBs. 33 , 44 Mechanical crop-thinning, like shoot-thinning, is nonselective. Particular berries or clusters are not targeted. The grapevine fruit zone is struck or beat with beater bars that detach the BB-sized berries from their rachises. Beating frequency must be adjusted to attain a predetermined production level. Rigorous crop estimation and field sampling should be conducted before implementing berry thinning to ensure appropriate levels of thinning. In the absence of a reliable, nondestructive crop estimation tool, growers need to keep very accurate records of yield. The grower’s aim is to balance crop level with exposed leaf area. 31 , 35 The immediate effect in the current season is a reduction in yield, 36 but an improvement in leaf area-to-fruit weight ratio is also realized because exposed leaf area is kept constant. 46 Ripening is hastened, especially soluble solids accumulation. 6 However, reducing the number of berries without affecting leaf area can result in an undesirable increase in berry size, 44 although the grower can manage this with water deficits. 25 Generally, a 25% reduction in number of berries by crop-thinning results in a 15 to 17% reduction in yield, due to compensatory growth of the remaining berries. 31 Mechanical grape harvesting Hand harvesting has some benefits including selectivity (rotten and unripe clusters can be avoided), but it is laborious and limits harvesting to daytime hours when temperatures are higher and can promote premature fermentation. Mechanical grape harvesting is less selective but requires far fewer people, proceeds much more quickly, and can be done at night. In general, 5 t of fruit can be picked in an hour at night under cooler temperatures, which maintains fruit quality and minimizes premature fermentation. About 90% of the winegrapes crushed in the United States are mechanically harvested. 32 If time is of the essence, mechanical harvesting is the best option. The speed of mechanical harvesters gives a grower better control over harvest timing than could be achieved with a hand-picking crew. To optimize mechanical harvesting, the trellis has to be installed correctly with correct vine and row spacing ( Tables 1 and 3 ). Other considerations for growers are logistics, i.e., delivery of large amounts of fruit to the winery and shipping times, and minimizing materials other than grapes (MOG) in the harvested fruit, which will be discussed later. The current rate for hand harvesting is $90 to $111/t of winegrapes. Five to six laborers are required to pick 1 t/hr. A study conducted in Fresno, CA, found that labor for harvesting a vineyard that produces on average 17.3 t/ha would cost $1384 to $1729/ha, whereas machine harvesting would cost $679 to $865/ha. 37 This is an ~50% savings in labor operations costs, and it ensures timely delivery to the winery. Harvester types . Grape harvesters are either pull-behind, requiring auxiliary power, or self-propelled. Pull-behind harvesters are the most economic choice for growers who already have a large tractor, so pull-behind mechanical harvesters are the most common. Some pull-behind harvesters have hydrostatic drive to assist in hill climbing. They can be fit with either canopy or trunk shaker heads to fit the needs of the grower. The advantages of this type of harvester include safer operations on slopes, lower initial cost, and the ability to select and substitute power units. The disadvantages of this type of harvester are more grapevine and trellis damage due to having an auxiliary power source, higher operation difficulty requiring more laborers, and their tendency to slide into the row if the speed of forward travel is not maintained constant. Self-propelled harvesters have a larger initial investment requirement than pull-behind harvesters. They have hydrostatic drivetrains and are engine-driven. Advantages of self-propelled harvesters include maneuverability, stability in most situations, ease of operation, and less grapevine and trellis damage. Disadvantages of self-propelled harvesters include initial high cost, fixed horsepower, and a power unit that is not easily substituted. Picking head types . Canopy shakers are the most common picking heads. A canopy shaker uses a bow rod head ( Figure 15B ). The picking head in this case compresses the canopy and transfers the force of the picking head to the canopy and trellis to detach the berries or clusters from the vine. The nylon picking rods are 2.54 cm diam, extruded, and formed into bowed rods. The rods are attached to the picking head with steel or aluminum holders. The following adjustments can be made to a canopy shaker machine bow rod picking head to accommodate a cultivar of trellis height and canopy girth: Stroke: 10 to 20 cm Rod spacing: 5 to 20 cm Rod tension: firm to very firm Throat width: best fit (250 to 1000 mm) Revolution speed: 300 to 450 rpm Forward speed: 1.6 to 5 km/hr Download figure Open in new tab Figure 15 Parts of a mechanical harvester, typically used in California, operating ( A ) at night, with ( B ) a canopy shaker head with the bow rods visible, ( C ) a trunk shaker head with the rails in the bottom above the catcher plates, ( D ) the catcher plates that go around the trunk and line posts, ( E ) the conveyor belt with food grade buckets that transports the berries to the top of the harvester to the kicker belt, ( F ) the over-the-row conveyor used to deliver fruit to receptacle bin in the neighboring row, and ( G ) passive mechanism for keeping canes and sticks out of the fruit buckets with the slider bar. Trunk shakers are used in winegrape vineyards with small clustered, short pedicel cultivars where transfer of force directly to the vine trunk is desired ( Figure 15C ). The trunk shaker type of picking head operates by transferring the force of two revolving counterweights to two rails that are moving in the opposite direction of harvester travel. Trunk shaker harvesters are quite heavy and need a large power unit to operate them. The following adjustments can be made to trunk shaker machines to accommodate characteristics of the trellis and the height of grapevines: Head tuning: counterweights must be in time Pinch pressure: best fit to prevent damage to the graft union Pinch spacing: best fit to prevent damage to the graft union Height of rails: >55 cm if vineyard is drip-irrigated Head revolution speed: 120 to 300 rpm Forward speed: 1.6 to 5 km/hr Fruit catching and delivery systems . Once grape berries are picked, they need to be caught and delivered cleanly to a bin. The grape berry catching systems in mechanical harvesters are constructed from Lexan (General Electric) or nylon and are referred to as “catcher plates.” They are designed like flower petals or overlapping leaves ( Figure 15D ). The plates open and close as needed to pass around vine trunks and trellis line posts. The catcher plates are unidirectional so the operator cannot back up. Berries on the collector plates are shed into a bucket conveying system ( Figure 15E ). The system consists of high-impact buckets attached to roller chains that are hydraulically driven. The berries are conveyed to the top of the harvester in an almost static state with no dragging, rolling, or unnecessary dumping. At the top of the harvester, the berries are dumped into an over-the-row conveyor (OTR) system ( Figure 15F ). The OTR system conveys the berries from the harvester, over the adjacent row, and into a companion gondola or bin trailer. The swing and height of the OTR are adjustable by the operator to fit vineyard conditions. Fruit cleaning systems . As the canopy is struck by bow rods or shaken by rails, some MOG inevitably fall into the catcher plate stream. MOG may include canes, leaves, trellis parts, and other debris found in the canopy. The allowable levels are relatively low, typically between 2 to 5%. The best way to minimize MOG is to properly set up the vineyard and adjust the harvester settings. However, some MOG is unavoidable, so harvesters are built with active and passive systems that remove MOG. The active MOG removal systems start at the point of transfer between the catcher plate and the bucket conveying system. A rotary MOG deflector removes loose debris such as leaves, canes, and green shoots by knocking them out of the stream, preventing buildup at this point. As the bucket conveying system transfers the fruit to the top of the harvester, a passive MOG cleaning system called a “MOG slider tube” ( Figure 15G ) guides large canes and sticks out of the buckets. As the berries are transferred from buckets to main kicker belts, the berries are directed inwards under primary cleaning fans ( Figure 16A ). A lower cross-conveyor belt directs fruit toward the OTR conveyor. The cleaning fan system consists of two or three fans that draw in air as the fruit drops ( Figure 16B ). Because leaves are lighter than fruit, they are sucked into the air stream, chopped up, and exhausted toward the rear of the harvesters. A final cleaning fan is mounted at the end of the cross conveyor before the berries are transferred to the OTR to clean any remaining debris. The OTR system also has a cleaning system for MOG. This cleaning system consists of a powerful magnet ( Figure 16C ) that removes ferrous materials such as pieces of wire, clips, wrenches, screwdrivers, etc., which can damage pumps and bladders at the winery. Download figure Open in new tab Figure 16 The cleaning system of a typical mechanical harvester indicating the location of ( A ) the kicker belt and cleaning fans, ( B ) the kicker belt and the drop cleaning fans, and ( C ) the ferrous material cleaner on the over-the-row conveyor before fruit is delivered to receptacle bin. Quality concerns . Mechanical harvesters do not remove 100% of the berries from grapevines. Up to 5% of berries may be left behind, and this is an acceptable level. The goal is not to over pick but to leave rotten and raisined berries on the grapevine. As the mechanical harvester shakes the canopy, leaves are also removed. It is desirable for 70 to 80% of the leaf area to remain intact after harvest to help the plant amass additional carbohydrates and recover mobile nutrients from the leaves, both of which help grapevines regrow the following season. Fruit temperature during harvest can affect must quality. Because a large mass of berries is picked and conveyed into a trailer (usually 6.7 t/bin), the temperature of the loads do not change rapidly. The time in transit should not take longer than 6 hr or the temperature of the load may increase and excessive oxidation may occur. Growers must consider that the same harvester, or harvester settings, will not work optimally in all vineyards. Adjustments need to be made to optimize harvest in each vineyard. Only certain varieties are easily harvested mechanically. Some varieties, especially those with short peduncles, can be difficult to harvest with machines. For example, Bordelais cultivars, such as Merlot or Cabernet Sauvignon, are easily harvested by machines. Conversely, Sauvignon blanc, or certain clones of Pinot noir are more challenging. Likewise, certain clones of Chardonnay with short peduncles can only be harvested reliably by trunk-shaker harvesters. Significance The rising cost and decreasing availability of farm labor in California and other important winegrape growing regions has heightened interest in mechanization. Currently, machines are available to mechanize the most laborious cultural practices used for producing winegrapes, including pruning, canopy management, and harvest. Making the best use of these machines requires a holistic approach that considers the vineyard site characteristics, grapevine cultivar, trellising, and equipment available. Vineyard accessibility and uniformity is important, and fewer, more well-trained staff are needed to calibrate and operate machines effectively. Machine use may be limited to one or two laborious tasks, but the opportunity exists to integrate machines into most tasks in a way that optimizes grape yield and quality. Footnotes Acknowledgments: The authors would like to acknowledge the American Vineyard Foundation, Bronco Wine Company, West Coast Grape Farming Incorporated, Constellation Brands US, V-Mech LLC, and Oxbo Inc. for their financial and in-kind support during the execution of the all the referenced trials that were conducted in California that made this review possible. Technical expertise and assistance of Gregory T.Berg, Saul Arriola, Tom Valdero, Jose Valdez, Brian Franzia, Carol Franzia, Michael Blaine, Daniel Bosch, Geoffrey Dervishian, Dave B. Terry, Jose Luis Huizar, Joseph Geller, Lydia Nida, Runze Yu, Michael Cook, Clinton Nelson, and especially Professor Emeritus Robert Wample, have been invaluable in conducting these works. The corresponding author is indebted to Mr. Freddie T. Franzia of Bronco Wine Co. for enabling vineyard mechanization research in California. By downloading and/or receiving this article, you agree to the Disclaimer of Warranties and Liability. The full statement of the Disclaimers is available at . If you do not agree to the Disclaimers, do not download and/or accept this article. Received December 2020. Revision received February 2021. Revision received March 2021. Revision received March 2021. Accepted April 2021. This is an open access article distributed under the CC BY license ( ).

Wine grape growers in California and elsewhere face increasing labor costs and severe labor shortages, making it difficult to manage and harvest a vineyard while maintaining profitability. Growers are increasingly turning to machines for pruning, canopy management and harvesting, but how well these practices are executed can substantially affect yield and quality. A new review by researchers at the University of California, Davis, published in the journal Catalyst, provides guidelines for growers to make the best use of machines. "Wine grape laborers have been virtually nonexistent. People don't want to work in vineyards anymore because it's remote, tough work," said Kaan Kurtural, professor of viticulture and enology and extension specialist at UC Davis. "There is now machinery available to do everything without touching a vineyard." Kurtural has designed a "touchless" experimental vineyard at the UC Davis Oakville Station to help growers understand how machines can help them cope with the labor shortage. While machines reduce the need for seasonal manual labor, they do not eliminate it. The degree of labor reduction depends on growing region, grapevine type and the number of practices growers mechanize. The review provides guidance on using machines for winter pruning, canopy management and harvesting as well as how to design a grape vineyard for machines before planting. Videos showing the operation of different types of machinery and practices can also be found in the review. Economic savings, quality grapes About 90% of the wine grapes crushed in the U.S. are mechanically harvested. Previous studies have found about a 50% savings in labor costs from using machines to harvest instead of hand harvesting. Credit: Catalyst "Using more mechanization in a vineyard beyond just harvesting can also reduce labor costs without affecting grape quality." Kurtural said. Mechanical pruning, for example, can save between 60% to 80% of labor operation costs per acre compared to manual pruning alone. One experiment in the San Joaquin Valley, where more than 50% of California's wine grapes are grown, also showed using mechanical canopy management machines to manage merlot grapes resulted in twice the amount of color. The more color, or higher anthocyanin concentrations, the better the quality. It can significantly improve returns from vineyards in California's heartland. Kurtural said there are machines available to manage canopies, including machines for leaf removal, shoot thinning and trunk suckering. Kurtural noted that the machines are American made, developed by researchers at the University of Arkansas and commercialized by manufacturers in Fresno and Woodland, California.

10.5344/catalyst.2021.20011

Space

Uncovering the mystery of early massive galaxies running on empty

K. Whitaker et al, Exhausted gas reservoirs drive massive galaxy quenching in the early universe (2021 Sept. 23), Nature, DOI: 10.1038/s41586-021-03806-7, www.nature.com/articles/s41586-021-03806-7 Preprint: public.nrao.edu/wp-content/upl … _Nature_Preprint.pdf Journal information: Nature

http://dx.doi.org/10.1038/s41586-021-03806-7

https://phys.org/news/2021-09-uncovering-mystery-early-massive-galaxies.html

Abstract Star formation in half of massive galaxies was quenched by the time the Universe was 3 billion years old 1 . Very low amounts of molecular gas seem to be responsible for this, at least in some cases 2 , 3 , 4 , 5 , 6 , 7 , although morphological gas stabilization, shock heating or activity associated with accretion onto a central supermassive black hole are invoked in other cases 8 , 9 , 10 , 11 . Recent studies of quenching by gas depletion have been based on upper limits that are insufficiently sensitive to determine this robustly 2 , 3 , 4 , 5 , 6 , 7 , or stacked emission with its problems of averaging 8 , 9 . Here we report 1.3 mm observations of dust emission from 6 strongly lensed galaxies where star formation has been quenched, with magnifications of up to a factor of 30. Four of the six galaxies are undetected in dust emission, with an estimated upper limit on the dust mass of 0.0001 times the stellar mass, and by proxy (assuming a Milky Way molecular gas-to-dust ratio) 0.01 times the stellar mass in molecular gas. This is two orders of magnitude less molecular gas per unit stellar mass than seen in star forming galaxies at similar redshifts 12 , 13 , 14 . It remains difficult to extrapolate from these small samples, but these observations establish that gas depletion is responsible for a cessation of star formation in some fraction of high-redshift galaxies. Main The 1.3 mm observations of dust emission were made with the Atacama Large Millimeter/submillimeter Array (ALMA), and the sample comprises six galaxies selected from the REsolving QUIEscent Magnified (REQUIEM) galaxy survey: MRG-M1341 (ref. 15 ), MRG-M0138 (ref. 16 ), MRG-M2129 (ref. 17 ), MRG-M0150 (ref. 16 ), MRG-M0454 (ref. 18 ) and MRG-M1423 (ref. 18 ) (Fig. 1 ). The targets are all strongly lensed, with magnification factors ranging from a factor of 2.7 (MRG-M1423) to 30 (MRG-M1341). Five out of the six galaxies are classified as quiescent owing to unusually low star-formation rates that reach down to 0.1 M ☉ yr −1 , as measured from fitting the optical to infrared spectral energy distributions ( Methods ). Although the most distant target, MRG-M1423, has a more typical star-formation rate of about 140 M ☉ yr −1 over the previous 100 Myr, consistent with normal star-forming galaxies at z = 3, its spectrum reveals classic post-starburst signatures that support a picture in which it has quenched rapidly within the last 100 Myr (ref. 18 ). These targets are qualitatively different to existing millimetre continuum/carbon monoxide (CO) spectroscopic observations tracing cold interstellar medium phases in quenched galaxies in that these galaxies have star-formation rates that are an order of magnitude lower for their stellar mass 2 , 4 , 5 , 6 , 8 , 11 , higher redshifts 3 , 7 , 9 , 10 , and uniquely deep flux limits facilitated by strong lensing magnification. Fig. 1: Images of six massive lensed galaxies for which star formation has been quenched. The panels are rank-ordered from z = 1.6 to z = 3.2 ( a – f ), showing a composite HST colour image ( i F814W , J F125W , H F160W generally, substituting J F110W for e ) and contours of ALMA/Band 6 dust continuum observations, with levels defined by signal to noise ratio (SNR). Each image is centred on the target galaxy, whose redshift is listed in the top-left corner. The dashed ellipse indicates the ALMA beam size, with the 1 σ noise level noted at the bottom of each panel in units of mJy per beam. Full size image For the redshift range of our sample, our 1.3 mm wavelength observations correspond to 300–500 μm rest frame on the Rayleigh–Jeans tail of the dust emission, which serves as a robust proxy for the cold molecular gas mass 19 . We clearly detect two of the sources in the dust continuum: MRG-M0138 at 0.27 ± 0.03 mJy and MRG-M2129 at 9.74 ± 0.16 mJy. Such direct detections of cold dust in individual quiescent galaxies outside the local universe, implying per cent-level molecular gas fractions, are scant owing to the extreme sensitivity requirements. In contrast with the extended stellar light profiles, and despite the enhanced resolution from strong lensing magnification, both sources remain unresolved with no evidence for missing extended flux ( Methods ). This suggests that they have high dust and molecular gas surface densities, as the dust continuum is centrally concentrated and significantly less extended than the stellar light (Fig. 1 ). Such a result has also been found in star-forming galaxies at similar redshifts 20 . The sensitive ALMA dust continuum imaging of the remaining four sources all yield strong upper limits, with the 3 σ detection limits ranging from 30−150 μJy before lensing corrections. We estimate the dust mass, M dust , by adopting a modified blackbody fit and making standard assumptions about dust temperature and emissivity ( Methods ). We show the redshift evolution of the dust fraction, f dust = M dust / M ★ , for our sample of lensed quenched galaxies in Fig. 2 . By adopting a ratio of the molecular gas mass to dust mass of 100 ( Methods ), we estimate M H2 directly from M dust and also show the inferred molecular gas fraction, f H2 = M H2 / M ★ (right axis). Even if we adopt an extremely conservative molecular gas to dust mass ratio that is a factor of ten higher, f H2 is still well below that of normal star-forming galaxies at this epoch 14 . Both of our unambiguously detected galaxies have low molecular gas fractions of 4.6 ± 0.5% and 0.6 ± 0.1%, respectively, with systematic uncertainties in dust temperature and molecular gas mass to dust mass ratio of a factor of 1.7. Strong upper limits from CO emission for these two targets rule out more exotic molecular gas-to-dust ratios in these particular cases, which would otherwise imply larger cold gas reservoirs ( Methods ). Although scaling relations adequately describe the cold gas content of quiescent galaxies in the local universe by construction 21 (for example, contours in Fig. 2 ), our observations reveal a population of massive galaxies at z > 1.5 that have molecular gas fractions more than an order of magnitude lower than empirical predictions at similar redshifts. Our measured f H2 is 0.9 ± 0.2 dex lower on average than scaling relation predictions for the given star-formation rates and stellar mass 14 . Fig. 2: Low dust masses for quenched galaxies. Measurements of f dust for distant lensed quiescent galaxies (circles) are extremely low given their sSFR (SFR per unit stellar mass). We compare existing dust continuum measurements in the literature of individual quiescent galaxies at z > 1.5 (refs. 5 , 6 ) (individual black symbols) and stacked quiescent galaxies 8 , 9 from JCMT/SCUBA and ASTE/AzTEC data out to z ~ 2 (large grey symbols), using identical conversions herein to our sample ( Methods ). The thick black error bars are the formal 1 σ measurement uncertainty in our 1.3 mm flux density and the thin black error bars represent systematic uncertainties when varying dust temperature. The smaller transparent symbols represent the predicted f H2 from empirical scaling relations given sSFR. The inferred f H2 (right axis) and scaling relations 14 for log( M ★ / M ☉ ) = 11 on the average log(SFR) − log( M ★ ) relation (solid), 1 dex (dashed) and 2 dex below (dotted) assume a molecular gas to dust mass ratio of 100. The shaded region shows the upper bound set by the lowest stellar mass in our sample (log( M ★ / M ☉ ) = 10.1), and vice versa for the highest stellar mass (log( M ★ / M ☉ ) = 11.7), with the literature dust/CO compilation out to z = 3 shown as a greyscale contour; note that local quiescent galaxies with f H2 ~ 1% at z = 0 are artificially high because the majority are upper limits. Source data Full size image Our program measures a broad range of (low) molecular gas masses in massive galaxies with suppressed star-formation rates (Fig. 3 ). A comprehensive literature search at 1.5 < z < 3.0 ( Methods ) demonstrates that galaxies typically form copious new stars (median specific star-formation rate, log(SFR/ M ★ ) = −8.6) and have a bountiful fuel supply, with a median value of f H2 = 51%. By comparison, our galaxies instead form two orders of magnitude fewer new stars (median log(SFR/ M ★ ) = −10.7) and have a median upper limit of f H2 < 1%. Until recently, such low molecular gas fractions have been measurable only in galaxies in the local universe 21 . Our new measurements confirm that the cold interstellar medium was already rapidly depleted at high redshift in at least some galaxies, not slowly consumed until the present day. Fig. 3: Low molecular gas masses compared to star forming galaxies. The molecular gas fraction f H2 is significantly lower at a given sSFR (sSFR≡SFR/ M ★ ) for distant lensed quiescent galaxies at z > 1.5 when compared to the compilation of existing CO and dust measurements of similarly massive star-forming galaxies (contours, Methods ). Our sample explores an order of magnitude lower sSFR and higher redshifts, finding median molecular gas fractions a factor of 10 lower than existing measurements for distant quiescent galaxies 4 , 5 , 6 , 8 ( Methods ). The thick horizontal error bars for the two new detections represent formal 1 σ measurement uncertainty in our 1.3 mm flux density and the thin horizontal error bars represent systematic uncertainties when varying dust temperature and molecular gas to dust ratio. Vertical error bars are 1 σ uncertainties. The data are largely consistent with rapid gas depletion, on average following the tracks for constant gas depletion timescales on the order of approximately 1 billion years (dotted lines). Source data Full size image Another study has already set the stage at high redshift, finding moderate cold gas reservoirs based on stacking dust continuum measurements in a mass-representative sample 8 . Although the cold gas reservoir of MRG-M2129 is consistent with these first results, all other sources remain in significant tension. The sample is too small to distinguish whether the subpopulation is biased, or whether contamination due to the significantly lower resolution of the earlier stacking study biases earlier measurements towards high redshifts. Our results also contradict the moderate cold gas reservoirs detected in recently quenched galaxies at lower redshifts that instead imply reduced star-formation efficiency 10 . In principle, differences in the ages of the stellar populations could explain this discrepancy 22 , but our sample includes both recently quenched (about 100–800 Myr) and older passively evolving galaxies (about 1.3–1.6 Gyr) 15 , 16 , 17 , 18 . Future observations constraining the distribution of dust temperatures may add clarity to these differences: because dust temperature changes the peak wavelength of the far-infrared dust bump, an overall hotter average dust temperature will decrease the millimetre flux density for a given total infrared luminosity, whereas it will increase for colder dust temperatures 9 . The large range in molecular gas (and dust) fractions observed at low star-formation rates across redshift, from less than 2–5% (refs. 2 , 3 , 4 , 5 , 6 , 7 ) to 10% (refs. 8 , 9 ) and up to 40% (refs. 3 , 10 , 11 ), may suggest a diversity in dust temperatures ( Methods ) or, more fundamentally, a diverse set of evolutionary pathways to quiescence. The galaxies in our sample either depleted their cold gas within the first few billion years of the Big Bang, or ejected it into the surrounding intergalactic medium. Chemical evolution arguments based on observed high metallicity and high alpha/Fe ratios in local early-type galaxies support the same interpretation, where massive galaxies must have consumed all of the available gas within roughly 1 billion years (ref. 23 ). Larger samples to similar or greater depth are needed to determine whether this scenario is generally applicable. Quiescent galaxies are spectroscopically confirmed as early as z = 4 (ref. 24 ). The existence of these early quiescent galaxies and the rapid and complete exhaustion of gas implied by our data are critical constraints on models of galaxy evolution, which currently struggle to produce realistic quiescent galaxies across redshift 21 . Predictions from cosmological simulations for the molecular gas leftover after star formation ceases span multiple orders of magnitude 25 , 26 . The essential problem is that high redshift dark matter halos contain enormous gas reservoirs 12 , 13 , 14 that should cool efficiently and maintain steady star formation over long timescales 27 , 28 . Indeed, many early massive galaxies do just that, with star-formation rates of order 100 M ☉ yr −1 (ref. 29 ) and sizeable molecular gas reservoirs 13 . Our observations show that the cessation of star formation for these galaxies is not caused by a sudden inefficiency in the conversion of cold gas to stars but due to the depletion or removal of their reservoirs. This lack of cold gas may be permanent. In the absence of a heating mechanism, the hot gas biding time in the halo of massive galaxies should theoretically cool and fall back onto galaxies within 1 billion years (ref. 30 ). Yet, we do not frequently observe rejuvenation in massive galaxies 31 . In light of this, there must be a physical mechanism that effectively blocks the replenishment of the cold gas reservoirs 32 . In the local universe, centrally driven winds observed in quiescent galaxies are known to clear the gas out of the system, and the central low-level active supermassive black hole has sufficient mechanical energy to heat the gas and suppress star formation 33 . Tentative evidence also exists at high redshifts for maintenance mode energy injection from central supermassive black holes 34 . This process may explain why quiescent galaxies are unable to effectively re-accrete cold gas in the subsequent 10 billion years of evolution to the present day, although there are other possibilities 35 . Our new data demonstrate a lack of dust, and by inference cold gas, indicating that such a physical process may have already occurred at significantly earlier times for some galaxies. With extremely sensitive limits on the dust continuum of individual massive quiescent galaxies at z ~ 2, our measurements imply extremely low f H2 of a few per cent or less. However, the use of the dust continuum as a proxy for the interstellar medium in massive galaxies with star-formation rates must be further investigated. In particular, while securing detections of both CO emission and dust continuum for the same high redshift quiescent galaxy is paramount, such observations are costly with our current generation of telescopes without the help of strong gravitational lensing magnification. Methods Cosmology and initial mass function assumptions Throughout this paper we assume a simplified cosmology of Ω M = 0.3, Ω Λ = 0.7 and H 0 = 70 km s −1 Mpc −1 when calculating physical parameters. Such values are commonly assumed to make literature comparisons easier, as the precise measured values evolve over time. We adopt the Chabrier 36 initial mass function throughout, correcting literature values where appropriate. Hubble and Spitzer Space Telescope observations The full details of the data reduction of the REQUIEM Hubble Space Telescope (HST) and Spitzer Space Telescope data are found in the REQUIEM methodology paper 37 . All targets have a minimum of 5 (up to 16) HST and 2 Spitzer Infrared Array Camera filters, covering λ rest ~ 1,000Å to ~1 μm. In addition to ground-based spectroscopic campaigns 15 , 16 , 17 , 18 , HST Wide-Field Camera 3 G141 grism spectroscopy exists for five out of the six targets, excluding MRG-M1423. Star-formation rate and stellar mass estimates Star-formation rates and stellar mass estimates are taken from the literature, derived from joint analyses of photometry and ground-based spectroscopy, modelling the rest-frame ultraviolet to near-infrared spectral energy distribution 16 , 18 . These papers adopt the Calzetti 38 dust attenuation curve and parameterized star-formation histories when fitting the stellar continuum with stellar population synthesis models 39 . Both exponentially decaying 16 and similar star-formation histories that allow linear growth before the exponential decay 18 yield consistent stellar mass and star-formation rate estimates and are generally well-suited to describe quiescent galaxies 40 . The procedures to fit the data to stellar population models marginalize over the spectroscopic redshift, velocity dispersion, age, metallicity, dust attenuation and the emission line parameters, including an analysis of systematic uncertainties introduced by the model assumptions. The choice of dust attenuation law and star formation history (SFH) will affect the inferred stellar masses and SFRs in particular. Studies suggest that the dust attenuation law is not universal 41 and that the Calzetti attenuation curve may not be representative at high redshift and/or low SFR per unit stellar mass (sSFR) 42 . Moreover, parametric SFHs are shown to yield systematically lower stellar masses owing to younger ages relative to non-parametric SFHs 43 . Although there remain significant uncertainties in the dust geometry given our centrally concentrated, unresolved dust continuum detections, it is valuable to test the effect of different dust attenuation assumptions and a non-parametric SFH on the global measurement of stellar mass and sSFR through a preliminary joint analysis of the HST and Spitzer photometry and HST grism spectroscopy using published Bayesian methodology 37 . Namely, we independently derive the measured stellar masses and sSFRs for the two ALMA-detected galaxies, MRG-M0138 and MRG-M2129. We adopt non-parametric SFHs, using the flexible stellar population synthesis models 44 with a two-parameter dust model 45 . Consistent with expectations, these tests yield higher stellar masses by 0.1–0.2 dex, and thus lower implied molecular gas and dust fractions. When including the 1.3 mm measurement in the fit for MRG-M0138, we find: (1) a (non-parametric) SFH that declines exponentially with old ages, low sSFRs and low dust, consistent with the ground-based spectroscopic results; and (2) dust temperatures that are preferentially warmer. A warm luminosity-weighted dust temperature of about 34 K is required to explain the low ALMA flux density, as most of the infrared energy escapes at shorter wavelengths. Conversely, for MRG-M2129 we find: (1) a younger post-starburst SFH with moderate dust attenuation and a steeper than Calzetti curve; and (2) a dust spectral energy distribution (SED) that is consistent with a very cold (about 14 K), though not maximally cold, temperature. Combined, these implied dust temperatures fall at the extremes of local observations for early-type galaxies (see 'Molecular gas mass estimates', below), and make testable predictions motivating future observations. However, it is important to note that the dust comes only from a yet unconstrained small central region, making it imperative to not overinterpret the global SED modelling. Regardless of the specific model adopted, the changes in stellar mass and sSFR for these galaxies do not impact the conclusions of this study, despite the significant challenges of constraining the SFR in the low sSFR regime in particular. If anything, our tests imply even more extreme cold gas depletion timescales on the order of 100 Myr (versus about 1 Gyr for previous SED modelling assumptions). So although we conservatively adopt the published values based on higher resolution ground-based spectroscopy 16 , 18 , derived completely independent from the dust masses, we note that our measurements may deviate even further from scaling relations under different modelling assumptions, which would only serve to strengthen our conclusions. Lens model assumptions The full details of the lens models for all strong lensed sources in this paper can be found in the original discovery papers 15 , 16 , 17 , 18 . The magnification factor was used to correct the stellar masses and star-formation rates. However, because the dust and molecular gas fractions and the specific star-formation rates, the main focus of this paper, are relative quantities, they are independent of the details of the lens models. Reduction of ALMA data ALMA 1.3 mm continuum observations were carried out in programs 2018.1.00276.S and 2019.1.0027.S. The observations were designed to reach limits of f H2 ~ 1%; due to the range in redshift and lensing magnification within the sample, the observations reach 1 σ depths of 9–56 μJy. The correlator was configured for standard Band 6 continuum observations, with 7.5 GHz total usable bandwidth. The data were reduced using the standard ALMA pipeline and imaged with natural weighting to maximize sensitivity. The observations were designed to avoid spatially resolving the target sources to the extent possible, and reach spatial resolutions of ~1.0–1.5 arcseconds. We create lower-resolution images of each source with a uv taper and find no evidence for extended emission in any source (see below). Flux densities for the two detected sources were measured from the peak pixel values in the images, appropriate for unresolved (pointlike) sources. For the remaining undetected sources, we place upper limits on the 1.3 mm emission using the image root mean squared values, under the assumption that the dust emission in the remaining sources would also be as compact as that in the two detected galaxies. We further verify that the submillimetre emission is unresolved in the two ALMA-detected objects in several different ways: by comparing peak to integrated flux densities, asymmetric tapered to untapered fluxes (such that the position angle of the resulting synthesized beam is aligned with the extended lensed arc), and uv plane and image-plane fitting, all of which agree that the two detected sources are indeed pointlike, with no evidence of extended emission. For the two detected galaxies, we carry out a further test for millimetre emission extended on scales as large as the rest-optical light. In brief, we create a series of mock ALMA observations using a model of the image-plane stellar light from the HST F160W images, renormalize the image to have either a known total flux density or known peak flux density (per beam), invert it, sample the Fourier transform at the uv coordinates of the real data, and add noise to the visibility model based on the noise properties of the real data. The two detected sources are representative of the others: MRG-M0138 is a highly extended arc, whereas MRG-M2129 is only slightly extended compared to the synthesized beam. By normalizing the model image to match the peak surface brightness of the real data, we test the extent to which the existing data rule out millimetre emission with comparable extent as the stellar light. By normalizing the model image to a known total flux density, we test our ability to recover known input signals and whether it is possible to find pointlike millimetre emission even if the true emission has the same structure as the stellar light. In both tests, we find that if the millimetre emission had identical structure to the stellar light, the resolved arc structure of the source would be clearly detected. We find that if the sources had the same total flux density as we measure in the real data, but this emission was distributed over the full extent of the stellar light, the input flux density would still be accurately recovered. Importantly, in this case the mock observed galaxies fail all of our previous tests for pointlike emission; the extended nature of the millimetre emission in the mock datasets would be easily discernible at the depth of our data. The millimetre emission in the detected sources is genuinely pointlike at our current spatial resolution, far less extended than the stellar light. Therefore, we conclude that there is no detectable dust emission extended on the same scales as the stellar light, in agreement with our finding that the submillimetre emission is pointlike at the current resolution. Galaxies that are not detected afford an opportunity to stack the dust continuum, reaching below the noise level for any individual map. Although large variations in the strong lensing magnification coupled with small number statistics complicate matters, we generate a weighted stack as a test under the following assumptions. For the four undetected REQUIEM-ALMA galaxies, each non-detection map is divided by the magnification and the individual maps’ demagnified root mean square defines the weight when averaging to generate a weighted stack. This methodology is similar to others in the literature for unresolved sources 46 , with our sample having roughly similar beam sizes that span 1.4–1.6 × 1.1–1.2 arcseconds. The same weights are used to calculate the average stellar mass and consequently the limit in f dust = M dust / M ★ for the stack. The resulting deep 3 σ limit in the dust continuum from the undetected REQUIEM-ALMA sources is 4.58 Jy at an average redshift of z = 2.59. For an average weighted stellar mass of log 10 (M ★ /M ☉ ) of 10.50, this corresponds to f dust of 1.8 × 10 -4 , largely driven by the highest magnification source, MRG-M1341, that also has the deepest ALMA 1.3 mm limits but the lowest stellar mass. Molecular gas mass estimates By probing the Rayleigh–Jeans tail at λ rest > 250 μm, the dust continuum can be used as a proxy for the mass of the molecular interstellar medium, M H2 . We estimate M dust from a modified blackbody fit 47 , assuming a dust temperature of 25 K, a dust emissivity index, β, of 1.8, and a dust mass opacity coefficient, κ 345GHz of 0.0484 m 2 kg −1 (ref. 48 ). By assuming a molecular gas to dust mass ratio, δ, of 100 (ref. 48 ), we can infer M H2 from M dust . In principle M dust could trace both neutral and molecular hydrogen, and quiescent galaxies at z ~ 0 are known to contain non-negligible neutral gas reservoirs 49 . Local studies show that the neutral hydrogen contribution varies widely 21 , 50 . Although we assume that all of the hydrogen gas is in the molecular form, a significant contribution from neutral hydrogen to our dust detection would only serve to strengthen our conclusion. For comparison, we also calculate M H2 explicitly following an empirical calibration 19 , finding an offset of 0.1 dex lower in M H2 , yielding even lower inferred molecular gas fractions. An alternative viable explanation of the null detections is that δ increases markedly for a significant fraction of early quiescent galaxies. There exists theoretical 51 and observational 52 evidence that in certain circumstances thermal sputtering by hot electrons could in principle efficiently destroy dust in dead galaxies. CO observations are required to rule out extreme molecular gas to dust ratios that would be necessary to reconcile our observations with higher, more typical values of f H2 . Although CO observations of quiescent galaxies at z > 1.5 are scant, such ratios are difficult to justify, as they imply that CO should be detectable 7 , 53 . At least in the case of our two detections, such exotic ratios are already ruled out by strong CO upper limits (A. Man, personal communication). We adopt a dust temperature of 25 K, which corresponds to a luminosity-weighted temperature of approximately 30 K. However, the cold interstellar medium of local quiescent galaxies is generally colder, with luminosity-weighted dust temperatures observed to be 23.9 ± 0.8 K (with a range from 16 K to 32 K) 54 . Adopting significantly colder dust templates would increase our estimates of molecular gas fraction, but our upper limits would still leave room for tension. The thin error bars for the two detected sources in Fig. 2 represent the systematic uncertainty in dust temperature from a Monte Carlo analysis adopting the observed temperature distribution of local quiescent galaxies 54 . Systematic uncertainties in Fig. 3 additionally include variation in the molecular gas to dust ratio by conservatively drawing from a uniform distribution ranging from 50 to 200. Star formation in quiescent galaxies at high redshift is generally less suppressed in comparison to that in local dead galaxies, and as such the expected dust temperature of the cold interstellar medium remains unclear. When including the measured ALMA 1.3 mm flux density in global SED modelling that assumes energy balance, as described above, we find that MRG-M0138 may be consistent with warmer dust with luminosity-weighted temperatures of about 34 K and MRG-M2129 with colder dust at about 14 K. Although we cannot draw conclusions on the dust temperature based on a single (unresolved) data point in the Rayleigh–Jeans tail with ample uncertainties looming in the dust geometry, the latter may support conclusions based on stacked observations 9 , whereas the former would represent a new extreme. It may be that once dust production from new star formation halts, dust is slowly removed by other physical processes; when the dust reservoirs are sufficiently depleted, the galaxy is optically thin and this dust may then be heated to higher temperatures. However, although the dust is still optically thick, self-shielding may effectively allow the dust to cool to very low temperatures 43 . Future observations and spatially resolved analyses will illuminate the dust morphology and temperature. Literature comparisons We include quiescent targets measured through dust continuum in Fig. 2 , both upper limits for individual galaxies 5 , 6 and for stacking 8 , 9 , as well as an individual quiescent CO upper limit measurement 4 in Fig. 3 ; all studies have a similar high redshift of z > 1.5. For the dust continuum measurements, all data are recalibrated using the same set of assumptions applied here, starting from the flux density and source redshift. We compare our results to a comprehensive compilation of 843 galaxies out to z = 3 from the literature with dust or CO measurements 6 , shown as contours in Fig. 2 . Within this sample, we highlight measurements of 188 (almost exclusively star-forming) galaxies at 1.5 < z < 3.0, tracing molecular gas via dust continuum 6 , 19 , 55 , 56 and CO 11 , 12 , 14 , 57 , 58 , 59 , 60 , 61 , 62 , 63 (contours presented in Fig. 3 ). Data availability Data that support the findings of this study are publicly available through the ALMA Science Archive under project codes 2018.1.00276.S and 2019.1.00227.S and the Barbara A. Mikulski Archive for Space Telescope under project code HST-GO-15663 (including additional archival data from project codes HST-GO-9722 , HST-GO-9836 , HST-SNAP-11103 , HST-GO-11591 , HST-GO-12099 , HST-GO-12100 , HST-SNAP-12884 , HST-GO-13459 , HST-SNAP-14098 , HST-GO-14205 , HST-GO-14496 , HST-SNAP-15132 and HST-GO-15466 ). All HST and ALMA mosaics are publicly available at . Derived data and codes supporting the findings of this study are available from the corresponding author upon request. Source data are provided with this paper.

Early massive galaxies—those that formed in the 3 billion years following the Big Bang—should have contained large amounts of cold hydrogen gas, the fuel required to make stars. But scientists observing the early universe with the Atacama Large Millimeter/submillimeter Array (ALMA) and the Hubble Space Telescope have spotted something strange: a half-dozen early massive galaxies that ran out of fuel. The results of the research are published today in Nature. Known as "quenched" galaxies—or galaxies that have shut down star formation—the six high-redshift galaxies that were selected for observation from the REQUIEM survey are inconsistent with what astronomers expect of the early universe. "The most massive galaxies in the universe lived fast and furious, creating their stars in a remarkably short amount of time. Gas, the fuel of star formation, should be plentiful at these early times in the universe," said Kate Whitaker, lead author on the study, and assistant professor of astronomy at the University of Massachusetts, Amherst. "We originally believed that these quenched galaxies hit the brakes just a few billion years after the Big Bang. In our new research, we've concluded that early galaxies didn't actually put the brakes on, but rather, they were running on empty." To better understand how the galaxies formed and died, the team observed them using Hubble, which revealed details about the stars residing in the galaxies. Concurrent observations with ALMA revealed the galaxies' continuum emission—a tracer of dust—at millimeter wavelengths, allowing the team to infer the amount of gas in the galaxies. The use of the two telescopes is by careful design, as the purpose of REQUIEM is to use strong gravitational lensing as a natural telescope to observe dormant galaxies with higher spatial resolution. This, in turn, gives scientists a clear view of galaxies' internal goings-on, a task often impossible with those running on empty. This composite image of galaxy cluster MACSJ 0138 shows data from the Atacama Large Millimeter/submillimeter Array (ALMA) and NASA’s Hubble Space Telescope, as observed by REsolving QUIEscent Magnified galaxies at high redshift, or the REQUIEM survey. The early massive galaxies studied by REQUIEM were found to be lacking in cold hydrogen gas, the fuel required to form stars. Credit: ALMA (ESO/NAOJ/NRAO)/S. Dagnello (NRAO), STScI, K. Whitaker et al "If a galaxy isn't making many new stars it gets very faint very fast so it is difficult or impossible to observe them in detail with any individual telescope. REQUIEM solves this by studying galaxies that are gravitationally lensed, meaning their light gets stretched and magnified as it bends and warps around other galaxies much closer to the Milky Way," said Justin Spilker, a co-author on the new study, and a NASA Hubble postdoctoral fellow at the University of Texas at Austin. "In this way, gravitational lensing, combined with the resolving power and sensitivity of Hubble and ALMA, acts as a natural telescope and makes these dying galaxies appear bigger and brighter than they are in reality, allowing us to see what's going on and what isn't." The new observations showed that the cessation of star formation in the six target galaxies was not caused by a sudden inefficiency in the conversion of cold gas to stars. Instead, it was the result of the depletion or removal of the gas reservoirs in the galaxies. "We don't yet understand why this happens, but possible explanations could be that either the primary gas supply fueling the galaxy is cut off, or perhaps a supermassive black hole is injecting energy that keeps the gas in the galaxy hot," said Christina Williams, an astronomer at the University of Arizona and co-author on the research. "Essentially, this means that the galaxies are unable to refill the fuel tank, and thus, unable to restart the engine on star production." The study also represents a number of important firsts in the measurement of early massive galaxies, synthesizing information that will guide future studies of the early universe for years to come. "These are the first measurements of the cold dust continuum of distant dormant galaxies, and in fact, the first measurements of this kind outside the local universe," said Whitaker, adding that the new study has allowed scientists to see how much gas individual dead galaxies have. "We were able to probe the fuel of star formation in these early massive galaxies deep enough to take the first measurements of the gas tank reading, giving us a critically missing viewpoint of the cold gas properties of these galaxies." Although the team now knows that these galaxies are running on empty and that something is keeping them from refilling the tank and from forming new stars, the study represents just the first in a series of inquiries into what made early massive galaxies go, or not. "We still have so much to learn about why the most massive galaxies formed so early in the universe and why they shut down their star formation when so much cold gas was readily available to them," said Whitaker. "The mere fact that these massive beasts of the cosmos formed 100 billion stars within about a billion years and then suddenly shut down their star formation is a mystery we would all love to solve, and REQUIEM has provided the first clue."

10.1038/s41586-021-03806-7

Medicine

Research uncovers how fructose in the diet contributes to obesity

Samuel R. Taylor et al, Dietary fructose improves intestinal cell survival and nutrient absorption, Nature (2021). DOI: 10.1038/s41586-021-03827-2 Journal information: Nature

http://dx.doi.org/10.1038/s41586-021-03827-2

https://medicalxpress.com/news/2021-08-uncovers-fructose-diet-contributes-obesity.html

Abstract Fructose consumption is linked to the rising incidence of obesity and cancer, which are two of the leading causes of morbidity and mortality globally 1 , 2 . Dietary fructose metabolism begins at the epithelium of the small intestine, where fructose is transported by glucose transporter type 5 (GLUT5; encoded by SLC2A5 ) and phosphorylated by ketohexokinase to form fructose 1-phosphate, which accumulates to high levels in the cell 3 , 4 . Although this pathway has been implicated in obesity and tumour promotion, the exact mechanism that drives these pathologies in the intestine remains unclear. Here we show that dietary fructose improves the survival of intestinal cells and increases intestinal villus length in several mouse models. The increase in villus length expands the surface area of the gut and increases nutrient absorption and adiposity in mice that are fed a high-fat diet. In hypoxic intestinal cells, fructose 1-phosphate inhibits the M2 isoform of pyruvate kinase to promote cell survival 5 , 6 , 7 . Genetic ablation of ketohexokinase or stimulation of pyruvate kinase prevents villus elongation and abolishes the nutrient absorption and tumour growth that are induced by feeding mice with high-fructose corn syrup. The ability of fructose to promote cell survival through an allosteric metabolite thus provides additional insights into the excess adiposity generated by a Western diet, and a compelling explanation for the promotion of tumour growth by high-fructose corn syrup. Main Humans in the Western world consume more fructose now than ever before in recorded history. Agricultural and industrial advances have improved the access to sweeteners like sucrose and high-fructose corn syrup (HFCS), which have tripled the total consumption of fructose and contributed to a burgeoning epidemic of obesity and related diseases 8 , 9 . The global rise in obesity is directly linked to an increase in obesity-related cancers such as colorectal cancer (CRC), the incidence and mortality of which are rising among young adults 10 , 11 . Several observations suggest that there is a causal relationship between fructose consumption and CRC. For example, fructose consumption is associated with the incidence and progression of gastrointestinal cancer 2 , 12 , 13 , 14 , and drives tumour growth and metastasis in mouse models of CRC 15 , 16 . As tumour growth is driven by hyperplasia and tumour cells frequently retain metabolic pathways from their tissue of origin, we hypothesized that fructose would promote hyperplasia of the normal intestinal epithelium just as it promotes growth in intestinal tumours. To assess this, we fed mice HFCS for four weeks and quantified the mean intestinal villus length using a high-throughput, unbiased image-segmentation-based approach (Extended Data Fig. 1a–f ). Mice of both sexes and a variety of ages and genetic backgrounds that were treated with HFCS showed a 25–40% increase in intestinal villus length in the duodenum and proximal jejunum compared to H 2 O-treated control mice (Fig. 1a , Extended Data Fig. 1h ). The increase in villus length correlated with increased weight gain and fat accumulation as well as lipid absorption (Extended Data Fig. 2a–l ). Fig. 1: Dietary fructose increases intestinal villus length and lipid absorption. a , Haematoxylin and eosin (H&E)-stained duodenum from mice that were fed normal chow with ad libitum H 2 O or 25% HFCS for four weeks. Scale bars, 3 mm (top); 200 μm (bottom). b , Relative change in the body mass of mice that were fed a control diet, a high-fat diet (45% kcal fat) (HF) or a high-fat, high-sucrose diet (HFHS) ( n = 5 mice per group). c , Mass of white adipose tissue from the gonadal depot after five weeks on each diet ( n = 5 mice per group; two depots per mouse). d , Relative duodenal villus length after five weeks on each diet ( n = 5 mice per group). e , Serum triglyceride levels in fasted mice after an oral gavage with olive oil ( n = 3 mice per group). f , BrdU immunohistochemistry (IHC) of duodenal sections from H 2 O or HFCS-treated mice 72 h after intraperitoneal BrdU injection. Scale bars, 200 μm. g , Duodenal villus length distal to the BrdU front ( n = 3 mice per group; 40 villi per mouse). h , i , IHC for CC3 ( h ) and pimodinazole ( i ) in duodenal sections from H 2 O-treated mice. Scale bars, 200 μm. b – e , One-way ANOVA followed by Holm–Sidak post-hoc test for multiple comparisons; g , two-sided Student’s t -test. NS, not significant; * P < 0.05, ** P < 0.01, **** P < 0.0001; exact P values are provided in the Source Data for all figures. All data are mean ± s.e.m. Source data Full size image We hypothesized that this increase in absorption would exacerbate weight gain in mice placed on a high-fat diet (HFD) that contained fructose. Over four weeks, mice were treated with a control diet that contained no fructose, a standard HFD (45% of calories from fat) that contained dextrose but no fructose, or an isocaloric HFD in which the dextrose was replaced by sucrose (Supplementary Table 1 ). Mice on the sucrose-fortified HFD gained significantly more weight and fat mass than those on the standard HFD, despite consuming and expending the same amount of energy (Fig. 1b, c , Extended Data Fig. 2m–v ). In agreement with the data from mice that consumed a normal chow (low-fat) diet, mice that were fed fructose in the form of sucrose had a similar small intestinal length but longer villi (Fig. 1d , Extended Data Fig. 2r ), exhibited increased levels of serum triglycerides after an oral lipid bolus (Fig. 1e ) and lost less energy in the faeces compared to isocaloric, sucrose-free control mice (Extended Data Fig. 2w–z ). These data suggest that dietary fructose increases intestinal villus length and nutrient absorption. Intestinal villus length is determined by a balance between the rates of proliferation and death of epithelial cells. Thus, the villus is constantly in a state of self-renewal as stem cells in the crypt divide to yield new intestinal epithelial cells (IECs), which then transit outward until they reach the villus apex and are extruded into the intestinal lumen 17 . To determine whether the longer villi resulted from an increased rate of migration (that is, proliferation) or increased cell survival, we conducted single- and dual-label tracing experiments using 5-bromo-2′-deoxyuridine (BrdU) and 5-ethynyl-2′-deoxyuridine (EdU) injections at several different time points. These assays showed that the duodenal villi of HFCS-treated mice had similar migration rates to those of H 2 O-treated control mice (Extended Data Fig. 3a-e ), but that they had more than twice as many IECs surviving longer than 72 h than did control mice (Fig. 1f, g ). There was also no change in cell proliferation as assessed by histologic Ki-67 staining (Extended Data Fig. 3f ). These data indicate that cell survival is a major determinant of the hypertrophy of villi in the presence of fructose. Cell transit up the intestinal villus terminates with cell death and extrusion into the intestinal lumen 17 . Indeed, in all cases in which extruding cells were captured in histological sections, staining for the apoptosis marker cleaved caspase-3 (CC3) or for terminal deoxynucleotidyl transferase dUTP nick-end labelling (TUNEL) was positive, regardless of diet (Fig. 1h , Extended Data Fig. 3f ). Because IECs migrate away from their blood supply during their transit along the villus, this cell death is likely to be influenced by tissue hypoxia. Consistent with this theory, pimonidazole staining, which is used to indicate tissues in which the partial pressure of oxygen is less than 10 mm Hg, correlated with distance from the muscularis layer in the small and large intestine of both H 2 O-treated and HFCS-treated mice (Fig. 1i , Extended Data Fig. 3g, h ). Despite this apparent similarity in hypoxia patterns between H 2 O- and HFCS-treated mice, we observed an increase in the hypoxia-inducible factor-1α (HIF-1α) target proteins enolase-1 (ENO1) and lactate dehydrogenase A (LDHA) in the intestinal epithelium of HFCS-treated mice, and a strong upregulation of the fructolytic proteins GLUT5 and ketohexokinase (KHK) (Extended Data Fig. 3i–k ). As hypoxia is a driver of cell death in a wide variety of tissues and contexts, we next examined whether fructose could also mitigate cell death in human CRC cell lines cultured in hypoxia. The addition of fructose did not affect the cell growth rate but did improve the survival of hypoxic HCT116 and DLD1 cells (Fig. 2a, b , Extended Data Fig. 4a–f ). Fructose also improved the survival of hypoxic mouse intestinal organoids. Hypoxia induced intense apoptosis in the organoid core (the morphological correspondent to the villus), which was reflected by both increased CC3 intensity and a decreased population of viable cells, and these changes were abrogated by treatment with fructose (Extended Data Fig. 4g, h ). We observed no increase in organoid proliferation with fructose treatment (Extended Data Fig. 4i ), indicating that the benefit of fructose is primarily attributable to cell survival in this context as well. Fig. 2: Fructose metabolism enhances hypoxic cell survival and decreases pyruvate kinase activity. a , Confluence of HCT116 cells grown in hypoxia with varying concentrations of fructose (Fru) ( n = 3 biological replicates per group). b , CytoTox viability dye intensity in HCT116 cells cultured in glucose medium with and without fructose. Stain intensity is reported as positive area per well normalized to the initial normoxic glucose control ( n = 3 biological replicates per group). Glc, glucose; Stau, stausporin control. AU, arbitrary units. In these and other cell viability assays, unless otherwise noted, glucose was replenished daily (see Methods ). c , d , Quantification of metabolites (F1P ion count ( c ); pyruvate to PEP ratio ( d )) in hypoxic HCT116 cells, via liquid chromatography–mass spectrometry (LC–MS) ( n = 3 biological replicates per group). e , f , Pyruvate kinase (PK) activity in hypoxic HCT116 and DLD1 cell lysates and in IEC lysates from mice that were fed the indicated diets (glucose or glucose plus fructose ( e ); H 2 O or HFCS ( f )) for four weeks ( n = 3 independent reaction wells per group; same final protein concentration in each well). g , h , Pyruvate kinase activity of recombinant pyruvate kinase isozymes (PKM1 ( g ); PKM2 ( h )) that were pre-incubated with the indicated metabolites ( n = 3 wells per group). i , Western blot against PKM2 using recombinant PKM2 samples cross-linked with disuccinimidyl glutarate ( n = 3 independent reaction wells per group). T, D and M indicate the putative sizes of tetrameric, dimeric and monomeric PKM2, respectively. a , b , g , h , One-way ANOVA followed by Holm–Sidak post-hoc test for multiple comparisons; c , d , f , two-sided Student’s t -test; e , two-way ANOVA followed by Holm–Sidak post-hoc test for multiple comparisons. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. All data are mean ± s.e.m. For gel source data, see Supplementary Fig. 1 . Source data Full size image Fructose is mainly transported into intestinal cells by the sugar transporter GLUT5, before it is phosphorylated by KHK to form fructose 1-phosphate (F1P). In human CRC cells, the total levels of KHK were consistent regardless of fructose or hypoxia exposure, whereas hypoxia induced an increase in the abundance of GLUT5 (Extended Data Fig. 5a ). Notably, treatment with cobalt chloride led to robust stabilization of HIF-1α, but did not induce the upregulation of GLUT5, suggesting that GLUT5 expression is independent of HIF-1α protein levels. We also noted strong expression of KHK-A in hypoxia in human CRC cells (Extended Data Fig. 5b, c ) and confirmed that exogenous fructose was converted to F1P, but that fructose carbon atoms were not incorporated into downstream glycolytic intermediates (Fig. 2c , Extended Data Fig. 6a ), consistent with previous studies in mouse intestinal tumours 15 . This agreed with our observation that fructose was not depleted from the medium of either human CRC cells or mouse intestinal organoids cultured in hypoxia (Extended Data Fig. 6d–d ). In fact, direct fructose metabolites explained only a small portion of the distinct metabolic signature that is associated with fructose exposure in hypoxia (Extended Data Fig. 6e , Supplementary Fig. 3 ). Upper glycolytic intermediates, however, were increased in hypoxic HCT116 cells and the ratio of pyruvate to phosphoenolpyruvate (PEP) was significantly lower, consistent with inhibition of pyruvate kinase (Fig. 2d , Extended Data Fig. 6f ). Pyruvate kinase is the final rate-limiting enzyme in glycolysis that converts pyruvate to PEP, and the activity of the M2 isoform (PKM2) is highly sensitive to changes in the intracellular metabolome 18 . Moreover, PKM2 expression is high in tissues that are subject to hypoxia, such as tumours and intestinal villi 15 , 19 . Using enzymatic assays, we confirmed the inhibition of pyruvate kinase in CRC cell lysates exposed to fructose and in IEC lysates from mice that were fed with HFCS (Fig. 2e, f , Extended Data Fig. 6g ). Because F1P is structurally similar to the endogenous regulator of PKM2, fructose 1,6-bisphosphate (FBP), we hypothesized that F1P might directly inhibit the activity of PKM2. FBP binds tightly in a regulatory pocket distant from the active site and stabilizes PKM2 in a highly active tetramer. Docking simulations showed that F1P can occupy the same pocket but lacks the outward-facing phosphate group that is necessary to interact with an adjacent peptide loop that is critical for tetramer formation 20 (Extended Data Fig. 7a ). Consistent with this mechanism, we found that F1P robustly inhibited PKM2 but did not inhibit PKM1, which lacks the FBP-binding pocket 18 (Fig. 2g, h ), and that this inhibition was accompanied by a decrease in the proportion of tetrameric to monomeric enzyme (Fig. 2i , Extended Data Fig. 7b, c ). By contrast, PKL—a pyruvate kinase isozyme with a modified type of FBP-binding pocket—was only partially inhibited by F1P (Extended Data Fig. 7d ). In an FBP-unresponsive mutant version of PKM2 (PKM2(R489L)) 20 , we still noted strong inhibition, suggesting that F1P not only competes with FBP for binding but also directly inhibits PKM2 once bound (Extended Data Fig. 7e ). In line with this, we found that the interaction between F1P and Ser519—a residue deep in the binding pocket—is critical for inhibition, as mutation of this serine residue to alanine ablated F1P inhibition while preserving FBP activation (Extended Data Fig. 7f–h ). The inhibitory effects of F1P could be also be overcome by using the small molecule TEPP-46 to activate PKM2 at a site remote from the FBP pocket 7 (Fig. 3a , Extended Data Fig. 7i ). Fig. 3: PK activation diminishes the effect of fructose on hypoxia survival. a , PK activity of recombinant PKM2 incubated with varying concentrations of F1P (F1PM indicates the molar concentration of F1P). FBP with or without TEPP-46 was added either before or after F1P. Half-maximal inhibitory concentration (IC 50 ) values are as follows. With FBP before F1P incubation: 3.3 mM (95% confidence interval: 1.1–9.6 mM); with FBP after F1P incubation: 0.35 mM (95% CI: 0.15–0.80 mM); with FBP and TEPP-46 after F1P incubation: 2.7 mM (95% CI: 1.7–4.4 mM) ( n = 2 wells per data point for FBP before; n = 4 for FBP after; n = 3 for FBP after + TEPP-46). b , Viability of HCT116 cells that were virally transduced with the indicated shRNAs (control scrambled shRNA (shScr) or shRNA targeting PKM2 (sh PKM2 )) and cultured in hypoxia with or without fructose ( n = 3 biological replicates per group). c , Viability of HCT116 cells that were cultured in hypoxia with or without fructose and with either TEPP-46 or control dimethyl sulfoxide (DMSO) ( n = 4 biological replicates per group). d , Relative luminescence of HCT116 cells that were transfected with firefly luciferase HIF-1α reporter (p2.1) and Renilla luciferase constitutive reporter (pRL) and incubated for 24 h in the indicated conditions ( n = 6 biological replicates per group for normoxia; n = 3 for hypoxia). e , ATP levels in HCT116 cells incubated for 24 h in the indicated conditions ( n = 3 biological replicates per group). PKM2 activators DASA-58 and TEPP-46 were used at 50 μM in the culture medium. f , Relative duodenal villus length in mice of the indicated genotypes after four weeks of ad libitum H 2 O or HFCS. Mean villus length is reported relative to H 2 O-treated controls for each genotype (mice per group: left to right: 5, 5, 5, 8, 6, 9). WT, wild type. g , Duodenal villi of wild-type mice that were treated for four weeks with the indicated diets. Scale bars, 200 μm. h , Change in serum triglyceride (TG) levels after an oral lipid bolus in mice treated through daily oral gavage for two weeks ( n = 8 mice per group for H 2 O and HFCS; n = 5 for HFCS + TEPP-46). i , Representative intestines from Apc Q1405X /+ mice treated with the indicated regimens and euthanized at 15 weeks old. Arrows indicate tumours. Scale bars, 2 mm. j , k , Total tumour area per histological section of large and small intestine ( j ) and red blood cell (RBC) count ( k ) in mice at 15 weeks (mice per group: left to right: 6, 5, 4, 6). b – f , j , k , Two-way ANOVA followed by Holm–Sidak post-hoc test for multiple comparisons; h , two-sided Student’s t -test at the 4-h and 7-h time points. NS, not significant; * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. All data are mean ± s.e.m. Source data Full size image To test whether PKM2 activity could influence the effects of fructose in cells and tissues, we first generated HCT116 cells in which PKM2 mRNA was knocked down using short hairpin RNA (shRNA), and exposed these cells to hypoxia in the presence or absence of fructose. In the setting of PKM2 depletion, fructose no longer improved hypoxic cell survival (Fig. 3b , Extended Data Fig. 8a ). Similarly, when we used TEPP-46 to activate PKM2 in cells, the effects of fructose were greatly diminished (Fig. 3c , Extended Data Fig. 8b ). These data suggest that PKM2 is a key mediator of fructose-induced cell survival. The ability to change activity and conformation in response to stimuli probably explains the high expression of PKM2 in rapidly dividing tissues, which must contend with nutrient and oxygen constraints to continue their growth. In low-oxygen states, for example, the inhibition of PKM2 mitigates reactive oxygen species to improve cell survival 5 . Consistent with this role, we observed that fructose reduced the total amount of H 2 O 2 in hypoxic cells, an effect that was abrogated by PK activation and was not observed in PKM2 -knockdown cells (Extended Data Fig. 8c–e ). The PKM2 monomer and dimer are also known to bind and transactivate HIF-1α, a key transcription factor for hypoxic adaptation 21 . Because F1P triggers the formation of these lower-order PKM2 units, we hypothesized that fructose would increase HIF-1α transactivation. Indeed, hypoxic and normoxic HCT116 cells exposed to fructose had increased HIF-1α transcriptional activity, and this result was abrogated by two structurally distinct PK activators 21 (Fig. 3d ). Moreover, HIF-1α activity correlated with intracellular ATP levels and AMP-activated protein kinase (AMPK) signalling, as well as lactate production in hypoxia (Fig. 3e , Extended Data Fig. 8g–i ), consistent with the role of HIF-1α in rewiring cellular metabolism. Together, our data identify F1P as an inhibitor of PKM2, which then amplifies the activity of HIF-1α to promote hypoxic cell survival. The importance of PKM2 in the physiological response to fructose was confirmed using genetic mouse models. Selective deletion of PKM2 in IECs ( Vil1 Cre/+ ;Pkm tm1.1Mgvh/tm1.1Mgvh mice; hereafter referred to as Vil1 Cre/+ ;Pkm2 f/f mice) led to a strong upregulation of PKM1 and increased pyruvate kinase activity in the epithelium (Extended Data Fig. 9a, b ), and also altered the nuclear localization of pyruvate kinase (Extended Data Fig. 9c ). The villi of these mice, as well as mice lacking KHK ( Khk −/− mice), were short and unable to elongate in the presence of HFCS (Fig. 3f ), indicating that F1P and PKM2 are both required. Furthermore, the mice deficient in PKM2 or KHK did not upregulate GLUT5 or HIF-1α target proteins (Extended Data Fig. 9d–f ) and were also protected from increased lipid uptake and fat accumulation after being fed with HFCS (Extended Data Fig. 9g–i ). Pharmacological activation of PKM2 also greatly modified the effects of fructose in the intestine. Even with doses far below those required to maintain effective serum levels of drug (2 mg per kg per day as opposed to 100 mg per kg per day) 7 , mice that were treated with TEPP-46 had higher intestinal pyruvate kinase activity (Extended Data Fig. 9b ) and were protected from the villus elongation induced by HFCS (Fig. 3g ). We repeated this experiment using mice that were administered a once-daily oral gavage of HFCS to more closely approximate typical human consumption 15 . In this model, HFCS-fed mice developed villus elongation extending to the ileum by day 10 of treatment, and this could be prevented and reversed with concurrent administration of TEPP-46 (Extended Data Fig. 10a–c ). As in the genetically altered mice, TEPP-46 protected against HFCS-induced increases in lipid absorption and fat accumulation (Fig. 3H , Extended Data Figs. 9 h, 10d ). Given the effects of PK activation on normal epithelial tissue, we hypothesized that this approach might also inhibit the growth of intestinal tumours. Intestinal tumours originate from IECs both in the crypt and in the villus, so hypoxic stress may be a limiting factor in their development and progression 22 . Consistent with this theory, we observed high expression of PKM2 and other HIF-1α targets in primary human CRC tumours compared to normal adjacent epithelium (Extended Data Fig. 10e, f ). In addition, the activity of pyruvate kinase in these tumours was inhibited relative to adjacent tissue (Extended Data Fig. 10g ), potentially providing a survival advantage in hypoxia. In mouse intestinal tumours, we found regions of hypoxia in the core and along the periphery of the tumour, with many apoptotic cells, as well as upregulation of GLUT5 (Extended Data Fig. 10h–o ). To test whether the activation of pyruvate kinase inhibits tumour growth, we fed HFCS with or without TEPP-46 to mice with a clinically relevant, tumour-predisposing mutation in one allele of the Apc gene ( Apc Q1405X/+ ) (ref. 23 ). In agreement with our previous findings 15 , HFCS led to a more severe tumour burden and more profound anaemia—a complication that is associated with more severe disease and worse survival in this model and in humans. These changes were both prevented by treatment with a low dose of TEPP-46 (Fig. 3i–k , Extended Data Fig. 10p–r ). Together, these findings indicate that fructose promotes hypoxic cell survival in the intestine. This conclusion adds to a growing body of evidence indicating that fructose metabolism is an important component of oxygen sensing in diverse biological contexts 24 , 25 . For example, endogenously produced fructose is critical to the survival of the naked mole-rat in hypoxic burrows and critical in mouse cardiomyocytes after ischaemia, yet the mechanisms behind these interactions are poorly understood 26 , 27 . The finding that fructose-derived F1P inhibits PKM2, an important enzyme in hypoxia adaptation 5 , 6 , offers additional insight into these observations. Given its relative scarcity in systemic circulation, endogenously produced fructose could serve as a highly specific signal for reprogramming cellular metabolism in response to hypoxia—a mechanism that we propose is leveraged (and targetable) when tissues such as intestinal villi and tumours are exposed to exogenous fructose. In addition, we find that the consequence of intestinal cell survival is an expansion of the intestinal surface area, which improves nutrient absorption. This finding may help to explain the growth-promoting effects of fructose in breast-fed infants, the increase in adiposity that occurs in fruit-foraging hibernating animals and the obesogenic properties of a Western-style diet 28 , 29 . Methods Mice and diets Six–eight-week-old male and female C57BL6/J, C57BL6/NJ, FVB and BALB/c mice were obtained from The Jackson Laboratory. Mixed-background male and female ‘G5H’ mice were provided by A. Dannenberg. Vil1 Cre/+ ;Pkm tm1.1Mgvh/tm1.1Mgvh ( Vil1 Cre/+ ;Pkm2 f/f ) mice were generated by crossing B6.Cg-Tg( Vil1 Cre /+ )997Gum (stock number 004586) and B6;129S- Pkm tm1.1Mgvh (stock number 024048) mice purchased from The Jackson Laboratory. Khk −/− mice lacking both KHK-A and KHK-C on the C57BL/6J background were provided by D. Bonthron (University of Leeds) and M. Lanaspa and R. Johnson (University of Colorado) 30 . Apc Q1405X /+ mice on the C57BL/6NJ background were provided by L. Dow (Weill Cornell Medical College) 23 . Unless otherwise indicated, all wild-type experiments used male C57BL/6J mice between the ages of 8 and 16 weeks. All genetically modified models ( Vil1 Cre/+ ;Pkm tm1.1Mgvh/tm1.1Mgvh , Khk −/− , Apc Q1405X /+ ) were equally weighted mixes of males and females between 10 and 20 weeks of age at the time of euthanasia. Mice were maintained in temperature- and humidity-controlled specific-pathogen-free conditions on a 12-h light–dark cycle and received rodent chow (PicoLab Rodent 20 5053 LabDiet) and free access to drinking water. HFCS was prepared by combining D-(+)-glucose (Millipore Sigma, Cat. #G8270) and D-(−)-fructose (Millipore Sigma, F0127) in a 45:55 ratio using tap water. Match Purina 5053 fructose-free control (D17011901), high-fat (D19090601) and high-fat/high-sucrose (D19090602) diets were purchased from Research Diets. Age-matched cohorts were treated with HFCS either by ad libitum delivery in the drinking water (25% HFCS in tap water) or by once-daily oral gavage (45 mg glucose + 55 mg fructose, total 400 μl in tap water). Control mice were treated with tap water in the water bottle or 400 μl of tap water via daily oral gavage. For drug trials, TEPP-46 (Millipore Sigma, 505487) dissolved in DMSO was added to the drinking water to a final concentration of 7.5 μg ml −1 such that the total daily dose for a 30-g mouse consuming 8 ml of water daily was 2 mg kg −1 . Fluid consumption was monitored weekly to confirm that similar amounts of drug were consumed in each cage. Control mice received equal volumes of DMSO in the drinking water. For oral gavage, TEPP-46 was administered at 2 mg kg −1 in HFCS or water. Control mice received an equal volume of DMSO dissolved in HFCS or water. Male and female Apc Q1405X /+ mice in a 1:1 ratio were initiated on their respective diets or treatments at 6 weeks and euthanized at 15 weeks of age. Other mice receiving treatment via the water bottle or diet were initiated on treatment at between 6 and 15 weeks of age and were euthanized after 4–6 weeks of intervention. Mice receiving oral gavage treatment were euthanized after 10–14 days of once-daily gavage. After euthanasia, tissues and intestines were collected, split into five sections (four of equal size for small intestine and one for the colon), Swiss-rolled and fixed in 4% paraformaldehyde (Santa Cruz Biotechnology, SC-281692) overnight at 4 °C. Tissues were then transferred to 70% ethanol and shipped to Histowiz for paraffin embedding, mounting, H&E staining and slide scanning at 40× magnification. All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) of Weill Cornell Medical College and maintained as approved by the Institutional Animal Care and Use Committee (IACUC) at Weill Cornell Medicine under protocol number 2012-0074. Mice were regularly monitored for lethargy, gross weight loss, pallor and rectal prolapse. Mice that exhibited greater than 20% weight loss from peak weight, had red blood cell counts below 1 × 10 9 ml −1 as determined from tail vein sampling or had rectal prolapse were euthanized. These limits were never reached in our experiments. Histological analysis Scanned H&E images of small intestine from each trial were downloaded from Histowiz as ScanScope Virtual Slide (SVS) files and divided into separate files for duodenum, proximal jejunum, distal jejunum and ileum. For manual villi analysis, each intestinal segment was further divided into 4 quadrants and 10 intact villi were measured from the distal edge of the crypt to the villi apex in each quadrant. A mean villus length was then calculated for the entire bowel segment. For semi-automated analysis, SVS files were opened in ImageJ and the length of the intestinal section was measured using the freehand measurement tool (Extended Data Fig. 1a ). Images were then stain-normalized to a standard H&E image using a custom MATLAB (release 2019b) script using a method described previously 31 . A random image from the set was then loaded into the MATLAB Colour Thresholder 32 and values were manually selected within the hue-saturation-value (HSV) colour space such that the intestinal villi, but not other tissues such as lymph nodes or pancreas, were selected (Extended Data Fig. 1b ). These values were entered into a batch processing script that performed this villi segmentation on every image in the set. This resulted in binary images of pixels identified as villi and pixels identified as non-villi. The pixels occupied by villi were converted to area in μm 2 using the embedded scale from the original SVS file. The villi area was divided by the bowel segment length to yield the average thickness of the intestinal villi layer. This measurement correlated well with manual measurements of villi length and provided improved intra-operator or inter-operator variation (Extended Data Fig. 1c–g ). Polyp number and area were determined from SVS files analysed using ImageJ software in a blinded manner. Body composition and lipid tolerance Body mass, fat mass and fat-free mass were measured and calculated using magnetic resonance spectroscopy using an EchoMRI Body Composition Analyzer as previously described 33 . Visceral fat and white adipose tissue were assessed by measuring the weight of the gonadal white adipose depot. To assess lipid tolerance, mice were fasted for 8 h then administered 200 μl olive oil via oral gavage (Whole Foods, Extra Virgin – Cold Processed). Tail blood serum was collected over time and measured via enzymatic assay (see ‘Biochemical analysis’). Mice resumed their diets after completion of the above testing and recovered for at least 48 h before euthanasia. To assess lipid absorption after blocking endogenous lipases, mice were treated with poloxamer 407 as previously described 34 . In brief, mice were fasted and then given an intraperitoneal injection of poloxamer 407. One hour later, triglyceride levels were measured from the serum and the mice were given a 200-μl oral olive oil bolus. Two hours later, serum triglyceride levels were measured again. Comprehensive metabolic monitoring Metabolic monitoring was conducted using a Promethion Metabolic Screening System (Promethion High-Definition Multiplexed Respirometry System for Mice; Sable Systems International) as previously described 35 . In brief, rates of oxygen consumption (VO 2 ) and carbon dioxide production (VCO 2 ) were acquired by indirect calorimetry with a sampling frequency of 1 s. Respirometry values were determined every 5 min; the dwell time for each cage was 30 s, with baseline cage sampling frequency of 30 s occurring every four cages. Values of respiratory exchange ratio were calculated as ratios of VCO 2 to VO 2 . Food intake and body mass were recorded continuously by gravimetric measurements within the cages. Physical activity was determined according to beam breaks within a grid of infrared sensors built into each cage. Energy expenditure was calculated using the Weir equation (energy expenditure = 3.941 kcal/l × VO 2 + 1.106 kcal/l × VCO 2 ) 36 . Energy expenditure is displayed as the total kcal per specified periods of time, with values adjusted by ANCOVA for body mass or corrected body mass using VassarStats. Faecal bomb calorimetry Nutrient absorption was quantified as described previously 37 . Faecal pellets were collected from cage bottoms over 24 h, during which mice were single-caged and housed at 22 °C. Faecal pellets were dehydrated for 48 h and then subjected to bomb calorimetry using a Parr 6725 Semimicro Calorimeter. Immunohistochemistry and immunofluorescence For BrdU tracing experiments, 100 μl of BrdU (100 mg kg −1 , Millipore Sigma, B5002) dissolved in sterile PBS (Corning, 21-040-CV) was injected intraperitoneally 72 h before mouse euthanasia as previously described 38 . For BrdU–EdU dual-labelling experiments, BrdU was injected 48 h before and EdU (10 mg kg −1 , Millpore Sigma, 900584) was injected 24 h before euthanasia. Pimonidazole, a 2-nitroimidazole that is reduced in hypoxic environments and then binds to thiol-containing proteins, was injected intraperitoneally 90 min before euthaniasia as per the manufacturer’s instructions 39 (Hypoxyprobe, HP1-100Kit). Immunohistochemistry was performed on formalin-fixed, paraffin-embedded tissues. Slides were deparaffinized with xylene and rehydrated in a graded ethanol series and water. Antigen retrieval was performed with 0.01 M citrate, pH 6.0 buffer by heating the samples in a pressure cooker for 10 min. Sections were blocked with avidin–biotin blocking for 30 min. Sections were incubated with primary antibody for 1 h at room temperature or overnight at 4 °C, followed by a 60-min incubation with biotinylated anti-rabbit IgG (goat, Vector Laboratories, PK6101, dilution 1:500) at room temperature for rabbit primaries. Mouse primaries on mouse tissues were assayed using a Mouse on Mouse Basic kit (Vector laboratories, BMK-2202) according to the manufacturer’s instructions. Detection was performed with the DAB detection kit (Vector Laboratories, SK-4100) according to the manufacturer’s instructions, followed by counterstaining with haematoxylin and cover slipping with Permount (Thermo Fisher Scientific, SP15-500). Immunofluorescence was performed on formalin-fixed, paraffin-embedded tissues using the same method as above up to the application of the primary antibodies, which were incubated together. Slides were then washed in PBS and incubated with Alexa-Fluor-488- and Alexa-Fluor-568-conjugated secondary antibodies (Thermo Fisher Scientific, A21202 and A10042) as per the manufacturer’s instructions. Slides were then washed and mounted with the TrueVIEW autofluorescence quenching kit with DAPI (Vector Laboratories, SP-8400-15) according to the manufacturer’s instructions. Organoids were stained as described previously 40 . Antibodies used for immunohistochemistry and immunofluorescence included Ki-67 (rabbit, Abcam ab15580, dilution 1:500), CC3 (rabbit, Cell Signaling Technologies (CST) 9661, dilution 1:200), PKM1 (rabbit, CST 7067, dilution 1:600), PKM2 (rabbit, CST 4053, dilution 1:800), BrdU (mouse, Santa Cruz Biotechnology sc-32323, dilution 1:250), pimonidazole adducts (mouse, Hypoxyprobe Mouse-Mab, dilution 1:50), and SLC2A5 (GLUT5, mouse, Invitrogen MA1-036, 1:500). EdU was visualized using the ClickiT Plus EdU Alexa Fluor 647 Imaging Assay Kit (Thermo Fisher Scientific, C10340) according to the manufacturer’s instructions. For BrdU tracing analysis, the total villi length and the length from crypt to the BrdU-labelled cells furthest from the crypt were measured from 40–50 villi in the duodenums of each mouse using ImageScope software (Leica Biosystems). For BrdU–EdU dual-labelled tracing, the difference between the lengths of BrdU- and EdU-stained areas was divided by the time between these two injections. TUNEL staining was performed on formalin-fixed, paraffin-embedded tissues by Histowiz. Imaging Images of fluorescent-stained sections were acquired on a Zeiss LSM 880 laser scanning confocal microscope. Raw TIF files were processed using Fiji (Image J) and/or Photoshop CS (Adobe Systems) to create stacks, adjust levels and/or apply false colouring. Biochemical analysis For measurement of hepatic triglyceride, frozen liver was weighed and digested in 6 volumes of alcoholic KOH (2:1 pure ethanol to 30% KOH) at 60 °C until the tissue was completely dissolved. Then 500 μl of digest was added to 540 μl of 1M MgCl 2 and mixed well. After a 10-min incubation on ice, samples were centrifuged for 30 min at maximum speed. The supernatant was aspirated into a new tube and glycerol content was measured using a calorimetric assay (Stanbio). This assay kit was also used to measure serum triglyceride. Glucose and fructose concentration in cell culture media were measured with the EnzyChrom glucose assay kit (BioAssay Systems, EBGL-100) and EnzyChrom fructose assay kit (BioAssay Systems, EFRU-100). For lactate determination, a previously described spectrophotometric enzymatic assay was adapted for 96-well plates 41 . Pyruvate kinase activity was measured in recombinant protein and cell or tissue lysates in the presence of the indicated allosteric activators or small molecules by a previously described, lactate dehydrogenase (LDH)-coupled reaction in which PEP is converted by pyruvate kinase to pyruvate, which is then rapidly converted to lactate by LDH 42 . LDH consumes NADH and this rate of change was measured using a microplate spectrophotometer (BMG Labtech). Each allosteric regulator was tested in varying concentrations of PEP and the resulting graph of reaction rate versus PEP concentration was fitted to a substrate velocity curve to derive kinetic parameters under each condition. Substrate–velocity curves were plotted using Prism software (GraphPad Software). To calculate PK activity relative to maximum, activity was measured in tissue lysates before and after incubation for 1 h at 37 °C with 1 mM FBP, and the ratio of initial versus activated activity was calculated. Unless otherwise mentioned, the PEP concentration in the final reaction was 0.5 mM. KHK activity was measured using a pyruvate kinase- and LDH-coupled reaction as previously described 43 , with volumes adjusted for the 96-well plate format. Pyruvate and ADP were used as positive technical controls. The fructose concentration was 10 mM in each reaction. Cell lines, cell culture, virus preparations, transfections and culture additives HCT116 and DLD1 cells were obtained from ATCC and cultured in DMEM (Corning, 10-013-CV) supplemented with 10% dialysed fetal bovine serum (dFBS) and 100 U ml −1 penicillin and 100 μg ml −1 streptomycin. All cells were cultured in a humidified incubator at 37 °C and 5% CO 2 unless otherwise stated. For assays not starting at confluence, the initial concentration of glucose was 25 mM. Cells were tested every two months for mycoplasma contamination. Hypoxia treatments were performed using a Forma Series 3 Water Jacketed CO 2 Incubator (Thermo Fisher Scientific). Oxygen was set to 2–4% depending on the meniscus height of the medium, using previously described calculations 44 , to ensure consistent cellular oxygen deficit. Any manipulations to cells requiring more than a 5-min exposure to ambient oxygen were performed in a InvivO2 400 hypoxia workstation set to the appropriate oxygen level (The Baker Company). For all hypoxia treatments the medium was supplemented to 10 mM with HEPES buffer (Corning, 25060CI). High cell density was defined as a seeding density of 1,000 cells per mm 2 . Experiments were performed in 6-well, 12-well or 48-well plates using 3 ml, 2 ml, and 400 μl of medium, respectively; 96-well plates were avoided owing to large meniscus effects on the medium height in this small well format. Glucose consumption rates in hypoxia for HCT116 and DLD1 cells were derived using medium samples from two different time points from confluent cells cultured in hypoxia. A known number of cells were plated the night before the experiment in 12-well plates, and confluence was confirmed the next morning. At the start of the experiment the growth medium was aspirated, and the cells were gently washed with warm PBS. Then, fresh DMEM with or without 10 mM fructose was added to the culture dishes. Samples from the initial medium were then frozen at −20 °C. Forty-eight hours later, the medium was collected from each well and frozen at −20 °C. Medium samples were subsequently tested for sugar content by an enzymatic assay, and the difference in sugar concentration between the final and initial time points was calculated and divided by the time between time points to establish a rate of decrease. This was then divided by the number of cells plated to calculate the consumption per 10 6 cells. Lactate production was calculated similarly using initial and final medium samples. For experiments plated at confluence lasting longer than 24 h, cells were plated in 10 mM glucose and glucose was replenished at a rate of 15 μmol per day per 10 6 initial cells unless otherwise noted. Cells used for metabolite labelling were plated at confluence in medium with 10 mM glucose ± 10 mM fructose. After 24 h, the cells were gently washed with warm PBS and given reduced-nutrient DMEM (Corning, 17-207-CV) supplemented with 5 mM glucose ± 5 mM fructose, 0.5 mM sodium pyruvate, 10 mM lactate, 1 mM glutamine and 10% dFBS to better simulate the tumour microenvironment during the final 8 h labelling period 45 . pLKO-shPKM2 was a gift from D. Anastasiou (Addgene, plasmid 42516) and scramble shRNA was a gift from D. Sabatini (Addgene, plasmid 1864). Lentiviruses were produced in 293T cells by co-transfection of plasmids expressing gag/pol, rev and vsvg with the respective pLKO. Selection was achieved with puromycin for at least 4 days. p2.1 and pRL-SV40 were procured from Addgene (Addgene, plasmids 27563 and 27163) and used as previously described 21 . Luciferase and Renilla activity were detected using the dual-luciferase reporter assay system (Promega, E1910) as per the manufacturer’s instructions. TEPP-46 (Millipore Sigma, 505487) was used at 1 μM and 50 μM in recombinant and cell-culture experiments, respectively, unless otherwise stated. DASA-58 (MedChemExpress, HY-19330), was used at 50 μM. N -acetylcysteine (Millipore Sigma, A9165) was diluted in medium, pH-balanced to 7.4 and used at 2.5 mM. Cell confluence, viability assays, ThiolTracker and reactive oxygen species Cell confluence, Annexin V Green (Essen BioScience, 4642), and Cytotox Red (Essen BioScience, 4632) measurements were conducted using an IncuCyte ZOOM Live Cell Analysis System (Essen BioScience) according to the manufacturer’s instructions. For Trypan Blue measurements, cells were trypsinized for 3 min at 37 °C and neutralized in complete medium. Resuspended cells were mixed 1:1 with Trypan Blue solution (Millipore Sigma, T8154) and analysed on a Cellometer Auto T4 bright field cell counter (Nexcelom Bioscience). For measurements of viability in adherent cells, the Cell Counting Kit-8 (Dojindo Molecular Technologies, CK04-05) was used according to the manufacturer’s instructions. For measurement of reduced thiols, confluent cell culture plates were incubated in hypoxia in reduced-nutrient DMEM (Corning, 17-207-CV) supplemented with 10% dFBS and 10 mM glucose or 5 mM glucose and 5 mM fructose. After 24 h, ThiolTracker Violet (Life Technologies, T10095) was used according to the manufacturer’s instructions at 10 μM and plates were analysed on a Synergy Neo 2 plate reader (BioTek Instruments). Total cell reactive oxygen species (ROS) measurements were performed using the ROS-Glo H2O2 assay (Promega, G8820) as per the manufacturer’s instructions. Values from normoxic and hypoxic plates were compared after correcting for cell-independent changes in ROS-Glo measurements using wells containing only culture medium. ATP measurements were acquired using the CellTiter-Glo Luminescent Cell Viability Assay (Promega, G7570) as per the manufacturer’s instructions. Isolation and culture of intestinal organoids Isolation, maintenance and staining of mouse intestinal organoids was performed as described previously 23 . For isolation, 15 cm of the proximal small intestine was removed and flushed with cold PBS. The intestine was then cut into 5-mm pieces, vigorously resuspended in 5 mmol l −1 EDTA-PBS using a 10-ml pipette and placed at 4 °C on a benchtop roller for 10 min. This was then repeated for a second time for 30 min. After repeated mechanical disruption by pipette, released crypts were mixed with 10 ml DMEM basal medium (advanced DMEM/F12 containing penicillin–streptomycin, glutamine, 1 mmol l −1 N -acetylcysteine containing 10 U ml −1 DNase I (Roche, 04716728001), and filtered sequentially through 100-μm and 70-μm filters. FBS (1 ml; final 5%) was added to the filtrate and spun at 1,200 rpm for 4 min. The purified crypts were resuspended in basal medium and mixed 1:10 with growth-factor-reduced Matrigel (BD Biosciences, 354230). Forty microlitres of the resuspension was plated per well in a 48-well plate and placed in a 37 °C incubator to polymerize for 10 min. Small intestinal organoid growth medium (250 μl basal medium containing 50 ng ml −1 EGF (Invitrogen, PMG8043), 50 nM LDN-193189 (Selleck Chemicals, S2618) and 500 ng ml −1 R-spondin (R&D Systems, 3474-RS-050)) was then laid on top of the Matrigel. For subculture and maintenance, the medium was changed on organoids every two days and they were passaged 1:4 every 5 to 7 days. To passage, the growth medium was removed and the Matrigel was resuspended in cold PBS and transferred to a 15-ml Falcon tube. The organoids were mechanically disassociated using a P1000 or a P200 pipette and pipetting 50 to 100 times. Seven millilitres of cold PBS was added to the tube and pipetted 20 times to fully wash the cells. The cells were then centrifuged at 1,000 rpm for 5 min and the supernatant was aspirated. The cells were then resuspended in GFR Matrigel and replated as above. For freezing, after spinning the cells were resuspended in basal medium containing 10% FBS and 10% DMSO and stored in liquid nitrogen indefinitely. For hypoxic organoid experiments, organoids from each independent line were dissociated and plated at a uniform density. After two days, the growth medium was changed to glucose-free basal medium supplemented with 10 mM glucose and 10 mM fructose, where indicated, and organoids were placed in normoxia, 4% O 2 or 1% O 2 conditions for 60 h at 37 °C. Every 24 h the concentration of glucose in the growth medium of each organoid well (3 ml total medium volume) was increased by 5 mM using sterile 1M glucose solution in a sealed hypoxia workstation. At the end of the hypoxic culture period, the medium was inoculated with EdU at 10 μM and organoids were returned to hypoxia to incubate for another 6 h. Then, medium samples were taken and frozen for later analysis of glucose and fructose depletion via enzymatic assay. Finally, the organoids were either dissociated and analysed via flow cytometry or fixed in-situ and analysed via confocal microscopy. Flow cytometry Organoid EdU flow cytometry was performed using the ClickiT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific, C10634). Each well of a six-well plate was broken up by pipetting vigorously 50 times in 1 ml PBS, then diluted in 5 ml of PBS. Cells were pelleted at 1,100 rpm for 4 min at 4 °C, then resuspended in 50 μl TrypLE and incubated at 37 °C for 5 min. Five millilitres of PBS was then added to inactivate the TrypLE, and cells were pelleted. Cells were resuspended in 250 μl of 1% BSA in PBS, transferred to a 1.7-ml tube, and then pelleted at 3,000 rpm for 4 min. Cells were then stained with live/dead fixable green viability dye (Thermo Fisher Scientific, L34969) as per the manufacturer’s instructions. Cells were then resuspended in 100 μl Click-iT fixative, and processed as instructed in the Click-iT Plus EdU protocol (starting with step 4.3). Wash and reaction volumes were 250 μl. After completion of staining, all cells from each well were resuspended in 250 μl 1% BSA in PBS and 200 μl of this suspension was analysed using an Attune NxT flow cytometer (Thermo Fisher Scientific). Viable and proliferating cells were identified by the gating strategy depicted in Supplementary Fig. 4 . Primary human tumour samples Frozen primary human colon tumour and matched normal epithelium were obtained after informed consent from the WCMC Digestive Disease Registry, a protocol approved by the Weill Cornell Institutional Review Board. No protected health information was provided to the research team. After resection, tissue samples were immediately embedded in optimal cutting temperature compound and frozen in liquid nitrogen. Tumour areas were identified by a board-certified pathologist and six 2-mm cores were obtained from the frozen block. Cores were maintained at −80 °C prior to lysis and downstream analysis. Cell lysis and immunoblotting Mouse tissues or pelleted cells were snap-frozen in a liquid nitrogen bath and stored at −80 °C until further processing. Lysis was performed in pyruvate kinase lysis buffer (50 mM Tris-HCl pH 7.5, 1 mM EDTA, 150 mM NaCl, 1% Igepal-630) supplemented freshly before usage with protease inhibitors (10 μg ml −1 phenymethylsulfonyl fluoride, 4 μg ml −1 aprotinin, 4 μg ml −1 leupeptin and 4 μg ml −1 pepstatin (pH 7.4)). For immunoblotting, lysates were mixed with SDS–PAGE loading buffer (50 mM Tris-HCl pH 8.8, 1% w/v SDS, 2.5% glycerol, 0.001% w/v bromophenol blue and 143 mM β-mercaptoethanol) and heated to 70 °C for 10 min. Samples were separated by electrophoresis on 4–12% NuPAGE Bis-Tris gels (Invitrogen) and transferred to 0.45-μm PVDF membranes with wet transfer cells (Bio-Rad Laboratories). After 1 h of blocking with Tris-buffered saline with 0.1% (v/v) Tween 20 (TBST) containing 5% (w/v) BSA, membranes were incubated overnight at 4 °C with primary antibody in 5% BSA followed by a TBST wash and the appropriate secondary antibody (1:6,000) for 1h at room temperature. Signal was detected using an Odyssey CLx imaging system (LI-COR). Antibodies In alphabetical order [target (species (Rb; rabbit; Ms, mouse), manufacturer, catalogue number, western blot dilution, IHC dilution (if applicable))]: ACC (Rb, CST, 3676, 1:1,000), pACC(s79) (Rb, CST, 3661, 1:1,000), aldolase A (Rb, CST, 8060, 1:1,000), aldolase B (Rb, Abcam, 153828, 1:1,000, 1:1,000), AMPKa (Rb, CST, 2532, 1:1,000), pAMPK (t172) (Rb, CST, 2535, 1:1,000), α-tubulin (Ms, CST, 3873, 1:1,000), β-actin (Ms, Abcam, 6276, 1:1,000), BCL-2 (Rb, CST, 2876, 1:1,000), BCL-XL (Rb, CST, 2764, 1:1,000), BrdU (IIB5) (Ms, SC, sc-32323, 1:1,000, 1:250), CC3 (Rb, CST, 9661, 1:1,000, 1:200), enolase 1 (Rb, CST, 3810, 1:1,000), GLUT5 (Ms, Invitrogen, MA1-036, 1:1,000), GLUT5 (Ms, SC, 271055, 1:1,000), HIF-1α (Rb, CST, 36169, 1:1,000), HK1 (Rb, CST, 2024, 1:1,000), HK2 (Rb, CST, 2867, 1:1,000), Hypoxyprobe (Ms, HP, Mouse-Mab, 1:1,000, 1:50), KHK (Rb, Abcam, 154405, 1:1,000, 1:500), KHK-A (Rb, SAB, 21708, 1:1,000, 1:500), KHK-C (Rb, SAB, 21709, 1:1,000, 1:500), LDHA (Rb, CST, 2012, 1:1,000), MCL-1 (Rb, CST, 5453, 1:1,000), pBAD (S136) (Rb, CST, 4366, 1:1,000), PDH (Rb, CST, 3205, 1:1,000), pPDH (s293) (Rb, CST, 31866, 1:1,000), PKLR (Rb, Abcam, 171744, 1:1,000), PKM1 (Rb, CST, 7067, 1:1,000, 1:600), PKM2 (Rb, CST, 4053, 1:1,000, 1:800). CST, Cell Signaling Technologies; SC, Santa Cruz Biotechnology; HP, Hypoxyprobe; SAB, Signallway Antibody. Metabolomics analysis Polar metabolites were extracted from cell pellets using a 40:40:20 mixture of ice-cold acetonitrile:methanol:water with 0.1M formic acid. Samples were then centrifuged at 4 °C for 15 min at 14,000 rpm. Supernatants were then evaporated and resuspended in deionized water for LC–MS analysis. Quantitative metabolome analysis was performed as described previously 15 . In brief, aqueous tissue extracts were separated by liquid chromatography on an Agilent 1290 Infinity LC system by injection of 10 μl of filtered extract through an Agilent ZORBAX Extend C18, 2.1 × 150 mm, 1.8 μm (Agilent Technologies) downstream of an Agilent ZORBAX SB-C8, 2.1 mm × 30 mm, 3.5 μm (Agilent Technologies) guard column heated to 40 °C. Solvent A (97% water/3% methanol containing 5 mM tetrabutylammonium hydroxide (TBA) and 5.5 mM acetic acid) and Solvent B (methanol containing 5 mM TBA and 5.5 mM acetic acid) were infused at a flow rate of 0.250 ml min −1 . The 24-min reverse phase gradient was as follows: 0–3.5 min, 0% B; 4–7.5 min, 30% B; 8–15 min, 35% B; 20–24 min, 99% B; followed by a 7-min post-run at 0% B. Acquisition was performed on an Agilent 6230 TOF mass spectrometer (Agilent Technologies) using an Agilent Jet Stream electrospray ionization source (Agilent Technologies) operated at 4,000 V Cap and 2,000 V nozzle voltage in high-resolution, negative mode. The following settings were used for acquisition: The sample nebulizer set to 45 psig with sheath gas flow of 12 l min –1 at 400 °C. Drying gas was kept at 325 °C at 8 l min −1 . Fragmentor was set to 125 V, with the skimmer set to 50 V and Octopole Vpp at 400 V. Samples were acquired in centroid mode for 1.5 spectra/s for m / z values from 50 to 1,500. Collected data from the above methods was analysed by XCMS and X13CMS 46 , 47 . Metabolites were identified from ( m / z , rt) pairs by both retention time comparison with authentic standards and expected isotopomer distributions. When indicated, cells were treated with d -[U-13C6]-glucose, d -[U-13C6]-fructose, l -[U-13C5]-glutamine, or l -[U-13C3]-lactate in place of the unlabelled nutrient (Cambridge Isotope Laboratories). Labelling proceeded for 8 h. The various fatty acids are represented by “Cx:y”, where x denotes the number of carbons and y the number of double bonds. For example, the symbol for palmitic acid is C16:0 and palmitoleic acid is C16:1. RNA extraction Total RNA was extracted from frozen cell pellets using the RNeasy Mini Kit (Qiagen) and following the manufacturer’s instructions. In brief, cells were lysed using Buffer RLT supplemented with 1% (v/v) β-mercaptoethanol and QIAshredder columns (Qiagen) were subsequently used to homogenize the cell lysates. To remove contaminating genomic DNA, on-column digestion with DNase I was performed using the RNase-free DNase Set (Qiagen) as per the manufacturer’s instructions. KHK amplicon generation Targeted isoform sequencing is a highly specific, amplification-based method used to characterize the diversity of expressed isoforms at a particular gene locus. In brief, 100 ng of high-quality, DNase-treated total RNA was primed with an oligodT primer and reverse-transcribed into single-stranded cDNA using the SuperScript IV First-Strand Synthesis System (Thermo Fisher Scientific). The reverse transcription (RT) reaction was treated with ribonuclease H (RNaseH) to remove any remaining RNA templates. The single-stranded cDNA was then supplied at 5% of the total PCR reaction volume of targeted amplification using KHK-specific primers and TaKaRa LA Taq Polymerase with GC Buffer I (Clontech). Three separate PCR reactions were performed per sample using three unique reverse primers designed at alternate 3′ exons based on NCBI reference annotations of KHK. The resulting full-length cDNA amplicons were then purified of excess nucleotides, adaptor dimers and buffers using 1× AMPure PB beads (Pacific Biosciences). KHK primer sequences [primer name (primer sequence 5′ → 3′)]: human KHK_F_cds (ATGGAAGAGAAGCAGATCCTGTG), human KHK_R_cds (TCACACGATGCCATCAAAG), human KHK_R_altCDS (TCACCCTAGCAGCCCCC), human KHK_R_altCDS_ex5 (CCTCATTCTGCAGAGGAAAA). PacBio library preparation and sequencing The purified, full-length cDNA amplicons were then prepped for PacBio single-molecule real-time (SMRT) sequencing using the Express Template Preparation Kit 2.0 (Pacific Biosciences) and following the manufacturer’s instructions. In brif, 100 ng of cDNA from each sample was treated with a DNA damage repair enzyme mix to repair nicked DNA, followed by an end repair and A-tailing reaction to repair blunt ends and polyadenylate each template. Next, barcoded overhang SMRTbell adapters were ligated onto each template and purified using 1× AMPure PB beads to remove small fragments and excess reagents (Pacific Biosciences). The completed SMRTbell libraries were further treated with the SMRTbell Enzyme Clean Up Kit to remove unligated templates and then were equimolar pooled. The final pooled library was then annealed to sequencing primer v4 and bound to sequencing polymerase 3.0 before being sequenced on one SMRTcell 1M on the Sequel I system with a 20-h movie. Targeted IsoSeq analysis After data collection, the raw sequencing subreads were imported into the SMRTLink 9.0 bioinformatics tool suite (Pacific Biosciences) for processing. Intramolecular error correcting was performed using the circular consensus sequencing (CCS) algorithm to produce highly accurate (above Q20) CCS reads, each requiring a minimum of three polymerase passes. The CCS reads were then passed to the lima tool to remove barcode sequences and orient the isoforms into the correct sense or antisense direction. The refine tool was then used to remove concatemers from the full-length reads, resulting in final consensus isoforms ready for downstream analysis. The full-length, non-chimaeric (FLNC) reads were subsequently aligned to the GRCh38 reference genome using GMAP (v.2020-09-12), a splice-aware aligner specifically designed to handle long-read cDNA sequences. Redundant transcripts were then collapsed down to representative isoforms by passing the uniquely mapped isoforms through the TAMA Collapse algorithm with default parameters. The representative isoforms were further processed using the SQANTI3 (v.1.6) tool suite, which identifies and removes RT-switching and intra-priming artefacts. The filtered isoforms were then annotated using SQANTI3 by comparing each isoform to the NCBI RefSeq gene annotation database and categorized as either a known or novel transcript of KHK. Novel isoforms were defined as having at least one novel splice junction not previously annotated by NCBI. Isoforms with more than one FLNC supporting read from each sample were then merged together using the TAMA Merge algorithm to form one final isoform set representing all isoforms expressed across all samples. Production of recombinant pyruvate kinase Human PKM1, PKM2 and its mutants were cloned into a pET28a vector (Novagen) at NdeI and BamHI sites and expressed as an N-terminal His6 tag fusion protein. The protein was expressed and purified by standard protocol. In brief, pET28a-PKM2 was transformed into BL21(DE3)pLysS cells and grown to an absorbance of 0.8 at 600 nm, then induced with 0.5 mM IPTG for 7 h at room temperature. Cells were lysed by lysozyme in lysis buffer (50 mM Tris pH 8.0, 10 mM MgCl2, 200 mM NaCl, 100 mM KCl, 20% glycerol, 10 mM imidazole, 1 mM PMSF) and cell lysate was cleared by centrifugation. Enzyme was purified by batch binding to Ni-NTA resin (Qiagen). The resin was then washed with lysis buffer containing 30 mM imidazole for 200 column volumes, and His6-tagged PKM2 was eluted with 250 mM imidazole. The protein was dialysed overnight at 4 °C to remove the imidazole. Human PKL was purchased from R&D Systems (8569-PK). Density gradient ultracentrifugation, cross-linking and size-exclusion chromatography Sucrose gradients were formed and analysed as described in detail elsewhere 48 . In brief, 10 ml 10–40% sucrose gradients were created using a Gradient Master (BioComp Instruments) with the indicated metabolites evenly distributed throughout the gradient at a concentration of 1 mM. Subsequently, 400 μg of recombinant protein was incubated with the indicated metabolites at 1mM for 30 min at 25 °C, then gently layered atop the gradients. The gradients were then centrifuged for 16 h at 4 °C and 237,000 g in an SW 55 Ti rotor and Beckman L-80 ultracentrifuge. A piston gradient fractionator (BioComp instruments) was then used to fractionate the separated protein complexes which were then analysed by western blot. For cross-linking, purified recombinant enzyme at a concentration of 10 μg ml −1 in 1× pyruvate kinase dilution buffer (50 mM Tris-HCl pH 7.5, 100 mM KCl and 5 mM MgCl 2 ) was incubated with the indicated metabolites for 30 min at 37 °C. Then, di( N -succinimidyl) glutarate (Millipore Sigma) was added to a final concentration of 1mM and incubated for 10 min at 25 °C. The reaction was quenched with 1M Tris-HCl to a final concentration of 50 mM and samples were analysed by SDS–PAGE and immunoblotting as previously described. For size-exclusion chromatography, recombinant PKM2 was incubated alone or with FBP (100 μM) or F1P (500 μM) for one hour in PK dilution buffer (50 mM Tris-HCl pH 7.5, 100 mM KCl and 5 mM MgCl 2 ) on ice. Protein was then run on a Superdex 200 Increase 10/300 column (GE Life Sciences) equilibrated with the same buffer and metabolite, and 0.5-ml fractions were collected and subjected to SDS–PAGE and Coomassie blue staining. Pyruvate kinase docking simulations The crystal structure of PKM2 20 was used to perform docking simulations for F1P using the Maestro software package (release 2019-2, Schrödinger). Statistics and reproducibility Statistical analyses were conducted using GraphPad Prism (v.9.1). Data are mean ± s.e.m. unless stated otherwise. Exact P values are provided with the Source Data. Experiments were repeated independently, with similar results obtained. Mouse cohort sizes were informed by a priori power calculations using the variation from initial villus investigations with the aid of G*Power software (G*Power, v.3.1). Mice were randomly assigned to treatment groups. Investigators were blinded during image analysis of villus length and tumour burden. Investigators were not blinded to allocation during experiments. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper. Data availability Additional data that support the findings of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper. Code availability Villi analysis code, licensing information, and instructions for use are available at .

Eating fructose appears to alter cells in the digestive tract in a way that enables it to take in more nutrients, according to a preclinical study from investigators at Weill Cornell Medicine and NewYork-Presbyterian. These changes could help to explain the well-known link between rising fructose consumption around the world and increased rates of obesity and certain cancers. The research, published August 18 in Nature, focused on the effect of a high-fructose diet on villi, the thin, hairlike structures that line the inside of the small intestine. Villi expand the surface area of the gut and help the body to absorb nutrients, including dietary fats, from food as it passes through the digestive tract. The study found that mice that were fed diets that included fructose had villi that were 25 percent to 40 percent longer than those of mice that were not fed fructose. Additionally, the increase in villus length was associated with increased nutrient absorption, weight gain and fat accumulation in the animals. "Fructose is structurally different from other sugars like glucose, and it gets metabolized differently," said senior author Dr. Marcus DaSilva Goncalves, the Ralph L. Nachman Research Scholar, an assistant professor of medicine in the Division of Endocrinology, Diabetes and Metabolism and an endocrinologist at NewYork-Presbyterian/Weill Cornell Medical Center. "Our research has found that fructose's primary metabolite promotes the elongation of villi and supports intestinal tumor growth." The investigators didn't plan to study villi. Previous research from the team, published in 2019, found that dietary fructose could increase tumor size in mouse models of colorectal cancer, and that blocking fructose metabolism could prevent that from happening. Reasoning that fructose might also promote hyperplasia, or accelerated growth, of the small intestine, the researchers examined tissues from mice treated with fructose or a control diet under the microscope. The observation that the mice on the high-fructose diet had increased villi length, which was made by first author Samuel Taylor, a Tri-Institutional M.D.-Ph.D. Program student in Dr. Goncalves' lab, was a complete surprise. And once he made the discovery, he and Dr. Goncalves set out to learn more. After observing that the villi were longer, the team wanted to determine whether those villi were functioning differently. So they put mice into three groups: a normal low-fat diet, a high-fat diet, and a high-fat diet with added fructose. Not only did the mice in the third group develop longer villi, but they became more obese than the mice receiving the high-fat diet without fructose. The researchers took a closer look at the changes in metabolism and found that a specific metabolite of fructose, called fructose-1-phosphate, was accumulating at high levels. This metabolite interacted with a glucose-metabolizing enzyme called pyruvate kinase, to alter cell metabolism and promote villus survival and elongation. When pyruvate kinase or the enzyme that makes fructose-1-phospate were removed, fructose had no effect on villus length. Previous animal studies have suggested that this metabolite of fructose also aids in tumor growth. According to Taylor, the observations in mice make sense from an evolutionary perspective. "In mammals, especially hibernating mammals in temperate climates, you have fructose being very available in the fall months when the fruit is ripe," he said. "Eating a lot of fructose may help these animals to absorb and convert more nutrients to fat, which they need to get through the winter." Dr. Goncalves added that humans did not evolve to eat what they eat now. "Fructose is nearly ubiquitous in modern diets, whether it comes from high-fructose corn syrup, table sugar, or from natural foods like fruit," he said. "Fructose itself is not harmful. It's a problem of overconsumption. Our bodies were not designed to eat as much of it as we do." Future research will aim to confirm that the findings in mice translate to humans. "There are already drugs in clinical trials for other purposes that target the enzyme responsible for producing fructose-1-phosphate," said Dr. Goncalves, who is also a member of the Sandra and Edward Meyer Cancer Center. "We're hoping to find a way to repurpose them to shrink the villi, reduce fat absorption, and possibly slow tumor growth."

10.1038/s41586-021-03827-2

Medicine

Arthritis-related gene also regenerates cartilage in joints and growth plates

Nancy Q. Liu et al, gp130/STAT3 signaling is required for homeostatic proliferation and anabolism in postnatal growth plate and articular chondrocytes, Communications Biology (2022). DOI: 10.1038/s42003-021-02944-y Journal information: Communications Biology

http://dx.doi.org/10.1038/s42003-021-02944-y

https://medicalxpress.com/news/2022-01-arthritis-related-gene-regenerates-cartilage-joints.html

Abstract Growth of long bones and vertebrae is maintained postnatally by a long-lasting pool of progenitor cells. Little is known about the molecular mechanisms that regulate the output and maintenance of the cells that give rise to mature cartilage. Here we demonstrate that postnatal chondrocyte-specific deletion of a transcription factor Stat3 results in severely reduced proliferation coupled with increased hypertrophy, growth plate fusion, stunting and signs of progressive dysfunction of the articular cartilage. This effect is dimorphic, with females more strongly affected than males. Chondrocyte-specific deletion of the IL-6 family cytokine receptor gp130, which activates Stat3, phenocopied Stat3-deletion; deletion of Lifr, one of many co-receptors that signals through gp130, resulted in a milder phenotype. These data define a molecular circuit that regulates chondrogenic cell maintenance and output and reveals a pivotal positive function of IL-6 family cytokines in the skeletal system with direct implications for skeletal development and regeneration. Introduction The continual growth of long bones and vertebrae is fueled by chondrogenic stem and progenitor cells in the growth plate 1 . Beginning in embryogenesis and continuing through adolescence, these chondroprogenitors provide a steady stream of progeny that undergo endochondral ossification and contribute to bone growth 2 . Chondrocytes enter a phase of rapid division as they leave the resting zone of the growth plate, forming characteristic columns of mitotic cells that begin to secrete increased levels of extracellular matrix (ECM) proteins such as collagen II and aggrecan 3 . Once their distance from the niche within the resting zone increases to its maximum extent, they become hypertrophic, deposit lower amounts of ECM, and begin secreting collagen X and alkaline phosphatase, which is essential for calcification of the ECM 4 . Once surrounded by calcified matrix, these chondrocytes either undergo apoptosis or transdifferentiate into osteoprogenitors 5 ; osteoprogenitors derived from the growth plate, or invading from the periosteum, differentiate into osteoblasts and secrete bone matrix. Endochondral ossification takes place during fracture repair, with similar coordinated cellular dynamics 6 . In humans, growth plates become inactive at the end of puberty, resulting in replacement with bone; this occurs earlier in the long bones than in the spine 7 . Females undergo growth plate fusion an average of 2 years earlier than males, suggesting sexually dimorphic regulation of chondroprogenitors 8 . In mice, growth plates in the long bones and spine persist throughout the lifespan of the animal, providing an opportunity to study how specific molecular pathways impact the dynamics of chondroprogenitors. Recently, Newton et al. defined a population of chondrogenic cells localized to the growth plate that generate clonal progeny and support long bone growth 9 . Their data supported previous work showing that Hedgehog (Hh) signaling is required for growth plate maintenance; if this pathway is inhibited, growth plates fuse and differentiate into trabecular bone 10 . These results highlight the need for controlled, sustained proliferation of chondroprogenitors to maintain active growth plates. Data from our group has shown that the transcription factor STAT3 regulates human chondrocyte proliferation at both fetal and adult stages 11 , 12 . Chondrocytes enriched for active STAT3 are more proliferative and are also enriched for SOX9 protein, the master regulator of chondrogenesis, while evidencing negligible levels of the osteogenic transcription factor RUNX2 12 . Inhibition of either Il6ST (gp130), the obligate co-receptor for all IL-6 family cytokines, or LIFR which heterodimerizes with gp130 to facilitate LIF signaling, greatly reduced chondrocyte proliferation and STAT3 activation, respectively 11 . Inhibition of STAT3 activity in human fetal chondrocytes greatly reduced proliferation while increasing apoptosis, suggesting an important role for this molecule and the signaling mediated by gp130 in developmental chondrogenesis 11 . This is supported by data from zebrafish, in which stat3 null fish evidence fully penetrant scoliosis, reduced proliferation and increased apoptosis 13 . Moreover, deletion of Stat3 in mesodermal and early skeletal tissues using T -Cre and Prrx1 -Cre, respectively, caused dwarfism, skeletal abnormalities and kyphosis; the authors did not assess proliferation, but did observe significantly reduced SOX9 levels in Stat3 −/− chondrocytes 14 . They went on to show that stimulation with the IL-6 family cytokine oncostatin M (OSM), which can signal through a LIFR/gp130 complex and/or a OSMR/gp130 complex 15 , could increase both STAT3 activation and SOX9 levels in vitro. Few of these animals survived past P7, precluding the analysis of chondroprogenitors at later stages of growth plate function. Together, these studies identify an important role for STAT3 in regulating developing chondrocyte proliferation and apoptosis. However, the function of STAT3 in postnatal developing chondrocytes, the upstream regulators of this transcription factor and the potential role of STAT3 in the sexual dimorphism of growth plate closure remain to be elucidated. Here we show that postnatal deletion of Stat3 in chondrocytes using Acan -Cre ERT2 results in decreased proliferation, increased hypertrophy and apoptosis, stunting and growth plate fusions in the long bones and the spine. Female mice are more strongly impacted than males. We also demonstrate that chondrocyte-specific deletion of Stat3 results in progressive dysfunction of articular cartilage, although this compartment appears to be less affected than the growth plate. Deletion of gp130, the obligate co-receptor for all IL-6 cytokines in postnatal chondrocytes mostly phenocopies deletion of Stat3 , while deletion of Lifr results in a milder phenotype. Conversely, Stat3 overexpression in postnatal chondrocytes resulted in hyperproliferation and skeletal abnormalities and rescued the thin growth plates observed in gp130 deletion mutants. Finally, we show that estradiol increases STAT3 levels and activity in vitro and inhibition of estrogen signaling in female mice reduces Stat3 levels and activation, suggesting a potential mechanism behind the dimorphic effects seen following Stat3 deletion in mice. Collectively, these data define Stat3 as a regulator of postnatal chondrocyte cell proliferation, differentiation and survival and implicates IL-6 family cytokine signaling as a positive regulator of chondrogenic cell proliferation and maintenance. Results Stat3 promotes proliferation and survival in postnatal growth plate chondrocytes During development, chondrocytes form a highly proliferative blastema to supply rapidly growing limbs 16 . We have recently published a survey of the human chondrocyte transcriptome throughout ontogeny, spanning early embryonic development through adult stages 12 . Interrogation of this longitudinal data set for IL-6 family cytokines and receptors, known regulators of STAT3 activity, demonstrated consistent expression of gp130 and stage-specific expression of LIF to fetal and adolescent stages, during which growth plates are active in humans (Supplementary Fig. 1 ). To understand the function of STAT3 during these stages, we knocked down STAT3 gene expression with lentiviral shRNA in human fetal chondrocytes isolated from the interzone at 15–17 weeks of development and compared their transcriptome to control cells (Fig. 1a ). Gene ontology analysis 17 , 18 of the genes significantly enriched (>1.5 fold, p < 0.05) in control vs. STAT3 -depleted cells yielded categories related to matrix anabolism and secretion as well as proliferation, confirming our earlier results 11 and further suggesting a relationship between STAT3 activity and chondrocyte proliferation and matrix protein production. Genes downregulated upon STAT3 knockdown included ACAN , COL2A1 , MATN3 and MATN4 (Supplementary Data 1 ). To assess the function of STAT3 in vivo, we crossed Stat3 fl/fl mice 19 with the Acan- Cre ERT2 strain 20 , enabling tamoxifen-inducible deletion of Stat3 in chondrocytes with a single allele of Acan- Cre ERT2 . Pups were administered tamoxifen at postnatal days 2 and 3 (P2 and P3) by gavage and then skeletal phenotypes were assessed at stages corresponding to sexual immaturity and maturity, skeletal maturity and early aging (1, 3, 6, and 9 months of age, respectively; Fig. 1b, c ). Compared to Stat3 fl/fl littermate controls, Stat3 mutants survived at equivalent rates but exhibited significant decreases in body weight and were proportionally smaller at all ages examined. Notably, females were more strongly affected then males at 3 and 9 months and evidenced a non-significant trend at 6 months as well (Fig. 1d ); levels of Acan mRNA and protein were not different between females and males at the time of induction (Supplementary Fig. 2 ). To assess if there were sex-specific differences in the levels of Stat3 protein, we stained the growth plates of wild type female and male mice for Stat3 using immunohistochemistry (IHC); these data demonstrated that not only was there more Stat3 in female growth plates, but in both sexes Stat3 was mostly localized to the resting and proliferative zones of the growth plate (Fig. 1e ). We then isolated sternal chondrocytes from the xiphoid process of both female and male wild type mice and quantified protein by Western blot. These data (Fig. 1f ) demonstrated a clear enrichment for both total and active (phosphorylated; p) Stat3 in female mice, which may contribute to the observed dimorphism in reliance upon Stat3 for growth. Fig. 1: STAT3 regulates anabolic genes in human fetal chondrocytes and body size in mice. a Altered gene expression in human fetal chondrocytes after knockdown of STAT3 by shRNA in vitro as determined by bulk RNA-Seq and GO analysis; n = 3. Postnatal deletion of Stat3 in chondrocytes via Acan -Cre +/ERT2 in vivo decreases body weight and size of female ( b ) and male ( c ) mice; n ≥ 7 for each sex and genotype per timepoint. Images of 6 month old mice are shown. d Reductions in body weight in Stat3 mutants compared to WT littermates were greater in female than in male mice. e Stat3 protein localized to the resting and proliferative zones of the growth plate at 1 month in WT mice. f Stat3 and pStat3 levels were higher in WT female mice as compared to WT male mice ( n = 8). Scale bars = 50 µm. Full size image Based on the size differences between Acan- Cre +/ERT2 ; Stat3 fl/fl mice and controls as early as 1 month, we measured the width of each zone in the growth plate in control and Stat3 deletion mutants at this age (Fig. 2a,b ); given the potential for insertion of Cre ERT2 into the Acan locus to disrupt skeletal development, we also verified that mice bearing a single allele of Acan- Cre ERT2 were not affected as has been previously published 21 , 22 . These data revealed no differences between Acan -Cre +/ERT2 ; Stat3 fl/fl and control animals in the resting zone where chondroprogenitors localize 9 ; however, the length of the proliferative zone was significantly reduced in Stat3 deleted animals, concurrent with an increase in the pre-hypertrophic and hypertrophic zones (Fig. 2a, b ). To validate the independence of the observed phenotype from the Cre line used to delete Stat3 , we also generated both Col2a1 -Cre; Stat3 fl/fl as well as Gli1 -Cre +/ERT2 ; Stat3 fl/fl mice. In the former, Cre is expressed constitutively in chondrocytes, resulting in Stat3 deletion during mid-embryogenesis. At 3 months, both Col2a1 -Cre; Stat3 fl/fl female and male mice were visibly smaller than Stat3 fl/fl control animals, and microCT measurements of femoral length demonstrated significant shortening of femurs in Stat3 deleted animals (Supplementary Fig. 3a ); there was a non-significant trend for females to be more impacted than males (Supplementary Fig. 3b ). Growth plates were also impacted, with significantly reduced widths and bony bridges evident (Supplementary Fig. 3c, d ). In Gli1 -Cre +/ERT2 ; Stat3 fl/fl animals induced with tamoxifen at P2/P3 analyzed at 1 month of age, mice were visibly smaller and growth plates were significantly thinner in both females and males as compared to controls (Fig. 2d–f ). Fig. 2: Postnatal loss of Stat3 in chondrocytes decreases the thickness of the growth plate at 1 month. a , b Following deletion of Stat3 at P2/P3 via Acan -Cre +/ERT2 , the proliferative zone of the proximal tibial growth plate was significantly reduced, while the pre-hypertrophic and hypertrophic zones were increased, in both female and male mice; the resting zone was not affected. Acan -Cre +/ERT2 animals treated with tamoxifen did not show any reduction in thickness of growth plate zones. n = 6 animals per genotype, p values were calculated with one-way ANOVA test. Deletion of Stat3 at P2/P3 in Gli1 + cells resulted in reduced body weight and significantly smaller growth plates in both female ( c , e ) and male ( d , e ) mice. Scale bars = 50 µm. Full size image To assess the cellular basis for the observed growth retardation and disruptions in the zonal architecture of the growth plate, we measured the thickness of the proximal tibial growth plates at multiple stages following early postnatal deletion of Stat3 . At one month of age, both female and male mice lacking Stat3 demonstrated reductions in total growth plate thickness; these deficits persisted with advancing age. Beginning at 3 months we observed growth plate fusions in Stat3 mutants of both sexes (Fig. 3a, b ), which were significantly more frequent in females at 3 months (Supplementary Fig. 3e ). These fusions were more frequent in older animals, suggesting a progressive loss of growth plate chondrocytes and potential replacement by bone. To validate this, we conducted microCT analysis of Acan -Cre +/ERT2 ; Stat3 fl/fl and Stat3 fl/fl control animals at 6 months following postnatal deletion at P2/P3; this analysis revealed significantly greater bone density in the proximal tibial epiphysis of Stat3 deleted females and males compared to controls (Supplementary Fig. 4 ). To better assess the cellular basis underlying thinner growth plates and the reduced size of the proliferative zone observed at 1 month, we injected EdU into 3 month old animals for 4 consecutive days before analysis (Fig. 3d ); these data showed significantly fewer proliferating cells in the growth plates of Acan -Cre +/ERT2 ; Stat3 fl/fl animals as compared to Stat3 fl/fl controls. Finally, based on the acknowledged function of Stat3 as an anti-apoptotic factor 11 , 23 we assessed the rate of cell death in growth plates of Stat3 deleted and control animals at 6 months following tamoxifen administration at P2/P3. Stat3 null chondrocytes evidenced significantly more apoptosis (Fig. 3e ). Together, these data support a role for Stat3 in promoting chondrocyte proliferation and survival to maintain growth rates of the long bones, and also suggest a role for Stat3 in inhibiting epiphyseal bone formation in the distal region of the growth plate. Fig. 3: Postnatal loss of Stat3 in chondrocytes decreases proliferation and increases apoptosis. Proximal tibial growth plates were greatly reduced in thickness following deletion of Stat3 at P2/P3 in ( a , c ) female and ( b , c ) male mice. Beginning at 3 months, growth plate fusions were observed with replacement by trabecular bone. Scale bars equal 50 µm. d EdU was injected intraperitoneally for 4 days before analysis of mice at 3 months of age. EdU + cells in the proximal tibial growth plate were scored. e Apoptosis of chondrocytes in the proximal tibial growth plate (3 month old) was detected by TUNEL. Boxed regions indicate the growth plate. For each group, n = 6. Scale bars = 50 µm. POC primary ossification center, SOC secondary ossification center. Full size image Stat3 is required for vertebral endplate integrity In the spine, chondroprogenitors located in the vertebral endplates produce chondrocytes that form growth plates 24 , 25 ; this mechanism is very similar to that found in long bones. Based on the shorter stature of Stat3 deleted mice (Fig. 1 ), we examined vertebral elements in Acan -Cre +/ERT2 ; Stat3 fl/fl and control Stat3 fl/fl mice at several ages following induction with tamoxifen at P2/P3. Acan- Cre ERT2 targets the endplates, annulus fibrosis, and nucleus pulposus 20 . Similar to the tibial growth plate, Stat3 protein activation was localized to the proliferative zone and was enriched in female vs. male mice at 1 month (Supplementary Fig. 5 ). Moreover, the height of the vertebral bodies was significantly shorter in Stat3 deleted mice as early as one month of age (Fig. 4a–d ). Control animals had organized vertebral structures at all stages examined between 1 and 9 months. Cartilaginous endplates had clear, continuous borders and the formation of annular lamellae was observed. The nucleus pulposus had a rich proteoglycan matrix with sparse notochordal cells. In contrast in Stat3 deleted mice, starting at 3 months we observed a loss in continuity of both cranial and caudal endplates (Fig. 4c, d ). Structural degradation of the annulus was evidenced by a loss of collagen-rich lamellae. Within the nucleus pulposus, degradation was characterized by the loss of proteoglycan matrix, an appearance of large void spaces and loss of apparent nucleus and notochordal cell morphology. The progressive loss of endplate integrity in Stat3 mutants, followed by alterations in the nucleus pulposus, were strikingly similar to those observed in mice with postnatal deletion of Sox9 via Acan- Cre ERT2 26 . Fig. 4: Postnatal deletion of Stat3 affects vertebral growth plates and skeletal morphology. a Coronal sections of caudal vertebrae of Acan -Cre +/ERT2 ; Stat3 fl/fl and control Stat3 fl/fl mice demonstrate shorter vertebral bodies in both female and b male mice, coupled with c loss of vertebral growth plates. Scale bars = 1 mm ( a , b ) and 50 µm ( c ); n = 6. AF annulus fibrosis, CEP cartilaginous endplate, NP nucleus pulposus. d MicroCT and X-ray imaging of the lumbar and sacral vertebrae of control female (top panel, left), Stat3 deleted female (top panel, right), control male (lower panel, left) and Stat3 deleted male (lower panel, right) mice at 6 months of age following postnatal deletion at P2/P3. Note the shorter, wider vertebral bodies in Stat3 deleted animals, resulting in overall shorter lumbar length (L). See Supplementary Table 2 for quantification. n = 2. e Principal component analysis (PCA) of sacra from Stat3 deleted and control mice based on distances between 34 anatomical landmarks (Supplementary Fig. 5 ). Animals clearly segregate based on genotype. f Deviations along PC1 are shown on 3D surface maps of control (left) and Stat3 deleted (right) male sacral regions. Maps are colored based on degree of deformation with respect to PC1 values, with red representing features more prominent in control and green more prominent in Stat3 deleted animals, respectively. Full size image We conducted microCT imaging to characterize the extent of the axial skeletal phenotype in Stat3 mutant mice. Stat3 deleted animals showed a reduction in the vertebral height in both lumbar and sacral regions, as well as changes in endplate structure (Fig. 4d ). This was consistent in both genders. Both anterior and posterior vertebral heights from L1 to L6 were reduced in the Stat3 mutants compared to controls (Supplementary Data 2 and Supplementary Table 1 ), ranging from 19-37% reduction. Mice lacking Stat3 had a larger endplate surface area for both cranial and caudal endplates compared to control mice (Supplementary Data 2 ), ranging from 9–31% larger. Intervertebral disc anteroposterior and transverse length did not differ between Stat3 deleted and control animals. We then conducted geometric morphometric (GM) mapping of the sacrum and pelvis to gain additional understanding of how loss of Stat3 in chondrocytes impacted overall skeletal form. For the sacrum, 34 anatomical landmarks were defined based on current best practices 27 , 28 , 29 and their exact 3D coordinates defined (see “Materials and methods”, Supplementary Fig. 6 ). Variation among landmark configurations and shape were explored via principal component analysis 30 (PCA; Fig. 4e ). Mice were clearly segregated along PC1 based on their genotype, confirming widespread changes in skeletal morphology following early postnatal deletion of Stat3 . To visualize these differences, thin-plate spline analysis 31 was used to project aberrations from normal on 3D renderings of control and Stat3 deleted male sacra (Fig. 4f ). A similar analysis of pelvic structure confirmed broad and consistent skeletal malformations in mutant animals (Supplementary Fig. 7 ). These “heat map” images provide a quantitative visual representation of how specific skeletal structures change following the loss of Stat3 , reinforcing the role of Stat3 in contributing to regulated chondrocyte proliferation and survival in growth plates throughout the body. Stat3 is required for maintenance of articular cartilage In postnatal mouse articular cartilage, chondrocyte proliferation rates are in the low single digit range 32 ; cartilage expands via increases in cell volume and matrix production rather than cell division. Our in vitro data from human chondrocytes suggest STAT3 increases expression of ECM genes (Fig. 1a ). Thus, we hypothesized Stat3 may function in articular cartilage maintenance by supporting matrix protein production and stability. Analysis of articular cartilage in Acan - Cre +/ERT2 ;Stat3 fl/fl mice revealed a progressive loss of proteoglycans at the articular surface compared to control mice (Fig. 5a, b ). This phenotype was exacerbated in female mice, in agreement with the results from the growth plate analysis. Loss of Safranin O was clearly evident beginning at 3 months of age in females, compared to 6 months in males. Almost all Safranin O + matrix was lost in Stat3 deleted mice by 6 months of age in females vs. 9 months for males. Concomitant with accelerated loss of proteoglycans, Stat3 deletion resulted in increased detection of aggrecan neoepitope, an antigen present when the proteoglycan aggrecan is cleaved and degraded, at 6 months of age (Fig. 5c, d ). In addition, substantial apoptosis was detected in both sexes at 9 months (Supplementary Fig. 8 ), We also detected increased expression of markers of hypertrophy in the absence of Stat3 including Runx2 and collagen X, indicating a mild degenerative process present in the joint 33 , coupled with decreased expression of Sox9 (Supplementary Fig. 8 ). However, no obvious signs of articular cartilage surface fibrillations, osteophytes or joint inflammation was observed at any of the investigated stages. Together, these data describe a role for Stat3 in articular chondrocytes in maintaining ECM homeostasis and preventing adoption of a hypertrophic, arthritic-prone state. Fig. 5: Postnatal Stat3 deletion results in subtle changes in articular cartilage of the knee joint. a Administration of tamoxifen at P2/P3 to Acan -Cre +/ERT2 ; Stat3 fl/fl and control Stat3 fl/fl female and b male mice elicited reduced proteoglycan staining in articular articular cartilage by 3 months of age in female mice and by 6 months of age in males. c Histological staining and quantitative assessment of pre-hypertrophic (Runx2, collagen 10, COLX) and degenerative changes (aggrecan neoepitope, ACAN NEO) in the articular cartilage of knee joints of female and d male Stat3 deleted and control mice at 6 months. In all panels, representative images are shown and scale bars = 50 µm. n = 4–6. Full size image Stat3 inhibits hypertrophy and maintains a proliferative state in Sox9 + chondrocytes Chondrogenesis is dependent on transcription factors of the Sox family. Sox9 is essential for chondrogenesis 34 , while Sox5 and Sox6 act downstream of Sox9 to promote proliferation and inhibit hypertrophy of chondrocytes 35 . In order to better assess the molecular consequences of postnatal Stat3 deletion, we sorted cells negative for blood and endothelial markers from Acan -Cre +/ERT2 ; Stat3 fl/fl and control Stat3 fl/fl knee joints and performed scRNA-seq; we first verified the dissection strategy would maximize capture of chondrocytes by examining Sox9 -GFP mice 36 (Fig. 6a ). Knee joints from 2–3 female mice of each genotype were digested and pooled, and live, lineage-negative cells were sorted (Fig. 6b ) and analyzed at the single-cell level. We focused our analysis on chondrocytes defined by expression of Sox9 and Col2a1 ; unbiased clustering of chondrocytes expressing both genes from Acan -Cre +/ERT2 ; Stat3 fl/fl and control Stat3 fl/fl animals via a tSNE plot suggested clustering independent of cofounding variables as well as some potential differences between the genotypes (Supplementary Fig. 9 and Fig. 6c ). Recently, Li et al. analyzed growth plate zone-specific gene regulation at the single-cell level at postnatal day 7 37 ; they identified genes whose expression was enriched in the proliferative and pre-hypertrophic/hypertrophic zones. Based on our data showing increases in the hypertrophic zone and bone density in the proximal epiphysis of Acan -Cre +/ERT2 ; Stat3 fl/fl mice compared to controls, we assessed expression levels of genes identified by Li et al., as well as known regulators of each zone, in the scRNA-seq data generated here. This analysis demonstrated highly significant enrichment of genes associated with chondrocyte maturation and hypertrophy in the absence of Stat3 (Fig. 6d, f ), while genes such as Sox5/6 that promote proliferation in immature chondrocytes were significantly enriched in control Stat3 fl/fl chondrocytes (Fig. 6e, f ). In parallel, we conducted unbiased k-means clustering to assess potential changes in chondrocyte populations upon deletion of Stat3 . This analysis defined 6 clusters, 4 of which evidenced some bias in their constitution by either genotype (Supplementary Fig. 10b, c ). Based on previously published data 38 , 39 , we speculatively annotated these clusters as columnar chondrocytes (clusters 2 and 4), resting chondrocytes (cluster 3) and osteoprogenitors (cluster 6); as expected from the growth plate phenotype, Stat3 deficient cells were underreprented in clusters 2 and 4 and overrepresented in cluster 6. Cluster 2 also demonstrated a proliferative signature, further supporting the function of Stat3 in promoting chondrocyte cell division (Supplementary Fig. 9e ); moreover, the moderate enrichment of Stat3 − /− cells in cluster 3 could suggest a defect in proliferation in the resting zone in mutant mice. Together, these data confirm at the transcriptional level that Stat3 is required to suppress chondrocyte maturation and promote continued output of immature chondrocytes from the resting zone. Fig. 6: scRNA-sequencng reveals loss of Stat3 in growth plate chondrocytes promotes premature differentiation. a Schematic of isolation, sorting and single-cell sequencing of chondrocytes from 1 month mouse knee joints. Epifluorescent analysis of Sox9 -GFP femurs confirmed the dissection strategy. b Live cells negative for markers of endothelium (CD31), red blood cells (Ter119) and white blood cells (CD45) were sorted (black box) from dissociated knee joints pooled from 2–3 female control Stat3 fl/fl or Acan -Cre +/Cre ; Stat3 fl/fl mice and subjected to scRNA-sequencing. c tSNE plot of single cells selected for expression of Sox9 and Col2a1 and colored by genotype of origin. Expression pattern of genes associated with growth plate chondrocyte maturation ( d ) as well as immature, proliferating chondrocytes ( e ) are shown. f Gene set enrichment analysis (GSEA) plots for chondrogenic maturation (top) and immature, proliferative chondrogenic (bottom) gene sets in Stat3 fl/fl (control) and Acan -Cre +/Cre ; Stat3 fl/fl (Stat3KO) female mice. The “barcode graph” (lower portion) of each plot shows the target genes rank-ordered (left to right) by their differential expression in Stat3KO mice as compared to control mice (indicated by Enrichment Score on the y -axis). Signal-to-noise ratio metric was used for ranking of gene expression patterns in cells. NES normalized enrichment score, FDR false discovery rate. Full size image Stat3 acts downstream of gp130 to maintain output from immature chondrocytes Many receptor tyrosine kinases, including EGFR 40 , PDGFR 41 and GHR 42 among others, can activate STAT3 through various signaling cascades. Based on our previous in vitro studies in articular chondrocytes 11 , we hypothesized that gp130, an obligate receptor for IL-6 family cytokines, may function upstream of Stat3 activation in immature chondrocytes. We found that RCGD 423, a small molecule modulator of gp130 that increases activation of Stat3, increased proliferation of articular chondrocytes 11 without activation of ECM catabolism. Analysis of gp130 expression in the growth plate of wild type mice at 1 month demonstrated enrichment in the resting and proliferative zones, similar to the expression pattern of Stat3 (Supplementary Fig. 11a ). Accordingly, we crossed Acan -Cre +/ERT2 mice to gp130 fl/fl mice 43 and administered tamoxifen at P2/P3 to ablate gp130 postnatally in chondrocytes. At all ages examined, there was significantly reduced growth plate thickness in both female and male gp130 deleted mice compared to gp130 fl/fl controls, including frequent disruptions in growth plate integrity by 3 months (Fig. 7a–c ); as was observed in Stat3 deleted animals, growth plate fusions were significantly more frequent in females at 3 months (Supplementary Fig. 11b ). Quantification of growth plate zones at 1 month yielded similar results to deletion of Stat3 in chondrocytes, demonstrating a highly significant decrease in the thickness of the proliferative zone and increases in the more mature pre-hypertrophic and hypertrophic zones in gp130 mutants (Supplementary Fig. 11c, d ). Additionally, vertebral bodies (Supplementary Fig. 12a, b ) were shorter with concomitant severe disruption of growth plates in Acan -Cre +/ERT2 ; gp130 flfl mice compared to gp130 fl/fl controls (Fig. 7d, e ). Analysis of the articular cartilage in gp130 deleted animals also demonstrated a similar phenotype as found in Acan -Cre +/ERT2 ; Stat3 fl/fl mice, with progressive loss of proteoglycans in both female and male mice, with earlier loss in females (Supplementary Fig. 13 ). These data, which mostly phenocopy the results of Stat3 deletion in chondrocytes, suggest a gp130/Stat3 signaling circuit that is required for chondrocyte proliferation, prevention of hypertrophy and maintenance of matrix synthesis. Fig. 7: Postnatal loss of gp130 in chondrocytes phenocopies Stat3 deletion. Proximal tibial growth plates were significantly reduced in thickness at all ages examined following deletion of gp130 at P2/P3 in female ( a , c ) and male ( b , c ) mice; severe growth plate disruptions were observed at 3 months. Vertebral body heights were reduced, coupled with loss of vertebral growth plates, in both female ( d ) and male ( e ) Acan -Cre +/ERT2 ; gp130 fl/fl mice as compared to gp130 fl/fl controls. AF annulus fibrosis, CEP cartilaginous endplate, NP nucleus pulposus. In all panels, scale bars = 50 µm; n = 4–6. Full size image In parallel, we also generated Acan -Cre +/ERT2 ; Lifr flfl mice, as data from us and others has clearly indicated a role for Lifr/gp130 signaling in skeletal development in mice 11 , 44 and humans 16 , 45 . Analysis of Lifr expression in growth plates confirmed a similar expression pattern in the resting and proliferative zones as defined for Stat3 and gp130 (Supplementary Fig. 11a ). Analysis of growth plates of Acan -Cre +/ERT2 ; Lifr flfl mice vs. Lifr fl/fl controls at all ages demonstrated reduced thickness in female and male mice (Fig. 8a–c ); at 1 month, the imbalance of the proliferative and hypertrophic zones of the growth plate observed in Stat3 and gp130 mutant animals was also present in Lifr deleted mice (Supplementary Fig. 11 ). However, in general the skeletal phenotype was milder than Stat3 and gp130 deleted mice, as bony bridges of the growth plate were infrequently observed even at 9 months and the articular cartilage, vertebral bodies and growth plates of Acan -Cre +/ERT2 ; Lifr fl/fl mice were not as severely impacted (Fig. 8d, e and Supplementary Figs. 12 and 13 ). These results suggest a role for Lifr to mediate IL-6 family cytokine-mediated activation of Stat3 in chondrocytes to promote maintenance of the growth plate, but demonstrate that other receptors are also required. Fig. 8: Postnatal loss of Lifr in chondrocytes partially phenocopies Stat3 deletion. Proximal tibial growth plates were significantly reduced in thickness at all ages examined following deletion of Lifr at P2/P3 in female ( a , c ) and male ( b , c ) mice; growth plate disruptions were not as severe as those observed in Stat3 and gp130 deletion mutants. Vertebral growth plates were mildly reduced in thickness, coupled with occasional disruption of integrity at later ages examined, in both female ( d ) and male ( e ) Acan -Cre +/ERT2 ; Lifr fl/fl mice as compared to Lifr fl/fl controls. AF annulus fibrosis, CEP cartilaginous endplate, NP nucleus pulposus. In all panels, scale bars = 50 µm; n = 6. Full size image Stat3 overexpression promotes excessive chondrocyte proliferation and rescues growth plate deficiencies observed in gp130 deleted mice Postnatal loss of Stat3 in chondrocytes resulted in premature closure of growth plates accompanied by reduced proliferation and survival of chondrocytes, suggesting this transcription factor is necessary to maintain appropriate mitotic rates. To assess the capacity of Stat3 to elicit a proliferative response, we crossed Acan -Cre +/ERT2 animals with mice carrying a conditional, constitutively active allele of Stat3 ( Stat3C ) under the control of the Rosa26 promoter 46 . In these animals, activation of Cre results in excision of a STOP cassette upstream of the Stat3C sequence. Analysis of Acan -Cre +/ERT2 ; Rosa26-Stat3C fl/+ mice treated with tamoxifen at P2/P3 and analyzed at 1 month of age revealed dramatic hypercellularity in proximal tibial growth plates (Fig. 9a ); hypercellularity was also apparent in the superficial zone and deeper layers in articular cartilage (Fig. 9b ). Hyperproliferation in the growth plate was confirmed by EdU incorporation in Acan -Cre +/ERT2 ; Rosa26-Stat3C fl/fl mice (Fig. 9c ), with significantly more EdU + chondrocytes present in Stat3C overexpressing mice as compared to Rosa26-Stat3C fl/fl controls. There are also more dividing cells in articular cartilage of Stat3C overexpressing animals at 3 months, although this was not as dramatic as in the growth plate (Supplementary Fig. 14 ). As Stat3 is a known mediator of signaling downstream of gp130 and our data demonstrated similar phenotypes in the growth plates of Stat3 and gp130 postnatal deletion mutants, we hypothesized that constitutive activation of Stat3 could rescue the defects in proliferation found in Acan -Cre +/ERT2 ; gp130 fl/fl mice. We generated Acan -Cre +/ERT2 ; gp130 fl/fl Rosa26-Stat3C fl/+ animals and compared their growth plates to Acan -Cre +/ERT2 ; gp130 fl/fl mice at 1 month. In both female and male mice, overexpression of Stat3C in chondrocytes could completely rescue the diminished growth plates in gp130 deleted mice (Fig. 9d ). These data indicate that gp130-mediated activation of Stat3 is required for chondrocyte proliferation, and show that Stat3 is sufficient to promote proliferation in a context-dependent fashion as increased mitosis was most prominent in chondrocytes that normally have the greatest capacity to divide. Fig. 9: Overexpression of constitutively active Stat3 (Stat3C) in chondrocytes promotes precocious proliferation in vivo and rescues growth plate defects in gp130 deletion mutants. Induction of Stat3C expression at P2/P3 resulted in significant increases in chondrocyte number in both growth plate ( a ) and articular cartilage ( b ) in Acan -Cre +/ERT2 ; Rosa26 -Stat3C fl/+ mice as compared to Rosa26 -Stat3C fl/+ control animals at 1 month. c EdU was injected intraperitoneally 4 h before harvesting knee joints at 1 month of age. EdU + cells in proximal tibial growth plate chondrocytes were scored. POC primary ossification center, SOC secondary ossification center. Boxed areas delineate the growth plate. d Induction of constitutively active Stat3 in Acan -Cre +/ERT2 ; gp130 fl/fl mice resulted in significantly thicker growth plates at 1 month of age. In all panels, scale bars = 50 µm; n = 4–6 for each group. Full size image Estrogen increases Stat3 expression and activation in chondrocytes A relationship between sex hormones and chondrogenesis is well established. In humans, estrogen is responsible in both females and males for accelerating growth at puberty and eventually driving the closure of growth plates; females tend to begin the pubertal growth spurt earlier than males and have higher levels of estradiol at this age (reviewed in 47 ). Based on the dimorphic difference in body size at 3 months in Acan -Cre ERT2 ; Stat3 fl/fl females and males, we hypothesized that estrogen may regulate Stat3 levels and/or activity. To address this, we isolated porcine articular chondrocytes from female and male animals and treated them with different levels of estradiol (E2) in vitro. These data demonstrated that both female (Fig. 10a ) and male (Fig. 10b ) articular chondrocytes upregulated STAT3 proteins levels and activity in response to E2, though female chondrocytes demonstrated a maximal response at 0.1 µM while male chondrocytes responded in a dose-dependent manner. To assess the impact of estrogen levels on Stat3 in vivo, the clinically-used small molecule aromatase inhibitor Letrozole (which results in lower estradiol levels) was injected intraperitoneally daily for 7 days into sexually mature female mice. Analysis of sternal chondrocytes revealed that Letrozole reduced both Stat3 levels and activity. Taken together, these data suggest that the enhanced estrogen levels present in growing female mice lead to a differential requirement for Stat3 to enhance skeletal growth. Fig. 10: Estrogen influences Stat3 levels and activity in vitro and in vivo. Addition of estradiol (E2) to female ( a ) and male ( b ) pig chondrocytes resulted in dose-dependent increases in Stat3 levels and activity; female chondrocytes are dose-limited in their response to E2 as evidenced by similar induction at the 0.1 and 1.0 µM doses. n = 3. c Intra-peritoneal Injection of Letrozole, an aromatase inhibitor that blocks the production of estrogen, reduced Stat3 levels and activity in female mouse sternal chondrocytes after 7 daily injections. n = 5 mice per group. Full size image Discussion Here we show that Stat3 is a critical factor controlling the proliferation, survival, maturation and matrix production of chondrocytes in growth plates and articular cartilage. Animals lacking Stat3 in chondrocytes following postnatal deletion are proportionately smaller than control animals and evidence significantly reduced EdU incorporation, coupled with thinner proliferative zones and increased apoptosis; by 9 months of age, all mice had abnormal growth plate fusions in both appendicular and axial skeletal elements, with increased frequency of fusions found in females. Articular cartilage was also affected, displaying a mild degenerative and pre-hypertrophic phenotype. Females tended to have a more severe phenotype than males, with a more reduced body size than males at 3 months, earlier and more frequent fusions in the growth plate and accelerated changes in the articular cartilage. Stat3 null chondrocytes were enriched for expression of genes previously associated with maturation in the growth plate and concurrently evidenced diminished expression of genes associated with immature, proliferative chondrocytes. We also showed that deletion of the obligate receptor for IL-6 family cytokines gp130 in chondrocytes postnatally phenocopied loss of Stat3 , nominating a role for gp130-Stat3 signaling in regulating immature chondrocyte maintenance; deletion of Lifr resulted in a much milder phenotype, potentially due to compensation by other alpha-receptors. Expression of a constitutively active form of Stat3 from the Rosa26 locus led to excessive proliferation of chondroprogenitors and chondrocytes and rescued the diminutive growth plates found in gp130 deleted mice, demonstrating Stat3 is both necessary and sufficient to promote cell division in permissive chondrogenic cells. Finally, we showed that estradiol can regulate Stat3 levels and activity in vitro and in vivo, identifying a potential mechanism of growth acceleration in females during longitudinal bone growth. Together, these data unveil IL-6 family cytokine signaling via gp130-Stat3 as a major regulator of postnatal chondrocyte biology. Regulation of postnatal bone growth via endochondral ossification has been investigated by several groups using transgenic mice. Deletion of Sox9 , the master regulator of chondrogenesis, at early postnatal stages resulted in death of the animals 48 ; however, deletion at P18 and later using Acan -Cre ERT2 produced mice with similar phenotypes to those we report here including growth plate shortening and fusions concomitant with reduced proliferation and increased apoptosis, compressed vertebral body heights and pre-arthritic changes in articular cartilage 48 . Some aspects of the phenotype they observed were more severe than in Stat3 deleted mice, including very rapid loss of matrix secretion proximate to deletion of Sox9 . Interestingly, Hall et al. demonstrated that STAT3 could bind to the Sox9 promoter and activate transcription, which may explain some of the phenotypic overlaps 14 . Indeed, deletion of Stat3 greatly reduced expression of Sox9 in the articular cartilage (Supplementary Fig. 8 ). Moreover, deletion of Sox9 using Acan -Cre ERT2 in mice 3 months of age resulted in loss of verterbral endplate integrity coupled with growth plate fusions followed by increases in fibrosis and apoptosis in the nucleus pulposus 26 ; these data are strikingly similar to the phenotypes reported here and reinforce the concept of a gene regulatory loop involving Stat3 and Sox9 that controls chondroprogenitor output. Although not inducible, another transgenic mouse that evidenced growth plate fusions and mild dwarfism was created by Sims et al., in which the intracellular motifs of gp130 that interact with Stat3 were deleted 49 . By 4 months of age, these mice had femoral lengths an average of 20% less than wild type controls, frequent growth plate fusions and significantly reduced proliferative zones in the growth plate. The authors did not note any vertebral defects, nor whether there were differences between males and females. Furthermore, given the constitutive nature of this knock-in model, it could not be determined whether the effects on cartilage were direct, and at what developmental stage of chondrogenesis the gp130/Stat3 circuit might be required. Additionally, manipulation of Hh signaling can also result in fusion of tibial and vertebral growth plates, either via antagonizing Shh 10 or genetic ablation of Ihh in chondrocytes 50 ; it remains to be determined whether there is interaction between Hh signaling and gp130/Stat3. The data presented here provide clear evidence that gp130/Stat3 signaling is required postnatally in chondrocytes to coordinate proliferation and differentiation, either independently of or potentially in concert with known regulators. Our data from Stat3 deleted mice reveal a sexually dimorphic response. Stat3 has been implicated in sex-specific responses to ischemia in both the heart 51 and brain 52 , with increases in Stat3 phosphorylation directly mediated by stimulation with estradiol. The data presented here suggest that estradiol may increase Stat3 levels and activity during the pubertal growth spurt; estradiol levels spike in female C57/B6 mice at 28–30 days 53 . In uninjured brains, Di Domenico et al. identified several differences in female vs. male neurons that depended on Stat3 using a proteomics approach; many of these proteins have been associated with cellular metabolism 54 , consistent with known functions of Stat3 in mitochondrial 55 and glycolytic metabolism 56 . Stat5a and Stat5b, other members of the same transcription factor family, have demonstrated roles in regulating sexually dimorphic gene expression downstream of growth hormone in the liver 57 . Together, these data support the feasibility of a sex-specific role for Stat3 in enhancing skeletal growth and potentially maintenance that should be explored further. Growth of the long bones and spine depend upon regulated output from chondroprogenitor cells located in growth plates. In humans, these structures support growth throughout childhood and adolescence, at which point increases in sex hormones are partially responsible for terminal differentiation of chondroprogenitors and closure of growth plates 58 . Abnormalities in the output from chondroprogenitors often manifest in visible alterations in skeletal size. Gigantism can occur when too much growth hormone is present during childhood and adolescence, resulting in excessive linear growth; this condition is treated clinically with a peptide that mimics somatotropin, a hormone that represses growth hormone secretion 59 . In contrast, chondrodysplasias that result in below average stature can result from myriad inherited or developed conditions. Mutations in the FGFR3 gene that produce constitutively active protein are associated with several skeletal conditions including achondroplasia and hypochondroplasia 60 , while inactivating mutations in LIFR or IL6ST (encodes gp130) genes cause Stuve-Wiedemann syndrome 45 , 61 . Moreover, short stature is frequently found in children with chronic inflammatory conditions such as Crohn’s disease and ulcerative colitis which are mediated by excessive levels of pro-inflammatory cytokines including IL-6 62 . Here we showed that Lifr deletion specifically in chondrocytes evidenced a noticeably milder phenotype than Acan -Cre ERT2 -mediated deletion of Stat3 or gp130 , as well as a much weaker skeletal phenotype than seen in global, constitutive deletion of Lifr in mice 44 or germline mutation in humans. We hypothesize that oncostatin M receptor (Osmr), through which Lif can also signal after heterodimerization with gp130, can compensate for Lifr deletion in chondrocytes. Our data from human ontogeny show abundant expression of OSMR at relevant stages of human ontogeny (Supplementary Fig. 1 ). Also, Lifr has clear functions in osteoblasts and osteoclasts 44 , and these functions outside the chondrogenic lineage could impact skeletal development. Finally, Stat3 is downstream of growth hormone receptor, FGFR3 and epidermal growth factor receptor (EGFR) activation as well as IL-6/gp130 signaling, reinforcing the relevance of the current work showing that the correct dosage of Stat3 activity is essential for regulating immature chondrocyte cell fate. Classically, IL-6/gp130/STAT3 signaling has been viewed as a negative regulator of chondrocyte biology and function and is broadly discussed in the context of arthritis and other inflammatory diseases. IL-6 signaling through IL-6R/gp130 suppresses chondrocyte proliferation 63 , promotes mineralization in articular cartilage 64 , downregulates expression of matrix proteins and increases expression of matrix-degrading proteases 65 . Moreover, blockade of IL-6 in vivo in mouse models of osteoarthritis (OA) has been shown to be chondroprotective 66 , 67 . Importantly, higher serum levels of IL-6 have been correlated with the development of OA in humans, highlighting the potential for systemic inflammation levels to contribute to local disease; a monoclonal antibody against IL-6R is currently in Phase III clinical trials for treatment of hand OA (NCT02477059). Furthermore, inhibition of Stat3 downstream of exogenous IL-6 is chondroprotective, reducing the severity of OA-like pathology in a mouse model 66 . However, analysis of phenotypes associated with Stat3 , gp130 and Lifr deletion in chondrocytes during homeostasis reveal essential roles for this pathway in chondroprogenitor proliferation and maintenance. This is reinforced by analysis of patients diagnosed with juvenile arthritis treated with the monoclonal anti-IL-6R antibody Tocilizumab; many of these individuals achieved normalized growth rates when their serum IL-6 levels normalized 68 . It remains to be determined if the Stat3 targets and overall functions in decease and development are identical, or if these effects are context specific with the final outputs determined by signaling interactions with other pathways which are clearly dissimilar in normal growth and diseases. Our results demonstrating the essential role of calibrated gp130/Stat3 signaling in developing chondrocytes suggest that modulation of this axis could have therapeutic benefit in other conditions where growth plate and/or articular cartilage maintenance is impacted. Materials and methods Mouse lines and breeding All procedures involving animals were approved by the Institutional Animal Care and Use Committee of USC. This study was compliant with all relevant ethical regulations regarding animal research. Mice with tamoxifen-regulated expression of the Cre recombinase under control of the chondrocyte-specific aggrecan (Acan) promotor ( Acan -Cre ERT2 ) were purchased from JAX (strain 019148 20 ) and used exclusively as heterozygotes ( Acan -Cre +/ERT2 ). To delete Stat3 , these mice were crossed with mice bearing floxed Stat3 alleles ( Stat3 fl/fl ), also purchased from JAX (strain 016923 23 ). To overexpress Stat3, mice carrying a floxed stop cassette followed by a constitutively active Stat3 allele in the Rosa26 locus (Rosa26-Stat3C fl fll ; a kind gift from Dr. Sergei Koralov of New York University School of Medicine 46 ) were crossed with Acan- Cre +/ERT2 mice. To delete Il6st (gp130), mice bearing floxed gp130 alleles ( gp130 fl/fl ) were purchased from the Knockout Mouse Project (KOMP) Repository of University of California, Davis ( Il6st tm1a(KOMP)Mbp ) and were crossed with mice expressing Flp recombinase under the control of the constitutive Rosa26 promoter (JAX strain 009086 69 ) to remove the Neo cassette. Animals negative for Flp recombinase were then crossed to Acan- Cre +/ERT2 mice. In vivo experiments were performed with cohoused littermate controls. Lifr fl/fl (Lifr tm1c(EUCOMM)Hmgu ) cryopreserved frozen sperm were purchased from MRC Harwell Institute, The Mary Lyon Centre, UK. The mice were re-derived at USC. Col2a1 -Cre mice were purchased from JAX (strain 003554 70 ). Gli1 -Cre +/ERT2 mice were purchased from JAX (strain 007913 71 ). All animals were on a C57BL/6 background. Genotyping was performed according to protocols provided by JAX, KOMP and MRC. STAT3 knockdown and gene expression analysis Human fetal chondrocytes from the interzone were isolated as previously described 16 from de-identified material from anonymous donors following informed consent; as no means of associating biospecimens with donors was possible, this does not meet the definition of human subjects research and is therefore exempt from ethical review. Passage 0 chondrocytes were cultured in vitro for 2–3 days before being infected with doxycycline inducible STAT3 shRNA or scrambled lentiviral particles (Dharmacon). One day later, doxycycline treatment commenced for a total of three days; fresh doxycycline was added daily. After three days, cells were sorted for RFP and mRNA was isolated with the RNeasy kit (Qiagen). Libraries (KAPA LTP kit, Roche) were prepared after RNA quality validation (Agilent Bioanalyzer 2100) and sequenced on a HiSeq3000 (Illumina). Raw sequencing reads were processed using Partek Flow Analysis Software. Briefly, raw unaligned reads were quality checked, trimmed and reads with a Phred quality score >20 were used for alignment. Trimmed reads were aligned to human genome version hg38 with Gencode release 29 annotations using STAR aligner (2.5.3a) with default parameters. Transcript levels were quantified to the annotation model using Partek’s optimization of the expectation-maximization algorithm using default parameters. Transcript counts were normalized using reads per kilobase million (RPKM) approach and transformed to log2(RPKM + 1). Differential expression gene expression analysis was performed by the DESeq2 R package 72 . RNA extraction and quantitative real-time PCR Total RNA was extracted using the RNeasy Mini Kit (Qiagen) and cDNA was amplified using the Maxima First Strand cDNA Synthesis Kit (Thermo Fisher). Power SYBR Green (Applied Biosystems) RT-PCR amplification and detection was performed using an Applied Biosystems Step One Plus Real-Time PCR machine. The comparative Ct method for relative quantification (2−ΔΔCt) was used to quantitate gene expression, where results were normalized to Rpl7 (ribosomal protein L7). Primer sequences used were: Col2a1 —GGGTCACAGAGGTTACCCAG, ACCAGGGGAACCACTCTCA; Acan —GTGGAGCCGTGTTTCCAAG, AGATGCTGTTGACTCGAACCT and Rpl7 —ACCGCACTGAGATTCGGATG, GAACCTTACGAACCTTTGGGC. Western blot analysis Non-treated xiphoid processes were carefully dissected from 2 month old male and female WT mice, ground and then dissolved in RIPA Lysis and Extraction Buffer (Pierce) containing protease inhibitors (Pierce) followed by sonication with a 15-s pulse at a power output of 2 using the VirSonic 100 (SP Industries Company). Protein concentrations were determined by BCA protein assay (Pierce) and boiled for 5 min with Laemmli Sample Buffer (Bio-Rad, Hercules, CA). Proteins were separated on acrylamide gels and analyzed by Western blot using primary antibodies: anti-STAT3 (9139), anti-pSTAT3 (9145) and anti-Histone H3 (9515; all from Cell Signaling and used at 1:1000). Histone H3 antibody was used as a loading control. Proteins were resolved with SDS-PAGE utilizing 4–15% Mini-PROTEAN TGX Precast Gels and transferred to Trans-Blot Turbo Transfer Packs with a 0.2-µm pore-size nitrocellulose membrane. The SDS-PAGE running buffer, 4–15% Mini-PROTEAN TGX Precast Gels, Trans-Blot Turbo Transfer Packs with a 0.2-µm pore-size nitrocellulose membrane were purchased from Bio-Rad. Nitrocellulose membranes were blocked in 5% nonfat milk in 0.05% (v/v) Tween 20 (Corning). Membranes were then incubated with primary antibodies overnight. After washing in PBS containing 0.05% (v/v) Tween 20 (PBST), membranes were incubated with secondary antibodies (31460 and 31430, Thermo Scientific; 1:3000). After washing, development was performed with the Clarity Western ECL Blotting Substrate (Bio-Rad) and images were quantified by ImageJ software. Immunohistochemistry (IHC) Limb and spine tissues were dissected and fixed in 10% formalin for 24 h. They were then decalcified with 14% EDTA, pH7.4, for 4–5 days at 4 °C, embedded in paraffin and cut at a thickness of 5 µm. Paraffin sections were deparaffinized and rehydrated by passage through xylene and 100, 95, and 70% ethanol. Antigen Retrieval was carried out in citrate buffer (pH 6.0) for 30 min at 60 °C. Sections were blocked with 2% normal horse serum in TBS for 1 h at room temperature and incubated with primary antibody in 1% BSA at 4 °C overnight. Slides were washed three times in TBS containing 0.05% (v/v) Tween 20 (TBST) for 5 min each and incubated at room temperature for an hour in secondary antibody-HRP (Vector Laboratories). Antibodies were then visualized by peroxidase substrate kit DAB (Vector Laboratories). Slides were viewed using a Zeiss Axio Imager.A2 Microscope and images were taken using an Axiocam 105 color camera with Zen 2 program. Standard microscope camera settings were used. Auto-exposure was used to normalize background light levels across all images. Protein expression was quantified using ImageJ by measuring number of positively labeled cells in each section normalized to total number of analyzed cells and expressed as a percentage of positive cells. Primary antibodies included the following: aggrecan neoepitope (Novus Biologicals; NB100-74350, 1:500), Runx2 (Santa Cruz Biotechnology; sc-390351, 1:100), collagen X (Abcam; ab58632, 1:1000), SOX9 (Abcam; 26414, 1:200), pStat3 (Cell Signaling; 9415, 1:50), Stat3 (Cell Signaling; 9139, 1:50; Cell Signaling; 12640, 1:100), gp130 (Biosis; bs-1459R, 1:300), Lifr (Biosis; bs-1458R, 1:800) and aggrecan (Abcam; ab36861, 1:200). For pStat3 IHC, EDTA antigen retrieval buffer, pH 8.0, at 90 °C for 40 min was used. Safranin O staining was performed as described 73 . Determination of apoptosis and proliferation TdT-mediated dUTP nick-end labeling (TUNEL) or EdU assays were performed by using in situ cell death detection kit (TUNEL, Promega Corporation, G3250) or EdU Click-iT® Assay Kit obtained from Thermo Fisher (C10639), respectively, as described in the manufacturer’s protocol. EdU (Abcam) was injected intraperitoneally daily at 25 mg/kg for 4 days ( Acan -Cre +/ERT2 ; Stat3 fl/fl mice) or 4 h ( Acan -Cre +/ERT2 ; Rosa26 -Stat3C fl/fl ) prior to sacrifice. Slides were counterstained with DAPI. Apoptosis and proliferation were evaluated by ImageJ analysis. MicroCT data collection and analysis Eight mice (2 per sex per genotype; Stat3 fl/fl and Acan -Cre +/ERT2 ; Stat3 fl/fl ) were microCT scanned with a GE phoenix nanotom m X-ray nanoCT scanner with the following x-ray parameters to assess axial skeleton changes: 70 kV energy, 160 µA current, 1440 projection along 360-degree rotation, 1 frame per second, averaging 2 frames and skipping 1. Three consecutive scans were performed to cover whole mouse at 0.02 mm voxel resolution. All three initial scans were reconstructed using embedded software in the scanner. Visualization of 3D morphological structures and acquisition of measurements for quantification were performed using VGSTUDIO MAX 3.3 (Volume Graphics GmbH). Landmark-based data were collected from the lumbosacral region and pelvic girdle. Eight measurements were collected for each lumbar vertebra (Supplementary Table 3 ). GM analysis was applied to assess phenotypic changes. A total of 34 sacral and 26 pelvic landmarks were obtained for the 3D GM analysis. The sacral landmarks utilized were: 1. The most cranioventral point on S1 (sacral vertebra) centrum on midsagittal plane; 2. The most craniodorsal point on S1 centrum on midsagittal plane; 3. The right ventrolateral point on S1 centrum outline cranially, where the ascendant part of ala starts at pedicle; 4. The left ventrolateral point on S1 centrum outline cranially, where the ascendant part of ala starts at pedicle; 5. The right dorsolateral point on S1 centrum outline cranially, where the ascendant part of ala starts at pedicle; 6. The left dorsolateral point on S1 centrum outline cranially, where the ascendant part of ala starts at pedicle; 7. The most cranial middle point on the right ala; 8. The most cranial middle point on the left ala; 9. The most cranioventral point on the right ala; 10. The most cranioventral point on the left ala; 11. The most craniodorsal point on the right ala; 12. The most craniodorsal point on the left ala; 13. The most caudal (deepest) point on the concave rim between the right ala and articular process; 14. The most caudal (deepest) point on the concave rim between the left ala and articular process; 15. The most craniomedial point on the right articular process; 16. The most craniomedial point on the left articular process; 17. The most craniodorsal point on the right articular process; 18. The most craniodorsal point on the left articular process; 19. The most cranial point on the dorsomedial S1 sacral canal wall; 20. The most cranioventral point on S2 centrum on midsagittal plane; 21. The most cranioventral point on S3 centrum on midsagittal plane; 22. The most cranioventral point on S4 centrum on midsagittal plane; 23. The most caudoventral point on S4 centrum on midsagittal plane; 24. The most craniodorsal point on S1 neural spine; 25. The most craniodorsal point on S2 neural spine; 26. The most caudodorsal point on S2 neural spine; 27. The most craniodorsal point on S3 neural spine; 28. The most caudodorsal point on S3 neural spine; 29. The most craniodorsal point on S4 neural spine; 30. The most caudodorsal point on S4 neural spine; 31. The most caudodorsal point on S4 centrum on midsagittal plane; 32. The right caudolateral point on S4 centrum on midcoronal plane; 33. The left caudolateral point on S4 centrum on midcoronal plane; 34. The most caudal point on the dorsomedial S4 sacral canal wall. Pelvic landmarks utilized were: 1. The central point of iliac spine rugosity, site attachment for m. rectus femori; 2. The caudodorsal point on the auricular surface (sacroiliac articular surface); 3. The ventral point on the lower ilium at the same transversal cross section as landmark 2, the transversal section corresponds to the longitudinal axis of pelvis; 4. The lateral point on the lower ilium at the same transversal cross section as landmark 2, the transversal section corresponds to the longitudinal axis of pelvis; 5. The medial point on the lower ilium at the same transversal cross section as landmark 2, the transversal section corresponds to the longitudinal axis of pelvis; 6. The deepest point on the ischium where its process meets acetabulum; 7. The ischial spine; 8. The caudodorsal point on ischial tuberosity; 9. The caudolateral point on the ischiopubic ramus; 10. The caudomedial point on the ramus of the ischium; 11. The most caudal point on the pubic symphysis; 12. The most cranial point on the pubic symphysis; 13. The ventral point of the ilio-pubic eminence; 14. The deepest point on the angulation between ventral point of ilio-public eminence and the longitudinal axis of ilium; 15. The most caudoventral point on the auricular surface; 16. The most ventral point on the iliac crest; 17. The most lateral point on the iliac crest; 18. The most dorsal point on the iliac crest; 19. The most cranial point on the rim of the obturator foramen; 20. The most ventral point on the rim of the obturator foramen; 21. The most dorsal point on the rim of the obturator foramen; 22. The cranial point on the rim of articular surface of acetabulum, which is defined as a parallel projection from landmark 1 along the major axis of pelvis; 23. The caudal point on the rim of articular surface of acetabulum, directly across from landmark 8 along the long axis of the ischium; 24. The caudoventral point of the lunate surface of the of acetabulum; 25. The deepest center point of the acetabulum. 26. The craniodorsal point of the articular surface of the acetabulum, which is defined as the extension of the line connecting landmark 24 and 25. Amira 2019.4 (FEI, Visualization Sciences Group) software was used to acquire the x , y and z coordinates for each landmark with accuracy of landmark size 0.1–0.2 mm. The visualization pattern of shape variance was performed using the interactive “MORPHOTOOLS” software 74 , 75 , 76 . The 3D coordinates of each landmarks were subjected to a generalized Procrustes analysis (GPA) 77 , 78 , which included translating, rotating, reflecting, and scaling them to a best least-squares fitting (GLS) 79 , 80 . Variation among landmark configurations and shape were explored via PCA 81 . Thin-plate spline analysis was used to warp surfaces on the basis of reference configuration and visualize the shape difference along the principal axis. Three mice per sex per genotype ( Stat3 fl/fl and Acan -Cre +/ERT2 ; Stat3 fl/fl ) were microCT scanned for epiphyseal trabecular bone analysis. Image data were acquired as above and focused on the knee joint to obtain 10 µm pixel size resolution images in order to be able to calculate trabecular properties on the proximal epiphyseal segment of tibia. The X-ray parameters used for these 10 µm pixel size resolution images included: 70 kV energy, 120 µA current, and a voxel resolution of 10 µm. Each specimen rotated 360 degrees about a vertical axis in front of the detector resulted in 1440 projections acquired at one frame per second. The 1440 projections were reconstructed into 16-bit unsigned images saved as.dicom formatted files using proprietary software of Nanotom GE phoenix (i.e., datosx 2 rec). The dicom files were used as the basis for data analyses. Visualization of 3D morphological structures and acquisition of measurements for quantification were performed using VGSTUDIO MAX 3.4 64 bit (Volume Graphics GmbH). After acquisitions of the initial image data on the knee joint, a subset of region of interest (ROI) in the proximal tibial epiphysis was created; cortical bone was removed from epiphysis up to the most distal limit of medial and lateral epicondylar transversal line. From this ROI trabecular BV/TV was calculated. For microCT analysis of Col2a1 -Cre; Stat3 fl/fl and control animals (3 per sex per genotytpe), a Skyscan1172 (Bruker MicroCT, Kontich, Belgium) machine using CTAn (v.1.13.2.1) and CTVol (v.2.3.2) software was employed. The microradiography unit was set to an energy level of 55 kVp, intensity of 181 µA, and 900 projections; specimens were scanned at an 8 µm voxel resolution. A three‐dimensional reconstruction was generated with NRecon software (Bruker MicroCT) from the set of scans. Femurs were measured directly from these reconstructions. Tissue collection and digestion Mouse knee joints were cut at the femur and the tibia and crushed lightly with a mortar and pestle and placed in digestion media (DMEM/F12 (Corning) with 10% Fetal Bovine Serum (FBS; Corning), 1% penicillin/streptomycin/amphotericin B solution (P/S/A; Corning), 1 mg/ml dispase (Gibco), 1 mg/ml type 2 collagenase (Worthington), 10 µg/ml gentamycin (Teknova) and 100 μg/ml primocin (Invivogen) in an Erlenmeyer flask with a spin bar in 4 °C overnight followed by 37 °C for 4–6 h. Cells were washed with DPBS (Corning) after digestion and strained through a 70 μm filter (Fisher) before FACS. FACS All FACS was performed on a BD FACSAria IIIu cell sorter. Mouse knee cells were washed twice in 1–2% FBS and stained with DAPI for viability. Populations of interest based on DAPI negativity expression and lack of lineage markers (APC-Cy7; CD31, Ter119, CD45 at 1 μl/10 6 cells; Biolegend) expression were directly sorted into DMEM/F12 containing 10% FBS with 1% P/S/A. Flow cytometry data was analyzed using FlowJo software. Single-cell sequencing using 10X Genomics Single-cell samples were prepared using Chromium Next GEM Single Cell 3' Reagent Kits v3.1 and chip Kit (10X Genomics) according to the manufacturer’s protocol. Briefly samples were FACS sorted using DAPI to select live cells followed by resuspension in 0.04% BSA-PBS. Nearly 1200 cells/µl were added to each well of the chip with a target cell recovery estimate of 10,000 cells. Thereafter Gel bead-in Emulsions (GEMs) were generated using GemCode Single-Cell Instrument. GEMs were reverse transcribed, droplets were broken and single stranded cDNA was isolated. cDNAs were cleaned up with DynaBeads and amplified. Finally, cDNAs were ligated with adapters, post-ligation products were amplified, cleaned-up with SPRIselect and purified libraries were sequenced on Novaseq at the UCLA Technology Center for Genomics and Bioinformatics. 10X sequencing data analysis Raw sequencing reads were processed using Partek Flow Analysis Software (build version 10.0.21.0602). Briefly, raw reads were checked for their quality, trimmed and reads with an average base quality score per position >30 were considered for alignment. Trimmed reads were aligned to the mouse genome version mm10 using STAR-2.7.3a with default parameters. Reads with alignment percentage >75% were de-duplicated based on their unique molecular identifiers (UMIs). Reads mapping to the same chromosomal location with duplicate UMIs were removed. Transcripts were quantified using mm10 Gencode M25 and UMI count matrix was generated with rows representing genes and columns representing each detected cell. QC metrics for genes and UMIs were visualized as violin plots using ggplot2 (v3.3.3). Thereafter “Knee” plot was constructed using the cumulative fraction of reads/UMIs for all barcodes. Barcodes below the cut-off defined by the location of the knee were assigned as true cell barcodes and quantified. Further noise filtration was done by removing cells having >5% mitochondrial counts and total read counts >24,000. Ambient mRNAs were removed using the R package SoupX 82 , resulting in 4908 cells in the Stat3 fl/fl sample and 1399 cells in the Acan -Cre +/ERT2 ; Stat3 fl/fl sample. Genes not expressed in any cell were also removed as an additional clean-up step. Cleaned-up reads were normalized using counts per million method followed by log-transformation generating count matrices for each sample. Samples were batch-corrected on the basis of expressed genes and mitochondrial reads percent. Count matrices were used to visualize and explore the samples in further details by generating tSNE plots, a non-linear dimensional reduction technique. This algorithm learns the underlying manifold of the data and places similar cells together in a low-dimensional space; 30 prinicipal components were used for clustering. Cells positive for Sox9 and Col2a1 expression were selected for clustering and further analysis (501 in the Stat3 fl/fl sample and 332 cells in the Acan -Cre +/ERT2 ; Stat3 fl/fl sample). K-means clustering was computed for identifying groups of cells with similar expression profiles using Euclidean distance metric based on the most appropriate cluster count. A maximum of 1000 iterations were allowed and the top marker genes (fold-change >2 and p value <0.05) for each cluster was determined. Top marker genes were overlapped with a list of cell cycle marker genes obtained from Tirosh et al. 83 . Statistical significance of this overlap was determined by a hypergeometric test assuming 21,000 mouse genes. Two-way Venn diagrams were generated using BioVenn 84 . Dot plots were generated in R (v 4.0.3) using ggplot2 (v3.3.3). Gene set enrichment analysis (GSEA) Single-cell sequencing data was analyzed using the GSEA-P desktop application 85 available at . The specific parameters used for the analyses are: number of permutations (1000), permutation type (gene set), enrichment statistic (weighted), metric for ranking genes (signal to noise). Enrichment score was calculated by walking down the rank-ordered gene list from top to last ranked gene, decreasing a running-sum statistic when a gene is encountered that is not part of a gene set, and increasing the score when a gene is encountered that is in the gene set. Hypertrophic gene set included genes Ptgis, Snai1, Cd200, Plin2, Foxa2, Ugp2, Adk, Lxn, Enpp2, Mef2c, Col10a1, Sp7, Alpl, Ihh and Col1a1 . Proliferative gene set included genes Itga4, Cdk1, Dcx, Sox5, Fn1, Cdk2, Sox6, Ptges3, Foxn2, Sdc4, Ncam1, Rps25, Lmna, Bmpr1b and Prg4 . Statistics and reproducibility Numbers of repeats for each experiment are indicated in the figure legends. Pooled data are represented as mean ± SEM unless otherwise indicated. Statistical analysis was performed using two-tailed Student’s t test to compare two groups unless noted, in which case ANOVA was used to compare three or more groups. p values <0.05 were considered to be significant. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability All scRNA-seq and bulk RNA-seq data are deposited in GEO and are available under accession numbers GSE184406 and GSE188353, respectively. Uncropped Western blot images are provided in Supplementary Fig. 15 . Source data files are provided in Supplementary Data 3 . All other data are available upon reasonable request. Change history 25 February 2022 A Correction to this paper has been published:

The IL-6 family of proteins has a bad reputation: it can promote inflammation, arthritis, autoimmune disease and even cancer. However, a new USC-led study published in Communications Biology reveals the importance of IL-6 and associated genes for maintaining and regenerating cartilage in both the joints and in the growth plates that enable skeletal growth in children. "We show, for the first time, that the IL-6 family, previously almost exclusively associated in the musculoskeletal field with arthritis, bone and muscle loss, and other chronic inflammatory diseases, is required for the maintenance of skeletal stem and progenitor cells, and for the healthy growth and function of the joints and spine," said the study's corresponding author Denis Evseenko, who is the J. Harold and Edna LaBriola Chair in Genetic Orthopedic Research, and an associate professor of orthopedic surgery, and stem cell biology and regenerative medicine at USC. "Our study establishes a link between inflammation and regeneration, and may explain why stem and progenitors are exhausted in chronic inflammation." In the study, first author Nancy Q. Liu from USC and her colleagues took a close look at a key gene activated by IL-6: STAT3. In both lab-grown human cells and in mice, the scientists demonstrated that STAT3 is critical for the proliferation, survival, maturation and regeneration of cartilage-forming cells in the joints and growth plates. When the gene ceased to function, cartilage-forming cells became increasingly dysfunctional over time, resulting in smaller body size, prematurely fused growth plates, underdeveloped skeletons and mildly degenerated joint cartilage. Mice experienced the same issues when they lacked a protein called glycoprotein 130 (gp130), which all IL-6 proteins use to activate Stat3. Deactivating another gene Lifr, which encodes a protein that works with gp130 to recognize one of the IL-6 proteins called Lif, produced similar but milder skeletal and cartilage changes. In mice lacking gp130, the scientists could restore normal growth plates by over-activating Stat3—although this also caused an overgrowth of cartilage that led to other skeletal abnormalities. Interestingly, the researchers noted significant sex-related differences: when Stat3 ceased to function, females experienced more severe cartilage and skeletal changes than males. To understand why, the researchers altered estrogen levels in mice, as well as in lab-grown pig cartilage cells. In both cases, estrogen increased the amount and activity of Stat3, suggesting that females might rely more heavily on this gene. The study has clinical implications for the use of existing drugs that inhibit STAT3 to curb inflammation in autoimmune diseases: these drugs may also interfere with growth and regeneration. Conversely, the Evseenko Lab has leveraged their understanding of the nuances of STAT3 and associated genes and proteins to develop a highly targeted drug with the potential to regenerate joint cartilage without triggering inflammation. This drug will soon be tested in human clinical trials. "Our findings really shift the paradigm and challenge the existing dogmas in the field about how IL-6, STAT3, and associated genes and proteins influence not only inflammation, but also regeneration," said Evseenko.

10.1038/s42003-021-02944-y

Medicine

Suppression of newly found protein could lead to future treatments to slow Alzheimer's progression

FAM222A encodes a protein which accumulates in plaques in Alzheimer's disease, Nature Communications (2020). DOI: 10.1038/s41467-019-13962-0 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-019-13962-0

https://medicalxpress.com/news/2020-01-suppression-newly-protein-future-treatments.html

Abstract Alzheimer’s disease (AD) is characterized by amyloid plaques and progressive cerebral atrophy. Here, we report FAM222A as a putative brain atrophy susceptibility gene. Our cross-phenotype association analysis of imaging genetics indicates a potential link between FAM222A and AD-related regional brain atrophy. The protein encoded by FAM222A is predominantly expressed in the CNS and is increased in brains of patients with AD and in an AD mouse model. It accumulates within amyloid deposits, physically interacts with amyloid-β (Aβ) via its N-terminal Aβ binding domain, and facilitates Aβ aggregation. Intracerebroventricular infusion or forced expression of this protein exacerbates neuroinflammation and cognitive dysfunction in an AD mouse model whereas ablation of this protein suppresses the formation of amyloid deposits, neuroinflammation and cognitive deficits in the AD mouse model. Our data support the pathological relevance of protein encoded by FAM222A in AD. Introduction Alzheimer’s disease (AD), the leading cause of dementia named for Dr. Alois Alzheimer, is characterized by pathologic hallmarks amyloid plaques and neurofibrillary tangles, and accompanied by other prominent pathological changes such as progressive atrophy of the brain, neuropil threads, dystrophic neurites, granulovacuolar degeneration, Hirano bodies, and cerebrovascular amyloid 1 . Amyloid plaques are spherical extracellular lesions composed of amyloid-β (Aβ) peptides, whereas neurofibrillary tangles are intracellular lesions made up of hyperphosphorylated form of the microtubule-associated protein tau. Although many risk factors such as aging, lifestyle, and environmental factors are usually considered for the pathogenesis, AD is increasingly proposed to be a genetically dichotomous disease in the early-onset familial form showing classical Mendelian inheritance with little influence from the environment (EOAD), or in the late-onset sporadic form inherited in a non-Mendelian fashion (LOAD) 2 . Less than 10% of AD cases are EOAD with only a small fraction caused by autosomal dominantly inherited genetic changes in amyloid precursor protein (APP), presenilin 1 (PS1) or presenilin 2 (PS2), all of which are responsible for the overproduction of Aβ and the earlier formation of amyloid plaques 3 . Though more than 90% of AD cases are LOAD referred to as sporadic AD without family history, they have the similar clinical and pathologic phenotypes as EOAD and are heritable 4 . In the past decade, intensive efforts have been made to identify over 25 genes associated with AD 5 . In support of the dominant amyloid cascade hypothesis suggesting Aβ deposition in the brain as the primary cause, a number of AD-associated genes are enriched in the APP processing pathway, and involved in Aβ overproduction and amyloid plaque deposition though their encoded proteins are usually not directly associated with amyloid plaques. Quantitative structural magnetic resonance imaging (MRI) has been extensively used for assessment of AD-related structural differences in selective brain regions 6 . Genome-wide association studies (GWAS) using MRI measures have identified several AD risk variants 7 , 8 . Likewise, the MRI changes in statistically-defined regions of interest (ROI) were found closely associated with reported AD risk variants 9 , 10 . Genetic loci harboring variants can be associated with multiple, sometimes seemingly distinct, traits 11 , 12 . The test for such associations, i.e., cross-phenotype association test, has been increasingly employed to investigate the genetic overlap between multiple traits and diseases. Our previous study has developed a cross-phenotype association analysis (CPASSOC) that can integrate association evidence from GWAS summary statistics of multiple traits, either correlated, independent, continuous, or binary traits, and has been successfully used to identify four loci associated with hypertension-related traits missed by a single-trait analysis 13 . In this study, we performed CPASSOC analysis of MRI measures and genetic datasets, and identified a possible link between FAM222A and AD-related regional brain atrophy. To understand its pathological role in AD, we investigated the protein encoded by FAM222A in patients with AD or transgenic mice for AD, and found its characteristic accumulation within the center of amyloid deposits. Further mechanistic study revealed that this protein could physically interact with Aβ and regulate Aβ aggregation and amyloid formation. Our results therefore identify a protein that likely plays an important role in amyloidosis, a finding providing perspective for AD pathogenesis. Results Susceptibility of regional brain atrophy to FAM222A in AD To identify brain atrophy-related imaging quantitative trait loci, we employed a genome-wide whole brain approach to analyze the imaging genetic dataset from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (Supplementary Fig. 1a ). After GWAS and the estimation of shared genetic contributions among 145 ROIs spanning the entire brain by linkage disequilibrium (LD) regression method 14 (Supplementary Fig. 1b, c ), we attempted to extract disease-related ROIs and detect genetic variants associated with them. With hierarchical clustering analysis on a genetic correlation network (Supplementary Fig. 1d–k ), 16 modules of ROIs with high within-module genetic correlation were generated (Supplementary Fig. 1l, m ). We further combined GWAS summary statistics of ROIs in each module using CPASSOC we developed 13 . Previously reported AD top markers, APOE single nucleotide polymorphism (SNP) rs429358 15 , TOMM40 SNP rs2075650 15 , APOC1 SNP rs12721051 16 , and rs117028417 on FAM222A were found in one module (Fig. 1 , Supplementary Figs. 2 , 3 and Supplementary Table 1 ), which consists of five ROIs including left hippocampus, right hippocampus, basal forebrain, entorhinal area, and planum polare, brain areas we know are affected by AD 17 , 18 , 19 and well predict AD (Supplementary Fig. 4 ). SNP rs117028417 had a minor allele A (frequency = 0.044) with positive effects for all 5 ROIs in the ADNI cohort ( P = 1.95 × 10 −8 for CPASSOC analysis; Supplementary Table 2 ), and was further validated to be associated with the mean volume of hippocampus, one of the earliest affected brain regions in AD, in the Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA) consortium cohort comprising MRI images of 30,717 individuals from 50 cohorts 20 with the same effect direction ( β = 38.1, P = 8.29 × 10 −3 ; Supplementary Table 2 ). Fig. 1: CPASSOC analysis of the ADNI cohort. Manhattan plot of CPASSOC analysis combining GWAS summary statistics of ROIs in the green-colored module (Supplementary Fig. 1m ) including left hippocampus, right hippocampus, basal forebrain, entorhinal area, and planum polare. The red line represents the threshold of P = 5 × 10 −8 for the genome-wide significance level. Arrows indicate loci associated with regional brain atrophy. Source data are provided as a Source Data file (Source Data for GWAS in Fig. 1). Full size image On a genome-wide scale, FAM222A rs117028417 was only marginally associated with AD diagnosis in the International Genomics of Alzheimer’s Project (I-GAP, P = 0.052) 21 . The positron emission tomography (PET) imaging using radiotracer 18 F-Florbetapir (AV45) provides quantitative measures of amyloid pathology in vivo 22 . To investigate whether SNP rs117028417 is associated with brain Aβ accumulation, we performed single SNP association test of rs117028417 with AV45 standard uptake value ratio (SUVR) measures for brain regions available in the PET scan dataset from ADNI. The minor allele A of rs117028417 was found significantly associated with longitudinal decrease but not baseline of AV45 SUVR measures in anterior or posterior cingulate ( P < 0.0125) (Supplementary Table 3 ), which is relatively consistent with the findings from brain volume associations, where the association direction was positive (i.e. higher volume). rs117028417 is located in an intergenic region that is approximately 5 kb downstream of FAM222A and 8 kb downstream of TRPV4 . We searched among published GWAS studies of AD or AD-related biomarkers and did not identify common variants previously reported on these two genes. We thus further performed gene-based burden and SKAT tests 23 , 24 of coding variants on both FAM222A and TRPV4 . All of the coding variants on these two genes available in ADNI whole-genome sequencing data were low-frequency or rare variants with minor allele frequency less than 0.05 (Supplementary Table 4 ). FAM222A , but not TRPV4 , displayed significant association with AV45 SUVR longitudinal changes in anterior or posterior cingulate or lateral parietal regions in the burden test after adjusting for 10 independent tests ( P < 0.005) (Supplementary Table 5 ), collectively suggesting a possible role of FAM222A in brain amyloid deposition. However, when we tested rs117028417 for genetic association with AD cerebrospinal fluid (CSF) Aβ and tau biomarkers, only nominal association of rs117028417 with total tau annual change could be discovered, and there was no association with baseline CSF Aβ and tau on both single SNP association tests and variant burden tests (Supplementary Tables 6 , 7 ), indicating that FAM222A variant may not have a strong genetic influence on AD biomarkers. FAM222A -encoded protein accumulates within amyloid plaques The large independent AD brain imaging dataset including GWAS studies is not available at this time, making it difficult to further validate the genetic link between FAM222A and AD-related brain atrophy. However, to elucidate the possible pathological role of FAM222A in AD, we carried out experimental validation to focus on its encoded protein, which we designated as Aggregatin. Aggregatin consists of 452 amino acids with a predicted molecular weight of 47 kD, and has not yet been characterized. Using a well-characterized specific antibody against Aggregatin (Supplementary Fig. 5a–e ), Aggregatin was found predominantly expressed in the central nervous system (CNS) including both the brain and the spinal cord, but not in other tissues such as heart, spleen, lung, kidney, or liver in mice or humans (Supplementary Fig. 5d, e ). There was a slight increase in the expression of Aggregatin in brain lysates from AD patients compared to age-matched control subjects (Supplementary Fig. 5f–h ). The most distinct pattern of Aggregatin immunostaining observed in AD was that Aggregatin was remarkably immunoreactive within the center of amyloid plaques, which were stained by the pan-Aβ antibodies 6E10 and 4G8, the N-terminal truncated and modified pyroglutamate Aβ species Aβ[N3pe] antibody 82E1, fibrillar Aβ dye thioflavin-S (Thio-S) or oligomer Aβ antibody NU-4 25 (Fig. 2a, b and Supplementary Fig. 6a–e ). In contrast, all control brain sections lacking detectable amyloid plaques demonstrated weak diffusive Aggregatin immunoreactivity without association with puncta (Fig. 2a ). Fig. 2: Aggregatin accumulates within the center of amyloid deposits. a Representative images of immunohistochemistry of Aggregatin (arrowheads) and amyloid plaques (stained by the 6E10 antibody) in adjacent sections (denoted by asterisks) of cortices of sporadic AD patients. b Representative fluorescent images of Aggregatin (Red), amyloid plaques (Green, stained by the 6E10 antibody) and DAPI nuclei staining (Blue) in cortices of sporadic AD. c Representative images of immunohistochemistry of Aggregatin (arrowheads) and amyloid plaques (stained by the 6E10 antibody) in adjacent sections (denoted by asterisks) of brains of 6-month-old 5XFAD mice. d Representative images of Aggregatin (Red), amyloid plaques (Green, stained by the 6E10 antibody) and DAPI nuclei staining (Blue) in brains of 6-month-old 5XFAD mice. e , f Representative dot blots of Aggregatin and Aβ (6E10) in serial fractions of amyloid plaques separated by differential centrifugation in sucrose gradient from sporadic AD patients ( e ) or 6-month-old 5XFAD mice ( f ). g , h Representative immunoblots of Aggregatin and Aβ (6E10) in the SDS-resistant insoluble core-enriched fraction from sporadic AD patients ( g ) or 6-month-old 5XFAD mice ( h ). Arrow heads point Aggregatin. Due to the presence of urea used for plaque core protein extraction, plaque core fractions show slight shifts compared to SDS soluble fraction. All experiments were independently performed at least three times. Source data are provided as a Source Data file (Source Data for Statistics and Blots). Full size image Robust Aggregatin staining of the central core of amyloid deposits was consistently observed in the brains of multiple mouse models for AD including 5XFAD 26 , TgCRND8 27 , APP/PS1 28 , Tg2576 29 , and 3xTg 30 transgenic mice overexpressing human mutant APP along with or without human mutant PS1 (Fig. 2c, d and Supplementary Fig. 6f–h ). With the exception of 5XFAD or Tg2576 mice in which Aggregatin-positive foci were connected with wispy fibrils, Aggregatin within amyloid deposits of other transgenic mice showed negligible projecting fibrillar structures, similar as in human plaques. Despite the general localization of Aggregatin large puncta to the core of amyloid deposits, they highly co-localized with Aβ in 5XFAD mice but not in AD patients or TgCRND8 mice, together indicating that the processes contributing to amyloid deposition may be different in human and different animal models. Notably, the formation of Aggregatin puncta occurred concurrently with amyloid deposition, but was not present in the pre-depositing young 5XFAD mice (Supplementary Fig. 6i ). The characteristic Aggregatin-positive core staining was abolished by the pre-absorption of primary antibodies with human recombinant Aggregatin protein (rAggregatin) purified by combined 10 K dialysis and size-exclusion chromatography, but not Aβ 1-42 peptides (Supplementary Fig. 7a ), further validating the specificity of the anti-Aggregatin antibody. To confirm the presence of Aggregatin within amyloid deposits, we isolated amyloid cores purified by sucrose density gradient fractionation of 2% sodium dodecyl sulfate (SDS) homogenized AD or 5XFAD mouse brains. Dot blot and immunoblot studies of proteins under native and denatured forms respectively confirmed the existence of full-length Aggregatin without noticeable cleaved products in the SDS-resistant insoluble core-enriched fractions positive for 6E10 (Fig. 2e–h ). Aggregatin physically interacts with Aβ The radioimmunoprecipitation assay buffer (RIPA) widely used for co-immunoprecipitation failed to extract Aggregatin from AD brains (Supplementary Fig. 7b ), making it difficult to examine the likely association between Aggregatin and Aβ in AD. To overcome this obstacle, we performed in vitro pull-down assays using synthetic Aβ 1–40 or Aβ 1–42 and rAggregatin. Dynamic light scatting (DLS), circular dichroism (CD), and SDS-PAGE assays of rAggregatin indicated that rAggregatin existed in the soluble partially folded monomeric state (Supplementary Fig. 7c–e ). Notably, rAggregatin co-precipitated with different forms of Aβ 1–40 or Aβ 1–42 (Fig. 3a and Supplementary Fig. 7f, g ). Consistently, immobilized monomeric Aβ 1–40 or Aβ 1–42 was also able to pull down rAggregatin (Supplementary Fig. 7h ). Further surface binding affinity assays revealed that immobilized Aβ 1–40 or Aβ 1–42 bound to rAggregatin, and similarly, immobilized rAggregatin bound to Aβ 1–40 or Aβ 1–42 all within the nanomolar ranges (Fig. 3b, c and Supplementary Fig. 7i, j ). In agreement with these results, surface plasmon resonance (SPR) measurements confirmed that Aβ 1–42 bound to immobilized rAggregatin at the low nanomolar dissociation equilibrium constant (Kd) (Fig. 3d ). Although no measurement was noted in blank or BSA-immobilized sensor chips (Supplementary Fig. 7k ), signal spikes produced in the SPR assays may be in proportion to the mass of Aβ aggregates, making dynamic measurements unlikely consistent with the surface binding affinity assessments at the steady state. To investigate the binding of rAggregatin to Aβ ex vivo, we performed an in situ binding assay in which fixed brain sections of AD patients or 5XFAD mice were incubated with Flag-tagged rAggregatin and stained by an anti-Flag antibody. Remarkably, all amyloid deposits were labelled by rAggregatin (Supplementary Fig. 7l–n ). Considering the widespread presence of Aβ in brains, it was not surprising that brain sections also showed background staining after rAggregatin incubation. Notably, amyloid deposits and the background binding of rAggregatin were completely abolished by pre-incubation of rAggregatin with Aβ 1–40 or Aβ 1–42 (Supplementary Fig. 7l–n ), confirming that rAggregatin binds amyloid deposits by interacting with Aβ. Collectively, these results highlight the pathological relevance of Aggregatin in AD, and show that Aggregatin is a Aβ binding protein with high-affinity. Fig. 3: Aggregatin interacts with Aβ. a Coimmunoprecipitation of purified Flag-tagged rAggregatin and Aβ 1–42 (pre-aggregated in vitro for 24 or 48 h). rAggregatin was immunoprecipitated using streptavidin magnetic beads and immunoblotted using the antibody to Flag. b Measurement of Aβ 1–42 levels bound to immobilized rAggregatin (normalized to maximal rAggregatin and Aβ 1–42 binding). n = 3 independent experiments. c Measurement of rAggregatin levels bound to immobilized Aβ 1–42 (normalized to maximal rAggregatin and Aβ 1–42 binding). n = 3 independent experiments. d Bio-layer interferometry measurement of the binding kinetics of monomeric Aβ 1–42 to immobilized rAggregatin. Curves are corresponded to Aβ 1–42 at 8320, 4160, 2080,1040, 520, 260, 130 and 65 nM from the top to bottom. e , f Representative immunohistochemistry ( e ) and quantification ( f ) of rAggregatin immunoreactivity (Flag antibody) in 5XFAD mouse brain sections after incubation with 100 nM indicated rAggregatin deletion mutants ( n = 6 independent experiments in each group). Blocks with blue color on the top of each immunohistochemistry image show NABD. g Coimmunoprecipitation analysis of purified Flag-tagged rAggregatin deletion mutations and Aβ 1–42 using streptavidin magnetic beads. Source data are provided as a Source Data file (Source Data for Statistics and Blots). Data are means ± s.e.m (± is the plus–minus sign). One-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test. **** P < 0.0001. ns, non-significant. Full size image Aggregatin binds to Aβ via its N-terminal region Next, we generated a series of rAggregatin deletion mutants to map the binding region for Aβ. Although rAggregatin alone does not form oligomers or aggregates, the composition of Aβ preparations at the micromolar range quickly changes over time due to the formation of higher order oligomers, which are expected to influence the Aggregatin and Aβ interaction. To quantitatively identify the binding strength of different rAggregatin deletion mutants, the in situ binding assay rather than pull-down assay was used for the binding motif mapping. The deletion of residues from 1 to 80 (designated as NABD, N-terminal Aβ binding domain), but not residues outside of this region, was found to greatly reduce the binding of rAggregatin to amyloid deposits (Fig. 3e–g and Supplementary Fig. 8a, b ). Recombinant NABD (rNABD) alone was able to bind to amyloid deposits or Aβ 1–42 similar as full-length rAggregatin, and caused a dose-dependent decrease in the association between rAggregatin and amyl deposits when co-incubated (Fig. 3b, c, e–g and Supplementary Fig. 8c, d ), together suggesting NABD as the domain both necessary and sufficient for Aβ binding. The residues from 61 to 80 appear to be a core motif for NABD though they alone were not sufficient to bind amyloid deposits (Fig. 3e–g and Supplementary Fig. 8a, b ). Notably, rNABD bound amyloid deposits in a length-dependent manner, and rAggregatin with partial deletions of every 5 amino acids within the core motif of NABD exhibited weaker but still strong interactions with amyloid deposits (Supplementary Fig. 8a, b ), further indicating that NABD may contain multiple sites cooperatively involved in Aβ binding. Aggregatin cross-seeds Aβ via direct binding Given the strong interaction between Aggregatin and Aβ, we further set out to determine whether Aggregatin would influence the Aβ aggregation process. Aβ aggregation kinetics were first monitored in vitro using Aβ 1–40 or Aβ 1–42 for the thioflavin T (ThT) based fluorescence assay. As illustrated by changes in ThT-associated fluorescence, Aβ self-aggregated only at high concentrations whereas rAggregatin alone did not produce any observable aggregate (Fig. 4a and Supplementary Fig. 9a ). Remarkably, once co-incubated with rAggregatin, Aβ was able to form aggregates at low concentrations even in the nanomolar range (Fig. 4a, b and Supplementary Fig. 9a–c ). With increasing concentrations of rAggregatin, the lag times of the aggregation reaction were greatly decreased (Fig. 4a and Supplementary Fig. 9b ). As a control, rAggregatinΔ61–80 had similar folding as wild type rAggregatin, but failed to induce Aβ 1–42 aggregation (Fig. 4a and Supplementary Fig. 7d, e ). These observations were confirmed using immunoblot and dot blot analyses for Aβ aggregation measurements under denatured and native conditions, which showed that Aggregatin but not rAggregatinΔ61–80 indeed promoted Aβ 1–42 oligomerization (Fig. 4c–f and Supplementary Fig. 9d ). Of note, due to the sensitivity of immunoblot, Aβ oligomer was only detectable with long exposure when Aβ 1–42 at the low micromolar but not nanomolar was applied. Consistently, transmission electron microscopy analyses revealed that soluble Aβ 1–42 protofibrils 31 were more abundant and have more complicated structures in the presence of rAggregatin during the early phase of incubation when Aβ fibrils were absent (Fig. 4g ). As expected, the low concentration of Aβ 1–42 only yielded very few short and un-branched fibrils after long periods of incubation under negative staining (Fig. 4g ), and rAggregatin alone did not form identifiable particles or large aggregates (Supplementary Fig. 9e ). Strikingly, co-incubation of low micromolar Aβ 1–42 with rAggregatin lead to the formation of large micrometer-long branched fibrils (Fig. 4g and Supplementary Fig. 9f ), which were Thio-S-positive and visible under the fluorescent microscopy (Fig. 4h ). Taken together, these data imply Aggregatin as a potent seeding factor for Aβ oligomerization and aggregation. Fig. 4: Aggregatin accelerates Aβ aggregation in vitro. a ThT-based assay measuring aggregation kinetics of 2.5 µM Aβ 1–42 in the presence of various concentrations of rAggregatin ( n = 5 independent experiments in each time points). b ThT-based assay measuring aggregation kinetics of various concentrations of Aβ 1–42 in the presence of 5 nM rAggregatin ( n = 5 independent experiments in each time points). c , d Representative immunoblot ( c , light exposure shown in Fig. S9d) and quantification (d) of Aβ 1–42 oligomers recognized by 6E10 in the 30 nM rAggregatin and 2.5 µM Aβ 1–42 mixture collected after 6-h co-incubation ( n = 4 independent experiments). Arrow head points to non-specific bands due to long exposure. e , f Representative dot blot ( e ) and quantification ( f ) of Aβ 1–42 oligomers recognized by the oligomer Aβ specific antibody NU-4 in the 30 nM rAggregatin and 2.5 µM Aβ 1–42 mixture collected after 6-hour co-incubation ( n = 4 independent experiments). g Negative staining electron microscopy of 2.5 µM Aβ 1–42 aggregates after 0.5-h, 6-h, 2-week, and 4-week co-incubation with or without 30 nM rAggregatin. h Representative 3D images of 2.5 µM Aβ 1–42 aggregates stained by Thio-S after 4-week co-incubation with or without 30 nM rAggregatin. Source data are provided as a Source Data file (Source Data for Statistics and Blots). Data are means ± s.e.m. One-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test. **** P < 0.0001. ns, non-significant. Full size image Aggregatin regulates amyloid deposition Aβ levels are low in young especially predepositing 5XFAD mice 26 . To examine the effect of extracellular Aggregatin on amyloid deposition with unrestricted access to predeposit-state Aβ, we performed intracerebroventricular (ICV) infusion of Flag-tagged rAggregatin or rAggregatinΔ61–80 into 5XFAD mice at 4-month-old, when Aβ rises to high levels 26 (Supplementary Fig. 10a ). Infusion did not cause the death of mice or histological abnormalities in the brain. Importantly, the levels of total Aβ, APP or BACE1 remained unchanged 4 weeks after rAggregatin infusion, indicating that rAggregatin did not affect Aβ production or degradation (Supplementary Fig. 10b, c ). ICV infused rAggregatin was detected in amyloid deposit (Fig. 5a ). Remarkably, compared to age-matched control mice infused with artificial cerebrospinal fluid (aCSF), rAggregatin-infused mice showed greatly increased amyloid deposition spreading the brain at 5 months of age, which could be completely blocked by the deletion of NABD core motif (Fig. 5b, c and Supplementary Fig. 10d, e ). As prominent AD pathological features, microgliosis and astrogliosis are closely associated with amyloid deposits in 5XFAD mice 26 , 32 . Corresponding to increased plaque load, 5XFAD mice infused with rAggregatin but not rAggregatinΔ61–80 exhibited more microgliosis and astrogliosis compared to aCSF-infused control 5XFAD mice (Fig. 5d and Supplementary Fig. 10f ). 5XFAD mice begin to show cognitive deficits at around 4-months-old 33 , 34 . Compared with NTG mice, FAD mice exhibited significantly impaired Y-maze and Barnes-maze performance, both of which were significantly exacerbated in transgenic mice with rAggregatin but not rAggregatinΔ61–80 infusion (Fig. 5e, f ). To further examine the role of neuronal Aggregatin in amyloid deposition, we injected adeno-associated virus serotype 1 encoding human Aggregatin or GFP alone under the neuron specific promoter eSYN (AAV1-Aggregatin or AAV1-GFP) into the hippocampus CA1 of young predepositing 5XFAD mice at 1.5-month-old (Supplementary Fig. 11a ). When analyzed at 5 months of age, in line with ICV infusion experiments, intrahippocampal injection of AAV1-Aggregatin significantly increased amyloid deposition without any effect on total Aβ levels in the GFP-positive hippocampal region, but not in the brain areas without AAV1-Aggregatin delivery (Fig. 5g, h and Supplementary Fig. 11b–f ), together suggesting that Aggregatin is sufficient to enhance amyloid deposition in vivo. Consistently, amyloid deposition associated microgliosis, astrogliosis, and cognitive deficits were also worsened by neuronal Aggregatin overexpression (Fig. 5i–k and Supplementary Fig. 11g ). To investigate whether Aggregatin was required for amyloid deposition, we performed intrahippocampal injection of AAV1 co-expressing GFP and a short hairpin RNA targeting Aggregatin (AAV1-shAggregatin) or control shRNAi (AAV1-shControl) in predepositing 5XFAD mice (Supplementary Fig. 5b, c ). It was observed that decreasing Aggregatin was not associated with neuronal loss or altered total Aβ (Supplementary Fig. 12a ). At 5 months of age, the injection of AAV1-shAggregatin significantly alleviated amyloid deposition in the GFP-positive areas of hippocampus compared to AAV1-shControl injection, but not in the GFP-negative brain areas (Fig. 5l, m and Supplementary Fig. 12b–e ). Likewise, Aggregatin reduction significantly alleviated amyloid deposit associated microgliosis, astrogliosis, and cognitive impairment (Fig. 5n–p and Supplementary Fig. 12f ). Taken together, these results further imply that Aggregatin is also an important factor necessary for amyloid deposition. Fig. 5: Aggregatin regulates amyloid deposits. 5-month-old 5xFAD mice were ICV infused with Flag-tagged rAggregatinΔ61–80 or rAggregatin for 4 weeks. a Representative images of Flag-tagged Aggregatin (Red) and amyloid plaques (Green, Thio-S) in the brain. b , c Representative images ( b ) and quantification ( c ) of plaque by NU-4 antibody in the total brain (Total), cortex or hippocampus ( n = 18 mice in each group). d Quantification of astrogliosis and microgliosis in hippocampus (representative images shown in Supplementary Fig. 10f ). e , f Y-maze ( e ) and Barnes maze ( f ) performance ( n = 15, 17, 18, 18, and 18 mice for NTG aCSF, NTG rAggregatin, 5XFAD aCSF, 5XFAD rAggregatinΔ61–80, and 5XFAD rAggregatin respectively). 5-month-old 5XFAD mice were injected with AAV1-GFP or AAV1-Aggregatin at 1.5 month-old. g , h Representative images ( g ) and quantification ( h ) of plaques stained by NU-4 in the hippocampus ( n = 18 mice in each group). i Quantification of astrogliosis and microgliosis in the hippocampus ( n = 18 mice in each group). j , k Y-maze ( j ) and Barnes maze ( k ) performance ( n = 18 mice in each group). Representative images are shown in Supplementary Fig. 11g . 5-month-old 5XFAD mice were injected with AAV1-shControl or AAV1-shAggregatin at 1.5-month-old. l , m Representative images ( h ) and quantification ( i ) of plaques stained by NU-4 in the hippocampus ( n = 18 mice in each group). n Quantification of astrogliosis and microgliosis in the hippocampus ( n = 18 mice in each group). Representative images are shown in Supplementary Fig. 12f . o , p Y-maze ( o ) and Barnes maze ( p ) performance ( n = 18 mice in each group). Source data are provided as a Source Data file (Source Data for Statistics and Blots). Data are means ± s.e.m. Student’s t -test or one and two-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test. * P < 0.05, # P < 0.05 (relative to aCSF, AAV1-GFP or shControl AAV1), *** P < 0.001, **** P < 0.0001. ns, non-significant. Full size image Discussion Here, we report on Aggregatin, the protein encoded by FAM222A , as a plaque core protein directly binding Aβ and facilitating Aβ aggregation, a process thought to be central in AD onset. Therefore, this work provides strong experimental evidence supporting a pathophysiological role for Aggregatin in AD. In people diagnosed with AD or mild cognitive impairment (MCI), a proportion of whom can progress to AD, FAM222A is associated with the module enriched for atrophy in AD-affected brain regions. FAM222A association with hippocampal volume could be validated in the replication ENIGMA cohort, together pointing to a potential mechanism by which FAM222A may affect regional brain atrophy. Notably, our cross phenotype association analysis also led to the identification of long-established AD risk genes APOE , TOMM40 , and APOC1 exclusively in the same module, suggesting possible genetic interplays between FAM222A and AD risking genes. Interestingly, although we only discovered marginal association between rs117028417 and AD diagnosis, FAM222A , but not the nearby gene TRPB4 , was found significantly associated with longitudinal increase of brain amyloid deposition. Along this line, as AD is a genetically complex and multifactorial disease with different etiological subtypes, FAM222A variants or pathogenic mutations strongly associated with AD may be present in subsets of AD patients. Nevertheless, although our genetic discovery study did not observe a strong influence of FAM222 variant on AD risk and biomarkers, the module enriched for FAM222A and previously reported AD risk variants likely represents a statistical AD-specific cluster worthy of further investigation using independent AD neuroimaging databases. Due to the limited sample size, only a slight, but not significant increase of Aggregatin mRNA level was observed in AD cortices collected in our laboratory (Supplementary Fig. 13a, b ). However, analysis of previously published microarray data of the Mount Sinai Brain Bank (MSBB) cohort 35 found significantly increased levels of Aggregatin mRNA in many brain regions especially cortices of AD, which were also tested to be associated with AD neuropathologies (Supplementary Fig. 13c, d and Supplementary Tables 8 , 9 ). DNA methylation is one of several epigenetic mechanisms regulating gene expression 36 and has been implicated in AD 37 . Interestingly, several methylation sites on FAM222A associated with AD could be identified (Supplementary Fig. 14 and Supplementary Table 10 ), indicating the likely involvement of FAM222A in AD pathogenesis through epigenetic regulation of its gene expression. Considering Aggregatin mRNA levels were usually measured at one time-point, future studies may be interesting to investigate longitudinal changes of Aggregatin gene expression and their relationship with neuropathologies during the progression of disease. While the relationship between FAM222A DNA methylation, transcription, translation, and posttranslational modification remains to be determined, these interesting findings provide further genetic evidence supporting the association of FAM222A to AD. Consistent with the genetic association of FAM222A with longitudinal brain Aβ deposition, pathologically accumulated Aggregatin, the protein encoded by FAM222A , is readily noted in plaques in AD and amyloid deposits in multiple APP transgenic mice, strongly illustrating the pathological function of Aggregatin. Of note, there are remarkable differences in the morphology of Aggregatin puncta and their co-localization with Aβ. Similarly, as plaques in AD patients are more complex structures than amyloid deposits in APP transgenic mice 38 , it could be expected that Aggregatin is also present differentially in amyloid core-enriched fractions from AD patients and 5XFAD mice. A number of explanations may account for the discrepancy regarding the pattern of Aggregatin puncta or presence of Aggregatin in plaques, including but not limited to differences in disease stages, the effects of Aβ clearance and degradation pathways or the length of time spent for plaque deposition. This notion is indeed supported by the observation that while only one or several condensed Aggregatin foci were present in single plaque in AD, amyloid deposits in cortex from patients with Down’s syndrome (DS), a complex genetic abnormality developing AD-like pathology, were largely associated with multiple foci (Supplementary Fig. 15 ). It is still unclear how Aggregatin becomes accumulated within the center of plaques without the ability for self-aggregation. Aggregatin appears to bind Aβ 1–40 and Aβ 1–42 with different affinities. Along this line, amyloid plaques are made up of different N or C-terminally truncated and modified Aβ species 39 . Interestingly, we found that Aggregatin was present in exosomes (Supplementary Fig. 16 ). Although Aggregatin has no signal sequence and is not predicted to be secreted, this data supports the possibility that Aggregatin can be exported into the interstitial fluid. Of note, the presence of exogenously expressed Aggregatin in exosomes of cultured cells is physiologic. There may be other mechanisms responsible for Aggregation secretion under pathological conditions. As Aggregatin protein levels were upregulated in AD, there may be a complex interplay among Aβ specific forms, Aggregatin expression, post-translational modification, extracellular secretion, and other unknown factors responsible for this. Nevertheless, on the basis of the facts that Aggregatin puncta appear concurrently with amyloid plaques and does not exist in the predepositing mice, Aggregatin should accumulate in plaques before or concurrent with rather than after the well formation of plaques. Aggregatin did not form intraneuronal accumulation in AD patients and 5XFAD mice. Not surprisingly, we did not observe the presence of Aggregatin puncta in neurons bearing neurofibrillary tangles (Supplementary Fig. 17a ). Along this line, intraneuronal APP and/or Aβ immunoreactivity assessed by 6E10 was not changed by Aggregatin in 5XFAD mice. Therefore, Aggregatin may not be involved in intraneuronal protein aggregation. Noteworthily, Aggregatin does not physically interact with tau and other previously reported plaque-associated proteins such as α-synuclein and APOE (Supplementary Fig. 17b–d ), further implicating the likely specific link between Aggregatin and Aβ. However, as AD is a multifactorial disease, further detailed investigation will still be needed to determine the spatiotemporal relationship between Aggregatin and other AD-related pathologies especially considering the presence of Aggregatin immunoreactivity outside of plaques. Aggregatin facilitates Aβ aggregation in vitro although it is not clear whether Aggregatin influences the primary or secondary nucleation. Increasing Aggregatin enhances, whereas reduced Aggregatin suppresses amyloid deposition and associated neuroinflammation and cognitive deficits. Of note, in addition to exacerbate Aβ pathology in adult 5XFAD mice, Aggregatin infusion causes further amyloid deposition in aged 5XFAD mice when amyloid deposit size and number largely plateau (Supplementary Fig. 18 ). Therefore, Aggregatin is likely an unrecognized co- or even limiting factor both necessary and sufficient for Aβ aggregating into the fibrils to form plaques. Although the bioinformatics analysis of Aggregatin amino acid sequence reveals that Aggregatin does not contain any known conserved functional motifs, our CD characterization of Aggregatin indicated it as at least a partially folded protein containing α-helix, β-sheet, and intrinsically disordered element(s) (Supplementary Fig. 7e ). While the structure and physiological function of Aggregatin is still under investigation, we found that Aggregatin was exclusively expressed in the CNS. The substantial loss of Aggregatin in hippocampus does not cause neuronal death, suggesting that Aggregatin may not be vital for neuronal survival. Future studies may be feasible to generate mice with global or neuronal specific deletion of Aggregatin to test whether the knockout of Aggregatin is sufficient to completely abolish amyloid deposition and further validate the pathological role of Aggregatin in amyloid plaque formation and disease progression. The genetic inhibition of Aggregatin-Aβ interaction was able to suppress Aggregatin-induced Aβ aggregation or amyloid deposits, suggesting that Aggregatin should directly interact with Aβ to regulate its pathology. Of note, although rNABD (i.e. Aggregatin 1–80 or Aggregatin Δ81–452) alone is able to bind Aβ, it does not induce Aβ 1–42 aggregation or promote amyloid deposits (Supplementary Fig. 19 ), suggesting that the C-terminal fragment is also required for Aggregatin-induced Aβ aggregation and plaque formation. The exact mechanism for Aggregatin-mediated Aβ aggregation is still under investigation. Noteworthily, likely due to the high Aβ binding affinity of Aggregatin, the specific anti-Aggregatin antibody used in this study does not dissociate the Aggregatin-Aβ interaction or prevent Aggregatin-induced aggregation of Aβ, and likewise, ICV infusion of the specific anti-Aggregatin antibody failed to alleviate Aβ pathologies in 5XFAD mice. Thus, the Aggregatin targeted immunotherapy for AD may require the generation of high-affinity monoclonal antibodies. The genetic manipulation or infusion of Aggregatin did not change APP or Aβ levels, suggesting that Aggregatin is unlikely involved in Aβ production or degradation. However, even though Aggregatin large puncta do not overlap with neurons, microglia or astrocytes, our results cannot rule out the possibility that Aggregatin may regulate amyloidosis indirectly through neuronal function or microglial or astrocytic Aggregatin, which is also worthy of further clarification. APOE4 is the strongest genetic risk factor for LOAD. Although the involvement of APOE in Aβ metabolism might complicate the interpretation of data 40 , future studies will also be interesting to investigate Aggregatin-mediated amyloidosis in vivo on the human ApoE knock-in or ApoE knockout background as previously reported 41 . In conclusion, we have reported FAM222A as a likely gene associated with AD-related regional brain atrophy, which encodes an amyloid plaque core protein pathologically involved in Aβ assembly and amyloid deposition. Our findings therefore not only inform future genetic studies of FAM222A , but also encourage detailed pathophysiological investigation of its encoded Aggregatin for AD and related dementia. Methods Samples, genotyping, and imputation Data used in the preparation of this article were obtained from the ADNI database ( ). The Illumina SNP genotyping data, demographic information, APOE genotype and baseline diagnosis information from 754 ADNI-1 participants, including 213 cognitive normal individual controls, 175 AD patients, and 366 patients with mild cognitive impairment (MCI) were downloaded from ADNI database. All participants provided written informed consent and study protocols were approved by participating sites’ Institutional Review Board. SNP genotyping of 620,901 markers on ADNI-1 participants were generated using Illumina BeadStudio 3.2 software from bead intensity data. All SNP genotypes are publicly available for download at the ADNI website. For genotype imputation analysis, only SNPs fulfilling the following criteria were included (1) per-SNP call rate ≥ 0.98; (2) minor allele frequency (MAF) ≥ 0.01; (3) P -value for Hardy-Weinberg equilibrium (HWE) ≥ 10 –6 in our sample set. Imputation was performed using the software MACH-ADMIX 42 using the 1000 Genomes Project Phase 3 V.5 ( ) as a reference panel. We excluded SNPs with R 2 < 0.3, MAF < 0.01 and all INDELs from the imputed genotype data to obtain genotypes for 7,512,167 SNPs for subsequent association analyses. MRI analysis and extraction of imaging phenotypes Dr. Christos Davatzikos’ group from University of Pennsylvanian analyzed the baseline MRI T1 scans of ADNI1 participants and generated the 145 ROIs spanning the entire brain by using the Multi-atlas region Segmentation (MUSE) framework 43 . In this framework, multiple atlases with semi-automatically extracted ground-truth ROI labels were first warped individually to the target image using non-linear registration methods 44 , 45 . To fuse the ensemble into a final segmentation, they adopted a spatial adaptive weighted voting strategy, in which a local similarity term was used for ranking and weighting ground truth labels from different atlases and an image intensity based term was used for modulating the segmentations at the boundaries of the ROIs according to the intensity profile of the subject image 43 . In validation experiments, the multi-atlas approach was showed to achieve significantly higher accuracy in comparison to single-atlas based segmentation 43 . In this study, we downloaded the volume measures of ROIs from ADNI. ROI-wise genome-wide association analysis in ADNI1 Autosomal chromosome SNP associations for volumes from 145 ROIs spanning whole brain were assessed by linear regression under the assumption of an additive genetic model. All models were adjusted for age, gender, education, handedness and 3 principal components to control population stratification. The genomic control for 145 GWASs ranged between 0.98 to1.02. Genetic correlation network analysis of brain ROIs in ADNI1 In multivariate quantitative genetics, a genetic correlation ( r g ) is the proportion of variance that two traits share due to additive genetic effects, which estimates the degree of pleiotropy or causal overlap 12 , 14 , 46 . The cross phenotype association analysis (CPASSOC) is a method proposed to integrate association evidence of multiple traits from multiple GWAS and detect cross-phenotype associations 13 . Thus, CPASSOC analysis of genetic correlated AD-related brain imaging traits could improve power to identify genetic variants associated with multiple AD-imaging traits. To identify groups of highly genetic correlated ROIs, we used the estimated pairwise ROI genetic correlations to define the brain genetic correlation network. In this network, nodes are brain ROIs while edges are estimated genetic correlations between ROIs. To extract modules from this network, we adopted a weighted gene co-expression network analysis (WGCNA) framework and used the method of topological overlap matrix (TOM) elements in hierarchical clustering to identify modular structures 47 . A flowchart for constructing a ROI genetic correlation network, extracting network modules and identifying genetic variants associated with modules using CPASSOC is presented in Supplementary Fig. 1a . Pairwise ROI genetic correlations were estimated by the technique of cross-trait LD score regression method 14 using the GWAS summary statistics of ROIs. For 10,400 pairs among 145 ROIs, genetic correlations were not correctly estimated for 3,255 pairs because the estimated values were either “NA”, above 1 or below −1, which might be driven by the small sample size, and these pairs were then filtered out. However, this filter may reduce power to identify variants associated with ROIs. The pairwise genetic correlations are presented in Supplementary Fig. 1 and we observed high genetic correlations among the ROIs. We used the ROI genetic correlation matrix and power adjacency function 47 to generate network adjacent matrix: $$a_{ij} = \left| {r_{gij}} \right|^\beta$$ (1) while r gij is the genetic correlation between nodes ROI i and ROI j , and a ij is the connection strength between two nodes. To choose the parameter β and genetic correlation P -value threshold, we used the scale-free network model to construct an image network. The scale-free network assumes that most nodes in a network are sparsely connected with the exception of a few hub nodes that are densely connected with other nodes 48 . In the scale-free network models, more connections are likely to occur for those hub nodes with already-high connectivity, which meet biological criteria 47 , 48 . We used the power law \(p(k) \sim k^{ - \gamma }\) to estimate the scale-free property, where k is the connectivity for each node and equals the number of its direct connections to other node. To generate the network, we assessed different power adjacency function parameter β = 2, 4, 6 and 8 and filtered the genetic correlation with different genetic correlation ( r g ) P -value thresholds of 0.5, 0.3, 0.2 and 0.1. For each P -value threshold, if the estimated genetic correlation P -value was larger than that, we set the genetic correlation to be 0. Using the four thresholds, we generated different networks for β = 2, 4, 6 and 8 and accessed their corresponding scale-free topology using linear regression model fitting index R 2 between log 10 ( p ( k )) and log 10 ( k ) for all nodes. We observed that a P -value threshold of 0.2 with β = 6 corresponded a network with the scale-free topology and had the largest R 2 of 0.61. The histogram of connectivity k and scale-free topology plots for networks with β = 6 and different P-value threshold were showed in Supplementary Fig. 2 . Thus, we used the network adjacent matrix generated under this criterion for further analysis. In this network, 40 out of 145 ROIs had k equal to 0 and 105 ROIs were carried out in module identification analysis. We adopted the methods introduced by WGCNA framework 47 to identify network modules. The adjacent matrix was transformed into a topological overlap matrix (TOM) with element defined as $$w_{ij} = \frac{{l_{ij} + a_{ij}}}{{\min \left\{ {k_i,k_j} \right\} + 1 - a_{ij}}}$$ (2) where \(l_{ij} = \mathop {\sum}\nolimits_u {a_{ij}a_{uj}}\) and \(k_i = \mathop {\sum}\nolimits_u {a_{iu}}\) is the node connectivity. TOM based dissimilarity measure was generated by $$d_{ij}^w = 1 - w_{ij}$$ (3) This dissimilarity matrix was used as the input for average linkage hierarchical clustering. The hierarchical clustering grouped the closet ROIs and formed the branches to identify module. For the genetic correlation network, we identified 16 modules spanning the whole brain with the largest module containing 17 ROIs and the smallest containing 3 ROIs (Supplementary Fig. 2 ). CPASSOC analysis within modules We applied the CPASSOC package developed by Zhu et al. 13 to combine association evidence of ROIs within each module. CPASSOC can integrate association evidence from summary statistics of multiple traits and improves power when variant is associated with at least one trait. CPASSOC provides two statistics, S Hom and S Het . S Hom is similar to the fixed effect meta-analysis method 49 but accounting for the correlation of summary statistics among cohorts induced by potential overlapped or related samples. In brief, assuming we have summary statistical results of GWAS from J cohorts with K phenotypic traits. In each cohort, single SNP-trait association was analyzed for each trait separately. Let T jk be a summary statistic for a SNP, j th cohort and k th trait. Let \({{{\mathbf{T}}}} = (T_{11}, \cdots ,T_{J1}, \cdots ,T_{1K}, \cdots ,T_{JK})^T\) represents a vector of test statistics for testing the association of a SNP with K traits. Let \({{{\mathbf{\beta }}}} = \left( {\beta _{11,...,}\beta _{J1,...,}\beta _{1K,...,}\beta _{JK}} \right)^T\) be the effect sizes of the SNP. The null hypothesis is H 0 : β = 0 and the alternative hypothesis H 1 is that at least one of the elements of β is not equal to zero. We used a Wald test statistic \(T_{jk} = \widehat \beta _{jk}/\widehat s_{jk}\) , where \(\hat \beta _{jk}\) and \(\hat s_{jk}\) are the estimated coefficient and corresponding standard error for the k th trait in the j th cohort. It is reasonable to assume that T follows a multivariate normal distribution with mean 0 and correlation matrix R under the null hypothesis. When the effect is homogeneous, the most powerful test statistic S Hom is defined as $$S_{{{{\mathrm{Hom}}}}} = \frac{{e^T({{{\mathrm{RW}}}})^{ - 1}T(e^T({{{\mathrm{RW}}}})^{ - 1}T)^T}}{{e^T({{{\mathrm{WRW}}}})^{ - 1}e}}$$ (4) which follows a χ 2 distribution with one degree of freedom, where \({{{\mathbf{e}}}}^{{{\mathbf{T}}}} = (1,...,1)\) has length J × K and W is a diagonal matrix of weights for the individual test statistics. We used the sample sizes for the weights, \(w_{jk} = \sqrt {n_j}\) , n j is sample size of the j th cohort. To further allow for different effect directions of a variant for different traits in different cohorts, we define S Het . We first let $$S(\tau ) = \frac{{e^T\left( {R(\tau )W(\tau )} \right)^{ - 1}T(\tau )\left( {R(\tau )W(\tau )} \right)^{ - 1}\left. {T(\tau )} \right)^T}}{{e^TW(\tau )^{ - 1}R(\tau )^{ - 1}W(\tau )^{ - 1}e}}$$ (5) Where T ( τ ) is the sub-vector of T satisfying \(\left| {T_{jk}} \right| > \tau\) for a given \(\tau > 0\) , \(R\left( \tau \right)\) is a sub-matrix of R representing the correlation matrix, and \(W\left( \tau \right)\) be the diagonal submatrix of W , corresponding to \({{{\mathbf{T}}}}\left( {{{\mathbf{\tau }}}} \right)\) . The test statistic is then $$S{}_{Het} = \mathop{\max}\limits_{\tau > 0}S(\tau )$$ (6) The asymptotic distribution of S Het does not follow a standard distribution but can be evaluated using simulation. S Het is an extension of S Hom but power can be improved when the genetic effect sizes vary for different traits. The distribution of S Het under the null hypothesis can be obtained through simulations or approximated by an estimated beta distribution. We applied both S Hom and S Het to combine summary statistics for ROIs within each module. The CPASSOC analysis of multiple genetic correlated traits in identified module would allow us to identify variants that are likely to be missed by conventional GWAS of single trait and reduce the multiple comparison burden in the genetic analysis of hundreds of neuroimaging traits. Finally, we identified 15 loci with CPASSOC test P -value less than 1 × 10 –7 in nine modules (Supplementary Table 1 ). Importantly, three previously reported AD-associated SNPs, rs429358, rs2075650 and rs439401 and the FAM222A SNP rs117028417 were exclusively found in one module, which were green colored in Supplementary Fig. 1M and Supplementary Table 1 . The Manhattan plots and Q-Q plots of CPASSOC analysis and single ROI GWAS for this module were showed in Fig. 1 and Supplementary Fig. 2 . Genetic analysis of AV-45 PET imaging 18 F-Florbetapir (AV-45) PET imaging was performed at baseline and 2-year follow-up for participants enrolled in the ADNI GO and two phases 22 . UC Berkeley extracted weighted AV-45 standardized uptake value ratio (SUVR) means for four main cortical regions: frontal, anterior, and posterior cingulate, lateral parietal and lateral temporal regions (version 2019.4.12) for ADNI-GO2 participants. They also calculated composite SUVR for cortical which is weighted SUVR mean in frontal, cingulate, parietal and temporal regions. These data can be downloaded from the ADNI database. We used the SUVR mean of composite region including whole cerebellum, pons/brainstem and eroded white matter as reference. Mean AV-45 SUVR of frontal, cingulate, lateral parietal, lateral temporal and composite cortical relative to the reference were calculated. The annual percent change in SUVR means at 2-year follow-up compared to baseline was used as the main quantitative phenotype for genetic analysis. The annual percent changes in AV-45 SUVR for all five brain regions were approximately normally distributed (Supplementary Fig. 20 ). We collected 369 individuals with both SUVR measures for baseline and 2-year follow-up and whole-genome sequencing data. The samples included 120 healthy people, 26 people with AD, 64 people with late mild cognitive impairment (LMCI) and 159 people with early mild cognitive impairment (EMCI) diagnosed at baseline. The samples characteristics and demographics for samples are shown in Supplementary Table 11 . WGS data from 817 ADNI participants were downloaded from the ADNI dataset. WGS was performed using blood-derived genomic DNA samples and sequenced on the Illumina HiSeq2000 using paired-end read chemistry and read lengths of 100 bp at 30–40X coverage 50 . As previously described using Broad GATK and BWA-mem, reads were mapped and aligned to the human genome (build 37), then variants were called 50 , 51 . For single SNP association test, association test of SNP rs117028417 with phenotypes were performed using linear regression under an additive genetic model in PLINK. Baseline age and gender were included as covariates. For gene-based association test, we extracted 8 and 6 functional coding variants defined as missense, in frame deletion/insertion, stop gained/lost, start gained/lost, splice acceptor/donor, or initiator/start codon for FAM222A and TRPV4 respectively. All of those variants are rare with minor allele frequency (MAF) < 0.01 in ADNI samples. Gene-based association tests were performed using burden and SKAT 52 , adjusting age and sex as covariates. Genetic analysis of CSF Aβ and Tau Collection and processing of ADNI CSF samples was described in the ADNI procedures manual ( ). We downloaded UPENNBIOMKs dataset.csv file from ADNI website. We collected 617 individuals with both CSF Aβ 42 , tTau and pTau at baseline level and WGS data. For baseline data, since raw CSF biomarkers were skewed or bimodal skewed distributed, rank normal transformations were conducted for each biomarker separately (Supplementary Fig. 21a–f ). To conduct CSF biomarkers longitudinal change genetic association, we collected 274 individuals with both baseline and 24-month follow-up CSF biomarkers and WGS data. The CSF biomarkers raw data at baseline and 2-year follow-up in 218 individuals were used to calculate annual changes in Aβ 42 , tTau and pTau separately. The annual changes of three CSF biomarkers were approximately normally distributed (Supplementary Fig. 21g–i ). The samples characteristics and demographics for CSF biomarker traits association analysis are shown in Supplementary Tables 12 , 13 . Association test of SNP rs117028417 with phenotypes were performed using linear regression under an additive genetic model in PLINK. Baseline age and sex were included as covariates. We extracted 8 and 15 coding variants defined as missense, in frame deletion/insertion, stop gained/lost, start gained/lost, splice acceptor/donor, or initiator/start codon for FAM222A and TRPV4 respectively (Supplementary Table 4 ). All of those variants are rare with minor allele frequency (MAF) < 0.01 in ADNI samples. Gene-based association tests were performed using burden and SKAT 52 , adjusting age and sex as covariates. Analysis of FAM222A mRNA in AD The development of the Mount Sinai Brain Bank (MSBB) cohort was described in the previous studies 35 , 53 . MSBB is a large AD cohort and now holds over 2,040 well-characterized human brains 53 . The datasets we used assessed a total of 125 human brains which was assembled after applying stringent inclusion/exclusion criteria and represents the full spectrum of cognitive and neuropathological disease severity 35 . Detailed sample demographic information and description of the cognitive and neuropathological traits can be seen in previously published paper by Dr. Bin Zhang’s lab 35 . We downloaded the normalized microarray data of MSBB Array Tissue Panel Study from the Synapse at . The RNA samples from 19 brain regions isolated from 125 MSBB specimens were collected and profiled on the Affymetrix 133AB and Affymetrix 133Plus2 platforms. RNA quality was assessed using a combination of a 260/280 ratio derived from resolution electrophoresis system (LabChipTM, Agilent Technologies, Palo Alto, CA, USA) and 3′–5′ hybridization ratios for GAPDH probes 35 . Not all brain regions for all subjects were available for analysis. There was an approximately 60 samples (40 AD, 20 controls) per brain region available for analysis. The array probes were annotated according to the Ensemble version 72 (genome build GRCh37) using the R/Biomart library. The raw microarray data were quantile normalized with all probe sets on the arrays using RMA 54 method implemented in the R/Bioconductor package affy (v1.44) with the default parameters. The data were then corrected for covariates including sex, postmortem interval (PMI), pH and race using a linear regression model. The FAM222A gene expression data was identified by probe set 226487_at . The processed FAM222A mRNA level means for groups of AD and control were compared using two-sided Welch t-test using R. Association analysis of FAM222A DNA methylation We downloaded two datasets, E-GEOD-45775 and E-GEOD-76105, with DNA methylation profiling from the European Bioinformatics Institute (EMBL-EBI) ArrayExpress website . Samples of dataset E-GEOD-45775 included 5 controls, 5 AD Braak stage I-II and 5 AD Braak stage V-VI (Supplementary Table 14 ). The methylation values were adjusted and normalized using BeadStudio software v3.2 to obtain normalized beta and average Beta detect P-value. The array used the HumanMethylation27_270596_v.1.2 design and one methylation site cg01335367 was identified located on chr12:109734355–109734404 ( GRCh38.p12 ), associated with FAM222A . We analyzed the association between methylation in cg01335367 with AD using logistic regression and adjusted for sex. We also performed one-way analysis of variance (ANOVA) to determine differences between methylation levels of control and different Alzheimer Braak stage groups. Study E-GEOD-70615 investigated DNA methylation profiling in the superior temporal gyrus (STG). Samples included 34 AD and 34 non-demented controls, which had 52 European, 8 Hispanic, 6 African, 1 Asian Americans and 1 unknown (Supplementary Table 15 ). The Beta values from the probes were quantile normalized using lumi package in R. We performed association analysis in 52 European Americans only. The association between methylation in those sites with AD were analyzed using logistic regression model adjusting age, gender and estimated cellular proportions (neuronal vs. glial). Mice and human tissues Mouse surgery and procedures were performed according to the NIH guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) at Case Western Reserve University. 5xFAD transgenic mice (B6.Cg-Tg(APPSwFlLon, PSEN1*M146L*L286V) 6799Vas/Mmjax, JAX#008730) were purchased from the Jackson Laboratory. The use of all human tissue samples was approved by the University Hospitals Institutional Review Board (IRB) for human investigation at University Hospitals Case Medical Center at Cleveland. Human brain tissues obtained postmortemly from University Hospitals of Cleveland were fixed, and 6-μm-thick consecutive sections were prepared. The information of fixed or frozen human tissues were listed in Supplementary Tables 16 , 17 . Immunocytochemistry, immunofluorescence and immunoblot Immunocytochemistry was performed by the peroxidase anti-peroxidase protocol. Taken briefly, paraffin embedded brain tissue sections were first deparaffinized in xylene and rehydration in graded ethanol and incubated in Tris Buffered Saline (TBS, 50 mM Tris-HCl and 150 mM NaCl, pH 7.6) for 10 min before antigen retrieval in 1X Immuno/DNA retriever with citrate (BioSB, Santa Barbara, CA) under pressure using BioSB’s TintoRetriever pressure cooker. Sections were rinsed with distilled H 2 O, and blocked with 10% normal goat serum (NGS) in TBS at room temperature (RT) for 30 min. Tissue sections were further incubated with primary antibodies in TBS containing 1% NGS overnight at 4 °C, and immunostained by the peroxidase-antiperoxidase based method. For double Immunofluorescence staining, paraffin embedded tissue sections were deparaffinized in xylene and re-hydrated in graded ethanol without H 2 O 2 incubation as described above. The sections were incubated in phosphate buffered saline (PBS) at RT for 10 min followed by block with 10% NGS in PBS for 45 min at RT. The sections were incubated with primary antibodies in PBS containing 1% NGS overnight at 4 °C. After being washed with 1% NGS in PBS for 10 min, the sections were incubated in 10% NGS for 10 min and followed by three quick washes with 1% NGS in PBS. Then, the sections were incubated with Alexa Fluor 488 or 568 dye labeled secondary antibodies (1:300, Invitrogen, Carlsbad, CA) for 2 h at RT in dark, washed three times with PBS, stained with DAPI, washed again with PBS for three times, and finally mounted with Fluoromount-G mounting medium (Southern Biotech, Birmingham, AL). For thioflavin-S staining, slides were incubated with 1% thioflavin-S (Santa Cruz Biotechnology, Dallas, TX) for 8 min, washed 2 times with 80% ethanol, and 1 time with 95% ethanol and PBS, then stained with DAPI. For immunoblot, human or mice tissue samples were all lysed with TBS plus 1 mM phenylmethylsulfonyl fluoride (PMSF) (Millipore, Burlington, MA), protease inhibitor cocktail (Sigma Aldrich, St. Louis, MO) and phosphatase inhibitor cocktail (Sigma Aldrich, St. Louis, MO). Equal amounts of total protein extract were resolved by SDS-PAGE and transferred to Immobilon-P (Millipore, Burlington, MA). Following blocking with 10% nonfat dry milk, primary and secondary antibodies were applied and the blots developed with Immobilon Western Chemiluminescent HRP Substrate (Millipore, Burlington, MA). Images were taken by ChemiDoc Touch Imager (Bio-rad, Hercules, CA). Primary antibodies used in this study are listed in Supplementary Table 18 . The dilution of antibodies used for IF or IHC. 4G8 (BioLegend, SIG-39220; IF, 1:1000), 6E10 (BioLegend, 803001; IF and IHC, 1:1000), 82E1 (IBL, 10323; IF, 1:1000), Aggregatin (Abcam, ab122626; IF/IHC, 1:100), Aggregatin (LifeSpan BioSciences, LS-C170630; IHC, 1:1000), Aggregatin (Aviva Systems Biology, ARP69038_P050; IHC, 1:1000), Flag (Sigma Aldrich, F1804; IF/IHC, 1:1000), Flag (Thermo Fisher, PA1–984B; IHC, 1:200), Flag (Cell Signaling Technology, 2368; IHC, 1:200), Flag-HRP (Proteintech, HRP-66008; IHC, 1:1000), GFP (Abcam, ab32146; IHC, 1:500), Myc (Thermo Fisher, MA1–21316; IHC, 1:1000), Myc (Cell Signaling Technology, 2276; IHC, 1:500), and Nu4 (Klein lab, IF/IHC, 1:2000). All uncropped and unprocessed blots are provided in the Source Data file (Source Data for Statistics and Blots). Expression vectors and recombinant proteins pcDNA3.1(+) (Invitrogen, Carlsbad, CA) plasmid was modified to express recombinant proteins to express recombinant proteins containing a 4xFlag-Twin-Strep-tag at their N-terminal. The cDNA of full length or truncated human Aggregatin were inserted into the modified pcDNA3.1(+) plasmid. All primers and cDNA constructs used in this study are listed in Supplementary Data 1 and Supplementary Table 19 . Eight micrograms plasmid was used to transfect one 10 cm dish of Lenti-293T cells with TransIT ® −293 Transfection Reagent (Mirus, Madison, WI). Cells were collected at 24 h after transfection and lysed by lysis buffer (100 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA and 1% NP40, pH 8.0) containing 1 mM PMSF (Millipore, Burlington, MA), protease inhibitor cocktail (Sigma Aldrich, St. Louis, MO) and phosphatase inhibitor cocktail (Sigma Aldrich, St. Louis, MO). The lysate was centrifuged at 14,000 g for 15 min at 4 °C. Supernatant was incubated with MagStrep type3 XT beads (IBA Lifesciences, Goettingen, Germany) overnight at 4 °C. Beads were washed three times with lysis buffer, and eluted with BXT buffer (IBA Lifesciences, Goettingen, Germany) overnight at 4 °C. At last, the eluted recombinant proteins were subjected to dialysis using 10 kD Slide-A-Lyzer™ Dialysis Cassettes (Thermo Fisher Scientific, Waltham, MA), concentration with 10 kD Spin Column (Abcam, Cambridge, MA) and purification by size-exclusion chromatography. Stereotaxic injection and ICV infusion Mice surgery were performed according to the NIH guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) at Case Western Reserve University. All AAVs with 10 13 genome copies per mL (GC per mL) were obtained from Vigene Biosciences (Jinan, China). For stereotaxic injection, mice were anesthetized with isoflurane and immobilized using the stereotactic frame equipped with a heating blanket to maintain body temperature throughout the procedure. After hair removal and the cleaning of the shaved area with betadine and alcohol, mice were injected with bupivacaine/lidocaine and a small incision was made to expose the skull surface. Two small holes were drilled in the skull (relative to bregma: anteroposterior −2.1 mm, medial lateral ±2 mm; Note that ± is the plus-minus sign throughout this study) followed by injection of 2 μl AAVs using Hamilton syringes into the hippocampal CA1 at dorsal ventral −1.45 mm. Injection speed was pump controlled at 0.2 μl per min. The needle was left in place for 5 min before it was slowly withdrawn. For ICV infusion, the mini-osmotic pump (Model 1004, Alzet, Cupertino, CA; flow rate of 0.11 μl per hour, 28 days) and brain infusion cannula attached with 2.5–3 cm catheter tubes (Brain infusion kit 3, Alzet, Cupertino, CA) were filled with recombinant protein in artificial cerebrospinal fluid (aCSF), followed by pump incubation in aCSF at 37 °C for 48 h according to the manufacturer’s instructions. For implant surgery, a hole was drilled in the skull (relative to bregma: anteroposterior −0.5 mm, medial lateral 0.75 mm). The cannula was positioned on the skull with the needle plug 2.5 mm into the ventricle. The cannula was fixed and secured by cyanoacrylate glue. Behavioral tests Mice behavioral tests were also performed according to the NIH guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) at Case Western Reserve University. The Barnes maze consisted of a white acrylic circular disk 92 cm in diameter with 20 equally spaced holes (5 cm in diameter) located 2 cm from the edge of the disk. The maze was illuminated by two 60 W lamps to provide an aversive, bright disk surface. An acrylic escape box (7 × 7 × 5 cm) could be fitted under any of the holes in the maze. The maze was raised 30 cm from the floor and rested on a pedestal that enabled it to be rotated 360° on a horizontal plane. An acrylic start bin with 15 cm diameter and 15 cm height was used. Trials were recorded using a webcam and analyzed by video tracking software (EthoVision XT, Noldus, Leesburg, VA). Each trial began with the start bin positioned in the center of the maze with the mouse placed inside. The mouse remained in the start bin for 30 s, providing a standard starting context for each trial and ensuring that initial orientation of the mouse in the maze varied randomly from trial to trial. Each mouse was allowed to explore the maze freely for 2 min. After the mouse entered the escape hole, the mouse was left in the escape box for 90 s before being returned to its home cage. If the mouse did not enter the escape box within 120 s, it was gently picked up by the experimenter and placed over the target hole and allowed to enter the escape box. After each trial, the maze and escape box were cleaned carefully with a 10% alcohol solution to dissipate odor cues and provide a standard olfactory context. Five training sessions consisting of two trials each were run on subsequent days and escape latencies were measured. For Y maze test, mice were placed in a Plexiglas Y maze (with arms 60 cm in length) and allowed to explore the maze freely for 10 min. When put in the Y maze, the mice were recorded using the ANY-maze tracking system, and the time and frequency in the spontaneous alteration ratio were counted automatically. All tests were performed at the Case Behavior Core, with the investigator blinded to mouse genotype. Plaque isolation Amyloid plaque cores were isolated as previously described 55 . Briefly, whole mouse brain or human brain gray matters were homogenized, boiled in lysis buffer (2% SDS, 50 mM Tris-HCl pH 7.5, 50 mM DTT), and centrifuged at 100,000 g for 1 h at 10 °C. The pellet was solubilized in fraction buffer (1% SDS, 50 mM Tris-HCl pH 7.5, 50 mM DTT) and centrifuged at 100,000 g for 1 h at 10 °C. The pellet was further suspended in fraction buffer and loaded on top of a discontinuous sucrose gradient (1.0, 1.2, 1.4 and 2.0 M sucrose in 50 mM Tris pH 7.5 containing 1% SDS), centrifuged at 220,000 g for 20 h at 10 °C and fractionated into sixteen fractions (300 µl per fraction). Plaque-core-enriched fraction #13 were further diluted in fraction buffer and centrifuged at 220,000 g for 2 h at 10 °C. The resulting pellet was dissolved in 70% formic acid and subsequently dried using a SpeedVac system. Solubilized proteins were further resuspended in 1X SDS sample buffer with 8 M Urea. Aβ preparation, pull-down, and co-sedimentation assay Synthetic human Aβ 1–42 and Aβ 1–40 peptides (GL Biochem, Shanghai) were dissolved in hydroxyl-fluro-isopro-panol (HFIP) and subsequently dried using a SpeedVac system. Both Aβ 1–42 and Aβ 1–40 monomers were prepared by dissolving the lyophilized Aβ in dimethyl sulfoxide (DMSO) at 5 mM, sonicated for 10 min and diluted in PBS buffer (NaCl 137 mM, KCl 2.7 mM, Na 2 HPO 4 10 mM, KH 2 PO 4 1.8 mM, pH 7.4) to different concentrations. Aβ 1–42 oligomers were prepared in DMSO/PBS and oligomerized by incubation at 4 °C for 24 or 48 h. Monomeric or oligomer Aβ 1–40 (100 μM) and Aβ 1–42 solutions (50 μM) supplemented with or without rAggregatin bound Strevdin-avdin beads were incubated in IP buffer (NaCl 300 mM, KCl 2.7 mM, Na2HPO4 10 mM, KH2PO4 1.8 mM, pH7.4) at RT with shaking for 2 h. After 4 times wash with IP buffer, beads were eluted by 1XSDS sample buffer (32.9 mM Tris-HCl pH6.8, 13% Glycerol, 1% SDS and 0.005 % bromophenol blue) and analyzed by 10–20% SDS/Tricine protein gels (Invitrogen, Carlsbad, CA). For Aβ 1–42 oligomer formation and co-sedimentation assay, HFIP dissolved synthetic Aβ 1–42 peptides were solubilized in 30 mM NaOH to a final concentration of 100 μM, diluted to 2.5 μM in PBS and incubated with and without 30 nM rAggregatin at 37 °C for different time points. After 10-minute centrifuge at 14,000 g, pellets and supernatants were collected and analyzed by 10–20% SDS/Tricine protein gels (Invitrogen, Carlsbad, CA). Dynamic light scattering Dynamic light scattering (DLS) experiments were carried out with DynaPro™ instrument from Wyatt technology with a wavelength of 633 nm and a scattering angle of 173°. The measurements of Aggregatin or Aggregatin Δ61–80 at 100 nM were performed at 25 °C after 2 min equilibration with correlation times defined on 10 s per run with 30 runs for each measurement. The results were plotted as intensity of distribution (%) of particles versus hydrodynamic radius (nm). Circular dichroisms The spectra were recorded over a wavelength range of 260–190 nm with standard sensitivity at the 50 nm per min scan speed with 1‐nm resolution and 1‐s time constant at room temperature using a spectropolarimeter (Jasco J-815). All the proteins were dissolved in phosphate buffer (pH8.0). The final concentration of all samples was 1 µM. The secondary structure content was calculated from the Circular dichroisms (CD) spectra using the online software K2D3. Surface plasmon resonance Surface plasmon resonance (SPR) was determined using BIAcore3000 (GE Healthcare Life Sciences, Pittsburgh, PA). rAggregatin (0.1 mg per ml) was immobilized on the CM5 sensor surface (GE Healthcare Life Sciences, Pittsburgh, PA) in 10 mM acetate buffer (pH = 4.5). Running buffer was 1% DMSO in PBS-P buffer (0.02 M phosphate, 2.7 mM KCl, 137 mM NaCl and 0.05% Tween 20). Binding of a dilution series comprising Aβ 1–42 monomers to rAggregatin was analyzed and fitted to the 1:1 binding model using BIAevaluation software (GE Healthcare Life Sciences, Pittsburgh, PA). Solid phase binding assay rAggregatin was coated onto Nunc MaxiSorp 96-well plates (Thermo Fisher Scientific, Waltham, MA) at 0.1 μg per well in PBS at 4 °C overnight. After blocking in 1% BSA in PBS for 2 h at RT, Aβ 1–42 at 6.25, 12.5, 25, 50, 100, or 200 nM or Aβ 1–40 at 0.5, 1, 2, 4, or 8, or 16 μM monomers were added to the plates at 4 °C overnight. Plates were washed with PBS 4 times and incubated with 6E10 antibody at 4 °C overnight, followed by 4 times PBS wash and development in TMB solution (Thermo Fisher Scientific, Waltham, MA). The reaction was stopped by sulfuric acid and assessed using a Synergy H1 microplate reader (BioTek, Winooski, VT). Likewise, 0.2 μg Aβ 1–42 or Aβ 1–40 monomers were immobilized on plates and incubated with 3.125, 6.25, 12.5, 25, 50, or 100 nM rAggregatin. Bound rAggregatin were detected by an anti-Flag antibody and developed in TMB solution as described above. ThT fluorescence assay HFIP treated Aβ 1–40 or Aβ 1–42 peptides were solubilized in 30 mM NaOH to a final concentration of 400 μM, sonicated for 5 min in a water bath and stored at −80 °C until further use. To monitor Aβ 1–40 and Aβ 1–42 fibrillization, a ThT assay was performed according previous studies 56 , 57 . Briefly, a stock solution of Aβ was diluted to in PBS with 20 μM ThT. Then rAggregatin were added at desired concentrations to the final volume of 100 µl. All samples were transferred to a black 96-well nonbinding Surface microplate with clear bottom (Corning, Corning, NY), and sealed with a polyester-based sealing film (Corning, Corning, NY). Samples were incubated at 37 °C with stirring. Real-time ThT fluorescence was measured every 5 min for at least 12 h at the excitation and emission wavelengths of 446 nm and 482 nm respectively by a Synergy H1 microplate reader (BioTek, Winooski, VT). Aβ 1–42 aggregates stained by Thio-S To evaluate Aβ aggregates formed in vitro, rAggregatin (30 nM) and 2.5 μM Aβ in PBS were incubated at 37 °C for 4 weeks. 20 μl of protein solution were applied to the glass slides and completely air dry for 30 min. After washing with PBS, the samples were stained by 1% Thio-S for 10 min. The 3D confocal images were analyzed by using Imaris (Bitplane, Concord, MA) and the structure surface were extracted by using the SURFACE tools following the manufacturer’s instructions. Negative electric microscopy HFIP dissolved synthetic Aβ 1–42 peptides were solubilized in 30 mM NaOH to a final concentration of 100 μM. Then diluted to 2.5 μM in PBS and incubated with and without 30 nM rAggregatin at 37 °C. Immediately following the indicated incubation time, 20 μl of protein solution were applied to the support surface of the grids, which were autoclaved by UV irradiation overnight. The grids were washed with 20 μl droplets of water 4 times, followed by a 20 μL droplet of uranyl acetate solution, then examined in an FEI Tecnai Spirit (T12) with a Gatan US4000 4kx4k CCD. Total Aβ measurement by ELISA Brains were homogenized in TBS Buffer (50 mM Tris-HCl and 150 mM NaCl, pH 7.6) containing 1 mM PMSF (Millipore, Burlington, MA), protease inhibitor cocktail (Sigma Aldrich, St. Louis, MO) and phosphatase inhibitor cocktail (Sigma Aldrich, St. Louis, MO). Total protein concentrations were determined using the BCA kit (Thermo Fisher Scientific, Waltham, MA). ELISA of total Aβ was carried out in 96-well high-binding microtiter plates. Monoclonal antibody 6E10 raised against residues Aβ1–16 was used as a capture antibody (diluted in PBS pH 7.4) and incubated over night at 4 °Cin a humid chamber. After removal of the capture antibody, the plate surface was blocking with 1% BSA for 1.5 h. After washing with PBS, 0.5 µg total protein were added and incubated at 4 °Covernight. Monoclonal antibody MOAB-2 coupled to horseradish peroxidase diluted in PBS were used as secondary antibodies and again incubated over night at 4 °C. After three times washing with PBS, 100 μl of TMB ELISA peroxidase substrate (Thermo Fisher Scientific, Waltham, MA) was added and incubated for 1–10 min at RT in darkness. The reaction was stopped with 100 μl 2 M H 2 SO 4 and absorbance was measured in a microplate reader at 450 nm. For generation of standard curves, synthetic Aβ1–42 peptides freshly dissolved in DMSO from 1 ng per µL to 10 pg per µL. Isolation of exosomes Lenti-293T cells were transfected with empty vector or pCDNA-4xFlag-Aggregatin using TransIT®−293 Transfection Reagent (Mirus, Madison, WI). Twenty-four hours after transfection, cells were cultured in the DMEM medium supplemented with exosome-free FBS. Forty-eight hours later, the cell culture medium was collected and centrifuged at 300 g for 15 min to remove cells and debris. The supernatant was further filtered through a 0.22 μm filter and centrifuged at 100,000 g for 2 h at 4 °C. The pellets enriched with exosomes were resuspended in the lysis buffer (100 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA and 1% NP40, pH 8.0) containing 1 mM PMSF (Millipore, Burlington, MA), protease inhibitor cocktail (Sigma Aldrich, St. Louis, MO), and phosphatase inhibitor cocktail (Sigma Aldrich, St. Louis, MO) followed by immunoblot analysis. Confocal microscopy and image analysis All fluorescence images were imaged on a Leica TCS SP8 gSTED confocal microscopy (Leica Microsystems, Buffalo Grove, IL) equipped with a motorized super Z galvo stage, two PMTs, three Hyd SP GaAsP detectors for gated imaging, and the AOBS system lasers including a 405 nm, Argon (458, 476, 488, 496, 514 nm), a tunable white light (470 to 670 nm), and a 592 nm STED depletion laser. Series of confocal images with optical thickness of 300 nm were collected using the ×100 oil objective. All 3D confocal images of plaque were reconstructed using Imaris (Bitplane, Concord, MA) after background subtraction. Quantification of Aggregatin foci in plaques and measurement of plaque load and size were performed with open-source image analysis programs WCIF ImageJ (developed by W. Rasband). Statistical analysis Statistical analysis was done with one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test or student-t-test using GraphPad Prism (GraphPad, CA). Data are means ± SEM. p < 0.05 was considered to be statistically significant. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The data that support the findings of this study are available on request from the corresponding author X.L.W. Case Western Reserve University supports the NIH Guidelines for the Sharing of Research Resources including “the Sharing of Biomedical Research Resources: Principles and Guidelines for Recipients of NIH Grants and Contracts”. If any intellectual property is pursued, the data will be shared and distributed following advice from the authorities of Case Western Reserve University. The source data underlying Figs. 1 , 2 g, 2 h, 3 a–d, 3 f, 3 g, 4 a–f, 5 c–f, 5 h–k, 5m–p , and Supplementary Figs. 3a –e, 4 , 5c –h, 7d , 7e , 7i –k, 8b , 8d , 9a –d, 10b , 10c , 10e , 11b –f, 12a , 12c –e, 13a –d, 14 , 15e –h, 16 , 17b –d, 18b , 18c , 19a , 19c , and 19e are provided as a Source Data file. Source Data for GWAS in Fig. 1 : 10.6084/m9.figshare.11336657 ( ). Source Data for GWAS in Supplementary Fig. 2a : 10.6084/m9.figshare.11336666 ( ). Source Data for GWAS in Supplementary Fig. 2b : 10.6084/m9.figshare.11337272 ( ). Source Data for GWAS in Supplementary Fig. 2c : 10.6084/m9.figshare.11337281 ( ). Source Data for GWAS in Supplementary Fig. 2d : 10.6084/m9.figshare.11337284 ( ). Source Data for GWAS in Supplementary Fig. 2e : 10.6084/m9.figshare.11337290 ( ). Change history 11 July 2022 A Correction to this paper has been published:

Researchers at the Case Western University School of Medicine say they have identified a previously unknown gene and associated protein which could potentially be suppressed to slow the advance of Alzheimer's disease. "Based on the data we have, this protein can be an unrecognized new risk factor for Alzheimer's disease (AD)," said Xinglong Wang, an associate professor of pathology at the School of Medicine. "We also see this as a potential novel therapeutic target for this devastating disease." Wang said proving the latter assertion, which has not yet been tested in humans, would require additional research to corroborate the function of the protein they have dubbed "aggregatin." Eventually, that would someday mean clinical trials with Alzheimer's patients, he said. "This protein characteristically accumulates, or aggregates, within the center of plaque in AD patients, like the yolk of an egg—which is part of the reason we named it "aggregatin," Wang said. A research team led by Wang and Xiaofeng Zhu, a professor of Population and Quantitative Health Sciences at the School of Medicine, has filed for a patent through the university's Office of Research and Technology Management for "novel Alzheimer's disease treatments and diagnosis based on this and related study," Wang said. "We're very excited about this because our study is likely the first systematic work combining the identification from a genome-wide association study of high dimensional brain-imaging data and experimental validation so perfectly in Alzheimer's disease," Zhu said. Their research was published this month by the scientific journal Nature Communications and supported by grants from the National Institutes of Health (NIH) and the Alzheimer's Association. Genomic and brain imaging data was obtained from the Alzheimer's Disease Neuroimaging Initiative, which is supported by the NIH. Alzheimer's Disease affects millions More than 5.7 million Americans have Alzheimer's disease, which is the primary cause of dementia and sixth-leading cause of death in the United States. That population is predicted to reach 14 million by the year 2050, according to the Alzheimer's Association. The relationship between Alzheimer's (and subsequent brain atrophy) and amyloid plaques—the hard accumulations of beta amyloid proteins that clump together between the nerve cells (neurons) in the brains of Alzheimer's patients—has been well-established among researchers. Less understood is precisely how that amyloid-beta actually leads to plaque formation—and where this new work appears to have broken new ground, Wang said. Further, while there has been much research into what genes might influence whether or not someone gets Alzheimer's, there is less understanding of genes that might be linked to the progression of the disease, meaning the formation of plaque and subsequent atrophy in the brain. The role of 'aggregatin' protein In the new work, the researchers began by correlating roughly a million genetic markers (called single-nucleotide polymorphisms, or SNPs) with brain images. They were able to identify a specific SNP in the FAM222, a gene linked to different patterns of regional brain atrophy. Further experiments then suggested that the protein encoded by gene FAM222A is not only associated with AD patient-related beta-amyloid plaques and regional brain atrophy, but that "aggregatin" attaches to amyloid beta peptide—the major component of plaque and facilitates the plaque formation. So when researchers injected mouse models with the "aggregatin" protein (made from the FAM222A gene), plaque (amyloid deposits) formation accelerated in the brain, resulting in more neuroinflammation and cognitive dysfunction. This happened, they report, because the protein was found to bind directly the amyloid beta peptide, thus facilitating the aggregation and placque formation, Wang said. Conversely, when they suppressed the protein, the plaques were reduced and neuroinflammation and cognitive impairment alleviated. Their findings indicate that reducing levels of this protein and inhibition of its interaction with amyloid beta peptide could potentially be therapeutic—not necessarily to prevent Alzheimer's but to slow its progression.

10.1038/s41467-019-13962-0

Earth

Climate change to bring longer droughts in Europe: study

Increased future occurrences of the exceptional 2018–2019 Central European drought under global warming, Scientific Reports (2020). DOI: 10.1038/s41598-020-68872-9 , www.nature.com/articles/s41598-020-68872-9 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-020-68872-9

https://phys.org/news/2020-08-climate-longer-droughts-europe.html

Abstract Since the spring 2018, a large part of Europe has been in the midst of a record-setting drought. Using long-term observations, we demonstrate that the occurrence of the 2018–2019 (consecutive) summer drought is unprecedented in the last 250 years, and its combined impact on the growing season vegetation activities is stronger compared to the 2003 European drought. Using a suite of climate model simulation outputs, we underpin the role of anthropogenic warming on exacerbating the future risk of such a consecutive drought event. Under the highest Representative Concentration Pathway, (RCP 8.5), we notice a seven-fold increase in the occurrence of the consecutive droughts, with additional 40 ( \(\pm \, 5\) ) million ha of cultivated areas being affected by such droughts, during the second half of the twenty-first century. The occurrence is significantly reduced under low and medium scenarios (RCP 2.6 and RCP 4.5), suggesting that an effective mitigation strategy could aid in reducing the risk of future consecutive droughts. Introduction Human-induced climate change is evident and it poses a great concern to society, primarily due to its potential to intensify extreme events around the globe 1 , 2 . In the past 2 decades, Europe experienced an increased frequency of droughts 3 , 4 with estimated loss of about EUR 100 billion 5 . One such devastating event was the drought in summer 2003, which was an exceptionally warm and dry year across most of central and western Europe. Historical reconstructions since 1500 C.E. suggest that it was one of the hottest summers 6 , and the event was estimated to result in a 30% reduction in gross primary production compared to previous years between 1998–2002 3 . Although, the 2003 drought event was rare and exceptional, even in a multi-centennial time window, its likelihood is expected to increase in the near future 7 , mainly due to the anthropogenic warming 8 , 9 , 10 , 11 . In the summer of 2018, temperature anomaly broke the record again in several locations across Europe, but with distinct spatial patterns. While in summer 2003 the increase in temperature was more concentrated in central and southern Europe (Fig. 1 a), summer 2018 was characterised by an anomalous increase in central and north-eastern Europe (Fig. 1 b). Unlike the 2003 event—where the temperature anomaly (Supplementary Fig. S1 ) and the ecosystem carbon and energy fluxes recovered early after the summer 12 , the extreme event of 2018 persisted to the subsequent year 2019 (Fig. 1 c). For all these years, the impact was strongest in the Central European region, where the increase in temperature was accompanied by concurrent significant reduction of summer precipitation (Fig. 1 d–f), which led to extreme drought conditions. Figure 1 Anomalies of climate and vegetation health index (VHI) during 2003, 2018 and 2019. ( a , b , c ) Mean summer (June–August) temperature anomalies ( \(^\circ \hbox {C}\) ) for 2003, 2018 and 2019 based on the 1980–2010 climatology, and ( d , e , f ) their corresponding precipitation anomalies (%). ( g , h , i ) Vegetation condition in terms of VHI during 2003, 2018 and 2019, respectively. ( j ) Yearly development of the summer time, percentage area with poor vegetation health (i.e., VHI \(\le 30\) ) estimated over the Central European region (depicted by a black rectangular region in the panel g ) during the period 2000–2019. The thick black line shows the year-wise weekly mean of VHI during summer months, and the pink bar represents the corresponding 95% confidence limit based on the sampling distribution of the mean. The years 2003, 2015, 2018 and 2019 experienced the deprivation in the vegetation health, where the poor vegetation health extends over more than 20% of the central European region. The gray shaded region highlights the years 2018 and 2019, during which the poor vegetation health persists over more than 20% of the central European area, consecutively in 2 years. (k) Yearly summer-time precipitation and temperature anomalies estimated over the central Europe region during the 254 years. Three exceptional years of 2003, 2018 and 2019 are shown by the red dots, where the mean summer temperature anomalies over the Central Europe reached the record extreme conditions of more than \(2\,^\circ \hbox {C}\) ; and precipitation anomalies show deficit of more than 20%. The maps in the figure are generated using Python version 3.7.3 ( ). Full size image The intensity and spatial extent of droughts significantly affects the plant and agricultural productivity 13 , 14 , underlying the severity of the drought impact in Central European region, where the focus on agriculture is strong 3 , 7 , 15 , 16 , 17 . With the use of remote sensing data-sets 18 , we find that the concurrent increased temperature with deficit precipitation impaired the condition for vegetation activities (Fig. 1 g) in the summer of 2003. We show this in terms of vegetation health index (VHI), which represents the vegetation stress due to the droughts (see methods for detailed description). Similar observations have been made during the summer 2018 as well (Fig. 1 h), when several countries suffered agro-economic shocks 19 . Further, the deprivation of vegetation health persisted and it is noticed even during the summer 2019 (Fig. 1 i). In the case of the exceptional 2003 and 2015 events 4 , 20 , the vegetation health recovered and returned to its normal condition during the following years. On the contrary, the impact of the 2018 drought on vegetation activities propagated to 2019 and the recovery is still underway, as shown in the time series of VHI (Fig. 1 j). Additionally, we note that the vegetation health being categorised in poor condition for at-least 20% of the Central European area in both 2018 and 2019 is unprecedented from the observations in the previous years of twenty-first century. Thus, it is with the utmost urgency that we need to recognise the importance of these persevering consecutive year events, and to develop a holistic framework to model the risk 21 . Results 2018–2019 Central European drought from the long-term observational records The historical reconstruction of composite 254-year long-term climatic database 22 , 23 indicates that although the precipitation anomaly exhibits a drier than average situation during the summer months of 2018 and 2019 across the Central Europe, its intensity is not that high and there are also many other years with similar range of precipitation anomalies (Fig. 1 k). On the other hand, 2018–2019 were two out of the three warmest summer periods in the record. To account for this joint effect of precipitation and temperature anomalies, we estimate the drought index based on the standardised precipitation evapotranspiration index (SPEI) 24 that considers the atmospheric water supply and demand (see Methods). Our analysis further demonstrates the usefulness of the SPEI estimates as relevant climate predictors for characterising the temporal variability of the summer-time vegetation activities (see Supplementary Fig. S2 ). While the spatial pattern of summer 2018 SPEI (Fig. 2 a) depicts severe drought conditions in the Central European region (SPEI \(\le 0.1\) ; see Methods), southern Europe (Balkan countries) experienced wetter than normal conditions (i.e., SPEI \(\ge 0.5\) ). Similar to 2018, a severe drought condition (SPEI \(\le 0.1\) ) was noticed during the summer of 2019 but the spatial extent of drought was substantially larger compared to the 2018 event (Fig. 2 b). In Central Europe, over 34% of the total land area is extensively used for agricultural purposes 19 . Our analysis suggests that more than 50% of the Central European region suffered severe drought conditions in the consecutive years of 2018 and 2019. To examine how frequently these consecutive extreme events have occurred in the long-term observational records, we computed and plotted the areal extent of drought ( \(A_t\) ) with SPEI( t ) \(\le 0.1\) for a given year ( t ) with the corresponding estimates for the next year ( \(A_{t+1}\) ) (Fig. 2 c). It is evident from the analysis that the 2018–2019 drought is a record breaking event in terms of the consecutive event in the last 254 years, with nearly 50% of the Central European area being classified under the extreme drought conditions. It is also worth mentioning that the 1949–1950 years ranked the second most large-scale consecutive drought years 25 . Nonetheless, in this case the spatial extent was considerably smaller (around 33%) than that of the 2018–2019 droughts. Figure 2 2-year droughts from the long-term observational records over the Central Europe. ( a , b ) Spatial distribution of the drought index estimated based on summer months (June–August) SPEI for 2018 and 2019. ( c ) Scatter plot showing the percent drought area over the Central Europe for the next year ( \(A_{t+1}\) ) as a function of current year drought area ( \(A_t\) ). Prominent drought years, viz., 1949, 1950, 2003, 2015; and the recent 2018 and 2019 years, during which the spatial extent of summer droughts are significantly higher than the rest are highlighted in red dots. The cumulative distribution of the percent drought area is shown in the right panel of ( c ), with highlighted major drought years. The green dashed lines in ( c ) depict the drought area threshold of 33.3%—i.e., one third of the Central Europe region. The 2018–2019 event stands alone as an exceptional event for the consecutive droughts during the last 254 years (1766–2019). The maps in the figure are generated using Python version 3.7.3 ( ). Full size image The large-scale atmospheric circulation during 2018–2019 was characterized by pronounced positive geopotential height anomalies and anticyclonic circulation pattern at 500 hpa, covering a large area centered over Central Europe and extending to the Northern European region (Supplementary Fig. S3 ). The complex evolution of these blocking conditions highlights its contribution to the exceptional observed temperature anomalies during 2018–2019. Further, the persistent occurrence of these atmospheric blocking conditions are responsible for the development of large-scale droughts and heat wave, and also triggers soil-moisture temperature feedbacks 6 , 26 , 27 , 28 , which could further exacerbate and prolong concurrent soil drought and atmospheric aridity 29 . Literature review suggests that the recent arctic warming is likely to be a main driving factor causing more frequent extreme weather events across the mid-latitudes regions in the Northern Hemisphere 30 , 31 , 32 , 33 . The major dynamical features for changing the mid-latitude weather due to arctic amplification is the position and structure of the jet stream and planetary wave activity. Jet streams are primarily driven by the difference in temperature between the polar and mid-latitudinal regions. However, the reduced temperature gradient between these two regions has been suggested to lead to a weaker zonal jet with larger meanders 32 and that this would cause weather systems to travel eastward rather slowly leading to more persistent weather patterns 34 . These movement activities are further going to be affected (more persistent) under future warming conditions with increased greenhouse emission 35 , 36 . Nevertheless, these theories/mechanisms are still being explored and debatable; and require further rigorous testing 37 . Moreover, in this study we restrict our focus on analysing (detecting) the exceptional 2018–2019 Central European drought from the long-term observational record point of view, and on the nature of their possible occurrences under warming worlds. Addressing the mechanism (attribution) of the 2018–2019 drought event itself is another line of research, which requires a comprehensive analysis and is beyond the scope of the present study. Future occurrences of 2-year droughts under global warming From our observations, it is clear that, 2018–2019 is an unprecedented 2-year drought event. We now use the ensemble of climate model simulations from the Coupled Model Intercomparison Project phase 5 (CMIP5) 38 (Table S1) to understand how the frequency of the 2-year drought event would change in the coming decades, and underpin the role of the anthropogenic warming in exacerbating such drought events (see Methods). In comparison to the simulations based on natural-only forced simulations (HistNat), the occurrence of the 2-year drought event shows a slight increase in the historical simulations (Hist) during the common period of 1850–2005 (Fig. 3 a). The differences in the temporal evolution of areas of Central Europe affected by drought among the two sets of simulations have become more apparent during the last 30 years (approx. post 1970; Supplementary Fig. S4 )—the period in which there are apparent indications of the role of anthropogenic activities exacerbating global warming 39 . Climate model simulations based on the Representative Concentration Pathway (RCP) 8.5 scenario further indicates a strong increase in areas under drought towards the end of the twenty-first century (Fig. 3 a). Under a moderate scenario of RCP 4.5, the increasing trend persists until the middle of the twenty-first century and stagnates thereafter, while there is apparently no increasing trend in the temporal evolution of area under drought under the more optimistic RCP 2.6 scenario. Based on the climate model simulation results, we find a seven-fold increase in the number of the 2-year drought events, covering at-least one third of Central European domain, in the second half of the century under the RCP 8.5 scenario as compared to the HistNat runs (inset of Fig. 3 a). As a result, the corresponding fraction of attributable risk (FAR 40 ; see Methods for more details) under RCP 8.5 is estimated to be nearly one, ascertaining a very strong anthropogenic contribution to exacerbating the occurrence probability of such drought events in the projected future period 2051–2100. Compared to the RCP 8.5 scenario, the number of 2-year droughts events reduces significantly by almost half under the RCP 4.5 scenario and to a very negligible number in RCP 2.6 for the projected period 2051–2100. The corresponding FAR values also reduce to 0.87 and 0.40 for the RCPs 4.5 and 2.6, respectively. These results clearly highlight the diverse role of anthropogenic activities in exacerbating the future occurrence of 2-year drought events; as well the possible benefits of mitigating measures to reduce carbon emissions (encoded as in RCP 2.6/4.5) in lowering the risk of the occurrence of consecutive drought events. Figure 3 2-year droughts from state-of-the-art climate model simulations and its implications on cropland and pasture. ( a ) Yearly development of the percent area of drought over the Central Europe based on the ensemble ( \(\hbox {N} = 11\) ) of climate model simulations from CMIP5 under different experimental scenarios: natural only historical (HistNat), all-forcings historical (Hist), and three future RCPs (2.6, 4.5, and 8.5). The thick solid lines show the multimodel means, and the filled areas represent the 95% confidence intervals based on the sampling distribution of the mean from 11 GCMs simulations. The inset plot in ( a ) represents the number of 2-year droughts, with an areal extent in each year covering at-least one third of the Central European region, estimated over the specified time-period for different experimental scenarios (i.e., 1850–2005 for the Hist and HistNat; and 2006–2050/2051–2100 for the RCPs). Shown are the ensemble mean and 95% confidence limits based on the sampling distribution of the mean, corresponding to the 11 climate model outputs. The top panel of ( a ) depicts the year in which any of the 11 climate models show the 2-year droughts. The bottom panel of ( b ) shows cropland area (in million hectares) affected by the consecutive droughts under different experimental scenarios. The top panel of ( b ) shows the corresponding estimates in terms of percent of total cropland areas over the Central Europe, affected by the 2-year droughts. ( c ) Same as ( b ), but for pasture lands. The colors and ensemble statistics (i.e, mean and confidence intervals) are estimated as mentioned above. Full size image Consistent with the previous studies 8 , 10 , 41 , our analysis also shows that anthropogenic warming will lead to an intensification of European droughts, and to a large extent on the occurrence of 2-year droughts in the future. Such events have substantial implications on many sectors including impacts on agro-phenology, crop water demand and vegetation health activities. Using the long-term historical and projected land use changes based on HYDE database 42 (see Methods), we find that drought affected cropland areas across the Central Europe will be nearly doubled (by \(20 \pm 5\) million ha) under the RCP 8.5 scenario in the second half of the Century in comparison to corresponding historical values (Fig. 3 b). This corresponds to the projection of nearly 60% of total cultivated areas being affected by drought in the Central Europe during 2051–2100. Adaptation strategies aiming at the amendment of global warming through the RCP 4.5/2.6 scenarios would significantly reduce the drought prone areas by almost 37%/60%, compared to RCP 8.5. A similar range of benefits in reducing the potential impacts of consecutive year droughts can be expected for areas covered with pastures (Fig. 3 c)—which are of high importance for sustaining livestock (i.e., grazing). Conclusions The present study analyses the occurrence of the consecutive droughts over the Central Europe in both historical and projected climate scenarios. The observational record suggests that the ongoing 2018–2019 European drought event is unprecedented in the last 250 years, with substantial implications for vegetation health. Our analysis based on an ensemble of climate model simulations suggests a strong increase in the occurrence of such a rare event, post 2050 under RCP 8.5 scenario. The frequency and the areal extent of these droughts strongly depend on the level of anthropogenic warming scenarios (as encoded in RCPs). Our analysis therefore demonstrates that the occurrences of the consecutive droughts as well their impact on crop and pasture areas can be significantly reduced, if the mitigation strategies leading to amendment of global warming are adopted. One of the major limitations of climate model simulation is its ability to reliably simulate the extreme events and the changes thereof 43 . Over Central Europe, we notice a general consensus between observations and climate models, especially post-1970, when the anthropogenic influences are apparent (Supplementary Fig. S5 ). Although, climate models have a relatively good ability to simulate historical past, larger uncertainties may still exist in projections 44 . Despite this limitation, the climate models are the only available tool to mechanistically understand the occurrence, processes and fate of future extreme events. Our study has mainly focused on detection and the future occurrence of the consecutive drought events. Although we show that under the increased global warming, the observed 2018–2019 droughts are going to increase in the future, an in-depth and separate (careful) analysis is required towards attributing the role of anthropogenic warming in modulating the occurrence of consecutive drought events. Further research is also needed to systematically understand driving mechanisms responsible for such consecutive droughts, whose value to climate adaptation can hardly be overemphasised. Methods Data The assessment of consecutive drought characteristics from 1766–2019 is performed over the central Europe using three types of observed gridded meteorologic datasets: Casty et al 22 for period 1766–1900, CRU TS dataset 23 for period 1901–1949; and E-OBS 45 for period 1950–2019. The composite dataset using monthly precipitation and air temperature is analysed at a spatial resolution of \(0.5^\circ \, \times 0.5^\circ \) , similarly as in to previous studies 25 , 46 . Furthermore, the E-OBS is used for correcting possible biases in the Casty and CRU data, which is trained on the overlapping period 1950–2015 25 , 46 . To examine the characteristics of the consecutive droughts in the past and future, we procure the state-of-the-art global climate model simulations from the Coupled Model Intercomparison Project phase 5 (CMIP5) 38 (detailed description is provided in Supplementary Table S1 ). To quantify the effect of human activities in the past (1850–2005), two types of monthly forcings are analysed: (1) natural-only (HistNat), and (2) historical (Hist). While the HistNat contains only the effects of natural forcing (e.g., changes in solar radiation, volcanic eruptions), the Hist considers both, natural and anthropogenic (i.e., greenhouse gas concentrations) effects. To assess possible future climate scenarios, we procure three Representative Concentration Pathway (RCP) scenarios (2.6, 4.5 and 8.5), which are available for 2006–2100. Here, we select 11 climate models based on the consistency in data availability along with the parity in the climate variables (precipitation, temperature and net-radiation) used to estimate SPEI across all the simulations (HistNat, Hist and RCP scenarios). Compared to observations, the climate model simulations were able to capture the overall trend and patterns of atmospheric demand, particularly post-1970–the period when the human influence on the global warming is relatively more apparent 39 (Supplementary Fig. S5 ). In this study, we did not apply bias corrections to the CMIP5 simulations. This is because the employed quantile-based SPEI estimates already account for systematic biases, particularly in the mean and standard deviation, as long as these do not lead to unrealistic P-E dynamics 47 , which is fairly well captured by climate model simulations (Supplementary Fig. S5 ). The vegetation health index (VHI) is one of the important proxies, which is frequently used to evaluate the impacts of drought on vegetation health 48 , 49 . This index is applicable for assessing the vegetation stress and to examine the vegetation response to the natural hazards, such in our case, drought 49 . The VHI for the summer months is obtained from remote sensing data-sets 18 at a weekly time step, where it is measured in percentile ranging from 0 to 100. A high value of VHI indicates healthy or unstressed vegetation condition, implying that these areas are not affected by drought conditions (i.e., lack of moisture conditions). The VHI of more than 50% shows above normal and/or healthy vegetation condition. Further, values ranging from 30 to 50% imply vegetation in the region suffering from moderate drought, and the VHI values less then 30% indicate a region experiencing severe drought leading to poor vegetation health conditions. In the present study, we considered VHI values \(\le 30\%\) as a proxy for poor vegetation health conditions. Subsequently, we inferred the drought-affected vegetation activities as percentage of Central European area (shown in a rectangular box in Fig. 1 g) exhibiting VHI \(\le 30\%\) . Drought analysis Recent studies 24 , 50 , 51 , 52 , 53 show the better performance of summer standardised precipitation evapotranspiration index (SPEI) 24 , 54 , 55 in capturing the drought impacts on hydrological, ecological and agricultural variables than the standard precipitation index (SPI) or the Palmer drought severity index (PDSI). The SPI is not particularly appropriate for our application where both temperature and precipitation are important, since it neither considers the effects of increasing temperature over the recent decades 6 , 56 , nor the much larger warming scenario which is expected under future climate change scenarios 7 . Therefore, the characteristics of drought during summer months (June-August) in Central Europe for both observations and climate model simulations are estimated using SPEI. We used the non-parametric kernel-based approach to estimate the SPEI that can efficiently handle the multi-modality of the sample dataset as compared to other traditional parametric distributions 8 , 57 ; and it can be represented as: $$\begin{aligned} {\mathrm {SPEI}}=F_{t}(x_{t}) \end{aligned}$$ (1) where, \(x_{t}\) denotes the difference between precipitation ( P ) and potential evapotranspiration ( \(E_{p}\) ) at a time t . \(F_{t}\) is the cumulative distribution function estimated using the kernel distribution \(f_{t}(x)\) of the corresponding time t . \(f_{t}(x)\) is estimated as $$\begin{aligned} f_{t}(x)=\frac{1}{nh}\sum _{t=1}^{n}K \left( \frac{x-x_{t}}{h}\right) \end{aligned}$$ (2) where K represents a Gaussian kernel function with a bandwidth h . The h is estimated by the Silverman approach 58 for each grid cell separately. The SPEI value using the above-mentioned non-parametric approach varies between 0 and 1, with values below 0.5 indicate drier conditions and above 0.5 the wet conditions. A grid cell in central European region at time t is considered to be in drought when \({\hbox {SPEI}}_{t} \, \le \, \tau \) . Here, \(\tau \) denotes that the SPEI in the particular grid cell is less than the values occurring \(\tau \, \times 100\%\) of the time, and the present study considers \(\tau \) as 0.1 (i.e., 1 in 10-year event or 20% of all dry events)—indicating the occurrence of severe drought event 8 , 59 . In case of climate model simulation we also use the non-parametric kernel density estimator, however, we fix the bandwidth with respect to natural-forced historical simulation and use the same for historical; and for all the RCP scenarios considered in the present study. We estimated the yearly development of drought area ( \(A_t\) ), considering all the cells of the total Central European region that are under drought ( \(\hbox {SPEI} \le 0.1\) ) for a given year ( t ). We marked a drought event as a 2-year consecutive event when \(A_t\) in both years crosses a certain threshold value (e.g., 33.3% or one-third of the Central European region). While estimating the number of consecutive drought events, especially in the RCP 4.5 and 8.5 scenarios, we notice many events with a common (overlapping) drought year. To account for the double counting effect, we counted those events as half which have an overlapping drought year between two (consecutive) events. Considering the availability of climate variables over long time period (1766–2019), we estimate the monthly potential evapotranspiration ( \(E_{p}\) ) based from the mean temperature and the approximations for extraterrestrial solar radiation 60 . Owing to the limitation on the estimation the temperature based \(E_{p}\) 61 , we check the consistency of this method with an alternative and more physically based \(E_p\) formulation. In this respect we use two \(E_{p}\) datasets derived based on the Penman–Monteith method using: (a) the CRU database 23 employing the mean, minimum and maximum temperature, vapour pressure, cloudiness and monthly climatology of wind speed available after 1901; and (b) the Princeton Global Forcing (PGF) 61 that employs full scale variability of all required meteorological variables (e.g., net radiation, temperature and wind-speed) provided by Sheffield et al. 61 for the period 1948–2008. Albeit different underlying meteorological databases being used (CRU 23 vs. PGF 61 ), in general, we notice a relatively good agreement among the three \(E_{p}\) values, especially in capturing the inter-annual variability over the Central European region (see Supplementary Figs. S6 for more details). We further check the consistency of our results based on the \(E_{p}\) estimates derived from an energy budget approach following Milley and Dunne 62 , as given by: $$\begin{aligned} E_{p} = 0.8 (R_{n} - G) \end{aligned}$$ (3) where \(R_{n}\) is net radiation at the surface, and G is ground heat flux. Here \(R_{n} - G\) are estimated using the energy balance as: \(R_{n} - G = L_{v}E + H\) , where \(L_{v}E\) and H are the latent and sensible heat flux, respectively. Using the climate model simulation outputs, our results show a high correspondence of \(E_{p}\) between the energy-based approach 62 and the Oudin et al. 60 for the study domain (see Supplementary Fig. S7a,b for more details). Furthermore, the robustness of our findings on the increased occurrence of the future 2-year consecutive droughts is confirmed, regardless of the employed \(E_{p}\) methods (see Supplementary Fig. S7c ). Fraction of attributable risk (FAR) The FAR has been used by many studies to quantify the anthropogenic influence on the occurrence of recent extreme events and its fate in projected scenarios. The FAR basically addresses the question of what fraction of extremes (in our case 2-year consecutive drought) occurring in Central European region is attributable to anthropogenic influence, and is given by, $$\begin{aligned} FAR = 1 - (P_{0}/P_{1}) \end{aligned}$$ (4) where \(P_{0}\) is the probability of exceeding a 2-year consecutive drought without anthropogenic influence (HistNat) and \(P_{1}\) is with the anthropogenic influence 40 (RCP scenarios). The FAR value near to 1 indicates the nearly certain human influence in causing the 2-year consecutive drought. Cropland and pasture areas The impact of droughts on cropland area and pastures are analysed using the dynamics of land use changes of land cover dataset 42 . This dataset consists of half-degree gridded historical and future fractional land-use patterns and underlying land-use transitions. The historical data uses the HYDE v3.1 historical data set for crop, pasture, and urban area 1500–2005, and the future land cover scenarios 2006–2100 are available for four Integrated Assessment Model (IAM) scenarios which reach different levels of radiative forcing by year 2100:, viz., MESSAGE (RCP 8.5), AIM (RCP 6.0), GCAM/minicam (RCP 4.5) and IMAGE (RCP 2.6). Further, each of these future projections are built by four different historical land-use products, all these are considered in our study. The cropland cover fraction over the Central Europe started to increase during post 1950, however, a drastic decrease in spatial extent happened after 1990 (Supplementary Fig. S8a ). In RCP 4.5 and RCP 8.5 scenarios, the Central Europe will experience a sharp decrease in the overall cropland area. This information is then combined with the fraction of total Central European area which is affected by droughts, as obtained from climate models (Supplementary Fig. S8c ). We notice a prominent increasing trend of cropland area affected by drought, especially in the RCP 8.5 scenario. Similar behaviour is projected for the pastures as well (Supplementary Fig. S8b,d ). With these observations, we notice a sharp increase in the areal effects both in cropland and pasture by the 2-year consecutive drought in the future, as shown in Fig. 3 b. These findings remained same even when we considered a fixed, not time varying, area corresponding to the year 2005 (Supplementary Fig. S8e,f ). Data availability Reconstructed historical precipitation and temperature (1766–1900) are available at ftp://ftp.ncdc.noaa.gov/pub/data/paleo/historical/europe/casty2007/ . The HadCRU TS product (1901–1950 for precipitation, temperature, potential evapotranspiration) is available from . The E-OBS data (1951–2019) are available from , the CMIP5 data from , the VHI data from , and HYDE (v3.1) landcover data from . Other processed datasets can be made available upon reasonable request from the corresponding author. Code availability The codes for estimating the SPEI based on kernel density approach can be acquired from PYTHON repository. Other processing codes can be procured from VH.

Punishing two-year droughts like the record-breaking one that gripped Central Europe from 2018 to 2019 could become much more frequent if the region fails to curb greenhouse gas emissions, researchers said Thursday, affecting huge swathes of its cultivated land. The five hottest years in recorded history have occurred in the last five years. This extreme heat was exacerbated in 2018 and 2019 by two consecutive summers of drought that affected more than half of Central Europe, according to a new study published in the Nature journal Scientific Reports. Researchers in Germany and the Czech Republic used data going back to 1766 to conclude the drought was the largest-scale and most severe dry spell ever recorded. "The observational record suggests that the ongoing 2018–2019 European drought event is unprecedented in the last 250 years, with substanprolongedtial implications for vegetation health," the study said. Researchers then sought to estimate whether prolonged droughts would become more frequent in the future by using global climate change models. Under a scenario where greenhouse gas emissions continue their inexorable rise, the researchers predicted that the number of extreme two-year droughts will increase sevenfold in Europe in the second half of this century. "This projection also suggested that drought-affected cropland areas across Central Europe will nearly double," said co-author Rohini Kumar, of the UFZ-Helmholtz Centre for Environmental Research, in Leipzig. This would result in a total of 40 million hectares of cultivated land affected—equivalent to 60 percent of all crop areas in the region. When researchers modelled for moderate emissions, the predicted number of two-year droughts halved compared to the worst case scenario, while the area expected to be hit by the drought also reduced. Kumar said this suggests a reduction in emissions could lower the risk of these damaging dry periods. The five hottest years in recorded history have occurred in the last five years Threat to agriculture The researcher said a two-year dry period presents a far greater threat to vegetation than the single-summer droughts of previous years because the land cannot recover as quickly. He said around a fifth of the Central European region had recorded poor vegetation health in the last two years. "Thus, it is with the utmost urgency that we need to recognise the importance of these persevering consecutive year events, and to develop a holistic framework to model the risk," he added. The study defined Central Europe as including parts of Germany, France, Poland, Switzerland, Italy, Austria, as well as Czech Republic, Belgium, Slovenia, Hungary, Slovakia. Over 34 percent of the total land area in the region is extensively used for agricultural purposes, it said. The 2015 Paris climate deal commits nations to capping temperature rises to "well-below" 2C (3.6 degrees Fahrenheit) above pre-industrial levels and to strive for a 1.5C limit if at all possible. With just 1C of warming so far, Earth is already buffeted by record-breaking droughts, wildfires and super storms made more potent by rising sea levels. To keep in line with the 1.5C target, the United Nations says global emissions must fall by 7.6 percent every year this decade.

10.1038/s41598-020-68872-9

Other

Did it keep its flavour? Stone-age 'chewing-gum' yields human DNA

A 5700 year-old human genome and oral microbiome from chewed birch pitch, Nature Communications (2019). DOI: 10.1038/s41467-019-13549-9 , www.nature.com/articles/s41467-019-13549-9 Press release Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-019-13549-9

https://phys.org/news/2019-12-flavour-stone-age-chewing-gum-yields-human.html

Abstract The rise of ancient genomics has revolutionised our understanding of human prehistory but this work depends on the availability of suitable samples. Here we present a complete ancient human genome and oral microbiome sequenced from a 5700 year-old piece of chewed birch pitch from Denmark. We sequence the human genome to an average depth of 2.3× and find that the individual who chewed the pitch was female and that she was genetically more closely related to western hunter-gatherers from mainland Europe than hunter-gatherers from central Scandinavia. We also find that she likely had dark skin, dark brown hair and blue eyes. In addition, we identify DNA fragments from several bacterial and viral taxa, including Epstein-Barr virus, as well as animal and plant DNA, which may have derived from a recent meal. The results highlight the potential of chewed birch pitch as a source of ancient DNA. Introduction Birch pitch is a black-brown substance obtained by heating birch bark and has been used as an adhesive and hafting agent as far back as the Middle Pleistocene 1 , 2 . Small lumps of this organic material are commonly found on archaeological sites in Scandinavia and beyond, and while their use is still debated, they often show tooth imprints, indicating that they were chewed 3 . Freshly produced birch pitch hardens on cooling and it has been suggested that chewing was a means to make it pliable again before using it, e.g. for hafting composite stone tools. Medicinal uses have also been suggested, since one of the main constituents of birch pitch, betulin, has antiseptic properties 4 . This is supported by a large body of ethnographic evidence, which suggests that birch pitch was used as a natural antiseptic for preventing and treating dental ailments and other medical conditions 3 . The oldest examples of chewed pitch found in Europe date back to the Mesolithic period and chemical analysis by Gas Chromatography-Mass Spectrometry (GC-MS) has shown that many of them were made from birch ( Betula pendula ) 3 . Recent work by Kashuba et al 5 . has shown that pieces of chewed birch pitch contain ancient human DNA, which can be used to link the material culture and genetics of ancient populations. In the current study, we analyse a further piece of chewed birch pitch, which was discovered at a Late Mesolithic/Early Neolithic site in southern Denmark (Fig. 1a ; Supplementary Note 1 ) and demonstrate that it does not only contain ancient human DNA, but also microbial DNA that reflects the oral microbiome of the person who chewed the pitch, as well as plant and animal DNA which may have derived from a recent meal. The DNA is so exceptionally well preserved that we were able to recover a complete ancient human genome from the sample (sequenced to an average depth of coverage of 2.3×), which is particularly significant since, so far, no human remains have been recovered from the site 6 . The results highlight the potential of chewed birch pitch as a source of ancient human and non-human DNA, which can be used to shed light on the population history, health status, and even subsistence strategies of ancient populations. Fig. 1 A chewed piece of birch pitch from southern Denmark. ( a ) Photograph of the Syltholm birch pitch and its find location at the site of Syltholm on the island of Lolland, Denmark (map created using data from Astrup 78 ). ( b ) Calibrated date for the Syltholm birch pitch (5,858–5,661 cal. BP; 5,007 ± 7). ( c ) GC-MS chromatogram of the Syltholm pitch showing the presence of a series of dicarboxylic acids (Cxx diacid) and saturated fatty acids (Cxx:0) and methyl 16-Hydroxyhexadecanoate (C 16 OH) together with the triterpenes betulin and lupeol, which are characteristic of birch pitch 3 . Full size image Results Radiocarbon dating and chemical analysis Radiocarbon dating of the specimen yielded a direct date of 5,858–5,661 cal. BP (GrM-13305; 5,007 ± 11) (Fig. 1b ; Supplementary Note 2 ), which places it at the onset of the Neolithic period in Denmark. Chemical analysis by Fourier-Transform Infrared (FTIR) spectroscopy produced a spectrum very similar to modern birch pitch (Supplementary Fig. 4 ) and GC-MS revealed the presence of the triterpenes betulin and lupeol, which are characteristic of birch pitch (Fig. 1c ; Supplementary Note 3 ) 3 . The GC-MS spectrum also shows a range of dicarboxylic acids and saturated fatty acids, which are all considered intrinsic to birch pitch and thus support its identification 7 . DNA sequencing We generated approximately 390 million DNA reads for the sample, nearly a third of which could be uniquely mapped to the human reference genome (hg19) (Supplementary Table 2 ). The human reads displayed all the features characteristic of ancient DNA, including (i) short average fragment lengths, (ii) an increased occurrence of purines before strand breaks, and (iii) an increased frequency of apparent cytosine (C) to thymine (T) substitutions at 5′-ends of DNA fragments (Supplementary Fig. 6 ) and the amount of modern human contamination was estimated to be around 1–3% (Supplementary Table 3 ). In addition to the human reads, we generated around 7.3 Gb of sequence data (68.8%) from the ancient pitch that did not align to the human reference genome. DNA preservation and genome reconstruction With over 30%, the human endogenous DNA content in the sample was extremely high and comparable to that found in well-preserved teeth and petrous bones 8 . We used the human reads to reconstruct a complete ancient human genome, sequenced to an effective depth-of-coverage of 2.3×, as well as a high-coverage mitochondrial genome (91×), which was assigned to haplogroup K1e (see Methods). To further investigate the preservation of the human DNA in the sample we calculated a molecular decay rate ( k , per site per year) and find that it is comparable to that of other ancient human genomes from temperate regions (Supplementary Table 3 ). Sex determination and phenotypic traits Based on the ratio between high-quality reads (MAPQ ≥ 30) mapping to the X and Y chromosomes, respectively 9 , we determined the sex of the individual whose genome we recovered to be female. To predict her hair, eye and skin colour we imputed genotypes for 41 SNPs (Supplementary Data 1 ) included in the HIrisPlex-S system 10 and find that she likely had dark skin, dark brown hair, and blue eyes (Supplementary Data 2 ). We also examined the allelic state of two SNPs linked with the primary haplotype associated with lactase persistence in humans and found that she carried the ancestral allele for both (Supplementary Data 1 ), indicating that she was lactase non-persistent. Genetic affinities We called 593,102 single nucleotide polymorphisms (SNPs) in our ancient genome that had previously been genotyped in a dataset of >1000 present-day individuals from a diverse set of Eurasian populations 11 , as well as >100 previously published ancient genomes (Supplementary Data 3 ). Figure 2a shows a principal component analysis (PCA) where she clusters with western hunter-gatherers (WHGs). Allele-sharing estimates based on f 4 -statistics show the same overall affinity to WHGs (Fig. 2b ). This is also reflected in the qpAdm analysis 12 (see Methods) which demonstrates that a simple one way model assuming 100% WHG ancestry cannot be rejected in favour of more complex models (Fig. 2c ; Supplementary Table 6 ). To formally test this result we computed two sets of D -statistics of the form D (Yoruba, EHG/Barcın; test, WHG) and find no evidence for significant levels of EHG or Neolithic farmer gene flow (Supplementary Fig. 7 ; Supplementary Tables 7 , 8 ). Fig. 2 Genetic affinities of the Syltholm individual. a Principal component analysis of modern Eurasian individuals (in grey) and a selection of over 100 previously published ancient genomes, including the Syltholm genome. The ancient individuals were projected on the modern variation (see Methods). b Allele-sharing estimates between the Syltholm individual, other Mesolithic and Neolithic individuals, and WHGs versus EHGs and Neolithic farmers, respectively, as measured by the statistic f 4 (Yoruba, X ; EHG/Barcın, WHG). c Ancestry proportions based on qpAdm 12 , specifying WHG, EHG, and Neolithic farmers (Barcın) as potential ancestral source populations. PWC Pitted Ware Culture, LBK Linearbandkeramik, GAC Globular Amphora Culture, LP Late Paleolithic, M Mesolithic, EN Early Neolithic, MN Middle Neolithic, LN Late Neolithic. Data are shown in Supplementary Tables 4 – 6 . Full size image Metataxonomic profiling of non-human reads To broadly characterise the taxonomic composition of the non-human reads in the sample, we used MetaPhlan2 13 , a tool specifically designed for the taxonomic profiling of short-read metagenomic shotgun data (see Methods; Supplementary Data 4 ). Figure 3a shows a principal coordinate analysis where we compare the microbial composition of our sample to that of 689 microbiome profiles from the Human Microbiome Project (HMP) 14 . We find that our sample clusters with modern oral microbiome samples in the HMP dataset. This is also reflected in Fig. 3b which shows the order-level microbial composition of our sample compared to two soil samples from the same site and metagenome profiles of healthy human subjects at five major body sites from the HMP 14 , visualised using MEGAN6 15 . Fig. 3 Metagenomic profile of the Syltholm birch pitch. a PCoA with Bray-Curtis at genera level with 689 microbiomes from HMP 14 and the Syltholm sample (see Methods). b Order-level microbial composition of the Syltholm sample compared to a control sample (soil) and metagenome profiles of healthy human subjects at five major body sites from the HMP 14 , visualised using MEGAN6 15 . Full size image Oral microbiome characterisation To further characterise the microbial taxa present in the ancient pitch and to obtain species-specific assignments we used MALT 16 , a fast alignment and taxonomic binning tool for metagenomic data that aligns DNA sequencing reads to a user-specified database of reference sequences (see Methods; Supplementary Data 5 ). As expected, a large number of reads could be assigned to oral taxa, such as Neisseria subflava and Rothia mucilaginosa , as well as several bacteria included in the red complex (i.e. Porphyromonas gingivalis , Tannerella forsythia , and Treponema denticola ) (see Table 1 ). In addition, we recovered 593 reads that were assigned to Epstein–Barr virus (Human gammaherpesvirus 4). We validated each taxon by examining the edit distances, coverage distributions, and post-mortem DNA damage patterns (see Supplementary Note 5 ). Table 1 List of non-human taxa identified in the Syltholm pitch, including the 40 most abundant oral bacterial taxa, viruses, and eukaryotes. Bacteria in the red complex are denoted with an asterisk. Depth (DoC) and breadth of coverage (>1x) were calculated using BEDTools 72 . Deamination rates at the 5’ ends of DNA fragments were estimated using mapDamage 2.0.9 59 . -Δ% refers to the negative difference proportion introduced by Hübler et al 79 . (see Supplementary Note 5 ). Full size table Pneumococcal DNA We also identified several species belonging to the Mitis group of streptococci (Table 1 ), including Streptococcus viridans and Streptococcus pneumoniae . We reconstructed a consensus genome from the S. pneumoniae reads (Fig. 4 ) and estimated the number of heterozygous sites (2,597) (see Methods) which indicates the presence of multiple strains. To assess the virulence of the S. pneumoniae strains recovered from the ancient pitch, we aligned the contigs against the full Virulence Factor Database 17 in order to identify known S. pneumoniae virulence genes (see Methods). We identified 26 S. pneumoniae virulence factors within the ancient sample, including capsular polysaccharides (CPS), streptococcal enolase (Eno), and pneumococcal surface antigen A (PsaA) (see Supplementary Data 6 ). Fig. 4 Streptococcus pneumoniae consensus genome reconstructed from metagenomic sequences recovered from the ancient pitch. From outer to inner ring: S. pneumoniae virulence genes (black, shared genes are shown in bold); S. pneumoniae coding regions on the positive (blue) and negative (red) strand; mappability (grey); sequence depth for the Syltholm pitch (orange), HOMP sample SRS014468 (light brown), SRS019120 (light blue), SRS013942 (turquoise), SRS015055 (blue), and SRS014692 (dark blue). Sequence depths were calculated by aligning to the S. pneumoniae TIGR4 reference genome and visualised in 100 bp windows using Circos 73 . Full size image Plant and animal DNA Lastly, we used a taxonomic binning pipeline specifically designed for ancient environmental DNA 18 to taxonomically classify the non-human reads in the sample that mapped to other Metazoa (animals) and Viridiplantae (plants). We only parsed taxa with classified reads accounting for >1% of all reads in each of the two kingdoms and a declining edit distance distribution after edit distance 0 (Supplementary Data 7 ). We then validated each identified taxon as described above (see Supplementary Note 5 ). Using these criteria, we identified DNA from two plant species in the ancient sample, including birch ( Betula pendula ) and hazelnut ( Corylus avellana ). In addition, we detected over 50,000 reads that were assigned to mallard ( Anas platyrhynchos ). Discussion We successfully extracted and sequenced ancient DNA from a 5700-year-old piece of chewed birch pitch from southern Denmark. In addition to a complete ancient human genome (2.3×) and mitogenome (91×), we recovered plant and animal DNA, as well as microbial DNA from several oral taxa. Analysis of the human reads revealed that the individual whose genome we recovered was female and that she likely had dark skin, dark brown hair and blue eyes. This combination of physical traits has been previously noted in other European hunter-gatherers 19 , 20 , 21 , 22 , suggesting that this phenotype was widespread in Mesolithic Europe and that the adaptive spread of light skin pigmentation in European populations only occurred later in prehistory 23 . We also find that she had the alleles associated with lactase non-persistence, which fits with the notion that lactase persistence in adults only evolved fairly recently in Europe, after the introduction of dairy farming with the Neolithic revolution 24 , 25 . From a population genetics point of view, the human genome also offers fresh insights into the early peopling of southern Scandinavia. Recent studies of ancient hunter-gatherer genomes from Sweden and Norway 23 have shown that, following the retreat of the ice sheets around 12–11 ka years ago, Scandinavia was colonised by two separate routes, one from the south (presumably via Denmark) and one from the northeast, along the coast of present-day Norway. This is supported by the fact that hunter-gatherers from central Scandinavia carry different levels of WHG and EHG ancestry, which reached central Scandinavia from the south and northeast, respectively 23 . Although we only analysed a single genome, the fact that the Syltholm individual does not carry any EHG ancestry confirms this scenario and suggests that EHGs did not reach southern Denmark at this point in prehistory. The Syltholm genome (5700 years cal. BP) dates to the period immediately following the Mesolithic-Neolithic transition in Denmark. Culturally, this period is marked by the transition from the Late Mesolithic Ertebølle culture (c. 7300–5900 cal. BP) with its flaked stone artefacts and typical T-shaped antler axes, to the early Neolithic Funnel Beaker culture (c. 5900–5300 cal. BP) with its characteristic pottery, polished flint artefacts, and domesticated plants and animals 26 . In Denmark, the transition from hunting and gathering to farming has often been described as a relatively rapid process, with dramatic shifts in settlement patterns and subsistence strategies 27 . However, it is still unclear to what extent this transition was driven by the arrival of farming communities as opposed to the local adaptation of farming practices by resident hunter-gatherer populations. Our analyses have shown that the Syltholm individual does not carry any Neolithic farmer ancestry, suggesting that the genetic impact of Neolithic farming communities in southern Scandinavia might not have been as instant or pervasive as once thought 28 . While the mtDNA we recovered belongs to haplogroup K1e, which is more commonly associated with early farming communities 29 , 30 , 31 , there is mounting evidence to suggest that this lineage was already present in Mesolithic Europe 32 , 33 , 34 . Overall, the lack of Neolithic farmer ancestry is consistent with evidence from elsewhere in Europe, which suggests that genetically distinct hunter-gatherer groups survived for much longer than previously assumed 35 , 36 , 37 . These WHG “survivors” might have triggered the resurgence of hunter-gatherer ancestry that is proposed to have occurred in central Europe between 7000 and 5000 BP 12 . In addition to the human data, we recovered ancient microbial DNA from the pitch which could be shown to have a human oral microbiome signature. Previous studies 38 , 39 , 40 have demonstrated that calcified dental plaque (dental calculus) provides a robust biomolecular reservoir that allows direct and detailed investigations of ancient oral microbiomes. However, unlike dental calculus, which represents a long-term reservoir of the oral microbiome built up over many years, the microbiota found in ancient mastics are more likely to give a snapshot of the species active at the time. As such, they provide a useful source of information regarding the evolution of the human oral microbiome that can complement studies of ancient dental calculus. The majority of the bacterial taxa we identified (Table 1 ) are classified as non-pathogenic, commensal species that are considered to be part of the normal microflora of the human mouth and the upper respiratory tract, but may become pathogenic under certain conditions. In addition, we identified three species ( Porphyromonas gingivalis , Tannerella forsythia , and Treponema denticola ) included in the so-called red complex, a group of bacteria that are categorised together based on their association with severe forms of periodontal disease 41 . Furthermore, we identified several thousand reads that could be assigned to different bacterial species in the Mitis group of streptococci, including Streptococcus pneumoniae , a major human pathogen that is responsible for the majority of community-acquired pneumonia which still causes around 1–2 million infant deaths worldwide, every year 42 . S. pneumoniae has a remarkable capacity to remodel its genome through the uptake of exogenous DNA from other pneumococci and closely related oral streptococci 42 . Understanding this process and the distribution of pneumococcal virulence factors between different strains can help our understanding of S. pneumoniae pathogenesis. We identified 26 S. pneumoniae virulence factors within our ancient sample, including several that are involved in host colonisation (e.g. adherence to host cells and tissues, endocytosis) and the evasion and subversion of the host’s immune response (Supplementary Data 6 ). While more research is needed to fully understand the evolution of this important human pathogen and its ability to cause disease, our capacity to recover virulence factors from ancient samples opens up promising avenues for future research. In addition to the bacterial taxa, we identified 593 reads that could be assigned to the Epstein–Barr virus (EBV). Previous studies 43 , 44 have demonstrated the great potential of ancient DNA for studying the long-term evolution of blood borne viruses. Formally known as Human gammaherpesvirus 4, EBV is one of the most common human viruses infecting over 90% of the world’s adult population 45 . Most EBV infections occur during childhood and in the vast majority of cases they are asymptomatic or they carry symptoms that are indistinguishable from other mild, childhood diseases. However, in some cases EBV can cause infectious mononucleosis (glandular fever) 46 and it has also been associated with various lymphoproliferative diseases, such as Hodgkin's lymphoma and hemophagocytic lymphohistiocytosis, as well as higher risks of developing certain autoimmune diseases, such as dermatomyositis and multiple sclerosis 47 , 48 . Lastly, we identified several thousand reads that could be confidently assigned to different plant and animal species, including birch ( B. pendula ), hazelnut ( C. avellana ), and mallard ( A. platyrhynchos ). While the presence of birch DNA is easily explained as it is the source of the pitch, we propose that the hazelnut and mallard DNA may derive from a recent meal. This is supported by the faunal evidence from the site, which is dominated by wild taxa, including Anas sp. and hazelnuts 6 , 49 . In addition, there is evidence from many other Mesolithic and Early Neolithic sites in Scandinavia for hazelnuts being gathered in large quantities for consumption 50 . Together with the faunal evidence, the ancient DNA results support the notion that the people at Syltholm continued to exploit wild resources well into the Neolithic and highlight the potential of ancient DNA analyses of chewed pieces of birch pitch for palaeodietary studies. In summary, we have shown that pieces of chewed birch pitch are an excellent source of ancient human and non-human DNA. In the process of chewing, the DNA becomes trapped in the pitch where it is preserved due to the aseptic and hydrophobic properties of the pitch which both inhibits microbial and chemical decay. The genomic information preserved in chewed pieces of birch pitch offers a snapshot of people's lives, providing information on genetic ancestry, phenotype, health status, and even subsistence. In addition, the microbial DNA provides information on the composition of our ancestral oral microbiome and the evolution of specific oral microbes and important human pathogens. Methods Sample preparation and DNA extraction We sampled c. 250 mg from the specimen for DNA analysis. Briefly, the sample was washed in 5% bleach solution to remove any surface contamination, rinsed in molecular biology grade water and left to dry. We tested three different extraction methods using between 20–50 mg of starting material: For method (1), 1 ml of lysis buffer containing 0.45 M EDTA (pH 8.0) and 0.25 mg/ml Proteinase K was added to the sample and left to incubate on a rotor at 56 °C. After 12 h the supernatant was removed and concentrated down to ~150 µl using Amicon Ultra centrifugal filters (MWCO 30 kDa), mixed 1:10 with a PB-based binding buffer 51 , and purified using MinElute columns, eluting in 30 µl EB. For method (2) the sample was digested and purified as above, but with the addition of a phenol-chloroform clean-up step. Briefly, 1 ml phenol (pH 8.0) was added to the lysis mix, followed by 1 ml chloroform:isoamyl alcohol. The supernatant was concentrated and purified, as described above. For method (3) the sample was dissolved in 1 ml chloroform:isoamylalcohol. The dissolved sample was then resuspended in 1 ml molecular grade water and purified as described above. DNA extracts prepared using a Proteinase K-based lysis buffer followed by a phenol-chloroform based purification step produced the best results in terms of the endogenous human DNA content (see Supplementary Table 1 ); however, following metagenomic profiling the extracts were found to be contaminated with Delftia spp., a known laboratory contaminant 52 . The contaminated libraries were excluded from metagenomic profiling. Negative controls We included no template controls (NTC) during the DNA extraction and library preparation steps. The NTCs prepared with the additional phenol-chloroform step were also found to be contaminated with Delftia spp., suggesting that the contaminants were introduced during this step. In addition, we included two soil samples from the site, weighing c. 2 g each, as negative controls. DNA was extracted as described above using 3 ml EDTA-based lysis buffer followed by 9 ml 25:24:1 phenol:chloroform:isoamyl alcohol mixture to account for the larger amount of starting material. The sequencing results are reported in Supplementary Table 1 . Library preparation and sequencing 16 µl of each DNA extract were built into double-stranded libraries using a recently published protocol that was specifically designed for ancient DNA 53 . One extraction NTC was included, as well as a single library NTC. 10 µl of each library were amplified in 50 µl reactions for between 15 and 28 cycles, using a dual indexing approach 54 . The optimal number of PCR cycles was determined by qPCR (MxPro 3000, Agilent Technologies). The amplified libraries were purified using SPRI-beads and quantified on a 2200 TapeStation (Agilent Technologies) using High Sensitivity tapes. The amplified and indexed libraries were then pooled in equimolar amounts and sequenced on 1/8 of a lane of an Illumina HiSeq 2500 run in SR mode. Following initial screening, additional reads were obtained by pooling libraries #2, #3, and #4 in molar fractions of 0.2, 0.4, and 0.4, respectively and sequencing them on one full lane of an Illumina HiSeq 2500 run in SR mode. Data processing Base calling was performed using Illumina’s bcl2fastq2 conversion software v2.20.0. Only sequences with correct indexes were retained. FastQ files were processed using PALEOMIX v1.2.12 55 . Adapters and low quality reads (Q < 20) were removed using AdapterRemoval v2.2.0 56 , only retaining reads >25 bp. Trimmed and filtered reads were then mapped to hg19 (build 37.1) using BWA 57 with seed disabled to allow for better sensitivity 58 , as well as filtering out unmapped reads. Only reads with a mapping quality ≥30 were kept and PCR duplicates were removed. MapDamage 2.0.9 59 was used to evaluate the authenticity of the retained reads as part of the PALEOMIX pipeline 55 , using a subsample of 100k reads per sample (Supplementary Fig. 6 ). For the population genomic analyses, we merged the ancient sample with individuals from the Human Origin dataset 11 and >100 previously published ancient genomes (Supplementary Data 1 ). At each SNP in the Human Origin dataset, we sampled the allele with more reads in the ancient sample, resolving ties randomly, resulting in a pseudohaploid ancient sample. MtDNA analysis and contamination estimates We used Schmutzi 60 to determine the endogenous consensus mtDNA sequence and to estimate present-day human contamination. Reads were mapped to the Cambridge reference sequence (rCRS) and filtered for MAPQ ≥ 30. Haploid variants were called using the endoCaller program implemented in Schmutzi 60 and only variants with a posterior probability exceeding 50 on the PHRED scale (probability of error: 1/100,000) were retained. We then used Haplogrep v2.2 61 to determine the mtDNA haplogroup, specifying PhyloTree (build 17) as the reference phylogeny 62 . Contamination estimates were obtained using Schmutzi’s mtCont program and a database of putative modern contaminant mitochondrial DNA sequences. Genotype imputation We used ANGSD 63 to compute genotype likelihoods in 5 Mb windows around 43 SNPs associated with skin, eye, and hair colour 10 and lactase persistence into adulthood (Supplementary Data 2 ). Missing genotypes were imputed using impute2 64 and the pre-phased 1000 Genome reference panel 65 , provided as part of the impute2 reference datasets. We used multiple posterior probability thresholds, ranging from 0.95 to 0.50, to filter the imputed genotypes. The imputed genotypes were uploaded to the HIrisPlex-S website 10 to obtain the predicted outcomes for the pigmentation phenotypes (Supplementary Data 3 ). Principal component analysis Principal component analysis was performed using smartPCA 66 by projecting the ancient individuals onto a reference panel including >1000 present-day Eurasian individuals from the HO dataset 11 using the option lsq project. Prior to performing the PCA the data set was filtered for a minimum allele frequency of at least 5% and a missingness per marker of at most 50%. To mitigate the effect of linkage disequilibrium, the data were pruned in a 50-SNP sliding window, advanced by 10 SNPs, and removing sites with an R 2 larger than 0, resulting in a final data set of 593,102 SNPs. D - and f -statistics D - and f -statistics were computed using AdmixTools 67 . To estimate the amount of shared drift between the Syltholm genome and WHG versus EHG and Neolithic farmers, respectively, we computed two sets of f 4 -statistics of the form f 4 (Yoruba, X ; EHG/Barcın, WHG) where “ X ” stands for the test sample. Standard errors were calculated using a weighted block jackknife. To confirm the absence of EHG and Neolithic farmer gene flow in the Syltholm genome and to contrast this result with those obtained for other Mesolithic and Neolithic individuals from Scandinavia, we computed two sets of D -statistics of the form D (Yoruba, EHG/Barcın; X , WHG) testing whether “ X ” forms a clade to the exclusion of EHG and Neolithic farmers (represented by Barcın), respectively. qpAdm Admixture proportions were modeled using qpAdm 12 , specifying Mesolithic Western European hunter-gatherers (WHG), Eastern hunter-gatherers (EHG) and early Neolithic Anatolian farmers (Barcın), as possible ancestral source populations. We present the model with the lowest number of source populations that fits the data, as well as the model with all three admixture components (see Supplementary Table 6 ). When estimating the admixture proportions for WHGs and EHGs, the test sample was excluded from their respective reference populations. MetaPhlan We used MetaPhlan2 13 to create a metagenomic profile based on the non-human reads (Supplementary Data 4 ). The reads were first aligned to the MetaPhlan2 database 13 using Bowtie2 v2.2.9 aligner 68 . PCR duplicates were removed using PALEOMIX filteruniquebam 58 . For cross-tissue comparisons 689 human microbiome profiles published in the Human Microbiome Project Consortium 14 were initially used, comprising samples from the mouth ( N = 382), skin ( N = 26), gastrointestinal tract ( N = 138), urogenital tract ( N = 56), airways and nose ( N = 87). The oral HMP samples consist of attached/keratinised gingiva ( N = 6), buccal mucosa ( N = 107), palatine tonsils ( N = 6), tongue dorsum ( N = 128), throat ( N = 7), supragingival plaque ( N = 118), and subgingival plaque ( N = 7). Pairwise ecological distances among the profiles were computed at genus and species level using taxon relative abundances and the vegdist function from the vegan package in R 69 . These were used for principal coordinate analysis (PCoA) of Bray–Curtis distances in R using the pcoa function included in the APE package 70 . Subsequently, we calculated the average relative abundance of each genus for each of the body sites present in the Human Microbiome Project and visualised the abundance of microbial orders of our sample and the HMP body sites with MEGAN6 15 . MALT To further characterise the metagenomic reads we performed microbial species identification using MALT v. 0.4.1 (Megan ALignment Tool) 16 , a rapid sequence-alignment tool specifically designed for the analysis of metagenomic data. All complete bacterial ( n = 12,426) and viral ( n = 8094) genomes were downloaded from NCBI RefSeq on 13 November 2018, and all complete archaeal ( n = 280) genomes were downloaded from NCBI RefSeq on 17 November 2018 to create a custom database. In an effort to exclude genomes that may consist of composite sequences from multiple organisms, the following entries were excluded: GCF_000922395.1 uncultured crAssphage GCF_000954235.1 uncultured phage WW-nAnB GCF_000146025.2 uncultured Termite group 1 bacterium phylotype Rs-D17 The final MALT reference database contained 33,223 genomes and was created using default parameters in malt-build (v. 0.4.1). The sequencing data for the ancient pitch sample, two soil control samples and associated extraction and library blanks were de-enriched for human reads by mapping to the human genome (hg19) using BWA aln and excluding all mapping reads. Duplicates were removed with seqkit v.0.7.1 71 using the ‘rmdup’ function with the ‘–by-seq’ flag. The remaining reads were processed with malt-run (v. 0.4.1) where BlastN mode and SemiGlobal alignment were used. The minimum percent identity (–minPercentIdentity) was set to 95, the minimum support (–minSupport) parameter was set to 10 and the top percent value (–topPercent) was set as 1. Remaining parameters were set to default. MEGAN6 15 was used to visualise the output ‘.rma6’ files and to extract the reads assigned to taxonomic nodes of interest for our sample. A taxon table of the raw MALT output for all samples and blanks, as well as species level read assignments to bacteria, archaea and DNA viruses for the ancient pitch sample are shown in Supplementary Data 5 , where reads listed are the sum of all reads assigned to the species node, including reads assigned to specific strains within the species. Reads assigned to RNA viruses were not considered for further analyses, since our dataset consisted of DNA sequences only. Due to the limited number of reads assigned to archaeal species (Supplementary Data 5 ), we did not consider Archaea in downstream analyses of species identification. To validate the microbial taxa, we aligned the assigned reads to their respective reference genomes and examined the edit distances, coverage distributions, and post-mortem DNA damage patterns (see Supplementary Note 5 ). Pneumococcus analysis We reconstructed a S. pneumoniae consensus genome (Fig. 4 ) by mapping all reads assigned to S. pneumoniae by MALT 16 to the S. pneumoniae TIGR4 reference genome (NC_003028.3). To investigate the presence of multiple strains we estimated the number of heterozygous sites using samtools 57 mpileup function, filtering out transitions, indels, and sites with a depth of coverage below 10. Coverage statistics of the individual alignments (MQ ≥ 30) were obtained using Bedtools 72 and plotted using Circos 73 in 100 bp windows. Mappability was estimated using GEM2 74 using a k-mer size of 50 and a read length of 42, which is comparable to the average length of the trimmed and mapped reads in the ancient pitch. Virulence genes were identified by assembling the ancient S. pneumoniae MALT extracts into contigs using megahit 75 . The contigs were aligned against known S. pneumoniae TIGR4 virulence genes in the Virulence Factor Database 17 (downloaded 22/11–2018) using BLASTn 76 . Only unique hits with a bitscore >200, >20% coverage, and an identity >80% were considered as shared genes (Supplementary Data 6 ). To identify all streptococcus virulence factors in the ancient pitch, we aligned the contigs against the full Virulence Factor Database 17 (downloaded 22/11–2018) using BLASTn 76 and the same filtering criteria as described above (Supplementary Data 6 ). To validate the approach we repeated the analysis with five modern oral microbiome samples (SRS014468; SRS019120; SRS013942; SRS015055; SRS014692) from the Human Microbiome Project (HMP) 14 using only the forward read (R1) (Supplementary Data 6 ). We find that the number of virulence genes we recovered directly correlates with sequencing depth (Supplementary Fig. 16 ). Holi For a robust taxonomic assignment of reads aligning to Metazoa (animals) and Viridiplantae (plants), all non-human reads were parsed through the ‘Holi’ pipeline 18 , which was specifically developed for the taxonomic profiling of ancient metagenomic shotgun reads. Each read was aligned against the NCBI’s full Nucleotide and Refseq databases (downloaded November 25th 2018), including a newly sequenced full genome of European hazelnut ( Corylus avellana , downloaded April 10th 2019) 77 . The alignments were then parsed through a naive lowest common ancestor algorithm (ngsLCA) based on the NCBI taxonomic tree. Only taxonomically classified reads for taxa comprising ≥1% of all the reads within the two kingdoms and a declining edit distance distribution after edit distance 0 were parsed for taxonomic profiling and further validation. To validate the assignments, we aligned the assigned reads to their respective reference genomes and examined the edit distances, coverage distributions, and post-mortem DNA damage patterns (see Supplementary Note 5 ; Supplementary Data 7 ). Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The sequencing reads are available for download from the European Nucleotide Archive under accession number PRJEB30280. All other data are included in the paper or available upon request.

Danish scientists have managed to extract a complete human DNA sample from a piece of birch pitch more than 5,000 years old, used as a kind of chewing gum, a study revealed Tuesday. The Stone-Age sample yielded enough information to determine the source's sex, what she had last eaten and the germs in her mouth. It also told them she probably had dark hair, dark skin and blue eyes. And genetically, she was more closely related to hunter-gatherers from the mainland Europe than to those living in central Scandinavia at the time, they concluded. "It is the first time that an entire ancient human genome has been extracted from anything other than human bones," Hannes Schroeder of the University of Copenhagen, told AFP. Schroeder is co-author of the study, which was published in the review Nature Communications. They found the sample during an archaeological dig at Syltholm, in southern Denmark, said Tehis Jensen, one of the other authors. "Syltholm is completely unique," he said. "Almost everything is sealed in mud, which means that the preservation of organic remains is absolutely phenomenal." Artistic reconstruction of Lola. Credit: Tom Björklund The researchers also recovered traces of plant and animal DNA—hazelnut and duck—confirming what archaeologists already know about the people who lived there at the time. But they were not sure why their subject chose to chew the bark: whether to turn it into a kind of glue, to clean her teeth, to stave off hunger—or simply as chewing gum.

10.1038/s41467-019-13549-9

Medicine

HIV treatment might boost susceptibility to syphilis, say researchers

A double edged sword: does highly active antiretroviral therapy contribute to syphilis incidence by impairing immunity to Treponema pallidum? Sexually Transmitted Infections, sti.bmj.com/lookup/doi/10.1136 … sextrans-2016-052870 Editorial: Syphilis and HIV: is HAART at the heart of this epidemic? Sexually Transmitted Infections, sti.bmj.com/lookup/doi/10.1136 … sextrans-2016-052940 Journal information: Sexually Transmitted Infections

http://sti.bmj.com/lookup/doi/10.1136/sextrans-2016-052870

https://medicalxpress.com/news/2017-01-hiv-treatment-boost-susceptibility-syphilis.html

Abstract Background and hypothesis Recently, the world has experienced a rapidly escalating outbreak of infectious syphilis primarily affecting men who have sex with men (MSM); many are taking highly active antiretroviral therapy (HAART) for HIV-1 infection. The prevailing hypothesis is that HAART availability and effectiveness have led to the perception among both individuals who are HIV-1 infected and those who are uninfected that HIV-1 transmission has become much less likely, and the effects of HIV-1 infection less deadly. This is expected to result in increased sexual risk-taking, especially unprotected anal intercourse, leading to more non-HIV-1 STDs, including gonorrhoea, chlamydia and syphilis. However, syphilis incidence has increased more rapidly than other STDs. We hypothesise that HAART downregulates the innate and acquired immune responses to Treponema pallidum and that this biological explanation plays an important role in the syphilis epidemic. Methods We performed a literature search and developed a mathematical model of HIV-1 and T. pallidum confection in a population with two risk groups with assortative mixing to explore the consequence on syphilis prevalence of HAART-induced changes in behaviour versus HAART-induced biological effects. Conclusions and implications Since rising syphilis incidence appears to have outpaced gonorrhoea and chlamydia, predominantly affecting HIV-1 positive MSM, behavioural factors alone may be insufficient to explain the unique, sharp increase in syphilis incidence. HAART agents have the potential to alter the innate and acquired immune responses in ways that may enhance susceptibility to T. pallidum . This raises the possibility that therapeutic and preventative HAART may inadvertently increase the incidence of syphilis, a situation that would have significant and global public health implications. We propose that additional studies investigating the interplay between HAART and enhanced T. pallidum susceptibility are needed. If our hypothesis is correct, HAART should be combined with enhanced patient management including frequent monitoring for pathogens such as T. pallidum . SYPHILIS HIV ANTERETROVIRAL THERAPY MATHEMATICAL MODEL INFECTION This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: googletag.cmd.push(function() { googletag.display("dfp-ad-mpu"); }); Statistics from Altmetric.com See more details Picked up by 23 news outlets Blogged by 2 Tweeted by 36 On 2 Facebook pages 107 readers on Mendeley Linked Articles Request Permissions If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways. ?xml version="1.0" encoding="UTF-8" ? Request permissions SYPHILIS HIV ANTERETROVIRAL THERAPY MATHEMATICAL MODEL INFECTION sextrans-2016-052870 Share Play Video Play Mute Current Time 0:00 / Duration Time 4:40 Loaded : 0% 0:00 Progress : 0% 0:00 Progress : 0% Stream Type LIVE Remaining Time -4:40 Playback Rate 1 Chapters Chapters descriptions off , selected Descriptions subtitles off , selected Subtitles captions settings , opens captions settings dialog captions off , selected Captions en (Main) , selected Audio Track Fullscreen This is a modal window. Caption Settings Dialog Beginning of dialog window. Escape will cancel and close the window. Text Color White Black Red Green Blue Yellow Magenta Cyan Transparency Opaque Semi-Transparent Background Color Black White Red Green Blue Yellow Magenta Cyan Transparency Opaque Semi-Transparent Transparent Window Color Black White Red Green Blue Yellow Magenta Cyan Transparency Transparent Semi-Transparent Opaque Font Size 50% 75% 100% 125% 150% 175% 200% 300% 400% Text Edge Style None Raised Depressed Uniform Dropshadow Font Family Proportional Sans-Serif Monospace Sans-Serif Proportional Serif Monospace Serif Casual Script Small Caps Defaults Done Close Modal Dialog This is a modal window. This modal can be closed by pressing the Escape key or activating the close button. Close Modal Dialog This is a modal window. This modal can be closed by pressing the Escape key or activating the close button. Video abstract Introduction STDs, HIV-1 and men who have sex with men In British Columbia (BC), Canada, from 2005 to 2014, infectious syphilis case reports (primary, secondary and early latent) rose 90.6% (288–549), chlamydia 39.9% (9540–13 348) and gonorrhoea 63.8% (1100–1802), corresponding to rate changes per 100 000 population of +72.4%, +33.9% and +47.6%, respectively. 1 Female syphilis cases decreased while male cases jumped from 202 to 524, accounting for 95% of all 2014 cases. Men who have sex with men (MSM) accounted for 60.4% (122) of cases in 2005, rising to 88.9% (466) in 2014. There was a fourfold increase in MSM syphilis from 2010 to 2014 (115–466 cases), including 112 reinfections (24%) in 2014. The HIV-1 coinfection rate in MSM during this period was 50%–75%. From 2005 to 2014, the male infectious syphilis rate increased 235% (9.7–22.8) compared with a much smaller increase for chlamydia (56.9%; 141.7–222.4) and gonorrhoea (42.1%; 39.7–56.4) in men. Comparatively larger increases in syphilis cases were also observed in the USA from 2005 to 2014. Primary and secondary syphilis cases rose 128.1% (8724–19 999), chlamydia 47.7% (976 445–1 441 789) and gonorrhoea 3.1% (339 593–350 062), corresponding to rate changes of +117.2%, +38.5% and −3.4%, respectively. 2 Males contributed an increasing proportion of syphilis cases, accounting for 91% in 2014. From 2007 to 2014, among the 27 US states that collected sex partner data for ≥70% of males, MSM cases increased steadily to 82.9%. In Los Angeles County between January and May, 2010, 76% of 537 early syphilis cases were MSM and ≥58% of these were HIV-1 positive. 3 HIV-1 infection in US MSM has also been statistically associated with repeat syphilis infection. 4 In the UK from 1998 to 2013, new syphilis cases in MSM rose from 23 (26.0% of all cases) to 2546 (71.3% of all cases). From 2009 to 2013 in England, the odds of being diagnosed with syphilis increased from 2.71 (95% CI 2.41 to 3.05, p<0.001) to 4.05 (95% CI 3.70 to 4.45, p<0.001) in HIV-1 positive relative to HIV-1 negative/undiagnosed MSM. 5 Results Highly active antiretroviral therapy and syphilis First-line highly active antiretroviral therapy (HAART) regimens may comprise (1) two nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs), commonly tenofovir (TDF) and emtricitabine (FTC) or lamivudine (3TC), plus a non-nucleoside reverse transcriptase inhibitor (NNRTI), commonly efavirenz, (2) two NRTIs and an integrase strand-transfer inhibitor (InSTI) or (3) two NRTIs and a boosted protease inhibitor (PI). PIs, InSTIs and fusion/entry inhibitors are used as second-line and third-line alternative agents. In BC, 86.1% of patients take two NRTIs with a boosted PI (34%), NNRTI (30%), InSTI (19%) or fusion/entry inhibitor (1%). 6 Common NRTIs include TDF (64%), FTC (59%) and 3TC (40%). An additional 11% (741 patients) take InSTIs in other HAART combinations. In total, 30% (2056) take InSTIs. HAART usage has grown steadily including a 288% increase in InSTI usage from 2010 to 2015 (530–2056 patients). Because HAART stimulates and supports immune system recovery from HIV-1-related immunosuppression, one might expect HAART to be associated with declining infection rates for most pathogenic organisms. However, several studies support the hypothesis that HAART may be associated with syphilis acquisition. Receiving HAART (adjusted HR=1.81 (95% CI 1.25 to 2.62, p<0.002)), older age and MSM status were independent risk factors for syphilis seroconversion by multivariate logistic regression analysis in 1010 people who were infected with HIV in Northeast China from 2009 to 2013. 7 Using Poisson regression analysis, Park and coauthors 8 found that the period-specific incidence rate of early syphilis in 539 patients receiving HAART in Korea significantly increased in proportion to the years after starting HAART (p<0.001). These two studies were conducted in stable cohorts but they did not control for specific sexual risk behaviours. Additional studies have reported a significant proportion of new syphilis cases in persons receiving HAART, including 32.7% of all 1089 new syphilis cases in France from 2000 to 2003 and 71% of the 502 cases who knew their HIV-1 positive status. 9 Among 104 early syphilis and 36 late or indeterminate latent syphilis cases in a Malaga study from 2004 to 2011, 65 of 85 (76.4%) with prior known HIV-1 infection were taking HAART. 10 A retrospective, descriptive study at the University of Alabama found 40 incident syphilis cases from 2004 to 2007 in 1544 patients who were HIV-1 positive. 11 Two-thirds were receiving HAART when syphilis was diagnosed including all five patients with primary syphilis. The proportion of the entire cohort on HAART was not reported. Among 3448 patients followed in a Parisian Hospital Infectious Disease Service from January 2000 to December 2002, 48 of 71 (67.6%) patients with a new diagnosis of syphilis were taking HAART. 12 For BC, over 8000 males have tested HIV-1 positive, and HAART usage in MSM is more than 80%. 1 , 6 However, the percentage taking HAART when syphilis was diagnosed is unknown. Behavioural change One explanation for high rates of syphilis in MSM is increased risky sex, often in the context of optimistic risk perception. Similarly, any association between syphilis and HAART might be a surrogate marker for risky sex in individuals treated with HAART rather than an effect of HAART itself. Seroadaptive behaviours to prevent HIV-1 transmission, such as serosorting (ie, selective unprotected sex with partners of the same HIV status), have also been implicated in increasing STD incidence including syphilis. 13 What then is the evidence for HIV-1 treatment optimism and behavioural change in MSM and persons on HAART? Several studies support treatment optimism including a 2013 literature review, which concluded that quantitative studies were ‘largely in support’ of an association between optimistic beliefs and HIV-1 transmission risk. 14 However, other studies have shown no change or decreased risky behaviour and/or no lessening of risk perception, as discussed later in this section. Many HIV-1-infected MSM believe they have a responsibility to protect their sex partners, 15 and many eliminate or reduce HIV-1 transmission behaviours after HIV-1 diagnosis. A 2005 US meta-analysis concluded that high-risk sexual behaviour with partners who were HIV-1 negative was significantly reduced (68%, 95% CI 59% to 76%, p<0.001) after HIV-1 diagnosis. 16 Persson used the Swiss Consensus Statement that people on effective HIV-1 treatment cannot transmit HIV-1 as a surrogate for HIV-1 treatment optimism in interviews of HIV-1 discordant couples. Participants were highly sceptical of the Statement's prevention message and not one described it as having any direct relevance to their sexual decision making. 17 Much of the HIV-1 treatment optimism literature focuses on behavioural change in persons on HAART and subsequent to HAART initiation. In 512 patients receiving HAART in Bangkok, unprotected sex risk was found in only 27 patients (5%) and multivariate analysis showed no association with beliefs about HIV-1 transmission while taking HAART. 18 A 2009 meta-analysis of unprotected anal intercourse (UAI) among HIV-1 diagnosed US MSM found no association with HAART 15 and a cross-sectional study of 420 London men who were HIV-1 positive showed that men on HAART had fewer sexual partners and less UAI. 19 Among 456 HIV-1 positive US MSM, Remien et al 20 found no increased sexual risk behaviour and substantial ongoing perception of HIV-1 transmission risk while on HAART. Using data from a prospective behavioural study nested in a randomised controlled trial of early HAART (Temprano), Jean et al 21 found significant decreases over 24 months in sexual activity (OR 0.72, 95% CI 0.57 to 0.92), multiple partnerships (OR 0.57, 95% CI 0.41 to 0.79), unprotected sex (OR 0.59, 95% CI 0.47 to 0.75) and risky sex (OR 0.58, 95% CI 0.45 to 0.76). We could find only one study that found a link between increasing syphilis incidence and seroadaptive behaviours such as serosorting in individuals treated with HAART with unsuppressed viral load, but this would not explain rising syphilis incidence out-of-proportion to other STDs. 13 Some studies have implicated core groups in this syphilis outbreak, 22 while other studies have shown little or no significant increase in risky sex, number of contacts or concurrency among core group members. In a retrospective study of syphilis among MSM in San Francisco, number of sex partners, illicit substance use, partner meeting venues and commercial sex were not associated with repeat syphilis infection. 4 In the Alabama, Bangkok and Temprano cohorts referenced earlier, similar core group characteristics were not associated with incident syphilis, unprotected sex risk or multiple partnerships, respectively. 11 , 18 , 21 Finally, a mathematical model of behaviour in a core group predicted a transient spike in unprotected sex that was counteracted by other pathways on longer timescales, leading to lower rates of unprotected sex on the whole. 23 Modelling studies conducted herein to test our hypothesis are in support of both behavioural change and HAART treatment being able to increase syphilis prevalence, with the combined effect being more than additive ( figure 1 ). Specifically, we developed a mathematical model of HIV-1 and Treponema pallidum coinfection in a population with two risk groups and assortative mixing between groups. Susceptible individuals acquire infection from a partner who is infected with a fixed probability per sexual encounter. In the case of HIV-1, the infection probability is lower if the partner who is HIV-1 positive is on HAART versus untreated, but it is higher if the partner who is either HIV-1 negative or HIV-1 positive is already infected with T. pallidum versus uninfected. Individuals who are infected receive treatment at a constant rate dependent on the pathogen. Individuals lose their treated status at a disease-dependent rate. Individuals infected with HIV-1 have a disease-imposed mortality rate that is lower for individuals on HAART. Download figure Open in new tab Download powerpoint Figure 1 A susceptible-infective-treated model of HIV-1 and Treponema pallidum epidemiology predicts that both highly active antiretroviral therapy (HAART) and behavioural change substantially boost syphilis prevalence above baseline. The model used two risk groups and assortative mixing between groups. Using baseline parameters from the literature, 24 we varied parameters quantifying HAART effects and partnership formation rates on susceptibility to and transmission of T. pallidum . We introduced HAART at 20 years and plotted syphilis prevalence without (left panel) and with (right panel) a threefold effect of HAART on susceptibility to T. pallidum , under different assumptions about the effect of HAART on partnership formation rates (for details and code, see ). Compared with the baseline of no effect (left panel, black line), the combined effect of increased susceptibility and behavioural change (right panel, coloured lines) is larger than the sum of the effects of behavioural change alone (left panel, coloured lines) and increased susceptibility alone (right panel, black line). We numerically simulated the model using baseline parameters taken from the literature. 24 We initialised each simulation with a small number of individuals who are infected and introduced HAART 20 years later. We explored effects on syphilis prevalence of HAART-induced changes in behaviour versus immunology. For behavioural change, we assumed that individuals on HAART adopt more risky sexual behaviours by increasing their partnership formation rates, which increases their ability to transmit T. pallidum . For immunological change, we assumed that individuals on HAART have a higher susceptibility to T. pallidum than individuals who are untreated. The simulation results show that either behavioural ( figure 1 , left panel, blue and red lines) or immunological ( figure 1 , right panel, black line) change alone can produce syphilis outbreaks with peak prevalence that is substantially higher than baseline. Strikingly, the peak prevalence of the syphilis outbreak produced by both behavioural and immunological changes ( figure 1 , right panel, blue and red lines) is larger than the sum of the peaks of outbreaks produced independently by either type of change ( figure 1 , left panel, blue and red lines; right panel, black line). Therefore, the immunological effects of HAART and HAART-induced behavioural change can in principle act synergistically to increase syphilis prevalence by amounts comparable with that observed in the ongoing outbreak. Discussion Biological plausibility Protection against the extracellular pathogen T. pallidum , the causative agent of syphilis, is dependent upon T cell expansion and the generation of an early Th1-stimulating, interferon γ (IFN γ)-producing host proinflammatory response that potentiates the primary clearance mechanism of T. pallidum , macrophage-mediated opsonophagocytosis. 25 The latter process is dependent on unperturbed mitochondrial function to ensure peak metabolic activity within macrophages, 26 opsonic antibody production and IFN γ-mediated macrophage activation. 27 Opsonic antibody quality is reduced in individuals infected with HIV-1 28 and certain HAART agents significantly suppress mitochondrial function, 26 the proinflammatory response 29 and macrophage activation, 30 leading to reduced treponemal clearance via opsonophagocytosis. InSTIs have been shown to suppress the proinflammatory response in cohort studies 29 and opsonophagocytosis is reduced in vitro following treatment of macrophages with NRTIs, consistent with mitochondrial damage. 26 Further, the well-documented depletion of CD4 + memory T cells in individuals infected with HIV-1 30 would enhance their susceptibility to syphilis reinfection. NRTIs, especially TDF, have been shown to inhibit telomerase activity leading to accelerated shortening of telomerase length in peripheral blood mononuclear cells (PBMCs), 31 which may lead to the accumulation of replicative senescent cells 32 with limited ability to protect against pathogens such as T. pallidum . Reciprocally, upregulation of monocyte expression of CCR5 receptors by treponemal lipoproteins enhances the susceptibility of monocytes to HIV-1 infection, 33 further weakening the innate and adaptive immune responses to T. pallidum . Collectively, these observations provide viable explanations for (1) an enhanced susceptibility of individuals infected with HIV-1, especially those on HAART, to syphilis infection and reinfection and (2) higher syphilis incidence among individuals treated with HAART compared with chlamydia and gonorrhoea, infections caused by pathogens that are less reliant on opsonophagocytosis for clearance. Potential approach for hypothesis testing HAART decreases the proinflammatory response in patients infected with HIV-1, 29 but this may result from an HAART-induced reduction of lymphocytes' ability to upregulate inflammatory markers, in line with our hypothesis, or from the disappearance of a major cause of inflammation (ie, a fall in HIV-1 viral load). These different effects can be teased apart using experiments in macaques, which have been used as models for human infection with T. pallidum 34 and develop AIDS-like disease following simian immunodeficiency virus (SIV) infection. T. pallidum challenge after specific HAART administration with or without prior SIV infection can provide insight into how HAART affects the proinflammatory response and syphilis susceptibility in the absence or presence of retrovirus. Prospective cohort studies can compare opsonophagocytic activity of macrophages and susceptibility to syphilis among healthy individuals who are HIV-1 negative receiving preventative pre-exposure prophylaxis (PrEP), healthy controls, untreated patients with HIV-1 and patients with HIV-1 receiving the same HAART as healthy individuals. To determine the effect of HAART on CD4 + memory T cells, one can measure the number of CD4 + memory T cells by flow cytometry in PBMCs of persons on and off HAART with particular attention to TDF. A retrospective case–control and/or a prospective cohort study comparing the prevalence and epidemiological features of infectious syphilis cases among patients who are HIV-1 positive and treated with HAART, patients who are HIV-1 positive and untreated and patients who are HIV-1 negative, including the usage of specific HAART agents, would be enlightening. Syphilis databases without HAART information could be linked to treatment databases to delineate overlapping factors including HAART usage. Conclusion Clinicians and researchers typically view HAART suppression of the proinflammatory response positively because it decreases HIV-1-associated pathological sequelae. However, this immune dampening may have detrimental effects including enhanced susceptibility to infection with T. pallidum and the occurrence of unusual clinical manifestations such as ocular syphilis. The possibility also exists that HAART-mediated immune dampening may predispose an individual to other conditions that are non-infectious in origin and depend upon a particular immune response for control, including certain types of cancer. In this regard, it is of interest to note that HAART treatment has been suggested to be associated with a higher risk of anal cancer and, potentially, other non-AIDS-defining cancers. 35 Overall, these findings suggest a possible link between HAART and an increased risk for selected diseases of infectious and non-infectious origin, a potential unforeseen consequence that warrants further study. If borne out, it will be imperative that the highly exciting and efficacious global implementation of PrEP be carried out with awareness towards the potential need for enhanced patient management. Key messages The number of new syphilis cases has risen dramatically worldwide in recent years, with repeat infections within a single individual a frequent occurrence. The syphilis outbreak has outpaced new cases of chlamydia and gonorrhoea and is primarily affecting men who have sex with men and who are taking highly active antiretroviral therapy (HAART). Mathematic modelling suggests both behavioural change and HAART contribute to the increasing syphilis prevalence, with the combined effect being more than additive. If experimentally confirmed, HAART administration and pre-exposure prophylaxis implementation will require enhanced patient management guidelines to mitigate the increased risk of developing selected infectious and non-infectious diseases.

The antiretroviral drugs used to treat HIV infection might inadvertently be boosting gay/bisexual men's susceptibility to the bacteria responsible for syphilis, Treponema pallidum, conclude researchers in the journal Sexually Transmitted Infections. This might explain why new and repeat cases in these groups have risen so sharply compared with other sexually transmitted infections over the past decade, they suggest. The availability of highly active antiretroviral therapy (HAART) for the treatment of HIV infection has meant that HIV is no longer the automatic death sentence that it once was. The absence of the fear factor is thought to have prompted higher rates of sexually transmitted infections as a result of risky, unprotected sex. But it's not clear why rates of syphilis among gay/bisexual men should be so much higher than those of chlamydia or gonorrhoea, especially as HAART boosts immune system activity, and so would be expected to lower susceptibility to infections, say the Canadian and South African researchers. They therefore reviewed the available evidence on the impact of HAART on behavioural and immune system change to come up with a numerical analysis to explore which might affect the prevalence of the infection the most. They used two risk 'models' to test the likelihood of syphilis infection: one (lower risk) which compared HAART with no treatment in an HIV positive partner; and the other (higher risk) which compared existing infection with no infection in a partner who was either HIV negative or positive. Behaviour change was taken to mean that HAART would result in more sexual partners; and immune system changes were taken to mean that HAART would boost susceptibility to Treponema pallidum. The calculations showed that either factor could produce outbreaks of syphilis that would be substantially higher than expected, but that both factors combined produced a peak in the number of infections that was greater than that associated with either factor alone—and equivalent to the sorts of figures seen in the current outbreak. This suggests that there's an interplay between behavioural change and immune system changes, say the researchers, who offer a possible explanation for a biological effect on the immune system. The body's clearance of T pallidum relies on an increase in the number of an immune system cell called a T cell and a cascade of chemicals that stimulate an inflammatory response. HAART dampens down these activities. Clearance of chlamydia and gonorrhoea is less reliant on these processes, say the researchers. The researchers also refer to other relevant associations, such as the increased risk of certain types of cancer that have been linked to HAART. "Overall, these findings suggest a possible link between HAART and an increased risk for selected diseases of infectious and non-infectious origin, a potential unforeseen consequence that warrants further study," they write. In a linked editorial, Drs Susan Tuddenham, Maunank Shah, and Khalil Ghanem from Johns Hopkins University School of Medicine, Baltimore, caution that the rise in syphilis cases may simply reflect inadequate screening for chlamydia and gonorrhoea. They also point to previous outbreaks of syphilis in straight men and women in the 1980s and 1990s before the advent of HAART. Similarly, they suggest that the calculations used by the researchers don't take account of the complexities of sexual relationships, such as relationship length or the number of concurrent partners, or decreased use of condoms in long term relationships, all of which might influence infection risk. But despite these caveats, they describe the researchers' hypothesis as "intriguing," and one that "warrants careful consideration." They continue: "We are living in an era where [antiretroviral therapy] is being used to effectively treat and prevent HIV infection. To some extent this seems to have tempered the urgency to control other [sexually transmitted infections]. As history has shown many times over, that would be a costly mistake." And they conclude: "Over the past 15 years, syphilis rates among [men who have sex with men] have been rising unabated. We are not aware of any recent intervention that has led to a sustained decline in [these rates] in this population. "If further investigations support a role for [antiretroviral therapy] in increasing susceptibility to syphilis, this will provide one more reason why screening, diagnosis and treatment of [sexually transmitted infections] in [men who have sex with men] must be prioritised."

sti.bmj.com/lookup/doi/10.1136 … sextrans-2016-052870

Chemistry

Researchers show RNA ribozymes able to cooperate to reassemble themselves

Spontaneous network formation among cooperative RNA replicators, Nature (2012) doi:10.1038/nature11549 Abstract The origins of life on Earth required the establishment of self-replicating chemical systems capable of maintaining and evolving biological information. In an RNA world, single self-replicating RNAs would have faced the extreme challenge of possessing a mutation rate low enough both to sustain their own information and to compete successfully against molecular parasites with limited evolvability. Thus theoretical analyses suggest that networks of interacting molecules were more likely to develop and sustain life-like behaviour. Here we show that mixtures of RNA fragments that self-assemble into self-replicating ribozymes spontaneously form cooperative catalytic cycles and networks. We find that a specific three-membered network has highly cooperative growth dynamics. When such cooperative networks are competed directly against selfish autocatalytic cycles, the former grow faster, indicating an intrinsic ability of RNA populations to evolve greater complexity through cooperation. We can observe the evolvability of networks through in vitro selection. Our experiments highlight the advantages of cooperative behaviour even at the molecular stages of nascent life. Journal information: Nature

http://dx.doi.org/10.1038/nature11549

https://phys.org/news/2012-10-rna-ribozymes-cooperate-reassemble.html

Abstract The origins of life on Earth required the establishment of self-replicating chemical systems capable of maintaining and evolving biological information. In an RNA world, single self-replicating RNAs would have faced the extreme challenge of possessing a mutation rate low enough both to sustain their own information and to compete successfully against molecular parasites with limited evolvability. Thus theoretical analyses suggest that networks of interacting molecules were more likely to develop and sustain life-like behaviour. Here we show that mixtures of RNA fragments that self-assemble into self-replicating ribozymes spontaneously form cooperative catalytic cycles and networks. We find that a specific three-membered network has highly cooperative growth dynamics. When such cooperative networks are competed directly against selfish autocatalytic cycles, the former grow faster, indicating an intrinsic ability of RNA populations to evolve greater complexity through cooperation. We can observe the evolvability of networks through in vitro selection. Our experiments highlight the advantages of cooperative behaviour even at the molecular stages of nascent life. Main The ‘RNA world’ is a plausible stage in the development of life because RNA simultaneously possesses evolvability and catalytic function 1 . An RNA organism that could evolve in such a fashion is likely to have been one of the Earth’s first life forms. A search is underway 2 , 3 for an RNA autoreplicase that relies on its individual genotype to compete for survival and reproduction by Darwinian-type evolution in a fitness landscape. Yet the transition from a prebiotic chemistry to this stage of life is not understood. Several authors have proposed that the most primitive life thrived less on discrete genotypes and instead on collections of molecular types more subject to systems chemistry than to straightforward selection dynamics 4 , 5 , 6 , 7 , 8 , 9 . In particular, it was suggested that webs of functionally linked, genetically related replicators were required in the earliest phases of life’s appearance to prevent informational decay (the so-called error catastrophe) 4 , 10 , 11 , 12 . An empirical demonstration of RNA replicator networks could illuminate critical features of this early stage of life. Ribozymes are good candidates for this because they can evolve outside of an organismal context, construct other RNAs, exhibit self-sustained reproduction, and explore sequence space in efficient ways 13 , 14 , 15 . However, their ability to form catalytic networks capable of expanding as predicted from theory has not yet been shown, despite the observation that collections of nucleic acids have the potential to manifest complexity 6 , 16 . Simulations show that molecular networks should arise, evolve and provide a population with resistance against parasitic sequences 8 . These results are robust within structured environments such as cells or on grids, but are less so in a solution phase. Recent experimental work in vitro has been very successful at demonstrating simple ecologies 17 , 18 , 19 , reciprocity between two species 6 , 16 , 20 , and sustained exponential growth via cross catalysis 15 . Empirical efforts to date have been limited by an inability to expand past reciprocal interactions between two species to prebiotically relevant systems that have the capacity to increase their complexity by expanding to three, and then more, members 18 , 21 . Specifically, the use of systems in which the recognition domain in the catalyst and the target domain in the substrate are co-located in each replicator has prevented networks of more than two members from forming. If this molecular feature could be circumvented, larger networks could be realized within RNA populations in the test tube and help demonstrate a potential escape from the error catastrophe problem that tends to plague selfish systems. The Azoarcus ribozyme system The ∼ 200-nucleotide (nt) Azoarcus group I intron ribozyme 22 can be broken into fragments that can covalently self-assemble by catalysing recombination reactions in an autocatalytic fashion 23 , 24 ( Supplementary Fig. 1 ). By allowing variation in the sequence recognition mechanism by which this assembly occurs, which is provided by the 3-nt internal guide sequence (IGS) at the 5′ end of the ribozyme, many such autonomously self-assembling ribozymes become possible. We sought to determine if these ribozymes could display cooperative behaviour if their IGS sequences target the assembly of other ribozymes, but not themselves. To create a cooperative network, we fragmented the Azoarcus ribozyme into two pieces in three different ways with the intent of observing how they could spontaneously reassemble via intermolecular cooperation ( Fig. 1a, b ). We manipulated the IGS (canonically GUG) and its target triplet to generate both matched and mismatched partners. We mixed various IGS and target pairs in two-piece constructs to test the ability of mismatched pairs to promote self-assembly ( Supplementary Fig. 2 ). From these data, we chose three mismatched pairs that exhibit relatively little autocatalysis: GUG/CGU, GAG/CAU, and GCG/CUU. These crippled pairs are denoted I 1 , I 2 and I 3 , respectively, meaning that they are informational subsystems, albeit weakly autocatalytic. Figure 1: Cooperative covalent assembly of recombinase ribozymes. a , Design of recombinase ribozymes capable of spontaneous cooperative covalent assembly from fragments. The Azoarcus ribozyme 25 can be broken at three loop regions to obtain four oligomers capable of self-assembling into a full-length molecule 26 , 27 . The grey box in W (magenta) is the internal guide sequence (IGS), whereas those at the 3′ ends of the W , X (lime) and Y (blue) fragments are recombination targets (tags) recognized by the IGS, which guides the catalysis of a covalent closure (•) of the loops. b , A cooperative system comprised of three subsystems, each created from partitioning the molecule into two pieces at different junctions: I 1 ( W + h • X • Y • Z ), I 2 ( W • X + h • Y • Z ) and I 3 ( W • X • Y + h • Z ). Numbers over arrows estimate the cooperative advantage for each step (see text). c , Electrophoretic observation of assemblies of E 2 and E 3 . The 5′ fragments of I 2 or I 3 were independently 5′-radiolabelled with 32 P (that is, *I 2 or *I 3 ). The reactions were performed by incubating 0.5 μM (for autocatalysis) or 0.05 μM (for direct assembly, cross catalysis and cooperation) of each fragment for 8 h. Where appropriate, the arrows identify the subsystems being assembled by the previous subsystems in the network, where the IGS and recombination tags match. d , Yields of individual E i ribozymes over time, measured every 30 min for 16 h when all six I i RNA fragments are co-incubated at 0.05 μM. PowerPoint slide Full size image We chose the triplet pairs so that when the three subsystems are mixed together, they should constitute a cyclical cooperative network in which the output of one subsystem can catalyse the replication of the next one in the cycle ( Fig. 1b ). This occurs because the IGS of one subsystem is matched to the target in the next subsystem, and the physical separation of the IGS and its target allows for cycles of more than two members. When the six RNAs ( W , h • X • Y • Z , W • X , h • Y • Z , W • X • Y and h • Z ; • indicates covalent bonding) are allowed to fold together and be co-incubated in equimolar ratios, we expect the subsystems first to form non-covalent versions of ribozymes, and then catalyse the formation of covalent versions of the next ribozyme in the cycle. To test whether cooperation between enzymes occurred, we took several approaches. First, for the cycle to exhibit positive feedback 4 , there should be a distinct advantage to being a covalently contiguous ribozyme (E i ), as opposed to remaining fragmented (I i ). Once covalent ribozymes are formed, they should further promote synthesis of their target ribozymes, at faster rates than the non-covalent versions would. When we tested each in isolation, we found that the E i ribozymes recombined their respective target substrates into products 1.3–6.3-fold more than the I i versions when assayed separately ( Supplementary Fig. 3 ). Second, by examining each subsystem in isolation or in pairs, we could compare the relative strengths of autocatalysis (E i synthesizing E i ), cross-catalysis (E i +1 synthesizing E i ), and what should be the most efficient, direct catalysis (E i synthesizing E i +1 ). When we incubated just the two RNAs from any one subsystem, such as I 2 , alone, there is minimal synthesis of the corresponding ribozyme E 2 ; after a few hours roughly 0.1% of W • X is converted into W • X • Y • Z . This low background level of autocatalytic synthesis reflects residual catalytic activity available to a mismatched IGS and IGS target, for example GAG with CAU 25 , showing that each I i subsystem has severely limited information-replication potential in isolation. Likewise, when the four RNAs of two subsystems were co-incubated, the cross-catalytic synthesis of the ribozyme corresponding to the preceding subsystem in the cycle is similarly poor, again hindered by an IGS–IGS-target mismatch ( Fig. 1c ). After only 1 h of incubation, the yield of E 3 from 0.5 μM I 3 is 0.10 ± 0.02% (autocatalysis), and the yield of E 3 from 0.5 μM I 3 and 0.5 μM E 1 is 0.7 ± 0.06% (cross-catalysis), but the yield of E 3 from 0.5 μM I 3 and 0.5 μM I 2 is 13 ± 0.5% (direct catalysis) (data not shown; errors given as s.e.m.). These differences are all statistically significant as measured by t -tests several planned comparisons ( P < 0.001). From these data we determined that direct catalysis is significantly more efficient than catalysis resulting from mismatched IGS sequences and their targets. When all six RNAs of all three subsystems are co-incubated, cooperation causes the synthesis of W • X • Y • Z to rapidly escalate, as expected. The composite yield of full-length RNA after 16 h when I 1 , I 2 and I 3 are mixed is 125-fold higher than the sum of the yields of the three subsystems in isolation ( Supplementary Fig. 4 ). This enhancement can be readily visualized after shorter periods of time ( Fig. 1c ). Each subsystem grows at a different rate ( Fig. 1d ). The synthesis of E 3 by E 2 is more rapid than that of the other two ribozymes, presumably because the non-covalent version of the enzyme (I 2 ) is nearly as efficient as the covalent version (E 2 ); it could also be because certain IGS–IGS target pairs are more efficient 25 . Importantly, we can detect two-step (relayed) cooperativity by comparing the yields with and without the intervening enzyme. In the case of E 1 for example, after 4 h the increase in yield of E 1 upon addition of I 2 to I 1 with I 3 present is 2.5%, whereas the increase in yield of adding I 2 to I 1 without I 3 present is only 0.02%, showing the operation of E 2 through E 3 onto E 1 ( Supplementary Table 1 ); this is supported by doping experiments ( Supplementary Fig. 5 ). To observe the advantage of cooperation in another way, we constructed a control system in which the I i molecules could act as catalysts, but could not be covalently assembled themselves because their target sequences were not a match for any catalyst in the system ( Supplementary Fig. 5 ). Cooperation would be manifest when enzymes synthesize other enzymes, and there is some benefit to being covalent. Thus we measured the yields of W • X • Y • Z molecules at 8 h in this control system and in our normal system (that is, Fig. 1b ). The yields in the control system were consistently worse, and we calculated the ratio (E i catalysis + I i catalysis) to (I i catalysis only) as the advantage of being covalent in each leg of the cycle. These ratios, indicated above the coloured arrows in Fig. 1b , are 1.73, 1.02 and 1.22 for i = 1, 2 and 3, respectively. Assuming these values are multiplicative, the cooperative benefit is about 2.2 for the entire cycle. An impediment to truly hyperbolic growth for such a system 4 is the occasional formation of non-productive complexes (for example, W – Y • Z ) through partially complementary base pairing ( Fig. 1a ). We can detect such complexes ( Supplementary Fig. 6 ), but when they are minimized by pre-folding each RNA separately, the yield after 2 h increases by 25–50% ( Supplementary Fig. 7 ). As shown by heat–cool regimes, reverse reactions that have the net effect of breaking down covalent ribozymes into fragments may also have a small role in preventing hyperbolic growth ( Supplementary Fig. 8 ). Cooperation versus selfishness Next we tested whether a three-membered cooperative system has the potential to have higher fitness than purely autocatalytic systems when placed in direct competition ( Fig. 2 ). To construct ‘selfish’ autocatalytic subsystems (S i ), we reverted the IGS–IGS target pairs within each subsystem so that they would match. To create S 1 we used GUG W CAU and h • X • Y • Z , to create S 2 we used GAG W • X CUU and h • Y • Z , and to create S 3 we used GCG W • X • Y CGU and h • Z . Each of these subsystems replicates well in isolation. Upon mixing of RNAs, we tracked selfish and cooperative ribozymes by the composition (matched or mismatched, respectively) of the W -containing fragments because these contain the IGS and hence the most crucial genetic element ( Fig. 2a ). When we compared the total yield of S 1 + S 2 + S 3 to that of I 1 + I 2 + I 3 , the former out-performed the latter at all time points (that is, selfishness wins in isolation). One reason for this result is that there would be less time delay in initiating covalent synthesis in the all-selfish system. However, when we placed all six subsystems (12 RNAs: I 1 + I 2 + I 3 + S 1 + S 2 + S 3 ) in the same reaction, the relative yields at later times are reversed, and the growth of the enzymes resulting from the cooperative network now exceeds those from the selfish subsystems (that is, cooperation wins in competition). These results are independent of the exact RNA fragments we chose, as the same result can be seen in other systems with different IGS and IGS targets (see Supplementary Fig. 9 ). The yield reversal upon mixing happens because the selfish enzymes now participate in—and effectively expand—the cooperative network ( Supplementary Fig. 10 ). This would be a mechanism for a network connectivity increase when the subsystems involved are competing for at least one shared resource, in this case the catalytic core ( Y – Z ), because all W -containing fragments can use the same 3′ fragments. Whereas selfish enzymes can also benefit from the network, the asymmetry in the proficiencies of the various IGS–IGS-target pairings creates potential for an asymmetry in the relative benefits of the various enzymes in the mixed environment. This feature would have been common in primordial genetic systems, allowing us to posit that cooperation could have been predisposed even in homogeneously mixed environments. Figure 2: Cooperative chemistry out-competes selfish chemistry when directly competed. a , Empirical results using cooperative (I 1 , I 2 and I 3 , that is, Fig. 1b ) and selfish subsystems (S 1 , S 2 and S 3 , where IGS and IGS targets were changed to be matching in each subsystem). Yields of total W • X • Y • Z RNA tracked the concentrations of cooperative (mismatched) or selfish (matched) W -containing RNAs (0.05 μM initial concentrations) over time either when the cooperative (green) and selfish (red) sets of subsystems were incubated separately (dashed lines) or together in the same reaction mixture (solid lines; upper left inset). Data points are averages of three independent trials. Error bars show the standard error of the mean (s.e.m.), and the yields of the cooperative trials in the mixed experiment are significantly greater than those of the selfish trials at the 10- and 16-h time points ( P < 0.05 by t -tests using Sidák’s correction for multiple a posteriori comparisons). b , Simulation of growth dynamics using a toy model of the network of cooperation and selfish interactions (see Supplementary Information ). Cooperative enzymes fare better in competition than do selfish enzymes, as demonstrated empirically in panel a . PowerPoint slide Full size image Modelling Empirical systems such as the one described above are subject to the particularities of chemical and methodological idiosyncrasies, so we sought to generalize these results by constructing mathematical models that show that under a certain set of parameters, the laboratory results should indeed be possible. First we constructed an ordinary differential equation (ODE) model for the three-membered network shown in Fig. 1b . We tracked the yield of each of the three E i ribozymes separately—using three identical replicates from the same initial reaction mixture—by taking aliquots every 30 min for 16 h ( Fig. 1d ). We used standard optimization techniques to find the rate constants of all the possible reactions in Fig. 1b that produced trajectories in the ODE system closest to the observed data ( Supplementary Information ). We used these estimated rate constants to construct a second ODE model that would mimic the cooperative growth of the three subsystems. In general, the non-covalent versions of the ribozymes form relatively tight complexes, with K d values in the low nanomolar range. When we built cooperative behaviour into the model by relying on differential equations of type dE j /dt = k ij [I j ][E i ], the experimental data were fit very well in all three subsystems ( Supplementary Fig. 11 ). When we removed direct catalysis from the model and inserted only autocatalysis instead, the quality of the fit decayed substantially such that the root mean squared error was 2.4-fold greater ( Supplementary Fig. 12 ), confirming these results. These data support the contention that replication of the subsystems is indeed cooperative. Next we constructed a toy model comparing the cooperative and selfish behaviours seen in Fig. 2a using the dynamical relationships that can exist among all enzymes ( Fig. 2b ). The ‘selfish’ enzymes perform some altruistic catalysis when alternative substrates become available. The empirical data display more striking yield differences than the model, perhaps because the time delays in bringing the results of the selfish catalytic events back to the selfish subsystems are exacerbated by physical processes such as diffusion. Again this result is general, at least within this network topology, and does not depend on the particular IGS–IGS-target pairings chosen. In essence, although the selfish replicators can parasitize the cooperators, the cooperative network benefits more by incorporating the selfish RNAs. Interestingly, the opposite is generally true in evolutionary dynamics: groups of cooperative individuals grow more quickly than groups of selfish individuals, but a group consisting of both types will eventually be dominated by the selfish 26 . One limitation to the experiment shown in Fig. 2a is that there is only a single iteration of selection. The RNAs used to seed the experiment limit its evolutionary potential; Supplementary Fig. 13 depicts joint genotype frequency changes over time. Experiments in a serial transfer format are needed to show the selection of one strategy over the other (see below), but we can use both our data and modelling to predict that cooperation would have been advantageous in simpler chemical systems that preceded organismal biology. Randomization experiment The system described in Fig. 1 is only one of a very large number of possibilities. To test the notion that cooperative networks of RNA could form spontaneously, we randomized the middle nucleotide of both the IGS (M) and its target triplet (N) in fragments of the ribozyme, generating both matched and mismatched partners within a population. We created three pools of randomized fragments containing the IGS on the 5′ end of the ribozyme: GMG W CNU , GMG W • X CNU and GMG W • X • Y CNU , plus three fragments containing the catalytic core and the 3′ end of the ribozyme: X • Y • Z , Y • Z and Z ( Fig. 3a ). Fourfold variation in M and in N, combined with threefold variation in the junction ( j ) where recombination occurs (before X , Y or Z ) leads to 48 genotypic possibilities ( Fig. 3a ). These assembled ribozymes can be distinguished by three variables: (1) the middle nucleotide of the IGS (M), (2) the location of the junction (x, y or z) and (3) the middle nucleotide of the target (N). We therefore denote each ribozyme with the three-letter code M j N, where j = x, y or z. Each of these ribozymes can be covalently assembled by any other ribozyme, itself covalently contiguous or not, provided that M in the catalyst is complementary to N in the substrate. Figure 3: The randomization experiment. a , Experimental design. The middle nucleotides of the IGS and the tags were randomized to create diverse RNA pools. A reaction of 300 pmol each (0.5 μM) of GMG W CNU , GMG W • X CNU , GMG W • X • Y CNU , X • Y • Z , Y • Z and Z was sampled at 0.5, 2, 4 and 8 h, and millions of recombined full-length W • X • Y • Z ribozymes were genotyped by nucleotide sequence analysis ( Supplementary Table 2 ). b , Comparison of growth curves from fixed and randomized RNAs. Yields over time were compared for the simple three-membered cycle (filled triangles, UxG + AyA + CzU; the sum of the three curves in Fig. 1d ) to that in the randomized format (filled circles, panel a ) when both were performed at the same RNA pool concentrations (0.05 μM). c , Proposed succession from simple to complex networks using genotype frequency data from experiment in panel a . Simple autocatalytic cycles where M and N are complementary were directly tracked by the sum of such W • X • Y • Z molecules (dashed line with crosses; for example, AzU). Reciprocal two-membered cycles were tracked by the sum (×10, for ease of presentation) of the joint frequencies of all genotypes that can potentially participate in such cycles (dashed line with squares; for example, AxA + UxU). The rise of three-membered cycles can be seen from the sum (×10,000 for ease of presentation) of joint frequencies of three sets of genotypes: Fig. 1b and its two permutations by junction (solid line; UxG + AyA + CzU; UyG + AzA + CxU; UzG + AxA + CyU). See Supplementary Information for calculation of the joint frequencies. d , The potential network of RNA genotypes. Each node is one of the 48 possible M j N genotypes; size scales with relative frequency in the 8 h pool. Nodes are autocatalysts (red) or those that must replicate cooperatively (green). Grey arrows show all possible direct catalytic events; orange arrows show reciprocal two-membered cycles in which the frequencies of both members at least double between 30 min and 2 h; green arrows show key three-membered networks: thick green is the system studied in depth ( Fig. 1b ), thin green are permutations by junction, dotted green is AxC + GyA + UyU. Starred genotypes can participate in a four-membered network. PowerPoint slide Full size image When we incubated equimolar amounts of these six RNA sets, all 48 possible full-length W • X • Y • Z Azoarcus ribozymes arose. The relative frequencies of the 48 possible full-length ribozymes recovered at each time point over an 8 h time course ( Supplementary Table 2 ) show that, in accordance with the above and published data 25 , recombination at the Y – Z junction is favoured, but no single genotype ever exceeded 13% of the total. Growth in the randomization experiment showed markedly greater yields (2–12-fold) than in our engineered three-membered system ( Fig. 3b ), indicating that far more productive interactions among RNA species are occurring in the former. From approximately three million W • X • Y • Z genotypes sampled at each time point, distinct trends portray indirect evidence of a rapid succession from smaller to increasingly larger networks of cooperators ( Fig. 3c, d ). Genotypes that could easily propagate by selfish autocatalytic replication peak at or before the first time point at 30 min ( Fig. 3c , dotted line with crosses). These are S i genotypes (for example, those in Fig. 2 ) where M and N are complementary. A prime example is CyG, which could increase in number from the association of GCG W • X CGU and Y • Z molecules, and this genotype rose in frequency from 4.8% to 7.2% between 30 min and 2 h. Out of the 48 possible product genotypes, twelve (25%) are of this type. After peaking early, the frequencies of autocatalysts dropped below random expectation and then slowly climbed. Because of extremely large sample sizes, these deviations are highly significant (two-tailed G -tests of independence; P ≪ 0.001). However, this later frequency increase may not be a consequence of autocatalysis per se, but of the incorporation of autocatalysts into higher-ordered networks, akin to the mechanism by which cooperative networks assimilate selfish replicators ( Fig. 2 and Supplementary Fig. 10 ). Analyses of the frequencies of the product genotypes cannot reveal the identities of the catalysts that made them, and thus do not provide direct evidence of replicator cycles. Nevertheless, we examined whether networks of two or more distinct members could be increasing over time. Some pairs of genotypes can cooperate with each other to form two-membered cycles (for example, AxC + GzU), whereas others cannot (for example, AxC + UzG). We noticed that the global joint frequencies of the members comprising all possible two-membered cycles peaked at 30 min, declined and recovered, although delayed with respect to the autocatalysts ( Fig. 3c ). Support for the succession from autocatalysts to these two-membered cycles is found in the frequencies of two possible partners for the autocatalysts GjC, which are CxG and CzG (autocatalysts themselves); the sum of these rose monotonically between 2–8 h (3.7% to 6.1%). At roughly 2 h, a succession to three-membered cycles may have occurred. Although there are hundreds of such possible assemblages, the joint frequencies of the members of diverse ones requiring synthesis at all three junctions (such as UxG + AyA + CzU) jump at the 2 h mark ( Fig. 3c , solid line). Many others peak then as well; the joint frequency of the AxC + GyA + UyU trio increases nearly 20-fold after the 30 min point. At 4 h and later the possibility of succession to even higher-ordered networks that subsume all simpler ones obfuscates individual trends. Visualization of all possible connections among genotypes underscores these conclusions ( Fig. 3d ). By 8 h the network is dominated by genotypes that can only be replicated via cooperation (green circles). In fact, the variance in the genotype frequencies drops monotonically over the course of the experiment, indicating that all genotypes increasingly participate in the network over time. Serial transfer of the randomized population The experiments depicted in Fig. 3 portray the dynamic changes that occur on a kinetic time scale as a batch of RNAs approaches equilibrium. In an actual prebiotic scenario, however, this effect would be iterated and perhaps magnified over several generations, as opposed to being an asymptotic value that results from mixing several RNAs in a single reaction vessel. To bring a stronger evolutionary flavour, we repeated the randomization experiment but in a serial transfer format. Starting with another aliquot of the exact same set of RNAs (that is, products from the same in vitro transcription), we carried a population through eight serial transfers, taking 10% of the population each hour into a fresh tube of fragments. In this manner the W • X • Y • Z molecules that spontaneously assemble are continually being fed with new fragments, such that selection will favour those molecules and networks that grow faster and persist over iterations. Given that the assembly that occurs each round can be strongly influenced by the actions of naive RNAs from the 90% fresh material, we opted to assay genotypic change by sampling only the most high-frequency genotypes: those present in an abundance greater than random chance (1/48). By manually sequencing the same number of genotypes (75) from transfers number 1 and 8 and enumerating those genotypes present more frequently than random expectation (2/75 > 1/48), we were able to observe the amalgamation of an RNA network over time ( Fig. 4 ). At the 1 h time point, no closed network was possible and autocatalysts were relatively frequent (33%), but by 8 h a reflexively autocatalytic set was present in which every reaction is catalysed by at least one molecule involved in any of the reactions of the set 27 . This set included nine genotypes and fewer autocatalysts (25%), although the latter drop is not quite statistically significant (one-tailed G -test of independence; P = 0.14). Such expansion of the network to add additional genotypes is a more general case than the direct competition that we described in Fig. 2 . As another indicator of the effect of serial transfer, the outcome of this experiment differed markedly from the batch assembly experiment ( Fig. 3 ). After 8 h in the batch experiment the genotypes were dominated by pyrimidine-containing IGSs and targets (YzY; Fig. 3d ). By contrast, the serial transfer experiment, although also reiterating the bias for the Y – Z junction, distinctly favoured IGS and target sequences containing purines (RzR; Fig. 4 ). Figure 4: The serial transfer experiment. The same RNA used to seed the randomization experiment ( Fig. 3 ) was also subjected to a serial transfer protocol. For the first iteration, 50 pmol each of GMG W CNU , GMG W • X CNU , GMG W • X • Y CNU , X • Y • Z , Y • Z and Z were incubated in a 100 μl volume. After 1-h time points, 10% of the reaction mixture was transferred to a new tube containing 90% fresh RNA with a total volume of 100 μl. The population was sampled via 5′ RACE and RT–PCR to capture variation in all positions of any W • X • Y • Z molecules present in the population. The 1 and 8 h populations were cloned, and genotype frequencies were obtained by manual sequence analysis of 75 clones each ( Supplementary Table 3 ). Any genotype present twice or more was included on this diagram (see text); size of the circles scales to relative frequencies within their respective populations. All possible catalytic interactions are shown with arrows among non-autocatalytic genotypes (green), with autocatalytic genotypes (red) not participating in the network. Grey genotype in the 1st iteration disappears. Genotypes with asterisks appear by the 8th iteration. PowerPoint slide Full size image Fragmentation into four pieces Lastly, we tested whether increased fragmentation of the RNA could provide additional complexity, and enhance the pre-biological relevance. We did this by breaking the molecule up into four pieces instead of two, creating four-piece versions of I 1 , I 2 and I 3 analogously to Fig. 1a, b . When we mixed the resulting 12 RNAs together, we observed two interesting phenomena ( Fig. 5 ). First, the growth curve was distinctly sigmoidal, indicating that when more fragments are involved, the cooperativity of the system becomes even more apparent. In the four-piece fragmentation, W • X • Y • Z ribozymes can be made via many pathways, including those in which more than one enzyme cooperates to construct the product: for example, an E 1 ribozyme could recombine the W – X junction, an E 2 ribozyme could recombine the X – Y junction, and an E 3 ribozyme could recombine the Y – Z junction. Second, analysis of the sequences of the product W • X • Y • Z ribozymes showed that such cooperation was common ( Supplementary Fig. 14 ). In fact 85% of all ribozymes required help from enzymes from at least two subsystems ( Fig. 5 ). Figure 5: Growth curve of a four-piece system. A more highly fragmented system based on that shown in Fig. 1b was created by breaking the ribozyme into four fragments for each I i subsystem. The resulting 12 RNAs were co-incubated at 0.5 μM each, and samples were removed over time for both yield analysis (plot) and nucleotide sequence analysis (frequencies). The W • X • Y • Z RNAs can be assembled from a minimum of one, two or three IGS-bearing enzymes (examples shown with diagrams); the high frequencies of the latter two classes demonstrates the system’s cooperativity. PowerPoint slide Full size image Discussion Our results illustrate a scenario in which simple autocatalytic cycles form easily but are later supplanted by more complex cooperative networks that take advantage of the autocatalysts. Our system describes the short-term kinetic phenomena that provide the foundation for evolutionary behaviour 10 in the presence of sequence variation throughout the ribozymes analogous to those described as “prelife” 9 . Features of the system described here that would make it relevant to early evolution are that it is comprised solely of RNA (although other polymers could display cooperative behaviour 17 , 18 ) and that the 3-nt IGS or IGS targets are essentially the tag sequences 28 that have been suggested as a means to form molecular coalitions that can partition genetic information in a homogeneous milieu. Closure of autocatalytic sets would have been facilitated by the cooperative aggregation of oligomers with related tags 21 . Subsequent expansion of cooperative networks as shown here is possible by invasion of the network by a new set with a distinct tag sequence, for example, moving from the three-membered cycle to a four-membered cycle such as by inclusion of a new IGS–IGS-target pair ( Fig. 3d , starred genotypes), and then well beyond four members ( Fig. 4 ). Longer-term evolutionary optimization would have required spatial heterogeneity 29 or compartmentalization 8 , 30 to provide lasting immunity against parasitic species or short autocatalytic cycles. Over time, a transition back to purely selfish replicators based on polymerization chemistry could proceed 7 . In our system, we show how RNA networks have the potential to arise spontaneously and to buffer informational decay. A key to the latter is the use of recombination for replication. Although allowing for some genotypic variability, it does not lead to the accumulation of deleterious mutations as does template-directed polymerization 31 . Highly interdependent networks of genetically related replicators as a means to circumvent the error catastrophe in nascent life have been proposed 11 . The three-membered cycle shown here resembles a hypercycle as envisioned previously 4 , 21 , 32 , but without hyperbolic growth. We prefer to focus on the observation that the cycle can be derived from simpler cycles and has the potential to expand to more complex ones as evidence that RNA molecular coalitions can show spontaneous order-producing dynamics, which already has theoretical support 27 . Molecular ecological succession is a plausible model for a bridge between selfish replicators and cooperative systems. Methods Summary Experimental Ribozyme assays or covalent self-assembly from oligomers were performed as described previously 24 , 25 . Briefly, RNA oligomers were incubated together in 100 mM MgCl 2 and 30 mM EPPS buffer (pH 7.5) for 5 min–16 h at 48 °C at a final concentration of 0.01–2.0 μM each. Visualization and quantification was possible via phosphorimaging when W -containing fragments were 5′-end-labelled with γ[ 32 P]ATP before use. For genotyping, ∼ 200-nt RNA was excised from a gel and subject to PCR with reverse transcription (RT–PCR) using W - and Z -specific primers. High-throughput sequence analysis on the Illumina platform was possible after 5′ RACE to capture the sequence variability in the IGS of assembled ribozymes. For manual sequence analysis, the PCR products were cloned into Escherichia coli and individual colonies were picked for colony PCR reactions. Resulting amplicons were either subjected to nucleotide-sequence analysis or restriction fragment length polymorphism (RFLP) analysis. Modelling The cooperative system was modelled as a set of six differential equations describing the concentrations over time of the six principal species (see Supplementary Information ). These equations are derived from the detectable catalysis reactions (encompassing six direct-catalysis reactions and three cross-catalysis reactions). The experimental time series data from the full three-component system and from the two-component subsystems yielding detectable product were fit simultaneously to the model by standard optimization techniques. Change history 31 October 2012 A minor typo in Fig. 1 was corrected.

(Phys.org)—A team of chemistry and applied sciences researchers from several universities in the United States has shown that RNA fragments torn apart in the lab work together to reassemble themselves. This finding, the team claims in their paper published in the journal Nature suggests that early life may have started with cooperative efforts between RNA molecules eventually leading to cooperative replication. The team based its work on mathematical theories proposed by Manfred Eigen, a chemist who suggested that because early RNA wouldn't be able to successfully evolve from short stranded molecules, they must have had help. That help, he said, may have come in the form of cooperative efforts between molecules. In earlier work team lead Niles Lehman had found that if long RNA molecules known as ribozymes were cut into fragments and then placed together in a Petri dish, they would over time reassemble themselves into their original configuration. In this new research, Lehman et al altered three ribozyme samples so that they were identical save for one letter that allowed for distinguishing among them. Each was cut into two pieces and placed in a Petri dish. The team found that if the ribozymes were placed together in a Petri dish, they reassembled themselves faster than if they were put in the dish alone. This occurred, they report, because one of the ribozymes helped another reassemble, who then helped a third reassemble who in turn helped the first reassemble, which formed a closed loop network. To see if the same result might be possible in a more chaotic environment, the researchers placed 48 cut ribozymenes in a test tube with millions of other RNA molecules and found that the original 48 found a way to locate their other parts and each other and helped one another reassemble; again much faster than any of them would have alone. The team suggests that a similar type of cooperation among short RNA molecules in Earth's primordial soup may have allowed them to replicate in a way that avoided the problem of building up mistakes when making copies that mathematical models have suggested would have led to evolutionary death. That would have allowed them to evolve into longer more complex RNA molecules and eventually into all the other molecules that exist today.

doi:10.1038/nature11549

Medicine

Diabetes drug may help symptoms of autism associated condition

Nature Medicine (2017). DOI: 10.1038/nm.4335 Journal information: Nature Medicine

http://dx.doi.org/10.1038/nm.4335

https://medicalxpress.com/news/2017-05-diabetes-drug-symptoms-autism-condition.html

Abstract Fragile X syndrome (FXS) is the leading monogenic cause of autism spectrum disorders (ASD). Trinucleotide repeat expansions in FMR1 abolish FMRP expression, leading to hyperactivation of ERK and mTOR signaling upstream of mRNA translation. Here we show that metformin, the most widely used drug for type 2 diabetes, rescues core phenotypes in Fmr1 −/y mice and selectively normalizes ERK signaling, eIF4E phosphorylation and the expression of MMP-9. Thus, metformin is a potential FXS therapeutic. Main Dysregulated mRNA translation is linked to core pathologies diagnosed in FXS, such as social and behavioral problems, developmental delays and learning disabilities 1 , 2 . In the brains of individuals with FXS and knockout mice ( Fmr1 −/y ; X-linked Fmr1 deletion in male mice), loss of the fragile X mental retardation protein (FMRP) results in hyperactivation of the mammalian/mechanistic target of rapamycin complex 1 (mTORC1) and the extracellular-signal-regulated kinase (ERK) signaling pathways 1 , 2 . Consistent with increased ERK activity, eukaryotic initiation factor 4E (eIF4E) phosphorylation is elevated in the brains of individuals with FXS and Fmr1 −/y mice, thereby promoting translation of the mRNA encoding matrix metalloproteinase 9 (MMP-9), which is also elevated in the brains of both individuals with FXS and Fmr1 −/y mice 1 , 2 , 3 , 4 , 5 . In accordance with these findings, knockout of Mmp9 rescues the majority of known cellular and behavioral phenotypes in Fmr1 −/y mice. MMP-9 degrades components of the extracellular matrix, including proteins that are important for synaptic function and maturation, which have been implicated in FXS and ASD. Recent observations indicate that metformin, a first-line therapy for type 2 diabetes, imparts numerous health benefits beyond its original therapeutic use, such as decreased cancer risk and improved cancer prognosis 6 . Metformin inhibits mitochondrial respiratory chain complex 1, leading to a decrease in the cellular energy state and, thus, activation of the energy sensor AMP-activated protein kinase (AMPK) 6 . Several AMPK-independent activities of metformin have also been reported 7 , 8 . As metformin suppresses translation by inhibiting the mTORC1 and ERK pathways, we reasoned that metformin could have beneficial therapeutic effects in Fmr1 −/y mice 9 . Adult (aged 8–12 weeks) wild-type (WT) and Fmr1 −/y mice were injected intraperitoneally (i.p.) with metformin (200 mg per kg bodyweight per day, a dose previously used in preclinical studies 8 ) or vehicle for 10 d ( Fig. 1a ). Metformin, as previously reported 10 , crossed the blood–brain barrier (BBB), with lower concentrations found in the brain than in plasma after acute and chronic injection ( Supplementary Figs. 1 and 2 ). Twenty-four hours after the last injection of metformin, mice were subjected to a social novelty test. Vehicle-treated Fmr1 −/y mice were impaired in their preference for social novelty, showing no preference for interaction with the novel stimulus (stranger 2) over the original social stimulus (stranger 1; Fig. 1b,c ). Metformin treatment restored the impaired preference of Fmr1 −/y mice for the novel stranger mouse, thus rescuing the social deficit. Next, we examined the effect of metformin on repetitive behavior, a core characteristic of individuals with FXS that is recapitulated in Fmr1 −/y mice as increased self-grooming 1 , 11 . Metformin reversed the increased grooming in Fmr1 −/y mice ( Fig. 1d ) and decreased the number of grooming bouts ( Fig. 1e ) measured 24 h after the last injection. Prolonged exposure to metformin was required to rescue behavioral deficits, as metformin treatment of Fmr1 −/y mice for 1 or 5 d failed to correct the core FXS phenotypes ( Supplementary Figs. 3 and 4 ). We tested several other behavioral phenotypes, including audiogenic seizures, hyperactivity and cognitive function, in Fmr1 −/y mice after metformin treatment and observed no cognitive impairment. Ten-day treatment with metformin reduced the incidence of seizures but did not have an effect on hyperactivity ( Supplementary Figs. 5 and 6 ). Figure 1: Chronic metformin treatment corrects the social deficit, repetitive behavior, aberrant dendritic spine morphology and exaggerated LTD in Fmr1 −/y mice. ( a ) Metformin (met) or vehicle (veh) was injected i.p. over 10 d (200 mg per kg bodyweight per day), followed by analysis of social behavior, grooming, dendritic spine morphology and LTD. ( b , c ) Preference for social novelty was assessed in the three-chamber social interaction test by measuring the time spent with the novel social stimulus (stranger 2, S2) or the previously encountered mouse (stranger 1, S1) ( b ) and the time spent in each chamber ( c ). Vehicle-treated WT mice ( n = 10), vehicle-treated Fmr1 −/y mice ( n = 10), metformin-treated WT mice ( n = 9), metformin-treated Fmr1 −/y mice ( n = 12). ( d , e ) Self-grooming test examining the total time spent grooming ( d ) and the total number of grooming bouts ( e ). Vehicle-treated WT mice ( n = 10), vehicle-treated Fmr1 −/y mice ( n = 10), metformin-treated WT mice ( n = 8), metformin-treated Fmr1 −/y mice ( n = 12). ( f ) Golgi–Cox staining of CA1 dendritic spines in metformin- and vehicle-injected WT and Fmr1 −/y mice. Scale bar, 2 μm; representative images shown ( n = 4 mice in each group). ( g , h ) Quantification of spine density, measured as the number of spines per 10-μm dendrite length ( g ), and spine subtype analysis (S/M, spiny or mushroom; F, filopodial), presented as the fraction of the total spines for each subtype ( h ) ( n = 4 mice in each group). ( i , j ) mGluR-dependent LTD was measured in CA1 in response to ( S )-3,5-dihydroxyphenylglycine (DHPG; 50 μM for 10 min) in slices prepared from vehicle-treated WT ( n = 4) and Fmr1 −/y ( n = 9) mice ( i ) and metformin-treated WT ( n = 5) and Fmr1 −/y ( n = 6) mice ( j ). ( k ) Field excitatory postsynaptic potential (fEPSP) slope during the last 10 min of recording. In b – e , each point represents data from an individual mouse, and in g , h and k each point represents data from 1–3 slices per mouse. Values are shown as mean ± s.e.m. *** P < 0.001, ** P < 0.01, * P < 0.05 versus all other groups; N.S., not significant; calculated by two-way ANOVA with Tukey's post hoc test. Full size image Neurons from individuals with FXS and Fmr1 −/y mice exhibit aberrant spine morphology 1 , 11 . We observed spine dysmorphogenesis in Fmr1 −/y mice, as evidenced by increased density of dendritic spines for CA1 hippocampal pyramidal neurons, along with fewer mature spines with a stubby or mushroom morphology and an increased number of immature, filopodia-like spines ( Fig. 1f–h ). Metformin administration for 10 d corrected the dendritic abnormalities in Fmr1 −/y mice ( Fig. 1f–h ). Fmr1 −/y mice also display exaggerated group 1 metabotropic glutamate receptor (mGluR)-dependent long-term depression (LTD) of synaptic transmission 1 , 12 . Metformin treatment for 10 d rescued exaggerated LTD ( Fig. 1i–k ) in Fmr1 −/y mice and also restored excitatory synaptic activity to WT levels in hippocampal slices from Fmr1 −/y mice ( Supplementary Fig. 7 ). A hallmark of post-adolescent male individuals with FXS and Fmr1 −/y mice is macroorchidism 11 , 12 . Metformin administration for 10 d also led to a partial reduction of the increased testicular weight in Fmr1 −/y mice ( Fig. 2a ). Figure 2: Chronic metformin treatment corrects macroorchidism, decreases translation and reduces the phosphorylation of upstream eIF4E effectors. ( a ) Mean testicular weight of vehicle- and metformin-treated WT and Fmr1 −/y mice. Vehicle-treated WT mice ( n = 6), vehicle-treated Fmr1 −/y mice ( n = 6), metformin-treated WT mice ( n = 6), metformin-treated Fmr1 −/y mice ( n = 7). ( b ) Immunoblots and quantification of lysates from hippocampal slices incubated with puromycin to measure basal rates of protein synthesis. β-tubulin was used as a loading control. Puromycin incorporation is presented as percentage change relative to vehicle-treated WT slices ( n = 7 in each group). ( c – j ) Representative immunoblots of lysates from vehicle- and metformin-treated WT and Fmr1 −/y mice and quantification of phosphorylation and total levels of MEK ( c ), ERK ( d ), eIF4E ( e ) and MMP-9 ( f ) in prefrontal cortex and MEK ( g ), ERK ( h ), eIF4E ( i ) and MMP-9 ( j ) in the hippocampus. GAPDH was used as a loading control. For quantification, the phospho-protein signal was normalized first against total protein signal and is presented relative to the signal for vehicle-treated WT mice ( n = 6 mice in each group in c – j , except for MMP-9 in the prefrontal cortex in f , where n = 5 mice). Full-length immunoblots are shown in Supplementary Figures 12–14 . Each point represents data from an individual mouse, and all values in a – j are shown as mean ± s.e.m. *** P < 0.001, ** P < 0.01, * P < 0.05 versus all other groups; calculated by two-way ANOVA with Tukey's post hoc test. Full size image Fmr1 −/y mice exhibit elevated mRNA translation 1 , 12 . Consistent with previous studies 1 , 12 , 13 , basal levels of protein synthesis were elevated in Fmr1 −/y mice, and metformin treatment for 10 d reduced the excessive translation ( Fig. 2b ). The ERK and mTOR signaling pathways are hyperactivated in Fmr1 −/y mice 1 , 2 , 12 , 13 . Metformin treatment for 10 d restored the levels of phosphorylated mitogen-activated protein kinase kinase (MEK), ERK and EIF4E and the total levels of MMP-9 in the prefrontal cortex and hippocampus of Fmr1 −/y mice ( Fig. 2c–j ), whereas the levels of phosphorylated S6 remained elevated in the hippocampus ( Supplementary Figs. 8a,b and 9 ). Similarly, metformin treatment for 10 d rescued increased phosphorylated ERK levels in the striatum, but not in the cerebellum ( Supplementary Fig. 10a,b ), of Fmr1 −/y mice and affected the known specific synaptic FMRP targets MAP2 and synapsin, with no effect on eEF2 and PUM2 levels 14 ( Supplementary Fig. 11 ). Apart from the brain, the level of ERK phosphorylation was increased in the liver but not in the gonads of Fmr1 −/y mice ( Supplementary Fig. 10c,d ). Metformin treatment for 10 d did not rescue the increased ERK phosphorylation in the liver ( Supplementary Fig. 10d ), implicating other pathways 12 or endocrine regulation outside the brain of Fmr1 −/y mice in the increased phosphorylation. Metformin administration for 10 d did not activate AMPK in the prefrontal cortex and hippocampus of Fmr1 −/y mice, as evidenced by the lack of increased levels of phosphorylated AMPK and its downstream substrates ACC1, TSC2, raptor and B-Raf (Ser729) in metformin-treated mice ( Supplementary Figs. 8c–k and 9a ). These findings are consistent with previous reports showing that chronic metformin administration does not increase AMPK phosphorylation in the brain 15 , 16 . It is not immediately clear why metformin administration for 10 d does not have this effect; however, in accordance with previous studies 17 , 18 , a single injection of 200 mg per kg bodyweight i.p. of metformin induced a transient increase in the levels of phosphorylated AMPK ( Supplementary Fig. 1c ). A plausible explanation is that the change in ERK signaling following chronic metformin treatment is due to the rescue of elevated expression of B-Raf and c-Raf in Fmr1 −/y mice ( Supplementary Fig. 9 ) 19 . Presently, there is no cure for FXS or ASD, and recently completed clinical trials in teenagers or adults with FXS were not promising 20 . Our data show that metformin, the most widely used US Food & Drug Administration (FDA)-approved anti-diabetic for patients aged 10 years and older, corrects most phenotypic deficits in the adult FXS mouse model. Thus, metformin, for which long-term safety and tolerability have been extensively documented in clinical practice, is one of the very few compounds that can be promptly repurposed as an FXS therapeutic for patients aged 10 years and older. Moreover, our data are in accordance with a recent finding that metformin treatment corrects circadian and cognitive deficits in a Drosophila melanogaster fragile X model 21 . We present a potential molecular mechanism to explain how metformin ameliorates FXS phenotypes by showing that chronic metformin treatment corrects enhanced Raf–MEK–ERK signaling and MMP9 expression in Fmr1 −/y mice ( Fig. 2 and Supplementary Fig. 9 ). Similarly, lovastatin, a drug that downregulates ERK signaling, also rescued audiogenic seizures, exaggerated mGluR-dependent LTD and decreased general protein synthesis in Fmr1 −/y mice 13 . Metformin, however, corrects a broader range of phenotypes than lovastatin. In combination, these findings bolster the idea that aberrant ERK activity has a critical role in engendering FXS-like phenotypes in FXS. Because Mmp9 mRNA translation is stimulated by eIF4E phosphorylation and knockout of Mmp9 reversed abnormal phenotypes in Fmr1 −/y mice 1 , 5 , it is highly likely that the rescue by metformin is selectively mediated via ERK- and eIF4E-dependent normalization of MMP-9 expression in the brain, providing a very strong mechanistic explanation for the action of metformin; however, we cannot exclude the possibility of an unidentified, peripherally mediated rescue mechanism, given that metformin inhibits gluconeogenesis and alters the gut microbiota 22 . Such peripheral phenotypes are linked to autism, intellectual disability and FXS and have been shown to affect brain plasticity 23 . Methods Knockout mice and metformin administration. Generation of Fmr1 −/y mice (the Fmr1 gene is on the mouse X chromosome; thus, male animals have a −/y genotype, where y corresponds to the mouse Y chromosome) on a C57BL/6J background (Jackson Laboratories, 003025) has been previously described 1 . Food and water were provided ad libitum , and mice were kept on a 12-h light/12-h dark cycle (7:00–19:00, light period). After weaning at postnatal day 21, mice were group housed (maximum of five mice per cage) by sex. Cages were maintained in ventilated racks in rooms with controlled temperature (20–21 °C) and humidity ( ∼ 55%). Standard corncob bedding was used for housing (Harlan Laboratories, Inc.). All animals received a 10-d chronic treatment with metformin (200 mg per kg bodyweight per day, i.p.) or vehicle (saline), except when indicated otherwise. Injection groups were randomized over all cages. All procedures were in compliance with Canadian Council on Animal Care guidelines and were approved by McGill University and Université de Montréal. Three-chamber sociability and preference for social novelty tests. The apparatus consisted of three Plexiglas chambers: the central chamber (36 cm × 28 cm × 30 cm) was divided from two side chambers (each chamber: 29 cm × 28 cm × 30 cm) by Plexiglas walls (Stoelting Co.) as previously described 1 , 24 . Each side was accessible to the mouse from the center through a doorway covered by a removable sliding door. A camera was mounted above the apparatus to record testing. The test consisted of three phases: habituation, sociability and preference for social novelty. In the first part, male mice aged 3 months were placed in the middle chamber and were allowed to explore all three empty chambers for 10 min. After this period of habituation, mice were gently guided back to the center chamber of the apparatus and the sliding doors to access the remaining two chambers were closed. In the second part of the test, an unfamiliar mouse (stranger 1), which was enclosed in a wire cage to ensure that only the test mouse could initiate social interaction, was placed into one of the two remaining side chambers. An empty wire cage that was identical to the wire cage housing stranger 1 was placed on the corresponding spot in the other side chamber. The side doors were then opened simultaneously to allow the test mouse to explore the three-chamber apparatus for 10 min to assess sociability. At the end of the 10-min period, the test mouse was gently guided to the central chamber and the sliding doors were closed. In the final part of the test, a new, unfamiliar mouse (stranger 2) was placed in the previously empty wire cage, and the test mouse could explore the three chambers for an additional 10 min to assess preference for social novelty. Stranger mice consisted of age- and sex-matched C57BL/6J mice that were group housed (four mice per cage) and were used in a counterbalanced way. The empty wire cages were alternated between side chambers for different test mice. Stranger 1 and stranger 2 mice always came from different home cages. Mice were tested in the morning during the light cycle. Time spent directly sniffing, defined as the time the test mouse spent in direct nose contact with wire cages, time spent in each chamber and the number of transitions into the chambers were scored manually. Data were scored in a manner that was blinded to mouse genotype—and, if possible, by a third party—using a stopwatch. Statistical analysis included mixed ANOVA with a Tukey's post hoc test for multiple comparisons. Self-grooming test. The setup consisted of a new Plexiglas cage equal in size to the home cage, containing approximately 1 cm of bedding material but no nesting material. A camera was placed vertically in front of the cage for recording. Fmr1 −/y and WT mice (males aged 3 months) were placed in the new Plexiglas cage and allowed to explore for 20 min. The first 10 min of the experiment was considered to be the habituation phase; the final 10 min was used to acquire self-grooming data. The total time spent grooming and the total number of grooming bouts were used to analyze grooming behavior. Data were manually scored in a manner blinded to mouse genotype—and, if possible, by a third party—using a stopwatch. All measures were analyzed with two-way ANOVA with Tukey's post hoc test. Audiogenic seizures. Mice (male, postnatal day (P) 21–24) were chronically injected for 10 d with metformin (200 mg per kg bodyweight) or vehicle before experimentation. Mice were individually habituated in an isolated, sound-insulated behavioral chamber made of transparent plastic (outside dimensions: 28 cm × 17 cm × 16 cm) for 2 min and were subjected to a 130-dB acoustic stimulus using a personal alarm (Vigilant) for 2 min; during this time, the incidences of wild running, tonic-clonic seizures and status epilepticus were recorded. Open-field exploration. Animals (male, aged 8–12 weeks) were first habituated to the dimly lit experimental room ( ∼ 15 lx) for 30 min and were then individually placed in an illuminated clear Plexiglas chamber (40 cm × 40 cm × 40 cm, ∼ 1,200 lx) with a white floor. Animals were allowed to explore freely for 10 min following an initial 1-min habituation phase. Total path length and velocity were calculated using ANY-maze (Stoelting Co.) as a measure of hyperactive behavior. Light–dark transition test. The test apparatus was composed of two adjacent chambers connected by a small opening: a dark-enclosed chamber made of black Plexiglas (20 cm × 40 cm × 40 cm) and a chamber with three clear Plexiglas walls with an open top. Mice (male, aged 8–12 weeks) were placed into the 'light' side ( ∼ 390 lx) and allowed to explore freely for 10 min. An entry was defined as the mouse placing all four feet into each zone. Morris water maze and reversal learning. Chronic daily metformin (200 mg per kg bodyweight) or vehicle (saline) administration started 5 d before training and lasted throughout the whole course of testing, for a total of 10 d. The circular water maze pool was 100 cm in diameter. The water was maintained at 22–23 °C and made opaque by addition of white tempera. The platform was 10 cm in diameter. Mice (male, aged 8–12 weeks) were handled daily for 3 d before the start of the experiment. During the experiment, mice were trained three times per day with an intertrial interval of 30 min over five consecutive days (days 1–5). Each trial lasted a maximum of 120 s or until the mouse found the platform. If the mouse did not find the platform in the assigned time, it was guided to the platform and stayed there for 10 s before being returned to the home cage. For the probe test (on day 6), the platform was removed and each mouse was allowed to swim for 30 s. For the test of the reversal learning paradigm, in which the hidden platform was relocated to the opposite quadrant (days 6 and 7), mice received the same training procedure as described above. The platform was removed for the probe test of reversal learning (day 8) to assess spatial retention. The experiment was recorded with a video tracking system (HVS Image), whereby the latency to reaching the platform during acquisition and the time spent in the target quadrant during the probe trials were determined. Contextual fear conditioning. During acquisition (5 min), two foot shocks of 0.7 mA for 1 s each that were separated by 60 s were administered after an initial 2-min period of context exploration. Twenty-four hours after training, mice (male, aged 8–12 weeks) were tested for contextual fear memory in the same context for 5 min, as assessed by the percentage of total time spent freezing in the conditioning context. Behavioral scoring was carried out for a 5-min period in 5-s intervals, in which animals were assigned as either 'freezing' or 'not freezing'. Freezing (%) indicates the number of intervals where freezing was observed divided by the total number of 5-s intervals. Novel object recognition. On day 1, mice (male, aged 8–12 weeks) were first habituated for 15 min in a square testing arena (40 cm × 40 cm) followed by 15 min in an opaque box, before being returned to their home cages. On days 2 and 3, mice were put back in the arena for 15 min and presented with two identical objects (familiar) within specific areas (counterbalanced for location of objects). Mice were first allowed to freely explore the arena and objects, followed by a 15-min interval in which the mice were in an opaque box; after that, mice were returned to their home cages. On day 4, one of the objects (used for days 2 and 3) was replaced with a third object (novel object), and the mice were allowed to explore the environment for 15 min. Time spent exploring each object was recorded. Object exploration was defined as the time spent interacting with an object—when the mouse was sniffing and touching the object. Total exploration time was quantified as the time spent interacting with both objects. To assess preferential attention to an object, a discrimination index was calculated ( t novel – t familiar )/( t novel + t familiar ). A positive index represents a preference for the novel object. Immunoblotting and antibodies. Brain tissue (from male mice aged 3 months) was homogenized in RIPA buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% NP-40, 0.1% SDS, 0.5% sodium deoxycholate, 5 mM EDTA pH 8.0, 1 mM EGTA pH 8.0, 10 mM NaF, 1 mM β-glycerophosphate and 1 mM sodium orthovanadate) containing protease inhibitors (Roche). Protein extracts were denatured with heat and resolved by SDS–PAGE or gradient precast gels (Thermo Fisher Scientific). Following electrophoresis, proteins were transferred to nitrocellulose membranes and immunoblotting was performed. Membranes were stripped in 25 mM glycine-HCl pH 2.0 and 1% SDS for 30 min at room temperature, followed by washing in TBST before reprobing. Immunoreactivity was detected by enhanced chemiluminescence (plus-ECL, PerkinElmer, Inc.) after exposure to X-ray film (Denville Scientific, Inc.). Quantification of immunoblots was performed using ImageQuant 5.2. Values were normalized against GAPDH. The following antibodies were used: eIF4E (610270, BD Transduction Laboratories); phospho-eIF4E (NB-100-79938, Novus Biologicals); ERK (sc-93, Santa Cruz); phospho-ERK (4370, Cell Signaling); MEK1/2 (4694, Cell Signaling); phospho-MEK1/2 (9154, Cell Signaling); FMRP (4317, Cell Signaling); MMP-9 (TP221, Torrey Pines); AMPK (2532, Cell Signaling); phospho-AMPK (2535, Cell Signaling); ACC1 (4190, Cell Signaling); phospho-ACC1 (11818, Cell Signaling); S6 (2217, Cell Signaling); phospho-S6 (2215, Cell Signaling); TSC2 (4308, Cell Signaling); phospho-TSC2 (5584, Cell Signaling); raptor (2280, Cell Signaling); phospho-raptor (2083, Cell Signaling); c-Raf (53745, Cell Signaling); B-Raf (ab33899, Abcam); phospho-B-Raf S729 (ab124794, Abcam); phospho-B-Raf S602 (PA5-38412, Thermo Fisher Scientific); synapsin (5297, Cell Signaling); eEF2 (2332, Cell Signaling); MAP2 (ab5392, Abcam); PUM2 (A300-202A, Bethyl Laboratories); GAPDH (sc-32233, Santa Cruz); β-actin (A5441, Sigma); secondary anti-mouse and anti-rabbit (GE Healthcare). Full-length immunoblots are shown in Supplementary Figures 12–17 . The dilutions at which each of the antibodies were used are described in Supplementary Table 2 . For statistical analysis of immunoblot results, we used two-way ANOVA with Tukey's post hoc test and one-way ANOVA with Tukey's post hoc test (p-AMPK in the hippocampus, single-injection metformin experiment). LTD recordings. For analysis of hippocampal LTD, male WT or Fmr1 −/y mice aged 31–34 d, treated with either saline or metformin (as described above), were used. After obtaining hippocampal slices (400 μm thick), the CA1 and CA3 hippocampal regions were isolated by surgical excision and incubated for 2 h at 32 °C in oxygenated artificial cerebral spinal fluid for recovery (ACSF; 124 mM NaCl, 5 mM KCl, 1.25 mM NaH 2 PO 4 , 2 mM MgSO 4 , 2 mM CaCl 2 , 26 mM NaHCO 3 and 10 mM glucose). Later, slices were placed in a recording chamber at 27–28 °C and perfused with ACSF for an additional 30 min. Glass electrodes (2–3 MΩ) were filled with ACSF and gently placed on the CA1 stratum radiatum to record fEPSPs evoked by stimulation of Schaffer collaterals. The stimulating concentric bipolar tungsten electrode was placed in the mid-stratum radiatum proximal to the CA3 region to deliver 0.1-ms pulses at 0.033 Hz. The intensity was adjusted to evoke fEPSPs with 60% maximal amplitude. mGluR-LTD was induced by perfusion with the group I mGluR agonist DHPG (50 μM; Tocris Bioscience) for 10 min in ACSF. fEPSPs were recorded for a total of 60 min after the onset of induction. The fEPSP slope was measured on digitized analog recordings using the Clampfit analyze function and by using the cursor to define quantifiable epochs between 10% and 90% of the maximal fEPSP amplitude. This setting excluded fiber volley and population spikes. Data were then analyzed using two-way ANOVA with Tukey's post hoc test. Miniature excitatory postsynaptic current recordings. Organotypic hippocampal slices were prepared from WT and Fmr1 −/y mice (P4–P6). The brain was removed and dissected in HBSS (Invitrogen)–based medium. Corticohippocampal slices (400 μm thick) were obtained with a McIlwain tissue chopper (Campden Instruments). Slices were placed on Millicell culture plate inserts (Millipore) and incubated in Opti-MEM (Invitrogen)-based medium in a humidified atmosphere of 5% CO 2 and 95% O 2 at 37 °C. Experiments were performed after 14–20 d in culture. Cultures were treated with metformin (50 μM) or vehicle (Opti-MEM medium) for 4–5 d before electrophysiology experiments, which were performed blinded to treatment. Whole-cell recordings were obtained from CA1 pyramidal neurons using borosilicate pipettes (3–6 MΩ) filled with intracellular solution containing (in mM) 132 CsMeSO 3 , 8 CsCl, 0.6 EGTA, 10 diNa-phosphocreatine, 10 HEPES, 4 ATP-Mg 2+ and 0.4 GTP-Na (pH 7.25–7.30 with CsOH, 275–280 mOsmol). Spontaneous miniature excitatory postsynaptic currents (mEPSCs) were recorded in the presence of TTX (5 nM; Abcam) in ACSF containing (in mM) 124 NaCl, 2.5 KCl, 1 NaH 2 PO 4 , 26 NaHCO 3 , 2 CaCl 2 , 2 MgSO 4 and 10 D -glucose (pH 7.37–7.41 with NaCl, 295–305 mOsmol). Recordings were obtained using a Multiclamp 700A amplifier and a 1,440A Digidata acquisition board (Molecular Devices). Signals were low-pass filtered at 2 kHz, digitized at 20 kHz and stored on a PC. mEPSCs were recorded in a whole-cell voltage clamp at a holding potential of −70 mV, and identification of mEPSCs was confirmed by application of CNQX (10 μM). Access resistance was routinely monitored, and recordings were only included if <30 MΩ and with <25% variation over the recording period. For analysis, mEPSC traces were filtered at 2.8 kHz (Bessel filter) using pClamp10 software (Molecular Devices), and miniature events were analyzed using MiniAnalysis (Synaptosoft). Two-way ANOVA with Tukey's post hoc test was used to assess statistical significance. Analysis of neuronal morphology by Golgi–Cox staining. Four male mice per genotype and treatment (aged 3 months) were used for morphological analysis. The Rapid GolgiStain Kit (FD NeuroTechnologies) was used for the staining procedure according to the manufacturer's instructions. Briefly, whole brains were isolated from each animal, rinsed once in Milli-Q water and quickly immersed in impregnation solution (A+B); they were then stored at room temperature in the dark for 3 weeks. Sections that were 120 μm thick were cut, processed and mounted following the protocol provided with the kit. Hippocampal sections were imaged on a confocal microscope (LSM710, Zeiss). Apical dendrites for five pyramidal neurons from the hippocampal CA1 area per animal were analyzed. To measure spine density on apical shaft dendrites, the number of spines on each successive 25-mm segment was counted, starting at the soma and continuing to the end of the dendrite. Densities for each segment and for each neuron were pooled to get an average spine density for each animal; the difference between genotypes was analyzed by two-way ANOVA with Tukey's post hoc test. For each neuron, the spine morphology was determined by the first ten spines in every 25-μm bin along the apical shaft. Spines were assigned to one of the five morphological categories based on published methods 1 , 24 , 25 : A, thin; B, stubby; C, mushroom; D, filopodia; E, branched. χ 2 analysis was used to compare the distribution of spines in these categories between genotypes. For statistical analysis, we used two-way ANOVA with Tukey's post hoc test. Measurement of de novo protein synthesis. To assess whether metformin corrected increased translation in Fmr1 −/y mice, we measured de novo protein synthesis in hippocampal slices using the SUnSET puromycin incorporation assay 1 , 12 . Transverse hippocampal slices (400 μm thick) were prepared from the brains of mice aged 5–6 weeks and were allowed to recover for at least 3 h. Puromycin labeling was then performed as described 1 , 12 , 26 . Briefly, the slices were incubated with puromycin (P7255, Sigma, 5 μg/ml in ACSF) for 45 min and then processed for immunoblotting, as described above, using an anti-puromycin antibody (EQ0001, KeraFast). Slices processed in parallel but not incubated with puromycin served as an unlabeled control. The quantity of protein synthesis was determined by measuring total lane signal from 15–250 kDa and subtracting signal from the unlabeled protein control. Signals were quantified using ImageJ, normalized to β-tubulin (T8328, Sigma) and presented as percentage change relative to control. For statistical analysis of immunoblot results, we used two-way ANOVA with Tukey's post hoc test. Metformin bioanalysis and LC–MS/MS. WT mice on a C57BL/6J background (Charles River Laboratories; males aged 8–10 weeks) were used for the study. Food and water were provided ad libitum , and mice were kept on a 12-h light/12-h dark cycle (light period, 7:00–19:00). For pharmacokinetic analysis, the mice received a single dose of metformin (200 mg per kg bodyweight, i.p.) and plasma and brain tissues were collected at 0, 0.5, 1, 2 and 4 h after drug administration. For the dose-response study, mice were treated for 10 d with 25, 50, 100 or 200 mg of metformin per kg bodyweight per day (i.p.) and plasma and brain tissues were collected 24 h after the last injection. Brain tissue homogenate and plasma concentrations of metformin were determined by protein precipitation and liquid chromatography with mass spectrometric detection (LC–MS/MS). Metformin powder (Sigma) was used to prepare a 1.00 mg/ml solution in DMSO, adjusting for the salt factor as applicable. Calibration spiking solutions were prepared at 10.0, 20.0, 50.0, 100, 200, 500, 1,000, 2,000, 5,000, 10,000, 20,000, 50,000 and 100,000 ng/ml in DMSO from the primary stock solution. Plasma and brain tissue samples were quickly collected and stored at −70 °C. Brain samples and blank tissues were homogenized with three parts distilled water per gram of tissue for a final processing dilution factor of fourfold. The resultant blank tissues were used for matrix calibration standards, which were prepared the same day as analysis on ice at 0.5, 1.0, 2.5, 5.0, 10.0, 25.0, 50.0, 100, 250, 500, 1,000, 2,500 and 5,000 ng/ml by spiking blank brain tissue homogenate and plasma matrices at 1:20 with the appropriate metformin spiking solution. Subsequently, aliquots of the matrix samples, matrix calibration standards and matrix blanks were taken and protein was precipitated by the addition of labetalol in 100% acetonitrile (1:4). The resultant matrix samples, matrix calibration standards and matrix blanks were vortexed for 1 min and centrifuged for 10 min at 1,150 g at 4 °C. Then, 100 μl of the resultant supernatant was transferred into a clean 96-well plate and diluted with aqueous solution (1:1). All matrices, plasma and brain tissue were processed independently and in discrete batches containing appropriate matrix study samples, matrix calibration standards and matrix blanks, respectively. The analysis for each discrete batch was performed on an LC–MS/MS system: AB Sciex QTRAP 6500 with a Shimadzu Nexera UPLC system using a ZIC-HILIC 2.1 mm × 50 mm analytical column (EMD Millipore) with a 3.5-μm pore size. An injection volume of 1.5 μl was used for all samples and standards with a flow rate of 1.0 ml/min. The mobile phases consisted of the following: mobile phase A, 10 mM ammonium acetate in water; mobile phase B, 0.1% formic acid (v/v) in acetonitrile. Mass spectrometry data were generated with positive electrospray ionization (ESI+) using multiple-reaction monitoring (MRM) of the following transitions: metformin, 130.324/60.100 Da; labetalol (IS), 329.200/311.200 Da. Subsequent least-squares linear regression was performed on matrix calibration standards, and the matrix sample concentrations were interpolated from the appropriate matrix curve. All dilution factors were accounted for in final sample data with concentration of metformin expressed in ng/ml and ng/g for plasma and brain tissue samples, respectively. Statistical analysis. Experimenters were blinded to genotype and treatment during testing and scoring. To determine the sample size in our behavioral, electrophysiological, imaging and biochemical experiments, we followed the standard sample sizes used in similar experiments in each of the relevant fields in the literature. The sample sizes in our behavioral studies were based on Figure 5b in ref. 27 . All experimental n values represent individual animals unless otherwise stated—technical replicates of some immunoblots were carried out. All data are presented as mean ± s.e.m. Statistical significance was set at 0.05. Statistical results, along with tests used (one-way ANOVA, two-way ANOVA and mixed ANOVA), are summarized in Supplementary Table 1 . SPSS (IBM), Statistica (Statsoft), Sigmaplot (Systat Software, Inc.) and GraphPad Prism (GraphPad Software) were used for statistical analysis. Supplementary Table 1 outlines the statistics used for each figure. Data availability. The data supporting the findings of this study are available from the corresponding author upon reasonable request. Additional information Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

A widely used diabetes medication could help people with a common inherited form of autism, research shows. Scientists found that a drug called metformin improves sociability and reduces symptomatic behaviours in adult mice with a form of Fragile X syndrome. Researchers say that metformin could be repurposed as a therapy for Fragile X syndrome within a few years - if clinical trials prove successful. Fragile X syndrome is caused by inherited defects in a gene called FMR1, which leads to excess protein production in the brain. This results in the breakdown of connections between brain cells, leading to changes in behaviour. The team led by the University of Edinburgh and McGill University in Canada looked at the effects of metformin on mice that lack the FMR1 gene. These mice usually have symptoms consistent with Fragile X syndrome - they exhibit repetitive behaviours such as increased grooming and do not socialise with other mice. After mice had treatment with metformin for ten days, protein production in the brain returned to typical levels, brain connections were repaired and they displayed normal behaviour patterns, the researchers found. The therapy also reduced the occurrence of seizures, which are reported to affect between 10 and 20 per cent of people with Fragile X. Fragile X Syndrome affects around 1 in 4,000 boys and 1 in 6,000 girls. It is the most common known cause of inherited intellectual disability. Affected children have developmental delays that impair speech and language, problems with social interactions and are often co-diagnosed with autism, anxiety and seizures. Metformin is already approved by the UK's Medicines and Healthcare products Regulatory Agency and the US Food and Drug Administration as a therapy for type 2 diabetes. Dr. Christos Gkogkas, Chancellor's Fellow at the University of Edinburgh's Patrick Wild Centre, said: "Metformin has been extensively used as a therapy for type 2 diabetes for more than 30 years, and its safety and tolerability are well documented. Our study suggests the drug could be a novel therapeutic for Fragile X syndrome, a common type of autism. We next plan to investigate whether metformin offers any benefits for other types of autism." Dr. Nahum Sonenberg, James McGill Professor at McGill University's Biochemistry Department, commented: "This is some of the most exciting research work in my career, as it offers great promise in treating a pernicious genetic disease for which there is no cure." Dr. Andrew Stanfield, Co-director of the Patrick Wild Centre for Research into Autism, Fragile X Syndrome and Intellectual Disabilities, said: "'These findings are particularly important as metformin is a commonly used drug for other conditions so we already know a lot about its safety profile. If clinical trials in people with Fragile X syndrome are successful then it could be in use much more quickly than would be the case for a brand new medication." The study is published in the journal Nature Medicine.

10.1038/nm.4335

Medicine

Injections may have passed on Alzheimer's 'seeds': study

"Evidence for human transmission of amyloid-β pathology and cerebral amyloid angiopathy." Nature 525, 247–250 (10 September 2015) doi:10.1038/nature15369 . www.nature.com/nature/journal/ … ull/nature15369.html Journal information: Nature

http://www.nature.com/nature/journal/v525/n7568/full/nature15369.html

https://medicalxpress.com/news/2015-09-alzheimer-seeds.html

Abstract More than two hundred individuals developed Creutzfeldt–Jakob disease (CJD) worldwide as a result of treatment, typically in childhood, with human cadaveric pituitary-derived growth hormone contaminated with prions 1 , 2 . Although such treatment ceased in 1985, iatrogenic CJD (iCJD) continues to emerge because of the prolonged incubation periods seen in human prion infections. Unexpectedly, in an autopsy study of eight individuals with iCJD, aged 36–51 years, in four we found moderate to severe grey matter and vascular amyloid-β (Aβ) pathology. The Aβ deposition in the grey matter was typical of that seen in Alzheimer’s disease and Aβ in the blood vessel walls was characteristic of cerebral amyloid angiopathy 3 and did not co-localize with prion protein deposition. None of these patients had pathogenic mutations, APOE ε4 or other high-risk alleles 4 associated with early-onset Alzheimer’s disease. Examination of a series of 116 patients with other prion diseases from a prospective observational cohort study 5 showed minimal or no Aβ pathology in cases of similar age range, or a decade older, without APOE ε4 risk alleles. We also analysed pituitary glands from individuals with Aβ pathology and found marked Aβ deposition in multiple cases. Experimental seeding of Aβ pathology has been previously demonstrated in primates and transgenic mice by central nervous system or peripheral inoculation with Alzheimer’s disease brain homogenate 6 , 7 , 8 , 9 , 10 , 11 . The marked deposition of parenchymal and vascular Aβ in these relatively young patients with iCJD, in contrast with other prion disease patients and population controls, is consistent with iatrogenic transmission of Aβ pathology in addition to CJD and suggests that healthy exposed individuals may also be at risk of iatrogenic Alzheimer’s disease and cerebral amyloid angiopathy. These findings should also prompt investigation of whether other known iatrogenic routes of prion transmission may also be relevant to Aβ and other proteopathic seeds associated with neurodegenerative and other human diseases. Main Human transmission of prion disease has occurred as a result of a range of medical and surgical procedures worldwide as well as by endocannibalism in Papua New Guinea, with incubation periods that can exceed five decades 12 , 13 . A well-recognized iatrogenic route of transmission was by treatment of persons of short stature with preparations of human growth hormone, extracted from large pools of cadaver-sourced pituitary glands, some of which were inadvertently prion-contaminated. Such treatments commenced in 1958 and ceased in 1985 following the reports of the occurrence of CJD amongst recipients. A review of all 1,848 patients who were treated with cadaveric-derived human growth hormone (c-hGH) in the United Kingdom from 1959 through 1985 found that 38 had developed CJD by the year 2000 with a peak incubation period of 20 years 1 . Multiple preparations using different extraction methods were used over this period and patients received batches from several preparations. One preparation (Wilhelmi) was common to all patients who developed iCJD and size-exclusion chromatography, used in non-Wilhelmi preparation methods, may have reduced prion contamination 1 . As of 2012, a total of 450 cases of iatrogenic CJD have been recognized worldwide after treatment with c-hGH or gonadotropin (226 cases), transplantation of dura mater (228) or cornea (2), and neurosurgery (4) or electroencephalography recording using invasive medical devices (2) 2 . In France, 119/1,880 (6.3%) recipients developed iCJD, in the UK 65/1,800 (3.6%) and in the USA 29/7,700 (0.4%) 2 , 14 . Since 2008, most UK patients with prion disease have been recruited into the National Prion Monitoring Cohort study 5 , including 22 of 24 recent patients with iatrogenic CJD (iCJD) related to treatment with c-hGH over this period, all of whom necessarily have very long incubation periods. Of this group of patients with iCJD, eight patients (referenced no.s 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , Supplementary Information ) aged 36–51 years, with an incubation period from first treatment to onset of 27.9–38.9 years (mean 33 years) and from last treatment to onset of 18.8–30.8 years (mean 25.5 years), underwent autopsy with extensive brain tissue sampling at our hospital. In all eight brain samples we confirmed prion disease with abnormal prion protein labelling of the neuropil, perineuronal network and in most cases microplaques as described previously 15 , 16 , 17 . However, four (no.s 4, 5, 6, 8) of the eight patients with iCJD also showed substantial amyloid-β (Aβ) deposition in the central nervous system parenchyma by histology ( Fig. 1 ) and immunoblotting ( Fig. 2 ). A further two brain samples (no.s 1, 3) had focal Aβ pathology in one of the brain regions; one showed Aβ entrapment in PrP plaques and only one was entirely negative for Aβ. Furthermore, there was widespread cortical and leptomeningeal cerebral Aβ angiopathy (CAA) 3 in three patients (no.s 4, 6, 8) and focal CAA in one patient (no. 5) ( Fig. 1 ). Such pathology is extremely rare in this age range, 10/290 in the equivalent 36–50 year age strata without CJD 18 , P = 0.0002, Fisher’s test. None of our patients with iCJD had pathogenic mutations in the prion protein gene ( PRNP ). We used a custom next generation sequencing panel 4 to exclude mutation in any of 16 other genes associated with early-onset Alzheimer’s disease, CAA, or other neurodegenerative disorders, and none carried APOE ε4 or TREM2 R47H alleles ( Supplementary Table 2 ). Although such observations are unprecedented in our wide experience of human prion diseases, we nevertheless considered whether prion disease itself might predispose to, or accelerate, Aβ pathology, for example by cross-seeding of protein aggregation or overload of clearance mechanisms for misfolded proteins. We therefore compared the Aβ pathology in the iCJD cohort with that of a cohort of 116 patients with other prion diseases who had undergone autopsy: sporadic CJD (sCJD) ( n = 85, age 42–83), variant CJD ( n = 2, age 25 and 36) and inherited prion diseases (IPD) ( n = 29, age 29–86). None of the patients in the control cohorts had comparable Aβ pathology (Consortium to Establish a Registry for Alzheimer’s disease (CERAD) score, P = 0.001, CAA, P = 0.005, topographical Aβ score P = 0.02, and cumulative Aβ score P = 0.02 (rank sum test) and digital Aβ quantification P = 0.04 ( t -test); all restricted to the strata aged 36–51 years ( n = 19)) ( Fig. 3 , and Extended Data Figs 1 and 2 show similar results in adjusted analyses in the full cohort). Indeed none of 35 prion cases aged 52–60 had significant Aβ pathology, with the exception of two cases at ages 57 and 58 positive for APOE ε4 alleles. Instead, the sCJD cohort shows Aβ pathology in parenchyma and blood vessels to a similar extent/severity as seen in iCJD, only in a much older age group ( Extended Data Figs 1 and 2 ), in keeping with the chance coincidence of late-onset Aβ pathology and sCJD as previously documented in a large study of 110 sCJD patients and 110 age-matched controls aged 27–84 (ref. 19 ) and a study of 2,661 individuals aged 26–95 (ref. 18 ). Further, we investigated whether prion and Aβ pathology co-localize in the iCJD cases. In our series there was a distinct absence of overlap of Aβ plaques and PrP ( Fig. 1d, e ) or Aβ CAA and vascular PrP ( Fig. 1b, c ), consistent with these pathologies developing independently. Figure 1: Aβ accumulation in central nervous system parenchyma and blood vessels (CAA) in iCJD. a , Frontal cortex with widespread diffuse Aβ deposition, formation of plaques, and widespread parenchymal and leptomeningeal CAA (patient no. 4). b , c , Non-colocalized deposition of Aβ and prion protein. Vessels with CAA do not entrap or co-seed prion protein. d, e , Adjacent histological sections stained for Aβ or prion protein show clearly separated plaques of both proteins (no. 5). f , An overlay with colour inversion of prion protein plaques highlights the separation. g , h , Dual labelling, confocal laser microscopy shows no co-localization of parenchymal Aβ plaques (no.s 5, 6) or CAA (no. 6). i , Aβ is detected in pituitary glands in patients with a high Aβ load in the brain. Scale bar corresponds to 200 μm in a , 100 μm in b–h , and 50 μm in i . PowerPoint slide Full size image Figure 2: Immunoblots of Aβ in iCJD patient brains. a – c , 10% (w/v) brain homogenates from patients with iCJD were analysed by enhanced chemiluminescence using anti-human Aβ monoclonal antibodies 6E10 that recognizes full-length APP and fragments that contain the epitope including Aβ ( a ) or 82E1 that specifically recognizes Aβ ( b ) or secondary antibody only ( c ). The identity of the patient brain sample is designated above each lane and the position of molecular mass markers is shown to the left. The equivalent of 5 µl 10% (w/v) brain homogenate was loaded per lane. The migration position of Aβ is indicated by the arrow. For gel source data, see Supplementary Fig. 1 . PowerPoint slide Full size image Figure 3: Early Aβ accumulation in the parenchyma and blood vessels in a subset of eight patients with iCJD aged 36–51 years, but not in controls (stratum aged 36–51 years) of 19 prion diseases of other aetiologies, suggests human transmission. a , Widespread, moderate-to-severe early-onset CAA in three, and focal, mild CAA in one iCJD patient but only one focal, mild CAA in 19 controls. b , Significant differences of parenchymal Aβ accumulation (all central nervous system regions, see supplementary material ). c , d , Cortical Aβ load was assessed semiquantitatively and quantitatively and again was significantly different between the iCJD and age-matched control cohort. For methods of quantification and calculations of significance levels see Supplementary Information . PowerPoint slide Full size image We then went on to examine pituitary glands for the presence of Aβ deposits. Pathological species of tau, Aβ and α-synuclein have been reported in the pituitary gland of patients with neurodegenerative disease and controls 20 . We examined 55 pituitary glands, 6 from patients without, and 49 from patients with cerebral Aβ pathology, and found in the latter group seven samples containing Aβ, confirming frequent Aβ in pituitaries of patients with Alzheimer’s disease-like pathology 20 ( Fig. 1i and Extended Data Fig. 3 ), consistent with the hypothesis that Aβ seeds have been iatrogenically transmitted to these patients with iCJD. There has been longstanding interest as to whether other neurodegenerative diseases associated with the accumulation of aggregates of misfolded host proteins or amyloids might be transmissible in a ‘prion-like’ fashion 21 , 22 . Experimental seeding of Aβ pathology has previously been demonstrated in primates and transgenic mice by central nervous system inoculation with Alzheimer’s disease brain homogenate 6 , 7 , 8 , 9 , 10 . Of particular interest with respect to our findings is that peripheral (intraperitoneal) inoculation with Alzheimer’s disease brain extract into APP23 (ref. 11 ) transgenic mice has been demonstrated. While ageing APP23 mice show mostly parenchymal deposits, the intraperitoneally-seeded mice showed predominantly CAA, a feature seen in patients with iCJD who had significant Aβ pathology. This experimental study and our findings suggest that there are mechanisms to allow the transport of Aβ seeds as well as prions (and possibly other proteopathic seeds such as tau 23 ) from the periphery to the brain 24 , 25 . While less than 4% of UK c-hGH treated individuals have developed iCJD, one out of eight patients with iCJD had focal, and three had widespread, moderate or severe CAA. Four patients had widespread parenchymal Aβ pathology and two further patients had focal cortical Aβ deposits. This might suggest that healthy individuals exposed to c-hGH are at high risk of developing early-onset Aβ pathology as this cohort ages. Although none of the iCJD cases with Aβ pathology had hyperphosphorylated tau neurofibrillary tangle pathology characteristic of Alzheimer’s disease, it is possible that the full neuropathology of Alzheimer’s disease would have developed had these individuals not succumbed to prion disease at these relatively young ages. An earlier study concluded that c-hGH recipients did not seem to be at increased risk of Alzheimer’s disease, but this was based on death certificates only without autopsy data 20 . However, the severe CAA seen in the patients with iCJD in our study is unquestionably concerning and individuals with such pathology would be at increasing risk of cerebral haemorrhages had they lived longer. At-risk individuals, including patients who had received dura mater grafts 26 could be screened by magnetic resonance imaging (MRI) for CAA-related pathologies (such as microbleeds) and by positron emission tomography (PET) for Aβ deposition 27 . It is possible, however, that prions and Aβ seeds co-purify in the extraction methods used to prepare c-hGH, which might mean that there would be a relatively higher occurrence of Aβ pathology in those with iatrogenic prion infection. Analysis of any residual archival batches of c-hGH for both prions and Aβ seeds might be informative in this regard 2 . While our data argue against cross seeding, we cannot formally exclude the possibility that prions somehow seed Aβ deposition but do not co-localize with Aβ deposits. While there is no suggestion that Alzheimer’s disease is a contagious disease and no supportive evidence from epidemiological studies that Alzheimer’s disease is transmissible, notably by blood transfusion 28 , 29 , our findings should prompt consideration of whether other known iatrogenic routes of prion transmission, including surgical instruments and blood products, may also be relevant to Aβ and other proteopathic seeds seen in neurodegenerative diseases. Aβ seeds are known, like prions, to adhere to metal surfaces and to resist formaldehyde inactivation and conventional hospital sterilisation 30 . Methods No statistical methods were used to predetermine sample size, the experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment. Patient recruitment and genotyping A national referral system for prion diseases was established by the Chief Medical Officer in the UK in 2004. UK neurologists were asked to refer all patients with suspected prion disease jointly to the National CJD Research and Surveillance Unit in Edinburgh and the NHS National Prion Clinic (NPC) in London. All patients with possible CJD who had received cadaver-derived growth hormone were referred to the NHS National Prion Clinic (London, UK) and since 2008 were recruited into the National Prion Monitoring Cohort study. Next-generation sequencing to exclude mutations known to be causal of Aβ pathology Deep next-generation sequencing using a custom panel was performed as described previously 31 . Analysis was done using NextGENe and Geneticist Assistant software (Softgenetics, USA). Variants were assessed for pathogenicity by reference to the published literature, control population allele frequencies (our primary database for allele frequency was the Broad Institute’s ExAC browser ( )) and in silico predictive tools. The analysis methodology has been validated for the detection of APP duplication 31 , which was important to exclude. No causal mutations for dementia or Aβ pathology were detected, see Supplementary Table 2 . As expected, several rare variants were detected which may modify the risk of various neurodegenerative diseases, see Supplementary Table 2 . Autopsies and tissue preparation Autopsies were carried out in a post mortem room designated for high risk autopsies. Informed consent to use the tissue for research was obtained in all cases. Ethical approval for these studies was obtained from the Local Research Ethics Committee of the UCL Institute of Neurology/National Hospital for Neurology and Neurosurgery. The anterior frontal, temporal, parietal and occipital cortex and the cerebellum (at the level of dentate nucleus) were dissected during the post mortem procedure and frozen. Samples of the following areas were taken and analysed: frontal, temporal, parietal, occipital, posterior frontal cortex including motor strip, basal ganglia, thalamus, hippocampus, brain stem including midbrain, and cerebellar hemisphere and vermis. Pituitary glands were taken in all cases. Tissue samples were immersed in 10% buffered formalin and prion infectivity was inactivated by immersion into 98% formic acid for one hour. Tissue samples were processed to paraffin wax and tissue sections were routinely stained with haematoxylin and eosin. Antibodies and immunohistochemistry The following antibodies were used: Anti-PrP ICSM35 (D-Gen Ltd, London, UK 32 , 33 1:1,000), Anti-phospho-Tau (AT-8, Innogenetics, 1:100) and anti-βA4 (DAKO 6F3D, 1:50). ICSM35 was stained on a Ventana Benchmark or Discovery automated immunohistochemical staining machine (ROCHE Burgess Hill, UK); βA4 and Tau were stained on a LEICA BondMax (LEICA Microsystems) or a Ventana automated staining instrument following the manufacturer’s guidelines, using biotinylated secondary antibodies and a horseradish-peroxidase-conjugated streptavidin complex and diaminobenzidine as a chromogen. Immunoblot detection of Aβ in iCJD brain Biochemical studies were carried out in a microbiological containment level 3 facility with strict adherence to safety protocols. Frozen brain tissue was available from seven of eight patients with growth hormone iCJD (cases 1 and 3–8). 10% (w/v) brain homogenates (grey matter; frontal cortex) were prepared in Dulbecco’s PBS lacking Ca 2+ or Mg 2+ ions using tissue grinders as described previously 34 . 20-μl aliquots were treated with 1 µl benzonase nuclease (purity >99%; 25 U ml −1 ; Novagen) for 15 min at 20 °C. Samples were then mixed with an equal volume of 2× SDS sample buffer (125 mM Tris-HCl, 20% (v/v) glycerol pH 6.8 containing 4% (w/v) SDS, 4% (v/v) 2-mercaptoethanol and 0.02% (w/v) bromophenol blue) and immediately transferred to a 100 °C heating block for 10 min. Electrophoresis was performed on 16% Tris-glycine gels (Invitrogen), run for 70 min at 200 V, before electroblotting to Immobilon P membrane (Millipore) for 16 h at 15 V as described previously 34 . Membranes were blocked in phosphate buffered saline (PBS) containing 0.05% (v/v) Tween 20 (PBST) and 5% (w/v) non-fat dried skimmed milk powder. Blots were then probed with anti-human Aβ monoclonal antibodies 6E10 (Covance) and 82E1 (IBL international, Hamburg, Germany) at final concentrations of 0.2 µg ml −1 in PBST for at least 1 h. After washing for 1 h with PBST the membranes were probed with a 1:10,000 dilution of alkaline-phosphatase-conjugated goat anti-mouse IgG secondary antibody (Sigma-Aldrich no. A2179) in PBST. After washing (90 min with PBST and 5 min with 20 mM Tris pH 9.8 containing 1 mM MgCl 2 ) blots were incubated for 5 min in chemiluminescent substrate (CDP-Star; Tropix Inc.) and visualized on Biomax MR film (Carestream Health Inc.). Anti-human Aβ monoclonal antibody 82E1 recognizes an epitope specific to the amino terminus of Aβ while 6Ε10 recognizes an epitope spanning residues 3–8 of Aβ and cross-reacts with full-length APP or APP fragments that contain the epitope. Examination of prion pathology In all iCJD cases there was variably prominent microvacuolar change in the neocortices, deep grey nuclei and cerebellar cortex. Immunostaining for the abnormal prion protein revealed synaptic labelling in all grey matter areas examined. In all but one case, there were also microplaques in all grey matter structures. Variability in the intensity of the immunoreactivity for the abnormal prion protein was evident but detailed comparison between the cases and separately within each case was not feasible as prolonged formalin fixation in some cases significantly attenuated the immunoreactivity. It was apparent that more prominent microvacuolar change and synaptic labelling for abnormal prion protein was more intense in the pre-central gyrus and parietal lobe when compared to the anterior frontal and occipital cortices. Deep cortical layers showed more severe changes. In all cases the microvacuolar degeneration and prion protein deposits in the deep grey nuclei and hippocampal formation was prominent. It was most severe in the caudate nucleus and putamen, and appeared less severe in thalamus and it was least prominent in the globus pallidus. In the cerebellar vermis there was marked granule cell atrophy and often widespread loss of Purkinje cells accompanied by severe Bergmann gliosis, while cerebellar hemispherical cortex showed only patchy loss of Purkinje cells and no significant granule cell loss. Microvacuolar degeneration in the molecular layer was more prominent in the vermis than in the cerebellar hemisphere. No apparent difference in prion protein deposition was seen in vermis and hemisphere. In the dentate nucleus variably intense synaptic prion protein immunoreactivity was present, while the cyto-architecture of the nucleus was well preserved. Examination, classification and quantification of Aβ pathology All brains were examined according to the ABC classification 35 , which assesses the topographic progression of Aβ pathology in the brain (Thal phases 36 ), topographic progression of Tau neurofibrillary tangle pathology (Braak and Braak 37 ) and the density of mature (senile), neuritic plaques in the neocortex (Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) criteria 38 , 39 ). To allow a more detailed assessment of neocortical Aβ the original Thal phases were modified as follows. Phase 0, no cortical Aβ; phase 0.5, 1–2 neocortical regions affected; phase 1, 3–4 neocortical regions involved; phases 2–5 were scored as published 36 . In addition we have carried out a semiquantitative assessment of neocortical Aβ load on a standardised region within frontal, temporal, parietal and occipital lobes, and scored as follows. 0, entirely negative; 1, a single small deposit; 2, multiple small deposits, disseminated; 3, multiple small deposits, plus an area with a larger patch; 4, diffuse moderate numbers of deposits; 5, diffuse, frequent numbers of deposits. For each case a cumulative score (0–20) of total semiquantitatively assessed Aβ load in the neocortex was calculated. Cerebral amyloid angiopathy (CAA) was graded (0–3) according to the Vonsattel criteria 3 . CAA was assessed in leptomeninges and parenchyma of all hemispheric lobes and cerebellum with summary score (0–30) calculated for each case. Image acquisition and processing Histological slides were digitised on a LEICA SCN400F scanner (LEICA Milton Keynes, UK) at ×40 magnification and 65% image compression setting during export. Slides were archived and managed on LEICA Slidepath (LEICA Milton Keynes, UK). For the preparation of light microscopy images, 1,024 × 1,024 pixel sized image captures were taken, after matching paired images (Aβ and prion staining) in Slidepath, and overlays in Fig. 1f were prepared using the colour conversion function in conjunction with the image overlay in Slidepath. Laser scanning microscopy of double immunofluorescent tissue preparations was on a ZEISS LSM710 confocal microscope (ZEISS Cambridge, UK). Publication figures were assembled in Adobe Photoshop. Data plots were generated using Prism 5 (GraphPad Software, Inc., La Jolla, USA). Digital image analysis for cortical Aβ quantification From all cases Aβ immunostained slides from frontal, temporal, parietal and occipital lobes were digitised as described above. Digital image analysis on 496 whole slides was performed using Definiens Developer 2.3 (Definiens, Munich, Germany). Initial tissue identification was performed at a resolution corresponding to 5× image magnification and stain detection was performed at ×10 resolution. Tissue detection and initial segmentation was done to identify all tissue within the image, separating the sample from background and non-tissue regions for further analysis. This separation was based on identification of the highly homologous relatively bright/white region of background present at the perimeter of each image. A composite raster image produced by selecting the lowest pixel value from the three comprising colour layers (RGB colour model) provided a greyscale representation of brightness. The mean brightness of this background region was used to exclude all background regions from further analysis. Stain detection (brown) is based on the transformation of the RGB colour model to a HSD representation 40 . This provides a raster image of the intensity of each colour of interest (brown and blue). A series of dynamic thresholds ( T x ) are then used to identify areas of interest ( A x ). Initially, following exclusion of intensely stained areas with values greater than 1 arbitrary unit (au) (values range from 0au to 3au in HSD images), the 5th centile ( ) of brown stain intensity was calculated as a baseline. This represents the T brown stain separating the top 5% of A tissue . The standard deviation ( C 5 δ ) within the lower 95% of A tissue was used to update the T brown stain as with all pixels above this threshold classed as ‘stain’ ( A stain ) and those below as ‘unstained’ ( A unstained ). A stain was excluded if the intensity of blue staining was not significantly lower than the level of brown stain (difference less than 0.1au) to remove generically dark areas. The remaining A stain were further categorised using thresholds based on the mean ( ) and standard deviation (B δ ) of brown staining within the A unstained : T brown = (lower threshold); T dark brown = (upper threshold), to give A unstained ≤ T brown > A light brown ≤ T dark brown > A Aβ deposit . Artefacts were then identified as A stain with area greater than 1 mm 2 , or an area greater than 0.1 mm 2 with a standard deviation of brown staining below 0.2au. These A artefacts were then expanded to include surrounding pixels with brown staining greater than . This excludes large areas of homogenous staining and areas of more diffuse, non-specific chromogen deposit. The white matter region within the tissue was then manually selected by an expert neuropathologist (Z.J., S.B.). This white matter was excluded from calculation of proportional coverage of A Aβ deposit within A tissue . Change history 11 September 2015 Due to an administrative error, an incorrect version of the Competing Financial Interests (CFI) statement was published in relation to this paper. Although the published CFI statement did reference the authors' affiliation with D-Gen, it did not contain all of the information provided to Nature by the authors about the interests of the company. As soon as Nature was made aware of the error, we acted quickly to correct the text and expand the CFI statement. The CFI statement for this paper as originally published was "J.C. is a Director and J.C. and J.D.F.W. are shareholders of D-Gen Limited, which supplies antibody ICSM35." The updated CFI statement is “J.C. is a Director and J.C. and J.D.F.W. are shareholders of D-Gen Limited, an academic spin-out company working in the field of prion disease diagnosis, decontamination and therapeutics. D-Gen supplied antibody ICSM35”.

People injected with hormones extracted from cadaver brains in a long-abandoned procedure may have received "seeds" of Alzheimer's disease, said a study Wednesday, urging research into possible risks for "accidental" medical transmission. Published in the journal Nature, the research claims to provide evidence for the hypothesis that the protein fragments which assemble into Alzheimer's-causing plaques, can be passed between humans via diseased tissue transfer. But this did not mean that Alzheimer's was contagious, the study authors and independent commentators stressed. "This relates to a very special situation where people have been injected with essentially extracts of human tissue," said co-author John Collinge of University College London (UCL). "I don't think there needs to be any alarm that we're saying in any way that you can catch Alzheimer's Disease." Further research, however, would be "prudent," he said during a telephone press briefing. "We should think about whether there might be accidental routes in which these diseases might be transmitted by medical or surgical procedures." While conducting research into an unrelated disease, Collinge and a team examined the brains of eight people who had received injections in childhood of a hormone to treat dwarfism. The hormone had been extracted from pituitary glands harvested from thousands of human cadavers. This practice was halted in 1985 when doctors realised it could transmit a variant of Creutzfeldt-Jakob disease (CJD)—the human version of "mad cow" disease. Eight subjects in the study in fact died from this ailment. Too young Collinge and colleagues, "very much to our surprise," found that seven of the eight had brain deposits of Alzheimer's-linked amyloid beta (Abeta) fragments—with four of them having high concentrations. Strikingly the patients were 36-51 years old, whereas such deposits are normally seen in elderly people. "We think the most likely explanation is that the growth hormone preparations with which these people were treated as children, in addition to being contaminated with CJD prions (a different protein type), was probably also contaminated with Abeta seeds." Previous laboratory studies showed that Abeta in Alzheimer's-ridden brain tissue, when transferred to mice or monkeys, could infect the host animal brain—even when it had been injected into their abdomens. "So there are mechanisms to transport these protein seeds to the brain," said Collinge. "We don't know what they are, but clearly it can happen. So that's consistent with these seeds spreading from an intramuscular injection in the children to their brains." Amyloid beta seeds, the team wrote, "are known, like prions, to adhere to metal surfaces and to resist... conventional hospital sterilisation." Experts who were not part of the study underlined there was no evidence of any modern-day medical treatment, including dental surgery or blood transfusions, raising the Alzheimer's risk. Cautious, not concerned For the time being, "this paper should make us cautious but not overly concerned," said Simon Lovestone of the University of Oxford. John Hardy, of UCL, added it seemed "relatively sure" that Abeta can be transferred by injection. "Does it have implications for... blood transfusions: probably not, but this definitely deserves systematic epidemiological investigation," he said via the Science Media Centre (SMC) in London. "Does it suggest Alzheimer's disease is infectious through contact? Almost certainly not." The study authors said the eight fatalities in the study did not have the full-blown features of Alzheimer's—they were missing the "tangles" caused by a different protein called Tau. It was impossible to know whether they would have gone on to develop the disease. "I wouldn't want to cause alarm on this. I don't think anyone should delay or reconsider having surgery on the basis of these data at all," said Collinge. "We've got no evidence that this is a risk to humans, but I think it would be prudent to do some research in this area going forward." Some 30,000 people, mostly children with growth deficiency, received the hormone injections, of whom over 200 developed CJD. The disease has a very long incubation period, and new diagnoses continue to be made.

www.nature.com/nature/journal/ … ull/nature15369.html

Physics

Skyrmions can fly!

Yijie Shen et al, Supertoroidal light pulses as electromagnetic skyrmions propagating in free space, Nature Communications (2021). DOI: 10.1038/s41467-021-26037-w Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-021-26037-w

https://phys.org/news/2021-10-skyrmions.html

Abstract Topological complex transient electromagnetic fields give access to nontrivial light-matter interactions and provide additional degrees of freedom for information transfer. An important example of such electromagnetic excitations are space-time non-separable single-cycle pulses of toroidal topology, the exact solutions of Maxwell’s equations described by Hellwarth and Nouchi in 1996 and recently observed experimentally. Here we introduce an extended family of electromagnetic excitation, the supertoroidal electromagnetic pulses, in which the Hellwarth-Nouchi pulse is just the simplest member. The supertoroidal pulses exhibit skyrmionic structure of the electromagnetic fields, multiple singularities in the Poynting vector maps and fractal-like distributions of energy backflow. They are of interest for transient light-matter interactions, ultrafast optics, spectroscopy, and toroidal electrodynamics. Introduction Topology of complex electromagnetic fields is attracting growing interest of the photonics and electromagnetics communities 1 , 2 , 3 , 4 , 5 , while topologically structured light fields find applications in super-resolution microscopy 6 , metrology 7 , 8 , and beyond 9 , 10 . For example, the vortex beam with twisted phase, akin to a Mobius strips in phase domain, can carry orbital angular momentum with tunable topological charges enabling advanced applications of optical tweezers, machining, and communications 9 , 10 , 11 , 12 . The complex electromagnetic topological strips, knots, and caustic structures were also proposed as novel information carriers 13 , 14 , 15 , 16 . Recently, skyrmions, as topologically protected quasiparticles in high-energy physics and magnetic materials 17 , were also studied in optical electromagnetic fields. Optical skyrmionic fields were first demonstrated in the evanescent field of a plasmonic surface 18 , 19 , followed by the spin field of confined free-space waves 20 , 21 , Stokes vectors of paraxial vector beams 22 , 23 , 24 , 25 , polarizations in momentum-space 26 , and pseudospins in photonic crystals 27 . The sophisticated vector topology of optical skyrmions holds potential for applications in ultrafast nanometric metrology 28 , deeply subwavelength microscopy 20 , and topological Hall devices 27 . While a large body of work on topological properties of structured continuous light beams may be found in literatures, works on the topology of the time-dependent electromagnetic excitations and pulses only start to appear. For instance, the “Flying Doughnut” pulses, or toroidal light pulses (TLPs) first described in 1996 by Hellwarth and Nouchi 29 , with unique spatiotemporal topology predicted recently 30 , have only very recently observed experimentally 31 . Fueled by a combination of advances in ultrafast lasers and metamaterials in our ability to control the spatiotemporal structure of light 32 , 33 together with the introduction of experimental and theoretical pulse characterization methods 34 , 35 , 36 , 37 , TLPs are attracting growing attention. Indeed, TLPs exhibit their complex topological structure with vector singularities and interact with matter through coupling to toroidal and anapole localized modes 38 , 39 , 40 , 41 . However, while higher order, supertoroidal modes in matter have been introduced in the form of the fractal iterations of solenoidal currents 42 , 43 , 44 , 45 , 46 , 47 , generalizations of free-space propagating toroidal pulses have not been considered to date. In this paper, we report that the Hellwarth and Nouchi pulses, are, in fact, the simplest example of an extended family of pulses that we will call supertoroidal light pulses (STLPs). We will show that supertoroidal light pulses introduced here exhibit complex topological structures that can be controlled by a single numerical parameter. The STLP display skyrmion-like arrangements of the transient electromagnetic fields organized in a fractal-like, self-affine manner, while the Poynting vector of the pulses feature singularities linked to the multiple energy backflow zones. Results Supertoroidal electromagnetic pulses Following Ziolkowski, localized finite-energy pulses can be obtained as superpositions of “electromagnetic directed-energy pulse trains” 48 . A special case of the localized finite-energy pulses was investigated by Hellwarth and Nouchi 29 , who found the closed-form expression describing a single-cycle finite energy electromagnetic excitation with toroidal topology obtained from a scalar generating function f ( r , t ) that satisfies the wave equation \(({\nabla }^{2}-\frac{1}{{c}^{2}}\frac{{\partial }^{2}}{\partial {t}^{2}})f\left({{{{{\bf{r}}}}}},t\right)=0\) , where r = ( r , θ , z ) are cylindrical coordinates, t is time, \(c=1/\sqrt{{\varepsilon }_{0}{\mu }_{0}}\) is the speed of light, and the ε 0 and μ 0 are the permittivity and permeability of medium. Then, the exact solution of f ( r , t ) can be given by the modified power spectrum method 29 , 48 , as \(f({{{{{\bf{r}}}}}},t)={f}_{0}/\left[({q}_{1}+i\tau ){(s+{q}_{2})}^{\alpha }\right]\) , where f 0 is a normalizing constant, s = r 2 /( q 1 + i τ ) − i σ , τ = z − c t , σ = z + c t , q 1 and q 2 are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while α is a real dimensionless parameter that must satisfy α ≥ 1 to ensure finite energy solutions. In particular, the parameter α is related to the energy confinement of the pulse with α < 1 resulting in pulses of infinite energy, such as planar waves and cylindrical waves, while α ≥ 1 leads to finite-energy pulses. Next, transverse electric (TE) and transverse magnetic (TM) solutions are readily obtained by using Hertz potentials. The electromagnetic fields for the TE solution can be derived by the potential \({{{{{\bf{A}}}}}}({{{{{\bf{r}}}}}},t)={\mu }_{0}{{{{{\boldsymbol{\nabla }}}}}}\times \widehat{{{{{{\bf{z}}}}}}}f({{{{{\bf{r}}}}}},t)\) as \({{{{{\bf{E}}}}}}({{{{{\bf{r}}}}}},t)=-{\mu }_{0}\frac{\partial }{\partial t}{{{{{\boldsymbol{\nabla }}}}}}\times {{{{{\bf{A}}}}}}\) and H ( r , t ) = ∇ × ( ∇ × A ) 29 , 48 . Finally assuming α = 1, the electromagnetic fields of the TLP are described by 29 : $${E}_{\theta }=-4i{f}_{0}\sqrt{\frac{{\mu }_{0}}{{\varepsilon }_{0}}}\frac{r({q}_{1}+{q}_{2}-2ict)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{3}}$$ (1) $${H}_{r}=4i{f}_{0}\frac{r({q}_{2}-{q}_{1}-2iz)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{3}}$$ (2) $${H}_{z}=-4{f}_{0}\frac{{r}^{2}-\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{3}}$$ (3) where E θ represents the azimuthally directed component of electric field, and H r and H z are the radially and longitudinally directed components of magnetic field, respectively. Note that the TE mode field does not possess other kinds of components except the three, and the TM mode is expressed by exchanging the electric and magnetic fields. Equations ( 1 )–( 3 ) derived by Hellwarth and Nouchi for α = 1 show the simplest example of TLPs. Here we explore the general solution for values of α ≥ 1. In the TE case, electric and magnetic fields are given by (see detailed derivation in Supplementary Notes 1 – 3 ): $${E}_{\theta }^{(\alpha )}=-2\alpha i{f}_{0}\sqrt{\frac{{\mu }_{0}}{{\varepsilon }_{0}}}\left\{\frac{(\alpha +1)r{({q}_{1}+i\tau )}^{\alpha -1}({q}_{1}+{q}_{2}-2ict)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +2}}-\frac{(\alpha -1)r{({q}_{1}+i\tau )}^{\alpha -2}}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +1}}\right\}$$ (4) $${H}_{r}^{(\alpha )}=2\alpha i{f}_{0}\left\{\frac{(\alpha +1)r{({q}_{1}+i\tau )}^{\alpha -1}({q}_{2}-{q}_{1}-2iz)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +2}}-\frac{(\alpha -1)r{({q}_{1}+i\tau )}^{\alpha -2}}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +1}}\right\}$$ (5) $${H}_{z}^{(\alpha )}=-4\alpha {f}_{0}\left\{\frac{{({q}_{1}+i\tau )}^{\alpha -1}\left[{r}^{2}-\alpha \left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +2}}+\frac{(\alpha -1){({q}_{1}+i\tau )}^{\alpha -2}\left({q}_{2}-i\sigma \right)}{{\left[{r}^{2}+\left({q}_{1}+i\tau \right)\left({q}_{2}-i\sigma \right)\right]}^{\alpha +1}}\right\}$$ (6) For α = 1, the electromagnetic fields in Eqs. ( 4 )–( 6 ) are reduced to that of the fundamental TLP Eqs. ( 1 )–( 3 ). Moreover, the real and imaginary parts of Eqs. ( 4 )–( 6 ), simultaneity fulfill Maxwell equations and therefore represent real electromagnetic pulses. While propagating in free space, toroidal and supertoroidal pulses exhibit self-focusing. Figure 1 shows the evolution of the fundamental ( α = 1) TLP and STLP ( α = 5) upon propagation through the focal point. In the former case, the pulse is single-cycle at focus ( z = 0) becoming \(1\frac{1}{2}\) -cycle at the boundaries of Rayleigh range z = ± q 2 /2 (Fig. 1 a). On the other hand, STLPs with α > 1 (Fig. 1 b, α = 5) exhibit a substantially more complex spatiotemporal evolution where the pulse is being reshaped multiple times upon propagation. The STLP also possesses a more complex singular vector field configuration than the fundamental TLP. For instance, the magnetic vector distribution (see insets to Fig. 1 a, b) exhibits vortex-type singularities (gray lines) and saddle points (full circles), resulting in skyrmions structures in the transverse plane (colored arrows). See Supplementary Videos 1 and 2 for dynamic evolutions upon propagation of the fundamental TLP and STLP, and Supplementary Video 3 for the evolution of TLP and STLP with different focused degrees versus values of q 2 / q 1 . Fig. 1: From toroidal to supertoroidal light pulses. a , b Isosurfaces for the electric fields of a the fundamental TLP Re[ E θ ( r , t )], and b a STLP \(\,{{\mbox{Re}}}\,[{E}_{\theta }^{(\alpha )}({{{{{\bf{r}}}}}},t)]\) of α = 5, at amplitude levels of E = ±0.1 and the Rayleigh range of q 2 = 100 q 1 , at different times of t = 0, ± q 2 /(4 c ), and ± q 2 /(2 c ). x – z cross-sections of the instantaneous electric field at y = 0. The insets in ( a ) and ( b ) are schematics of spatial topological structures of magnetic vector fields at focus ( t = 0) for the fundamental TLP and STLP, respectively. The gray dots and rings mark the distribution of singularities (saddle points and vortex rings) in magnetic field, large pink arrows mark selective magnetic vector directions, and the smaller colored arrows show the skyrmionic structures in magnetic field. Full size image Electric field singularities Figure 2 comparatively shows the instantaneous electric fields for the TE single-cycle fundamental TLP and STLP ( α = 5) with q 2 = 20 q 1 at the focus ( t = 0). In all cases, there are always central singularities on z -axis ( r = 0; see vertical solid black lines in Fig. 2 a, b) owing to the azimuthal polarization. For the fundamental TLP, the electric field possesses two shell-like singular surfaces symmetrically distributed along axis. The singular shells divide the parts with opposite chiralities of azimuthal polarization, as shown in Fig. 2 a. Thus, the electric field forms counter-clockwise vortices around the z axis at z = q 1 and z = 35 q 1 (Fig. 2 a1, a3). On the other hand, at z = 5 q 1 (Fig. 2 a2) the electric field vanishes on a singular shell, while it changes its orientation from clockwise close to the z -axis to counter-clockwise away from the axis. For the STLP case, a more complex matryoshka-like structure emerges with multiple nested singularity shells, replacing the single shell of the fundamental TLP, as Fig. 2 b shows. The electric field configuration close to the singularity shells can be examined in detail at transverse planes at z = q 1 , 5 q 1 , 35 q 1 (Fig. 2 b1–b3). In this case, at transverse planes close to z = 0, the electric field changes orientation from counter-clockwise close to r = 0 to clockwise away from the z -axis (see Fig. 2 b1). On the other hand, on transverse planes close to z = 5 q 1 (see Fig. 2 b2), two singular rings (corresponding to the two singular shells) emerge as a cross-section of the multi-layer singularity shell structure, separating space in three different regions, in which the electric field direction alternates between counter-clockwise ( r / q 1 < 7), to clockwise (7 < r / q 1 < 15), and again to counter-clockwise ( r / q 1 > 15). Fig. 2: Electric fields of toroidal and supertoroidal light pulses. a , b The isoline plots of the electric field in the x – z plane for a the fundamental TLP, Re[ E θ ( r , t = 0)], and b the STLP of α = 5, \(\,{{\mbox{Re}}}\,[{E}_{\theta }^{(\alpha = 5)}({{{{{\bf{r}}}}}},t=0)]\) , in logarithmic scale. The bold black lines represent the zero-value singular lines. The dashed purple lines represent the positions of propagation distance corresponding to the transverse plots at right side. Panels a1 – a3 and b1 – b3 show the electric field distributions in the transverse planes. The field magnitude is plotted as contours (in logarithmic scale), while the field orientation is presented by arrow plots. Electric field zeros are marked by the black solid bold lines and black dots. Blue and red arrows represent the two opposite azimuthal directions of the electric fields. Unit for coordinates: q 1 . Full size image In general, the pulse of higher order of α is accompanied by a more complex multi-layer singular-shell structure, see the dynamic evolution versus the order index in Supplementary Video 4 . Although the above results of electric fields are instantaneous at t = 0, we note that the multi-layer shall structure propagation of supertoroidal light pulse is retained during propagation, see such dynamic process in Supplementary Video 5 . Magnetic field singularities The magnetic field of STLPs has both radial and longitudinal components, \({{{{{\bf{H}}}}}}={H}_{r}\widehat{{{{{{\bf{r}}}}}}}+{H}_{z}\widehat{{{{{{\bf{z}}}}}}}\) , which lead to a topological structure more complex than the one exhibited by the electric field. Figure 3 comparatively shows the instantaneous magnetic fields for the TLP and the STLP of α = 5. For the fundamental TLP (Fig. 3 a), the magnetic field has ten different vector singularities on the x – z plane, including four saddle points on z -axis and six vortex rings (the surrounding vector distribution forming a vortex loop) away from the z -axis. We note that we only consider the singularities existed at an area containing 99.9% of the energy of the pulse. While the singularity existed at the region far away from the pulse center with nearly zero energy can be neglected. The magnetic field distribution in the vicinity of the singularities is shown in more detail in Fig. 3 a1. Due to the axial symmetry of the pulse, the three singularities away from the z -axis correspond to three rings with vortices rotating clockwise ( z = 0) or counter-clockwse ( z > 0 and z < 0). For instance, Fig. 3 a2 presents the magnetic field distribution in the z = 0 plane, where the orientation of the magnetic field changes from parallel to the z -axis within the vortex ring to anti-parallel outside the singularity ring. For the STLP (Fig. 3 b), more vector singularities are unveiled in the magnetic field with six saddle points on z -axis and six off-axis singularities. The singularity structure of the STLP is presented in Fig. 3 b1. The orientation of the magnetic field around the on-axis saddle points is alternating between “longitudinal-toward radial-outward point” and “radial-toward longitudinal-outward”, similarly to the on-axis singularities of the TLP. Moreover, the off-axis singularities at z = 0 become now saddle points contributing to the singularity ring in the z = 0 plane. The remaining off-axis singularities are accompanied by clockwise and counter-clockwise magnetic field configurations at x > 0 and x < 0, respectively as shown Fig. 3 b2. Fig. 3: Magnetic fields of toroidal and supertoroidal light pulses. a , b Isoline and arrow plots of the magnetic fields in the x – z plane for a the fundamental TLP and b the STLP of α = 5, in logarithmic scale. Magnetic field singularities are marked by black dots with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panels a1 and b1 present the zoom-in plots corresponding to regions of blue boxes in ( a ) and ( b ), respectively. Panels a2 and b2 present transverse distributions of magnetic amplitude (in logarithmic scale) and normalized magnetic vectors at z = 0 planes, the positions marked by the black dashed lines in ( a ) and ( b ), respectively, where the magnetic fields vanish along the circular solid black lines with red arrows marking the styles of singularities (vortex for a2 and saddle for b2 ). Skyrmionic structures in magnetic fields of toroidal and supertoroidal light pulses: c Various textures of Néel-type skyrmionic structure observed at various transverse planes (see dashed purple lines in a and b ) for the fundamental TLP ( c1 – c2 ) and the STLP of α = 5 ( c3 – c6 ), which are demonstrated by the arrows with color-labeled longitudinal component value of magnetic field. The up-right insert of each panel shows the basic texture of the skyrmionic structure. Unit for coordinates: q 1 . Full size image Skyrmionic structure in magnetic field An intriguing feature of the topology of STLPs involves the presence of skyrmionic quasiparticle configurations, which can be observed in the magnetic field topology of STLPs. The skyrmion is a topologically protected quasiparticle with a hedgehog-like vectorial field, that gradually changes orientation as one moves away from the skyrmion center 49 , 50 , 51 . Recently skyrmion-like configurations have been reported in electromagnetism, including skyrmion modes in surface plasmon polaritons 18 and the spin field of focused beams 20 , 22 . Here we observe the skyrmion field configurations in the electromagnetic field of propagating STLPs. The topological properties of a skyrmionic configuration can be characterized by the skyrmion number s , which can be separated into a polarity p and vorticity number m 51 . The polarity represents the direction of the vector field, down (up) at r = 0 and up (down) at r → ∞ for p = 1 ( p = −1), the vorticity controls the distribution of the transverse field components, and another initial phase γ should be added for determining the helical vector distribution, see “Methods” for details. For the m = 1 skyrmion, the cases of γ = 0 and γ = π are classified as Néel-type, and the cases of γ = ± π /2 are classified as Bloch-type. The case for m = −1 is classified as anti-skyrmion. Here the vector forming skyrmionic structure is defined by the normalized magnetic field H / ∣ H ∣ of the STLP. Two examples of two skyrmionic structures in the fundamental TLP are shown in Fig. 3 c1 ( p = m = 1, γ = π ) and 3 c2 ( p = m = 1, γ = 0) occurre at the two transverse planes marked by purple dashed lines c1 and c2, which are both Néel-type skyrmionic structures, where the vector changes its direction from “down” at the center to “up” away from the center. In the case of the STLPs with more complex topology, it is possible to observe more skyrmionic structures. The STLP pulse ( α = 5) exhibits not only the clockwise ( p = m = 1, γ = π ) and counter-clockwise ( p = m = 1, γ = 0) Néel-type skyrmionic structures (Fig. 3 c3, c4), but also those with p = −1, m = 1, γ = π and p = −1, m = 1, γ = 0, in Fig. 3 c5, c6. In general, as the value of α increases, toroidal pulses show an increasingly complex magnetic field pattern with skyrmionic structures of multiple types, see Supplementary Video 4 . We also note that the topology of the STLP is maintained during propagation, see Supplementary Video 5 . Energy backflow and Poynting vector singularities The topological features of electromagnetic fields in supertoroidal pulses are linked to anomalous behavior of energy flow as represented by the Poynting vector S = E × H . An interesting effect for the fundamental TLP is the presence of energy backflow: the Poynting vector at certain regions is oriented against the prorogation direction (blue arrows in Fig. 4 a) 30 . Such energy backflow effects have been predicted and discussed in the context of singular superpositions of waves 52 , 53 , superoscillatory light fields 7 , 54 , and plasmonic nanostructures 55 . The Poynting vector map reveals a complex multi-layer energy backflow structure, as shown in Fig. 4 b. The energy flow vanishes at the positions of the electric and magnetic singularities and inherits their multi-layer matryoshka-like structure. Poynting vector vanishes at z = 0 plane, along the z -axis, and on the dual-layer matryoshka-like singular shells (marked by the black bold lines in Fig. 4 b). Importantly, energy backflow occurs at areas of relatively low energy density, and, hence, STLP as a whole still propagates forward. For the temporal evolution of the energy flow of the pulse see Supplementary Videos 4 and 5 . Fig. 4: Poynting vector fields of toroidal and supertoroidal light pulses. a , b Contour and arrow plots of the Poynting vector fields in the x – z plane, in logarithmic scale, for a ) the fundamental TLP and b the STLP of α = 5. Panels a1 and b1 present the zoom-in plots of the regions of blue boxes in ( a ) and ( b ), respectively. Poynting vector field zeros are marked by the black solid lines and black dots. Red and blue bold arrows highlight the regions of energy forward flow and backflow, respectively. Unit for coordinates: q 1 . Full size image Fractal patterns hidden in electromagnetic fields As the order α of the pulse increases (see Supplementary Video 4 ), the topological features of the STLP appear to be organized in a hierarchical, fractal-like fashion. A characteristic case of the STLP of α = 20 is presented in Fig. 5 . For the electric field, the matryoshka-like singular shells involve an increasing number of layers as one examines the pulse at finer length scales, forming a self-similar pattern that seems infinitely repeated. For the magnetic field, the saddle and vortex points are distributed along the propagation axis and in two planes crossing the pulse center, respectively. The distribution of singularities becomes increasingly dense as one approaches the center of the pulse, resulting in a self-similar pattern. A similar pattern can be seen for the Poynting vector map (see Supplementary Videos 4 and 5 ). Fig. 5: Fractal-like patterns in electromagnetic fields of supertoroidal pulses. a The isoline plot of the electric field in the x – z plane for the STLP of α = 20, \(\,{{\mbox{Re}}}\,[{E}_{\theta }^{(\alpha = 20)}({{{{{\bf{r}}}}}},t=0)]\) , in logarithmic scale. Electric field zeros are marked by the black solid lines and black dots. Panel a1 presents the zoom-in plot of region highlighted by blue box in ( a ). b The isoline plot of magnetic field amplitude (in logarithmic scale) and arrow plot of normalized magnetic vectors in the x – z plane for the STLP of α = 20. Magnetic field singularities are marked by black dots with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panel b1 presents the zoom-in plot of region highlighted in ( b ). Subwavelength features of skyrmionic structures: c1 – c4 The skyrmionic distributions of magnetic field at several transverse planes marked by dashed lines marked by “c1–c4” in ( b1 ). d1 – d4 The distribution of normalized magnetic field and its absolute value versus x for the skyrmionic structures in c1 – c4 . Insets illustrate the Subwavelength features at the regions highlighted by gray bands. Unit for coordinates: q 1 . Full size image Deep-subwavelength features of skyrmionic structures The fractal-like pattern of vectorial magnetic field of a high-order STLPs results in skyrmionic configurations with features changing much faster than the effective wavelength q 1 . Figures 5 c1–c4 show the four skyrmionic structures of the high-order STLP ( α = 20) at the four transverse planes marked by the dashed lines in Fig. 5 b1 at positions of z / q 1 = 14, 10, 6, 3.5, correspondingly. Here the observed skyrmionic structure is similar to the photonic skyrmion observed in ref. 20 . However, in contrast to the latter case, where skyrmionic structures were observed in the evanescent plasmonic field, here skyrmionic field configurations are observed in free-space propagating fields. Moreover, similarly to the “spin reversal” effect observed in deeply subwavelength scales in plasmonic skyrmionic fields 20 , here we demonstrate “subwavelength” features in propagating skyrmionic fields at scales much smaller than the effective wavelength (cycle length) of the STLPs. The four skyrmionic structures we obtained Fig. 5 c1–c4 have two different topologies with topological numbers of ( p , m , γ ) = (1, 1, π ), for c1 and c3, and (−1, 1, 0), for c2 and c4. In addition, they exhibit an effect of “spin reversal”, where the number reversals is given by \(\bar{p}=\frac{1}{2\pi }{[\beta (r)]}_{r = 0}^{r = \infty }\) ( β is defined as the radially-variant angle, see details in Method), e.g., \(\bar{p}=1,2,3,4\) for the skyrmionic structures in Fig. 5 c1–c4, respectively. Each reversal corresponds to a sign change of H z , which takes place over areas much smaller than the effective wavelength of the pulse ( q 1 ). The full width at half maximum of these areas for the four skyrmionic structures is 1/6, 1/10, 1/30, 1/50 of the effective wavelength, respectively. Conclusively, the sign reversals become increasingly rapid in transverse planes closer to the pulse center ( z = 0), see Fig. 5 d1–d4. Similarly, increasing the value of α leads to increasingly sharper singularities. Notably, in contrast to the fundamental TLP, the skyrmionic configurations in STLP occur at areas of higher energy density, and thus we expect that they could be observed experimentally. The topological structure of the STLP is directly related to the distribution of on-axis saddle points in its magnetic field. Indeed, the latter mark the intersection of the E -field singular shells with the z -axis, which in turn results in the emergence of different skyrmionic magnetic field patterns (see Fig. 5 and Videos 4 and 5 ). The number and position of on-axis magnetic field saddle points are defined by the supertoroidal parameter α . This is illustrated in Fig. 6 a, where we plot the number of on-axis H -field singularities as a function of alpha for a STLP with q 2 = 20 q 1 . The number of singularities is generally increasing with increasing α apart for values around α = 5.6 (marked by blue dashed line in Fig. 6 ). Moreover, the number of singularities increases in a ladder-like fashion, where only specific values of alpha lead to additional singularities. The origin of this behavior can be traced to changes in the pulse structure as α increases (see Fig. 6 b). For specific values of α , additional singularities appear away from the pulse center ( z = 0) and then move slowly towards it. On the other hand, the irregular behavior at α = 5.6 is a result of two singularities disappearing (see blue dashed line in Fig. 6 b). The topological structure of the STLP can be tuned also by the degree of focusing, which is quantified by the ratio q 2 / q 1 . In particular, tightly focused pulses exhibit a more complex topological structure at finer scales as opposed to collimated pulses (see Supplementary Video 6 ). Fig. 6: Evolution of on-axis singularity distribution versus supertoroidal order. a , b The numbers ( a ) and positions ( b ) of on-axis saddle-singularities of the magnetic field of the STLP ( q 2 = 20 q 1 ) within the range of z ∈ [0, 80 q 1 ], versus α . The blue dashed line marks where the number of singularities decreases. Full size image Discussion STLPs exhibit complex and unique topological structure. The electric field exhibits a matryoshka-like configuration of singularity shells, which divide the STLP into “nested” regions with opposite azimuthal polarization. The magnetic field exhibits skyrmionic structures with multiple topological textures at various transverse planes of a single pulse, related to the distribution of multiple saddle and vortex singularities. The electric and magnetic fields can be exchanged respecting to the difference of TE and TM modes. The instantaneous Poynting vector field exhibits multiple singularities with regions of energy backflow. The singularities of the STLP appear to be hierarchically organized resulting in self-similar, fractal-like patterns for higher-order pulses. The main challenges for the generation of supertoroidal pulses involve its toroidal topology, broad bandwidth (single-cycle duration), and complex spatially-dependent spectral structure (see Supplementary Note 5 ). We argue that supertoroidal pulses can be generated similarly to the generation of fundamental toroidal pulses 31 , 32 , i.e., by conversion of ultrashort linearly polarized pulses in a two-stage process. This process shall involve the linear-to-radial polarization conversion of an ultrashort laser pulse, followed by the spatio-spectral modification of the pulse in a multi-layered gradient metasurface. We anticipate that the requirement for the single-cycle temporal profile will be possible to be met if we use attosecond laser pulses as input. Alternatively, in the THz range, single-cycle pulses can be routinely generated by optical rectification of femtosecond optical pulses. In conclusion, to the best of our knowledge, STLPs are so far the only known example of free-space propagating skyrmions in electromagnetic field. Indeed, higher-order STLPs exhibit a range of different skyrmionic field configurations, which will be of interest in probing the topology of electromagnetic excitations in matter. Moreover, information can be encoded in the increasingly complex topological structure of the propagating pulses, which could be of interest for optical communications. Finally, the subwavelength features of the singular structures of the STLPs may lead to advanced approaches of super-resolution imaging and nanoscale metrology. Methods Solving the supertoroidal pulses The first step is to solve the scalar generating function f ( r , t ) that satisfies the wave equation \(({\nabla }^{2}-\frac{1}{{c}^{2}}\frac{{\partial }^{2}}{\partial {t}^{2}})f\left({{{{{\bf{r}}}}}},t\right)=0\) , where r = ( r , θ , z ) are cylindrical coordinates, t is time, \(c=1/\sqrt{{\varepsilon }_{0}{\mu }_{0}}\) is the speed of light, and the ε 0 and μ 0 are the permittivity and permeability of medium. The exact solution of f ( r , t ) can be given by the modified power spectrum method as \(f({{{{{\bf{r}}}}}},t)={f}_{0}/\left[({q}_{1}+i\tau ){(s+{q}_{2})}^{\alpha }\right]\) , where f 0 is a normalizing constant, s = r 2 /( q 1 + i τ ) − i σ , τ = z − c t , σ = z + c t , q 1 and q 2 are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while α is a real dimensionless parameter that must satisfy α ≥ 1 to ensure finite energy solutions. The next step is constructing the Hertz potential. For fulfilling the toroidal symmetric and azimuthally polarized structure, the Hertz potential should be constructed as \({{{{{\bf{A}}}}}}({{{{{\bf{r}}}}}},t)={\mu }_{0}{{{{{\boldsymbol{\nabla }}}}}}\, \times\, \widehat{{{{{{\bf{z}}}}}}}f({{{{{\bf{r}}}}}},t)\) . Then, the exact solutions of transverse electric (TE) and transverse magnetic (TM) modes are readily obtained by using Hertz potential. The electromagnetic fields for the TE solution can be derived by the potential as \({{{{{\bf{E}}}}}}({{{{{\bf{r}}}}}},t)=-{\mu }_{0}\frac{\partial }{\partial t}{{{{{\boldsymbol{\nabla }}}}}}\times {{{{{\bf{A}}}}}}\) and H ( r , t ) = ∇ × ( ∇ × A ) 29 , 48 , see Supplementary Notes 1 – 3 for more detailed derivations. Characterizing topology of skyrmions A skyrmion is a topologically stable 3D vector field confined within a 2D domain, noted as n ( x , y ), which can be represented as the vector distribution unwrapped from the vectors on a spiny sphere parametrized by longitude and latitude angles, α and β . The topological properties of a skyrmionic configuration can be characterized by the skyrmion number defined by 24 , 51 : $$s=\frac{1}{4\pi }\iint {{{{{\bf{n}}}}}}\cdot \left(\frac{\partial {{{{{\bf{n}}}}}}}{\partial x}\times \frac{\partial {{{{{\bf{n}}}}}}}{\partial y}\right)\,{{\mbox{d}}}x{{\mbox{d}}}\,y$$ (7) that is an integer counting how many times the vector \({{{{{\bf{n}}}}}}(x,y)={{{{{\bf{n}}}}}}(r\cos \theta ,r\sin \theta )\) wraps around the unit sphere. For mapping to the unit sphere, the vector can be given by \({{{{{\bf{n}}}}}}=(\cos \alpha (\theta )\sin \beta (r),\sin \alpha (\theta )\sin \beta (r),\cos \beta (r))\) . Also, The skyrmion number can be separated into two integers: $$s = \frac{1}{4\pi }\int\nolimits_{0}^{\infty }\,{{\mbox{d}}}r\int\nolimits_{0}^{2\pi }{{\mbox{d}}}\varphi \frac{{{\mbox{d}}}\beta (r)}{{{\mbox{d}}}r}\frac{{{\mbox{d}}}\alpha (\theta )}{{{\mbox{d}}}\,\theta }\sin \beta (r)\\ = \frac{1}{4\pi }{[\cos \beta (r)]}_{r = 0}^{r = \infty }{[\alpha (\theta )]}_{\theta = 0}^{\theta = 2\pi }=p\cdot m$$ (8) the polarity, \(p=\frac{1}{2}{[\cos \beta (r)]}_{r = 0}^{r = \infty }\) , represents the direction of the vector field, down (up) at r = 0 and up (down) at r → ∞ for p = 1 ( p = −1). The vorticity number, \(m=\frac{1}{2\pi }{[\alpha (\theta )]}_{\theta = 0}^{\theta = 2\pi }\) , controls the distribution of the transverse field components. In the case of a helical distribution, an initial phase γ should be added, α ( θ ) = m θ + γ . For the m = 1 skyrmion, the cases of γ = 0 and γ = π are classified as Néel-type, and the cases of γ = ± π /2 are classified as Bloch-type. The case for m = −1 is classified as anti-skyrmion. See Supplementary Note 6 for some theoretically simulated results of skyrmions with various topological indices. Data availability The data from this paper can be obtained from the University of Southampton ePrints research repository . Code availability The code from this paper and details on the code used can be obtained from the University of Southampton ePrints research repository .

Topology in optics and photonics has been a hot topic since 1890 where singularities in electromagnetic fields have been considered. The recent award of the Nobel prize for topology developments in condensed matter physics has led to renewed surge in topology in optics with most recent developments in implementing condensed matter particle-like topological structures in photonics. Recently, topological photonics, especially the topological electromagnetic pulses, hold promise for nontrivial wave-matter interactions and provide additional degrees of freedom for information and energy transfer. However, to date the topology of ultrafast transient electromagnetic pulses had been largely unexplored. In their paper Nat. Commun., physicists in the UK and Singapore report a new family of electromagnetic pulses, the exact solutions of Maxwell's equation with toroidal topology, in which topological complexity can be continuously controlled, namely supertoroidal topology. The electromagnetic fields in such supertoroidal pulses have skyrmionic structures as they propagate in free space with the speed of light. Skyrmions, sophisticated topological particles originally proposed as a unified model of the nucleon by Tony Skyrme in 1962, behave like nanoscale magnetic vortices with spectacular textures. They have been widely studied in many condensed matter systems, including chiral magnets and liquid crystals, as nontrivial excitations showing great importance for information storing and transferring. If skyrmions can fly, open up infinite possibilities for the next generation of informatics revolution. "This is the first know examples of propagating skyrmions," says Prof. Nikolay Zheludev, the project supervisor, "the fundamental topological constructs previously observed as spin formations in solids and localized electromagnetic excitations in the nearfield of metamaterial patterns." The supertoroidal pulse is as a generalization of the so-called "Flying Doughnut," a toroidal single-cycle pulse with space-time non-separable structure with links to vector singularities and non-radiating anapole excitations. The supertoroidal pulse has increasingly complex fractal-like toroidal topological structures, exhibiting electromagnetic field configurations with matryoshka-like singular shells, skyrmionic structures of various skyrmion numbers, and multiple singularities in the Poynting vector field accompanied by multi-layer energy backflow effects. And the topological complexity can be controlled by increasing a supertoroidal order of the pulse increases. These results put forward supertoroidal pulses as a playground for the study of topological field configurations and their dynamics. The topological features of the supertoroidal pulses presented here provide additional degrees of freedom that could find applications in a number of fields, such as information encoding/decoding schemes involving structured light, optical trapping, manufacturing by light, and particle acceleration. "We believe this is the first time that the skyrmionic structure is proposed in ultrafast structured pulses, and the multiple skyrmionic structure with various textures exist in the instantaneous electromagnetic field of a supertoroidal pulse. Such photonics skyrmionic structures harness intriguing sharp spatial features, promising the potential applications in high-precision metrology and superresolution imaging," says Dr. Yijie Shen, the lead author of the paper. This work opens many intriguing opportunities for the study of light-matter interaction, ultrafast optics, and topological optics with supertoroidal light pulses (e.g. coupling to electromagnetic anapoles and localized skyrmions) and their applications in superresolution metrology and imaging, information and energy transfer.

10.1038/s41467-021-26037-w

Nano

In borophene, boundaries are no barrier—researchers make and test atom-thick boron's unique domains

Xiaolong Liu et al, Intermixing and periodic self-assembly of borophene line defects, Nature Materials (2018). DOI: 10.1038/s41563-018-0134-1 Journal information: Nature Materials , Nature Nanotechnology

http://dx.doi.org/10.1038/s41563-018-0134-1

https://phys.org/news/2018-07-borophene-boundaries-barrierresearchers-atom-thick-boron.html

Abstract Two-dimensional (2D) boron (that is, borophene) was recently synthesized following theoretical predictions 1 , 2 , 3 , 4 , 5 . Its metallic nature and high in-plane anisotropy combine many of the desirable attributes of graphene 6 and monolayer black phosphorus 7 . As a synthetic 2D material, its structural properties cannot be deduced from bulk boron, which implies that the intrinsic defects of borophene remain unexplored. Here we investigate borophene line defects at the atomic scale with ultrahigh vacuum (UHV) scanning tunnelling microscopy/spectroscopy (STM/STS) and density functional theory (DFT). Under suitable growth conditions, borophene phases that correspond to the v 1/6 and v 1/5 models are found to intermix and accommodate line defects in each other with structures that match the constituent units of the other phase. These line defects energetically favour spatially periodic self-assembly that gives rise to new borophene phases, which ultimately blurs the distinction between borophene crystals and defects. This phenomenon is unique to borophene as a result of its high in-plane anisotropy and energetically and structurally similar polymorphs. Low-temperature measurements further reveal subtle electronic features that are consistent with a charge density wave (CDW), which are modulated by line defects. This atomic-level understanding is likely to inform ongoing efforts to devise and realize applications based on borophene. Main Direct synthetic pathways in the atomically thin limit enable 2D materials that are not layered in the bulk 8 . For example, atomically thin phases of boron, collectively referred to as borophene 1 , 2 , have been realized experimentally. Theoretical calculations predict multiple possible borophene polymorphs that possess similar formation energies, but different arrangements of hollow hexagons (HHs) in an otherwise triangular lattice 3 , 4 , 5 . Indeed, experimental studies have revealed multiple borophene phases whose relative occurrence depends on the growth conditions 1 , 2 . The growth substrate plays a major role in stabilizing borophene and determining its structural properties, such as substrate-induced undulations 9 and the shift of the structural ground state when grown on substrates compared with vacuum 3 . Practically, borophene stability is an important factor towards realistic applications, which include the transfer of borophene onto other substrates. Theoretically, both dynamic stability 10 and instability 11 of free-standing borophene, as indicated by the absence and presence of imaginary portions of the phonon spectrum, respectively, have been reported. Compared with semiconducting bulk boron, borophene is an anisotropic metal 1 , 2 , 9 , 12 with unique surface chemistry that strongly influences molecular self-assembly 13 . Moreover, experimental evidence for massless Dirac fermions 14 and theoretically predicted superconductivity 15 expand its desirable characteristics, which makes borophene a highly promising platform for novel material physics with potential utility in next-generation electronic technologies. Although most theoretical predictions have assumed perfect crystalline structures, defects are inevitable in as-grown borophene, which necessitates atomic-scale studies of defects. Although extensive research has explored point and line defects in graphene 16 , 17 and transition metal dichalcogenide monolayers 18 , 19 , 20 , borophene defects remain essentially unexplored. Here we perform an atomically resolved study of borophene under growth conditions that concurrently yield multiple borophene phases and thus relatively high concentrations of line defects. Figure 1a shows an STM image of submonolayer borophene after the deposition of boron on Ag(111) in UHV. The bright dots correspond to small boron particles that are typically present as minority species after borophene growth 1 , 2 , 13 . Due to the convolution of the topographic and electronic structure in STM imaging, the borophene islands appear as depressions under these imaging conditions, consistent with previous reports 1 , 2 , 13 . Atomically thin island structures have also been observed for the submonolayer coverage of metals deposited on surfaces 21 and nitridized metal surfaces 22 . However, those islands are either alloys with the substrate or covalently bonded to the substrates, which is distinct from the weaker non-covalent interaction between borophene and the silver substrate. The borophene islands of the same region are more clearly distinguished in the STS (d I /d V ) map (Fig. 1b ) as brighter domains. At substrate growth temperatures between 440 and 470 °C, two distinct borophene phases are consistently observed, and correspond to the v 1/6 and v 1/5 structures with HH concentrations of v = 1/6 and 1/5, respectively ( v = n / N , where n is the number of HHs in an otherwise triangular lattice with N lattice sites) 1 , 2 , 13 . The temperature dependence of borophene growth over a wider temperature window is given in Supplementary Fig. 1 , which shows a gradual transition from the v 1/6 to the v 1/5 phases of borophene as the growth temperature is increased from 350 to 500 °C. DFT calculations show similar chemical potentials and thus similar thermal stabilities of these two phases ( v 1/6 , −6.359 eV atom –1 ; v 1/5 , −6.357 eV atom –1 (Supplementary Fig. 2 )), in agreement with previous studies 3 , 4 . Fig. 1: Growth of pristine borophene on Ag(111) thin films. a , b , Large-scale STM topography ( a ) and STS map ( b ) of the density of states (DOS) of borophene grown on Ag(111). c , d , Atomic-resolution STM images of v 1/6 ( c ) and v 1/5 ( d ) borophene sheets with inter row spacings of 0.54 ± 0.03 nm and 0.45 ± 0.02 nm, respectively. e , Six STS spectra taken on different positions of the v 1/6 and v 1/5 borophene sheets show a metallic behaviour. f , g , In situ XPS spectra of the B 1 s ( f ) and O 1 s ( g ) core levels indicate pristine borophene growth. STM bias voltages ( V s ) = –1 V in a and b , –0.35 V in c and –1.2 V in d . a.u., arbitrary units. Full size image Atomic-scale STM images of the two phases are provided in Fig. 1 c,d, respectively. Although the repeating rectangular structural units are aligned in adjacent rows in the v 1/6 sheet (indicated with red rectangles; inter-row spacing of 0.54 ± 0.03 nm), these units are staggered and resemble a brick-wall pattern in the v 1/5 sheet (indicated with blue rectangles; inter-row spacing of 0.45 ± 0.02 nm), in agreement with previous work 2 . Both sheets are metallic, as evidenced by the d I /d V curves taken on each phase at 300 K (Fig. 1e ). In Fig. 1f , the in situ X-ray photoelectron spectrum of the B 1 s core level is fit with two subpeaks (red, 187.6 eV; blue, 188.9 eV) that correspond to distinct B–B bonds after background substraction (green curve). The absence of oxidized boron peaks at around 192 eV (ref. 23 ) coupled with a minimal O 1 s core-level signal (Fig. 1g ) confirms the pristine nature of the borophene. Figure 2a shows a spatial derivative image of the STM topography (Fig. 2b ), which more clearly reveals that the larger borophene island is composed of both v 1/6 and v 1/5 sheets with respective line defects indicated by the blue and red arrow heads (magnified images are provided in Supplementary Fig. 3 to support this assignment). The line defects in the v 1/6 sheet resemble the brick-wall structure of the v 1/5 sheet, which suggests non-unique assignments of the phase boundaries. This phase intermixing is schematically shown in Fig. 2c , in which the v 1/6 and v 1/5 sheets are coloured red and blue, respectively, and the corresponding line defects are coloured blue and red. Additional images of these line defects are provided in Supplementary Fig. 4 . Despite the relatively high density of line defects, STS mapping of the same region at 300 K reveals a minimal contrast (Fig. 2d ). DFT calculations confirm the metallic nature of the line defects (Supplementary Fig. 5 ), which implies a minimal impact on the electronic properties of metallic borophene at room temperature. Fig. 2: Presence of line defects in borophene. a , b , The derivative image ( a ) of the STM topography ( b ) of borophene islands reveals the presence of parallel line defects. The v 1/6 and v 1/5 borophene regions are labelled, and their respective line defects are indicated by the blue and red arrow heads. c , Schematic depiction of the distribution of line defects in v 1/6 (red) and v 1/5 (blue) borophene domains based on the image in a . The line defects are coloured oppositely. d , An STS map of the same region shows minimal electronic contrast within the borophene sheets. V s = –0.65 V in b and d . Full size image Atomically resolved STM images are provided in Fig. 3a,b , in which the line defects in v 1/6 and v 1/5 sheets are indicated by blue and red arrow heads, respectively. The line defect in the v 1/6 ( v 1/5 ) sheet is represented by staggered (aligned) blue (red) rectangles in Fig. 3a,b , and suggests the line defect in the v 1/6 domain resembles the unit of a v 1/5 sheet and vice versa. In addition, the respective inter-row spacings of the line defects in v 1/6 and v 1/5 sheets are 0.45 ± 0.02 nm (denoted as d 2 ) and 0.54 ± 0.03 nm (denoted as d 1 ), which match those of the v 1/5 and v 1/6 sheets. Fig. 3: Atomic structures of borophene line defects. a , b , STM images of line defects in v 1/6 and v 1/5 borophene sheets indicated by blue and red arrow heads, respectively. c , Regions of the blue, grey and green boxes in a and b that correspond to the boundary of the v 1/6 and v 1/5 sheets, line defect in v 1/6 and line defect in v 1/5 , respectively. d , Schematic representations of the three regions in c . e , Relaxed structure models (top) and enlarged structures (bottom) that correspond to the three regions in c . The v 1/6 and v 1/5 structures are shaded red and blue, respectively. f , Cross-sectional views of the relaxed structures in e show no out-of-plane buckling. g , Simulated STM images based on the structures in e show a close agreement with the experimental data. V s = –0.38 V in a and V s = –0.91 V in b . Full size image We further compare our experimental observations with DFT modelling that takes into account the contributions from the Ag substrate (details in Methods ). The regions enclosed by the blue, grey and green boxes in Fig. 3 a,b are displayed in Fig. 3c . These images are schematically depicted using the aforementioned red and blue rectangle structural units in Fig. 3d for a clear comparison. The corresponding structural models were generated using a large supercell (up to 800 atoms), and the DFT-relaxed structures are shown in Fig. 3e . The templating effect of the substrate results in epitaxial growth of the two phases with parallel HH rows. The registry of the two lattices along the horizontal rows further allows the v 1/6 and v 1/5 sheets (shaded red and blue, respectively) to connect seamlessly and exhibit an atomically smooth phase boundary, as shown by the enlarged images in Fig. 3e . In other words, each line defect is a row of the v 1/5 or v 1/6 structure inserted in the v 1/6 or v 1/5 sheet, respectively. The cross-sectional views of the DFT-relaxed structures on Ag(111) are shown in Fig. 3f and reveal no observable out-of-plane distortion, even at the phase boundaries, which contributes to their high stability. The calculated values d 1 = 0.51 nm and d 2 = 0.43 nm agree well with the experimentally measured values. In addition, the simulated STM images in Fig. 3g based on the structures in Fig. 3e are in good agreement with the experimental STM images, and thus corroborate the proposed structures. The perfect lattice match along the HH rows at the phase boundaries results in a negligible interface energy, in contrast to substantial interface energies in other configurations due to large lattice mismatches (schematically shown in Fig. 4a ). This anisotropic intermixing phenomenon is in contrast to silicene, which is also a synthetic 2D material with multiple phases 24 , 25 . Despite the subtle structural differences among silicene phases, the phase boundaries are complex, which leads to isolated domains without phase intermixing 25 , 26 , which can be attributed to the rotationally distinct registry of each silicene phase with the substrate and the weak capability of silicon to form variable chemical bonds. However, the v 1/6 and v 1/5 borophene sheets are highly anisotropic and share the same lattice parameters and rotational registry with Ag(111) along HH rows 2 , 12 , which emphasizes the importance of the Ag substrate in templating the coherent assembly of different borophene rows with sharp interfaces and the resulting well-defined line defects. This finding also highlights how structural anisotropy in 2D materials not only results in anisotropic materials properties, but also influences the nature of crystal defects. Fig. 4: Self-assembly of borophene line defects and formation of new phases. a , Schematic models of interfacing v 1/6 and v 1/5 sheets (shaded red and blue, respectively) with mismatched and perpendicular boron rows. The large lattice mismatch causes high interfacial energies. b , The structures of v 1/6 and v 1/5 rows (top and middle), and an example of a new boron phase formed by assembling v 1/6 and v 1/5 rows (bottom). c , A borophene sheet that contains domains with different periodic assemblies of v 1/6 and v 1/5 rows, which include two new phases of borophene ( v 7/36 and v 4/21 sheets). V s = –1.3 V. Full size image The lattice registry along the HH rows suggests the v 1/6 and v 1/5 rows can act as building blocks to assemble into additional 2D borophene sheets (Fig. 4b ), partially facilitated by the non-covalent nature of the interaction between borophene and the substrate. Although this framework seemingly accommodates arbitrary combinations of v 1/6 and v 1/5 rows, line defects often show a local periodicity (Fig. 2a and Supplementary Fig. 6 ). This observation is explored with DFT calculations, which included finite temperature contributions and substrate interactions, by shifting one line defect (for example, a v 1/6 row in a v 1/5 sheet) off its periodic position (Supplementary Note 1 , Supplementary Fig. 7 and Supplementary Table 1 ). The total energy increases as the line defect is displaced from its periodic position, which renders the structure with periodic line defects as the ground state, consistent with the different amounts of charge transfer from the Ag substrate to the v 1/5 and v 1/6 rows due to their different HH concentrations 5 . As crystals are defined by atomic ordering and structural periodicity, borophene domains with periodic assemblies of v 1/6 and v 1/5 rows can be equivalently viewed as new borophene phases. For example, Fig. 4c shows a borophene domain that displays different regions defined by line defects with distinct periodic lengths, as separated by the white dashed lines. The HH concentration of a periodic assembly of m v 1/6 and n v 1/5 units in a supercell is v = ( m + n )/(6 m + 5 n ). In this manner, we identify two new borophene polymorphs in Fig. 4c with supercells that are composed of one v 1/6 unit assembled with three and six v 1/5 rows, with v = 4/21 and 7/36, respectively (models and simulated STM images are given in Supplementary Fig. 8 ). With the increased energy resolution at low temperatures (~4 K), a gap at the Fermi level is revealed in the STS spectra of the v 1/5 and v 1/6 phases of borophene (Fig. 5a ) with two shoulders at about –45 and 45 mV. Moreover, periodic modulations appear in the STS map (Fig. 5c ) of the v 1/5 phase borophene (Fig. 5b ), which are better visualized in the Fourier transforms of the topography and the STS map, as respectively shown in Fig. 5d,e . The points indicated by the yellow arrow heads correspond to the inter-row spacing of 0.45 nm, and the blue arrow heads indicate points from a superstructure with a ~1.33 nm periodicity that corresponds to three boron rows. The borophene orientation is schematically shown in the inset of Fig. 5b . As the unit cell of the v 1/5 phase borophene contains two boron rows (Supplementary Fig. 2 ), this superstructure can be categorized as 3 × 2. This periodicity is independent of energy and is inconsistent with a Moiré pattern or Friedel oscillations, as detailed in Supplementary Fig. 9 . Given these observations and the metallic nature of borophene, the superstructure is consistent with a CDW, which is corroborated by DFT analysis (Supplementary Fig. 10 ). Figure 5f,g shows the topography and STS map, respectively, of a mixed-phase borophene region. Although the borophene in the yellow rectangle is within a larger v 1/5 region, the borophene in the green rectangle is confined by two line defects. The electronic modulation is enhanced in the confined region (Fig. 5g ), but with the same periodicity. The topographies of the two regions are directly compared in Fig. 5h , in which the superstructure is readily visible in the topography of the confined v 1/5 region as indicated by the blue arrow heads. In the v 4/21 phase borophene, the 3 × 2 modulation is absent, which is expected given the very narrow v 1/5 regions (Supplementary Fig. 11 ). Fig. 5: Electronic modulations at low temperature. a , STS spectra of the v 1/6 and v 1/5 phases of borophene measured at ~4 K reveal a gap feature at the Fermi level. b , c , STM topography ( b ) and STS map ( c ) of a v 1/5 borophene domain. The inset in b shows the borophene orientation. d , e , Fourier transforms of the images in b and c , respectively. The yellow and blue arrow heads indicate points that correspond to the inter-row spacing and the 3 × 2 superstructure, respectively. f , g , STM topography ( f ) and STS map ( g ) of a mixed-phase borophene domain. h , Direct comparison of the v 1/5 borophene regions enclosed by the yellow and green rectangles in f and g . The blue arrow heads indicate the apparent 3 × 2 superstructure. V s = 50 mV in b and –40 mV in f and h . Full size image In summary, we performed an atomic-scale characterization of the line defects in v 1/6 and v 1/5 borophene sheets and revealed that the line defects in each phase adopt the structure of the other phase and preferentially self-assemble into periodic arrays, which results in the formation of new phases of borophene. This work thus establishes line defects (that is, v 1/6 and v 1/5 rows) as the building blocks for a multitude of borophene crystalline phases with a single tuning parameter—the mixing ratio of the two rows. This blurring of the definition of crystalline domains and line defects in borophene is enabled by the similarities of the v 1/6 and v 1/5 lattice constants, the atomic registry with the Ag(111) substrate and a high structural anisotropy. As the line defects in borophene possess a similar metallic structure to the pristine v 1/6 and v 1/5 sheets, at room temperature the electronic properties of borophene are relatively unperturbed by its underlying structural complexity, but subtle electronic modulations consistent with a CDW are observable in the low temperature limit. In addition, faceted nanotubes with highly tunable electronic properties based on 2D materials 27 , 28 have been predicted theoretically but not yet realized. Two strategies proposed with phosphorus involve laterally stitched 2D polymorphs 27 and a single phase 2D sheet with parallel line defects 28 . Mixed-phase borophene addresses both of these strategies concurrently as it offers not only parallel line defects, but also equalizing line defects with borophene polymorphs. The periodic arrangement of line defects may further lead to highly symmetric nanotube cross-sections, which are otherwise difficult to engineer. Given the differences in physical (for example, bending stiffness 12 and thermal conductivity 10 ) and electrochemical (for example, ion adsorption and migration on borophene-based battery anodes 29 ) properties of the two phases, borophene sheets with tunable characteristics could be imagined under various periodic line-defect arrangements. For example, mixed-phase borophene may offer structurally tunable near-infrared plasmonic devices 30 . Overall, this study reveals the unique nature and implications of line defects on the structure and properties of borophene, which will inform future fundamental studies, such as investigations of the interactions between CDWs and defects, the predicted superconductivity 15 and the observed Dirac cones 14 in mixed-phase borophene, in addition to emerging efforts to realize borophene-based applications. Methods Growth of borophene The growth of multiphase borophene was achieved by electron-beam evaporation (SPECS EBE-1 and FOCUS EFM 3) of a solid boron rod (ESPI metals, 99.9999% purity) onto ~300 nm thick Ag(111) on mica (Princeton Scientific Corp.) held at 440–470 °C in a UHV preparation chamber (pressure ~10 −9 torr). The flux of boron during the deposition was maintained at 20–25 nA with a filament current of ~5.8 A (SPECS) or ~1.5 A (FOCUS) and an accelerating voltage of 1.3–1.6 kV. A deposition time of 20–30 min was needed to achieve a submonolayer coverage of borophene. Prior to borophene growth, the Ag substrates were prepared by repeated 30 min Ar ion sputtering at 3.3 × 10 −6 torr followed by 30 min of annealing at 550 °C. STM and spectroscopy Room-temperature STM/STS characterization (all images in the main text) was performed in a home-built UHV system 31 (~10 −10 torr) with a Lyding-design microscope 32 interfaced with Nanonis (SPECS) control electronics. Low-temperature STM/STS characterization was performed on a Scienta Omicron LT STM (~10 −11 torr) at ~4 K interfaced with Nanonis (SPECS) control electronics. Electrochemically etched PtIr tips (Keysight) were used in both cases. A lock-in amplifier (SRS model SR850) was used for STS measurements with amplitudes of 30 and 2 mV r.m.s. (r.m.s., root mean square) and modulation frequencies of ~8.5 and 0.8 kHz for room temperature and low temperature measurements, respectively. Gwyddion software was used for image processing. X-ray photoelectron spectroscopy (XPS) In situ XPS measurements were performed in a UHV XPS system (3 × 10 −10 torr) that was directly interfaced with the UHV preparation and STM chambers. The system was equipped with an Omicron DAR 400 M X-ray source (Al Kα), XM 500 X-ray monochromator and EA 125 energy analyser. The energy resolution was 0.6 eV with a pass energy of 20 eV. Using Avantage (Thermo Scientific) software, the modified Shirley backgrounds were subtracted. Given the trace amount of adventitious carbon for the clean Ag(111) surface prepared in UHV, all the core-level peaks were fit after calibrating the spectra to the Ag 3 d 5/2 core-level peak (368.2 eV). DFT calculations The theoretical calculations were performed by employing ultrasoft pseudopotentials for the core region and spin-unpolarized DFT based on the generalized gradient approximation of the Perdew–Burke–Ernzerhof functional, as implemented in the VASP code. A kinetic energy cutoff of 400 eV was chosen for the plane-wave expansion. In all the structures, the vacuum region between two adjacent periodic images was fixed to 20 Å to eliminate spurious interactions. The Brillouin-zone integration was densely sampled based on the supercell size, ensuring approximately the same k -point density among different-sized supercells. The model systems consisted of a boron sheet on flat Ag(111) that consisted of three atomic layers. Commensurability is attained by using large supercells (with atoms up to 800) in which the boron sheet is strained by less than 2% for most of the considered structures (<0.5% in mixed-phase borophene and borophene with line defects). The atomic positions in the topmost two metal layers plus the entire B sheet were fully relaxed until the force on each atom was less than 0.01 eV Å –1 . A dipole correction was included in calculating the total energies. All the STM images were simulated in a constant current mode ( V s = −1.5 V). The stability of different boron layers was evaluated by comparing the total energies of boron, defined as E B = ( E sys – N Ag E Ag )/ N B , where E sys is the total energy of the entire system, E Ag is the energy per atom in the Ag substrate, and N i ( i = B, Ag) is the number of i atoms. Data availability All the data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Information . Additional data related to this paper may be requested from the authors.

Borophene, the atomically flat form of boron with unique properties, is even more interesting when different forms of the material mix and mingle, according to scientists at Rice and Northwestern universities. Scientists at the institutions made and analyzed borophene with different lattice arrangements and discovered how amenable the varied structures are to combining into new crystal-like forms. These, they indicated, have properties electronics manufacturers may wish to explore. The research led by Rice materials theorist Boris Yakobson and Northwestern materials scientist Mark Hersam appears in Nature Materials. Borophene differs from graphene and other 2-D materials in an important way: It doesn't appear in nature. When graphene was discovered, it was famously yanked from a piece of graphite with Scotch tape. But semiconducting bulk boron doesn't have layers, so all borophene is synthetic. Also unlike graphene, in which atoms connect to form chicken wire-like hexagons, borophene forms as linked triangles. Periodically, atoms go missing from the grid and leave hexagonal vacancies. The labs investigated forms of borophene with "hollow hexagon" concentrations of one per every five triangles and one per every six in the lattice. Scanning tunneling electron microscope images of line defects in 1-to-6 and 1-to-5 borophene, indicated by blue and red arrowheads, respectively, show how the defects align in a way that preserves the synthetic material's metallic nature. Scientists at Rice and Northwestern universities made the first detailed analysis of ordered defect structures in borophene. Credit: Hersam Research Group/Northwestern University These are the most common phases the Northwestern lab observed when it created borophene on a silver substrate through atomic boron deposition in an ultrahigh vacuum, according to the researchers, but "perfect" borophene arrays weren't the target of the study. The lab found that at temperatures between 440 and 470 degrees Celsius (824-878 degrees Fahrenheit), both 1-to-5 and 1-to-6 phases grew simultaneously on the silver substrate, which acts as a template that guides the deposition of atoms into aligned phases. The labs' interest was heightened by what happened where these domains met. Unlike what they had observed in graphene, the atoms easily accommodated each other at the boundaries and adopted the structures of their neighbors. These boundary adjustments gave rise to more exotic—but still metallic—forms of borophene, with ratios such as 4-to-21 and 7-to-36 appearing among the parallel phases. "In graphene, these boundaries would be disordered structures, but in borophene the line defects, in effect, are a perfect structure for each other," said Rice graduate student Luqing Wang, who led a theoretical analysis of atom-level energies to explain the observations. "The intermixing between the phases is very different from what we see in other 2-D materials." "While we did expect some intermixing between the 1-to-5 and 1-to-6 phases, the seamless alignment and ordering into periodic structures was surprising," Hersam said. "In the two-dimensional limit, boron has proven to be an exceptionally rich and interesting materials system." A scanning electron microscope image (top) shows two periodic assemblies of borophene, a synthetic, two-dimensional array of boron atoms, that join at a line defect. Computational models in the middle and bottom images correspond to the regions, with 1-to-6 borophene in red and 1-to-5 in blue. Researchers at Rice and Northwestern universities determined that phases of borophene line up in such a way that the material's conductive, metallic nature is maintained. Credit: Luqing Wang/Rice University Wang's density functional theory calculations revealed the metallic nature of the line defects; this implied that unlike insulating defects in otherwise metallic graphene, they have minimal impact on the material's electronic properties at room temperature. At low temperature, the material shows evidence of a charge density wave, a highly ordered flow of electrons. Theoretical calculations also suggested subtle differences in stiffness, thermal conductivity and electrochemical properties among borophene phases, which also suggested the material can be tuned for applications. "The unique polymorphisms of borophene are on full display in this study," Yakobson said. "This suggests intriguing interplay in the material's electronic structure through charge density waves, which may lead to tantalizing switchable electronics." "As an atomically thin material, borophene has properties that should be a function of the substrate, neighboring materials and surface chemistry," Hersam said. "We hope to gain further control over its properties through chemical functionalization and/or integration with other materials into heterostructures." Yakobson and Hersam also co-authored a recent Nature Nanotechnology perspective about "the lightest 2-D metal." In that piece, the authors suggested borophene may be ideal for flexible and transparent electronic interconnects, electrodes and displays. It could also be suitable for superconducting quantum interference devices and, when stacked, for hydrogen storage and battery applications.

10.1038/s41563-018-0134-1

Medicine

The PANoptosome: A new frontier in innate immune responses

AIM2 forms a complex with pyrin and ZBP1 to drive PANoptosis and host defence, Nature (2021). DOI: 10.1038/s41586-021-03875-8 , www.nature.com/articles/s41586-021-03875-8 Journal information: Nature

http://dx.doi.org/10.1038/s41586-021-03875-8

https://medicalxpress.com/news/2021-09-panoptosome-frontier-innate-immune-responses.html

Abstract Inflammasomes are important sentinels of innate immune defence, sensing pathogens and inducing cell death in infected cells 1 . There are several inflammasome sensors that each detect and respond to a specific pathogen- or damage-associated molecular pattern (PAMP or DAMP, respectively) 1 . During infection, live pathogens can induce the release of multiple PAMPs and DAMPs, which can simultaneously engage multiple inflammasome sensors 2 , 3 , 4 , 5 . Here we found that AIM2 regulates the innate immune sensors pyrin and ZBP1 to drive inflammatory signalling and a form of inflammatory cell death known as PANoptosis, and provide host protection during infections with herpes simplex virus 1 and Francisella novicida . We also observed that AIM2, pyrin and ZBP1 were members of a large multi-protein complex along with ASC, caspase-1, caspase-8, RIPK3, RIPK1 and FADD, that drove inflammatory cell death (PANoptosis). Collectively, our findings define a previously unknown regulatory and molecular interaction between AIM2, pyrin and ZBP1 that drives assembly of an AIM2-mediated multi-protein complex that we term the AIM2 PANoptosome and comprising multiple inflammasome sensors and cell death regulators. These results advance the understanding of the functions of these molecules in innate immunity and inflammatory cell death, suggesting new therapeutic targets for AIM2-, ZBP1- and pyrin-mediated diseases. Main Inflammasomes are critical components of the innate immune response 1 . Following exposure to PAMPs and DAMPs, inflammasome sensors form a multi-protein complex that activates caspase-1, leading to cleavage of its downstream substrates, inflammatory signalling and inflammatory cell death. Individual inflammasome sensors detect and respond to specific PAMPs and DAMPs, but emerging evidence suggests that multiple inflammasome sensors can be activated in response to disease, particularly live pathogen infection, owing to the presence of multiple PAMPs and DAMPs 2 , 3 , 4 , 5 . The AIM2 inflammasome is known to sense double-stranded DNA (dsDNA) and has essential roles in development, infectious diseases, inflammatory diseases and cancer 6 , 7 , 8 , 9 , 10 , 11 . However, several critical functions of AIM2 beyond its canonically described role in inflammasome formation and pyroptosis have been observed that cannot be explained by our current understanding of the AIM2 inflammasome 2 . Herpes simplex virus 1 (HSV1), a dsDNA virus that causes lifelong incurable, recurrent pathologies, and Francisella , a Gram-negative bacterium that can cause rapid lethality upon infection, are two diverse pathogens that are known to activate the AIM2 inflammasome and cell death 6 , 10 , 11 , 12 , 13 . To investigate whether cells exposed to numerous PAMPs and DAMPs during these viral and bacterial infections can simultaneously engage multiple inflammasome sensors, we infected wild-type bone marrow-derived macrophages (BMDMs) and BMDMs singly deficient in several major inflammasome sensors with HSV1 or F. novicida . Infection with HSV1 or F. novicida induced AIM2-dependent and NLRP3- and NLRC4-independent cleavage of caspase-1, release of the inflammasome-dependent cytokines interleukin (IL)-1β and IL-18, and cell death (Fig. 1a–o , Extended Data Fig. 1a–i ). However, inflammasome activation and cell death were also partially reduced in Mefv –/– BMDMs during HSV1 and F. novicida infections (Fig. 1p–t , Extended Data Fig. 1j–l ), suggesting that these infections activate pyrin to drive AIM2-mediated inflammatory signalling and cell death. Fig. 1: HSV1 induces AIM2-, pyrin- and ZBP1-mediated caspase-1 activation, cytokine release and cell death. a , Immunoblot analysis of pro-caspase-1 (P45) and cleaved caspase-1 (CASP1) (P20) in HSV1-infected or poly(dA:dT)-transfected wild-type (WT) or Aim2 −/− BMDMs. b , c , IL-1β ( b ) and IL-18 ( c ) release following HSV1 infection. d , Cell death in BMDMs after HSV1 infection for 16 h, measured by SYTOX Green uptake assay. Red indicates dead cells. e , Quantification of the cell death in d . f – j , Immunoblot analysis of CASP1 ( f ), release of IL-1β ( g ) and IL-18 ( h ), cell death images at 16 h after infection ( i ) and quantification of cell death ( j ) from WT or Nlrp3 −/− BMDMs after HSV1 infection or lipopolysaccharide (LPS) plus nigericin (Ni) treatment. k – o , Immunoblot analysis of CASP1 ( k ), release of IL-1β ( l ) and IL-18 ( m ), cell death images at 16 h after infection ( n ) and quantification of cell death ( o ) from WT or Nlrc4 −/− BMDMs after HSV1 or Salmonella Typhimurium infection. p – t , Immunoblot analysis of CASP1 ( p ), release of IL-1β ( q ) and IL-18 ( r ), cell death images at 16 h after infection ( s ) and quantification of cell death ( t ) from WT or Mefv −/− BMDMs after HSV1 infection or C. difficile toxin AB + supernatant treatment. u – y , Immunoblot analysis of CASP1 ( u ), release of IL-1β ( v ) and IL-18 ( w ), cell death images at 16 h after infection ( x ), and quantification of cell death ( y ) from WT or Zbp1 −/− BMDMs after HSV1 or influenza A virus (IAV) infection. a , f , k , p , u , Data are representative of at least three independent experiments. b , c , g , h , l , m , q , r , v , w , Data are mean ± s.e.m. NS, not significant; **** P < 0.0001 (two-tailed t -test; n = 6 from 3 biologically independent samples). d , i , n , s , x , Images are representative of at least three independent experiments. Scale bars, 50 μm. e , j , o , t , y , Data are mean ± s.e.m. NS, not significant; **** P < 0.0001 (two-tailed t -test; n = 8 from 4 biologically independent samples). Exact P values are presented in Supplementary Table 1 . For gel source data, see Supplementary Figure 1 . Source data Full size image Because we observed residual AIM2-mediated caspase cleavage in Mefv –/– BMDMs after HSV1 and F. novicida infections (Fig. 1p , Extended Data Fig. 1j ), we screened several innate immune sensors to determine their involvement. Loss of the sensor ZBP1, which regulates inflammatory cell death in infectious and sterile conditions 14 , 15 , 16 , resulted in reduced caspase-1 cleavage, release of IL-1β and IL-18 and cell death compared with wild-type BMDMs after HSV1 and F. novicida infections (Fig. 1u–y , Extended Data Fig. 1m–o ), whereas loss of other sensors or adaptors had no effect (Extended Data Fig. 2 ), suggesting that ZBP1 is also involved in HSV1- and F. novicida -induced inflammatory signalling and cell death. To determine whether pyrin and ZBP1 act cooperatively or redundantly to decrease caspase-1 activation and cell death, we next used colchicine to inhibit pyrin activation 17 , 18 in Zbp1 –/– BMDMs. We found that colchicine treatment in Zbp1 –/– BMDMs further reduced cell death, cleavage of caspase-1 and release of IL-1β and IL-18 during HSV1 and F. novicida infections, mirroring the levels observed in Aim2 –/– BMDMs (Fig. 2a, b , Extended Data Fig. 3 ), suggesting that ZBP1 cooperates with pyrin to drive the AIM2-dependent responses. We also generated Mefv –/– Zbp1 –/– mice and found that cell death, cleavage of caspase-1 and release of IL-1β and IL-18 during HSV1 and F. novicida infections were further reduced in Mefv –/– Zbp1 –/– BMDMs compared with Mefv –/– or Zbp1 –/– BMDMs, mirroring the levels observed in Aim2 –/– BMDMs (Fig. 2c–j , Extended Data Fig. 4a, b ). These results suggest that pyrin and ZBP1 cooperate to induce AIM2-mediated inflammatory signalling and cell death during both viral infection with HSV1 and bacterial infection with F. novicida . Similarly, we also observed activation of cell death and caspase-1 in human THP-1 macrophages during HSV1 infection, and this activation was inhibited by short interfering RNA (siRNA) knockdown of AIM2 or combined knockdown of ZBP1 and MEFV (Extended Data Fig. 4c–e ), suggesting that the AIM2- and pyrin and ZBP1-mediated cell death also occurs in human cells, the natural host of HSV1. Furthermore, the extent of cell death and caspase-1 cleavage was similar between wild-type and Mefv –/– , Zbp1 –/– or Mefv –/– Zbp1 –/– BMDMs treated with the synthetic AIM2 inflammasome ligand poly(dA:dT) (Extended Data Fig. 4f–h ), suggesting that infection is a unique trigger causing pyrin and ZBP1 to cooperate to facilitate AIM2-mediated caspase-1 activation, cytokine release and cell death. Fig. 2: ZBP1 cooperates with pyrin to drive AIM2-mediated caspase-1 activation, cytokine release and cell death. a , b , Immunoblot analysis of caspase-1 in BMDMs infected with HSV1 or F. novicida with or without colchicine (Col). Data are representative of at least three independent experiments. c , Cell death in wild-type, Aim2 −/− or Mefv −/− Zbp1 −/− BMDMs at 16 h after infection with HSV1 or F. novicida . Red indicates dead cells. Data are representative of at least three independent experiments. Scale bar, 50 μm. d , Quantification of the cell death from c . Data are mean ± s.e.m. (one-way ANOVA with Dunnett’s multiple comparisons test; n = 9 from 3 biologically independent samples). e , f , Immunoblot analysis of CASP1 in BMDMs infected with HSV1 ( e ) or F. novicida ( f ). Data are representative of at least three independent experiments. g – j , Release of IL-1β ( g , i ) or IL-18 ( h , j ) following HSV1 ( g , h ) or F. novicida ( i , j ) infection. Data are mean ± s.e.m. (one-way ANOVA with Dunnett’s multiple comparisons test; n = 6 from 3 biologically independent samples). Exact P values are presented in Supplementary Table 1 . For gel source data, see Supplementary Figure 1 . Source data Full size image To understand how pyrin is activated during HSV1 and F. novicida infections, we first evaluated RhoA-GTP activity, which is known to be inhibited to activate pyrin in response to bacterial toxins such as the Clostridium difficile toxin TcdB 17 , 18 , 19 , 20 , 21 . We found that TcdB-mediated inhibition of RhoA-GTP activity occurred in an AIM2- and ZBP1-independent manner (Extended Data Fig. 5a ), whereas HSV1-mediated inhibition was dependent on AIM2 but independent of ZBP1 (Extended Data Fig. 5a ). We next examined RhoA-GTP levels in Aim2 –/– , Asc –/– and Casp1 –/– BMDMs during HSV1 and F. novicida infections. Whereas RhoA-GTP was absent in wild-type BMDMs upon infection, it was present in Aim2 –/– , Asc –/– and Casp1 –/– BMDMs (Extended Data Fig. 5b, c ). By contrast, treatment with the synthetic AIM2 inflammasome ligand poly(dA:dT) did not inhibit RhoA-GTP activity in any of the genotypes tested (Extended Data Fig. 5d, e ), suggesting that infection has a distinct ability to integrate the AIM2-mediated-signalling and RhoA-signalling cascades to influence pyrin activation. We then investigated how ZBP1 was activated during HSV1 and F. novicida infections. In the context of influenza A virus infection, ZBP1 Zα domains sense nucleic acids to drive RIPK3-, caspase-8- and NLRP3-mediated inflammatory cell death 14 , 15 , 22 . Similarly, we found that HSV1- and F. novicida -induced cell death was reduced in Zbp1 ∆Za2/∆Za2 BMDMs compared with wild-type BMDMs and was similar to that of Zbp1 –/– BMDMs (Extended Data Fig. 5f–i ). Our data show that HSV1 and F. novicida can activate multiple inflammasome sensors. Many pathogens can also induce multiple programmed cell death pathways, including pyroptosis, apoptosis and necroptosis 14 , 23 . Recent studies have found extensive crosstalk between programmed cell death pathways 14 , 24 , 25 , 26 , 27 , establishing the concept of PANoptosis 14 , 28 , 29 , 30 , 31 . Although inflammasome activation is primarily associated with gasdermin D (GSDMD)-mediated pyroptosis, there is emerging evidence of a contribution of inflammasome components in driving PANoptosis 14 , 15 , 26 , 27 , 29 , 30 , 32 . Therefore, we biochemically characterized the cell death induced by HSV1 and F. novicida infections. In wild-type cells, HSV1 and F. novicida activated key molecules involved in pyroptotic, apoptotic and necroptotic pathways (Fig. 3 ). However, we observed reduced activation of the pyroptotic molecules caspase-1, GSDMD and GSDME (Fig. 3a ), the apoptotic molecules caspase-8, -3 and -7 (Fig. 3b ) and the necroptotic molecules RIPK3 and MLKL (Fig. 3c ) in Mefv –/– and Zbp1 –/– BMDMs compared with wild-type BMDMs following HSV1 or F. novicida infection, and a complete loss of activation in Aim2 –/– and Mefv –/– Zbp1 –/– BMDMs. Collectively, these data suggest that HSV1 and F. novicida infections induce inflammatory cell death, PANoptosis, in a manner dependent on AIM2 and the coordinated activation of pyrin and ZBP1. Fig. 3: AIM2, pyrin and ZBP1 promote inflammatory cell death in response to HSV1 and F. novicida infections. a – c , Immunoblot analysis of pro- (P45) and activated (P20) caspase-1 (CASP1), pro- (P53) and activated (P30) GSDMD, pro- (P53) and activated (P34) GSDME ( a ); pro- (P55) and cleaved (P18) caspase-8 (CASP8), pro- (P35) and cleaved (P17/P19) caspase-3 (CASP3), pro- (P35) and cleaved (P20) caspase-7 (CASP7) ( b ); phosphorylated MLKL (pMLKL), total MLKL (tMLKL), phosphorylated RIPK3 (pRIPK3) and total RIPK3 (tRIPK3) ( c ) in wild-type, Aim2 −/− , Mefv −/− , Zbp1 −/− or Mefv −/− Zbp1 −/− BMDMs after HSV1 or F. novicida infection. Data are representative of at least three independent experiments. For gel source data, see Supplementary Figure 1 . Full size image We next sought to understand the regulatory relationship between AIM2, pyrin and ZBP1 during infection with HSV1 or F. novicida . Loss of AIM2 completely abrogated the inflammatory cell death, whereas loss of pyrin or ZBP1 resulted in a partial reduction, suggesting that AIM2 functions upstream of pyrin and ZBP1. Protein expression of pyrin and ZBP1 was reduced in Aim2 –/– BMDMs after HSV1 or F. novicida infection (Fig. 4a, b ). Additionally, their expression was also reduced in Asc –/– , Casp1 –/– and enzymatically inactive Casp1 C284A/C284A BMDMs as well as in wild-type BMDMs treated with the caspase-1 inhibitor VX-765 (Fig. 4c, d ), but not in Gsdmd –/– , Casp8 DA/DA , Casp7 –/– , Casp3 –/– or Casp6 –/– BMDMs after HSV1 and F. novicida infections (Extended Data Fig. 6a–c ). Furthermore, expression of pyrin and ZBP1 was not affected during influenza A virus infection in Aim2 –/– , Asc –/– , Casp1 –/– and Nlrp3 –/– BMDMs (Extended Data Fig. 6d ). Therefore, AIM2-mediated signalling functions as an upstream regulator for the expression of pyrin and ZBP1 in response to HSV1 and F. novicida infections. Fig. 4: AIM2-mediated signalling acts as an upstream regulator of pyrin and ZBP1, which are required to form the AIM2 PANoptosome. a , b , Immunoblot analysis of ZBP1, pyrin and AIM2 in wild-type, Aim2 −/− , Mefv −/− and Zbp1 −/− BMDMs infected with HSV1 ( a ) or F. novicida ( b ). c , Immunoblot analysis of caspase-1 (CASP1) activation and pyrin and ZBP1 expression in wild-type or Asc −/− BMDMs after HSV1 or F. novicida infection. d , Immunoblot analysis of CASP1 activation and pyrin and ZBP1 expression in wild-type, Casp1 −/− and Casp1 C284A/C284A ( Casp1 C284A ) BMDMs and wild-type BMDMs treated with the CASP1 inhibitor VX-765 (20 μM; VX) after HSV1 or F. novicida infection. e , Immunoprecipitation (IP) in wild-type BMDMs with IgG control antibodies or anti-ASC antibodies after HSV1 or F. novicida infection or poly(dA:dT) transfection. a – e , Data are representative of at least three independent experiments. f , Immunofluorescence images of wild-type BMDMs at 12 h after HSV1 infection. Scale bars, 5 μm. Arrowheads indicate the ASC speck. Images are representative of three independent experiments. g , Quantification of the percentage of cells with ASC + AIM2 + pyrin + ZBP1 + specks among the ASC speck + cells. Data are mean ± s.e.m. (one-way ANOVA with Dunnett’s multiple comparisons test; n = 6 from 3 biologically independent samples). h , Pulmonary viral titre at day 3 after infection with HSV1 (WT, n = 6; Aim2 −/− , n = 5). Each symbol represents one mouse. Data are pooled from two independent experiments. Data are mean ± s.e.m. *** P < 0.001 (two-tailed t-test). i , Survival of wild-type, Aim2 −/− , Mefv −/− and Zbp1 −/− mice infected intranasally with 5 × 10 7 plaque-forming units (PFU) of HSV1. Survival data are pooled from three independent experiments. P values for survival of each genotype versus wild type are shown in the key. ** P < 0.01; *** P < 0.001; **** P < 0.0001 (log-rank (Mantel–Cox) test). Exact P values are presented in Supplementary Table 1 . For gel source data, see Supplementary Figure 1 . Source data Full size image To investigate the molecular mechanism underlying how AIM2, ASC and caspase-1 activity-dependent signals promote the expression of pyrin and ZBP1 in infected cells, we assessed transcription and observed that HSV1- or F. novicida -infected Aim2 –/– , Asc –/– and Casp1 –/– BMDMs had reduced transcription of Zbp1 and Mefv compared with the levels in wild-type cells (Extended Data Fig. 7a–d ). It was previously shown that expression of pyrin and ZBP1 can be induced by interferon (IFN) signalling 14 , 33 . Therefore, we investigated whether AIM2 could induce IFN production to drive the expression of pyrin and ZBP1. We found that wild-type BMDMs produced IFN-β in response to HSV1 (multiplicity of infection (MOI) 10, 6 h after infection) or F. novicida infection (MOI 1, 3 h after infection), whereas Aim2 –/– , Asc –/– and Casp1 –/– BMDMs did not (Extended Data Fig. 7e, f ). Additionally, the expression of pyrin and ZBP1 was restored upon complementation of IFN-β in the AIM2-, ASC- and caspase-1-deficient cells (Extended Data Fig. 7g, h ). The differences in IFN production observed here are in contrast to previous studies with F. novicida , in which these knockouts have not been reported to have a marked loss of type I IFN production. These differences might be due to differing timepoints and MOIs: previous studies showed that Asc –/– BMDMs have slightly reduced IFN-β release 5 h after infection 34 (MOI 100) and similar Ifnb mRNA levels 9 h after infection 35 (MOI 100); Aim2 –/– BMDMs have significantly increased IFN-β release 6 h after infection 11 (MOI 250); and Asc –/– and Casp1/11 –/– BMDMs have significantly increased IFN-β release 6 h after infection 36 (MOI 50). These differences suggest that type I IFN release in response to F. novicida may be timepoint- and/or MOI-dependent. Next, we sought to understand the molecular relationship between AIM2, pyrin and ZBP1 in inducing inflammatory cell death and PANoptosis during HSV1 and F. novicida infections. We hypothesized that these molecules would all be part of the molecular scaffold that enables contemporaneous engagement of key proteins from pyroptosis, apoptosis and necroptosis 28 , 29 , 31 , 32 . We observed interactions of ASC with AIM2, pyrin, ZBP1, caspase-1, caspase-8, RIPK3, RIPK1 and FADD by immunoprecipitation following HSV1 and F. novicida infections (Fig. 4e ), and these interactions were not observed in Aim2 –/– , Mefv –/– Zbp1 –/– , Ripk3 –/– Casp8 –/– or Ripk3 –/– Fadd –/– BMDMs (Extended Data Fig. 8 ). By contrast, treatment with poly(dA:dT) allowed interactions between endogenous ASC, AIM2 and caspase-1, but not pyrin, ZBP1, caspase-8, RIPK3, RIPK1 or FADD (Fig. 4e ), suggesting that infection has a distinct ability to form this AIM2 multi-protein cell death-inducing complex. We have termed this complex the AIM2 PANoptosome. We also observed that ASC specks colocalized with AIM2, pyrin and ZBP1 collectively in the same cell at 12 h after infection with HSV1 or F. novicida (Fig. 4f, g , Extended Data Fig. 9a, b ). Similarly, ASC specks also colocalized with caspase-8 and RIPK3 in the same cell (Extended Data Fig. 9c–f ). Furthermore, the formation of this complex was required for cell death, as its disruption through the deletion of key components inhibited cell death in response to HSV1 and F. novicida infections (Fig. 2c, d , Extended Data Fig. 9g, h ). We then extended our findings to in vivo models of HSV1 and F. novicida infection. Lung lysates from infected wild-type mice showed activation of key pyroptotic, apoptotic and necroptotic molecules and increased expression of pyrin and ZBP1, but this activation and increased expression was not found in Aim2 –/– mice (Extended Data Fig. 10a, b ). The number of plaque- and colony-forming units for HSV1 and F. novicida , respectively, was increased in Aim2 –/– , Mefv –/– , Zbp1 –/– and Mefv –/– Zbp1 –/– BMDMs and in the tissue of Aim2 –/– mice (Fig. 4h , Extended Data Fig. 10c–e ). Additionally, all mice lacking AIM2 succumbed to HSV1 infection within 8 days and mice lacking pyrin or ZBP1 succumbed within 13 days, whereas around 70% of the wild-type mice survived (Fig. 4i ). Similarly, all mice lacking AIM2 succumbed to F. novicida infection within 4 days and mice lacking pyrin or ZBP1 succumbed within 9 or 11 days, respectively, whereas around 80% of the wild-type mice survived (Extended Data Fig. 10f ). These results highlight an important role for the AIM2-regulated expression of pyrin and ZBP1, inflammatory cell death and PANoptosis in host defence against HSV1 and F. novicida infections. In summary, AIM2, pyrin and ZBP1 form a multi-protein complex termed the AIM2 PANoptosome via interactions with ASC to induce PANoptosis during HSV1 and F. novicida infections. Loss of AIM2 reduces the expression of pyrin and ZBP1 during these infections, indicating that AIM2-mediated signalling functions as an upstream regulator of pyrin and ZBP1 to control assembly and activation of the AIM2 PANoptosome. The ability of HSV1 and F. novicida to activate AIM2 while also engaging the ZBP1 Zα domain and inhibiting Rho-GTP activity gives these pathogens the unique ability to form this AIM2-, pyrin- and ZBP1-containing PANoptosome complex. Other pathogens that can engage multiple sensors may also form similar complexes with shared central components. Furthermore, we observed roles for ZBP1 in innate immunity. ZBP1 is known to sense influenza viral Z-RNA to promote binding of RIPK3, caspase-6 and the NLRP3 inflammasome to form the ZBP1 PANoptosome complex and drive PANoptosis 14 , 15 , 22 , 31 . The multi-protein complex identified here is mediated by AIM2 but may be similar to this ZBP1 PANoptosome, as it shares the same cell death effectors, including caspase-8 and RIPK3, and also leads to the induction of PANoptosis, characterized by activation of pyroptotic, apoptotic and necroptotic molecules. Additionally, we discovered that ZBP1 also interacts and localizes with ASC and acts as an inflammasome sensor to induce caspase-1 cleavage and inflammatory cell death in response to HSV1 and F. novicida infections. Furthermore, AIM2 regulated the expression of pyrin and ZBP1 during HSV1 and F. novicida infections, but not during influenza A virus infection. Overall, our findings identify a critical interaction between AIM2, pyrin and ZBP1 that drives innate immune responses during pathogen infection. This regulatory mechanism defines the molecular basis of how infection with live pathogens that liberate numerous PAMPs or DAMPs can trigger the coordinated activation of innate immune and cell death signalling components to form a multi-protein complex and cause inflammatory cell death and cytokine release to shape the immune response and host defence. Methods Mice C57BL/6 J (wild type), Mefv −/− (ref. 21 ), Aim2 −/− (ref. 34 ), Nlrp3 −/− (ref. 37 ), Nlrc4 −/− (ref. 38 ), Zbp1 −/− (ref. 39 ), Tlr3 −/− (ref. 40 ), Trif −/− (ref. 41 ), Mda5 −/− (ref. 42 ), Mavs −/− (ref. 43 ), Nlrp6 −/− (ref. 44 ), Nlrp12 −/− (ref. 45 ), Asc −/− (ref. 46 ), Casp1 −/− (ref. 47 ), Casp1 C284A/C284A (ref. 48 ), Gsdmd −/− (ref. 49 ), Casp8 DA/DA (ref. 50 ), Casp7 –/– (ref. 51 ), Casp3 –/– (ref. 52 ), Casp6 −/− (Jackson Laboratory, 006236; ref. 31 ), Ripk3 –/– (ref. 53 ), Ripk3 –/– Casp8 –/– (ref. 54 ) and Ripk3 –/– Fadd –/– (ref. 55 ) mice have been described previously. Mefv −/− mice were crossed with Zbp1 −/− mice to generate Mefv −/− Zbp1 −/− homozygous knockouts. All mice were bred and maintained in a specific pathogen-free facility at the Animal Resource Center at St Jude Children’s Research Hospital and were backcrossed to the C57BL/6 background (J substrain) for at least 10 generations. Both male and female mice were used in this study; age- and sex-matched 6- to 8-week-old mice were used for in vivo and 6- to 12-week-old mice were used for in vitro studies. Cohoused mice were used for in vivo analyses. Mice were maintained in 20–23.3 °C and 30–70% humidity with a 12 h light/dark cycle and were fed standard chow. Animal studies were conducted under protocols approved by the St Jude Children’s Research Hospital committee on the Use and Care of Animals. Cell culture Primary BMDMs were cultivated for 6 days in DMEM (Thermo Fisher Scientific, 12440-053) supplemented with 10% FBS (Biowest, S1620), 30% L929-conditioned medium, 1% non-essential amino acids (Thermo Fisher Scientific, 11140-050) and 1% penicillin and streptomycin (Thermo Fisher Scientific, 15070-063). BMDMs were then seeded into 12-well plates at a density of 1 million cells per well and incubated overnight before use. THP-1 cells (ATCC TIB-202) were grown in RPMI 1640 with 10% FBS and differentiated into macrophages in RPMI 1640 medium containing 20% FBS and 100 ng ml −1 phorbol 12-myristate 13-acetate (PMA) for 2 days. The THP-1 cell line was purchased directly from ATCC and was not further authenticated in our laboratory. Cells were tested for mycoplasma contamination using mycoplasma detection PCR and were found to be negative for mycoplasma contamination. Virus and bacteria culture Human herpes simplex virus 1 (HF strain) (ATCC; VR-260) was propagated in Vero cells, and the virus titre was measured by plaque assay in Vero cells. The influenza A virus (A/Puerto Rico/8/34, H1N1 (PR8)) was prepared as previously described 31 and propagated from 11-day-old embryonated chicken eggs by allantoic inoculation. Influenza A virus titre was measured by plaque assay in MDCK cells. Salmonella Typhimurium strain SL1344 was inoculated into Luria-Bertani (LB) broth (MP Biomedicals, 3002-031) and incubated overnight under aerobic conditions at 37 °C. S . Typhimurium was then sub-cultured (1:10) for 3 h at 37 °C in fresh LB broth to generate bacteria grown to log phase. C. difficile strain r20291 AB - and AB + strains (provided by N. Minton 56 ) were streaked onto brain heart infusion agar (BD Biosciences, 211065) and incubated overnight at 37 °C in an anaerobic chamber. Single colonies were inoculated into tryptone-yeast extract medium and grown overnight at 37 °C anaerobically. F. novicida strain U112 was prepared as previously described 6 and was grown in BBL Trypticase Soy Broth (TSB) (211768, BD Biosciences) supplemented with 0.2% l -cysteine (Fisher) overnight under aerobic conditions at 37 °C. Bacteria were subcultured (1:10) in fresh TSB supplemented with 0.2% l -cysteine for 3 h and resuspended in PBS. Gentamicin (50 μg ml −1 , Gibco) was added 3 h after infection to kill extracellular F. novicida . Cell stimulation and infection For ligand-mediated AIM2 inflammasome activation, 2 μg of poly(dA:dT) (InvivoGen, tlrl-patn) was resuspended in PBS and mixed with 0.6 μl of Xfect polymer in Xfect reaction buffer (Clontech Laboratories, 631318). After 10 min, DNA complexes were added to BMDMs in Opti-MEM (ThermoFisher Scientific, 31985-070), followed by incubation for 5 h. For NLRP3 inflammasome activation, BMDMs were primed for 4 h with 500 ng ml −1 ultrapure lipopolysaccharide (LPS) from Escherichia coli (0111:B4) (Invivogen, tlrl-3pelps) and then stimulated for 45 min with 20 μM nigericin (Sigma, N7143). For NLRC4 inflammasome activation, S . Typhimurium was infected at an MOI of 1 for 2 h. For pyrin inflammasome activation, toxin AB + C. difficile (2 × 10 7 colony-forming units (CFU) ml −1 ) were spun down, and the supernatant was sterilized using 0.22 μm filters and then added to BMDMs for 12 h. For HSV1 (6, 12 or 16 h; MOI 10) and IAV (16 h; MOI 10) infections, cells were infected in DMEM plain medium (Sigma, D6171). For F. novicida infection, cells were infected in DMEM (ThermoFisher Scientific, 11995-065) at a MOI of 50 (12 or 16 h) to determine caspase, GSDMD, and GSDME cleavage and for cell death analyses, at a MOI of 150 (24 h) to determine phosphorylated MLKL and RIPK3 levels or at a MOI of 1 (3 h) for IFN analyses. For inhibition of the pyrin inflammasome, 30 μM colchicine (Sigma, C9754) was added to BMDMs 1 h before infection. For inhibition of caspase-1, 20 μM VX-765 (Chemietek, CT-VX765) was added to BMDMs 1 h before the infection. For the IFN-β treatment, 50 ng ml −1 of IFN-β (PBL Assay, 12400-1) was added to BMDMs 3 h after infections. To measure in vitro viral titre, the indicated BMDMs were infected with HSV1 (MOI 10) for 12 h, and the supernatants were used for plaque assays in Vero cells. To measure in vitro F. novicida levels, cells were infected at an MOI of 100, and cells were collected with PBS containing 0.5% Triton X-100 at 6 or 12 h after infection. The F. novicida were then streaked onto TSB plates supplemented with 0.2% l -cysteine and grown overnight at 37 °C anaerobically. siRNA-mediated gene silencing The Accell human siRNA SMARTPools against AIM2 (E-011951-00-0010), MEFV (E-011081-00-0010) and ZBP1 (E-014650-00-0010) genes were purchased from Horizon. THP-1 macrophages were transfected with siRNA using Accell siRNA delivery media according to the manufacturer’s instructions (Horizon). As a negative control, non-targeting control siRNA (D-001910-01-50) was used. Real-time PCR analysis RNA was extracted using TRIzol according to the manufacturer’s instructions (Life Technologies). Isolated RNA was reverse transcribed to cDNA using the First-Strand cDNA Synthesis Kit (Life Technologies). Real-time PCR was performed on an ABI 7500 real-time PCR instrument with 2× SYBR Green (Applied Biosystems). Primer sequences used in this study were as follows: 5′-AAGAGTCCCCTGCGATTATTTG-3′ and 5′-TCTGGATGGCGTTTGAATTGG-3′ for Zbp1 ; 5′-TCATCTGCTAAACACCCTGGA-3′ and 5′-GGGATCTTAGAGTGGCCCTTC-3′ for Mefv ; 5′-CGTCCCGTAGACAAAATGGT-3′ and 5′-TTGATGGCAACAATCTCCAC-3′ for Gapdh . Measurement of active RhoA-GTP levels BMDMs (2 × 10 6 cells) were seeded in 6-well plates and transfected with 0.5 μg ml −1 TcdB, 0.5 μg poly(dA:dT) or infected with HSV (MOI 10) or F. novicida (MOI 100). RhoA-GTP levels were measured using the RhoA G-LISA Activation Assay kit (BK121, Cytoskeleton) and normalized to total RhoA levels, which were measured using the Total RhoA ELISA kit (BK150, Cytoskeleton) according to the manufacturer’s instructions. For pull-down assays, a GST-tagged Rho binding domain (RBD) in the Rho effector protein was used, which has been shown to bind specifically to the GTP-bound form of RhoA 17 , 19 . RhoA-GTP in BMDMs (2 × 10 6 cells) was pull-down with Rhotekin–RBD beads using the RhoA activation Assay Biochem Kit (BK036, Cytoskeleton) according to the manufacturer’s instructions. In vivo infection Age- and sex-matched, 6- to 8-week-old wild-type wild-type and Aim2 –/– , Mefv –/– and Zbp1 –/– mice were used for infections. For HSV1 infection, mice were anaesthetized with 250 mg kg −1 avertin and then infected intranasally with HSV1 in 50 μl PBS containing around 5 × 10 7 PFU. Infected mice were monitored over a period of 14 days for survival. For the F. novicida infection, F. novicida strain U112 was grown in TSB supplemented with 0.2% l -cysteine overnight at 37 °C and then 1:10 subcultured for 4 h before infection. Mice were infected subcutaneously with 5 × 10 5 CFU of F. novicida in 200 μl PBS. Infected mice were monitored over a period of 12 days for survival. Lungs, livers and/or spleens collected at the indicated time points were homogenized in 1 ml PBS for viral titres to be enumerated by plaque assays or for bacterial titres to be streaked onto TSB supplemented with 0.2% l -cysteine plates and grown overnight at 37 °C anaerobically. Immunoblot analysis Immunoblotting was performed as described previously 57 . In brief, for caspase analysis, BMDMs were lysed along with the supernatant using 50 μl caspase lysis buffer (1× protease inhibitors, 1× phosphatase inhibitors, 10% NP-40 and 25 mM DTT) followed by the addition of 100 μl 4× SDS loading buffer. For signalling analysis, the BMDM supernatants were removed at the indicated time points, and cells were washed once with PBS, after which cells were lysed with RIPA buffer. Proteins from lung, spleen and liver tissues were extracted using RIPA buffer supplemented with protease and phosphatase inhibitors (Roche), and 30 μg per sample was loaded on the gel. Proteins were separated by electrophoresis through 8%–12% polyacrylamide gels. Following electrophoretic transfer of proteins onto PVDF membranes (Millipore, IPVH00010), non-specific binding was blocked by incubation with 5% skim milk; then membranes were incubated with the following primary antibodies: anti-caspase-1 (AdipoGen, AG-20B-0042, 1:1,000), anti-caspase-3 (CST, 9662, 1:1,000), anti-cleaved caspase-3 (CST, 9661, 1:1,000), anti-caspase-7 (CST, 9492, 1:1,000), anti-cleaved caspase-7 (CST, 9491, 1:1,000), anti-caspase-8 (CST, 4927, 1:1,000), anti-cleaved caspase-8 (CST, 8592, 1:1,000), anti-pRIPK3 (CST, 91702 S, 1:1,000), anti-RIPK3 (ProSci, 2283, 1:1,000), anti-pMLKL (CST, 37333, 1:1,000), anti-MLKL (Abgent, AP14272b, 1:1,000), anti-GSDMD (Abcam, ab209845, 1:1,000), anti-GSDME (Abcam, ab215191, 1:1,000), anti-pyrin (Abcam, ab195975, 1:1,000), anti-AIM2 (Abcam, ab119791, 1:1,000), anti-ZBP1 (AdipoGen, AG-20B-0010, 1:1,000), anti-ASC (Millipore, 04-147, 1:1,000 or AdipoGen, AG-25B-006-C100, 1:1,000), anti-RIPK1 (CST, 3493, 1:1,000), anti-FADD (Millipore, 05-486, 1:1,000), anti-β-actin (Proteintech, 66009-1-IG, 1:5,000), human anti-caspase-1 (R&D systems, MAB6215, 1:1,000), human anti-β-actin (CST, 4970, 1:1,000). Membranes were then washed and incubated with the appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies (1:5,000 dilution; Jackson Immuno Research Laboratories, anti-rabbit (111-035-047), anti-mouse (315-035-047)) for 1 h. Proteins were visualized by using Luminata Forte Western HRP Substrate (Millipore, WBLUF0500), and membranes were developed with an Amersham imager; images were analysed with ImageJ (v1.53a). Real-time cell death analysis Real-time cell death assays were performed using an IncuCyte S3 imaging system (Essen Biosciences). BMDMs or THP-1 cells were seeded in 12-well plates (10 6 BMDMs or 5 × 10 5 THP-1 cells per well) and stimulated. After infection, 100 nM SYTOX Green (Thermo Fisher Scientific, S7020) was added. The images were acquired every 1 h at 37 °C and 5% CO 2 . The resulting images were analysed using the software package supplied with the IncuCyte imager, which counts the number of Sytox Green-positive BMDM nuclei (Sytox + BMDM nuclei) present in each image. Immunoprecipitation Immunoprecipitation was performed as described previously 58 . In brief, after HSV1 or F. novicida infection or poly(dA:dT) treatment, wild-type BMDMs were lysed in a buffer containing 20 mM Tris-HCl (pH 7.4), 100 mM NaCl, 30 mM KCl and 0.1% NP-40. After centrifugation at 16,000 g for 10 min, the lysates were incubated with either IgG control antibody (CST, 3900 S) or anti-ASC antibody (AdipoGen; AG-25B-006-C100) with protein A/G PLUS-Agarose (Santa Cruz Biotechnology) overnight at 4 °C. After washing with the above buffer, the immunoprecipitated proteins were collected by boiling in 1× SDS loading buffer at 100 °C for 5 min. Immunofluorescence staining Immunofluorescence staining was performed as described previously 59 . In brief, after infection, BMDMs were fixed in 4% paraformaldehyde (ChemCruz, sc-281692) for 10 min and permeabilized with PBS containing 0.5% Triton X-100 for 3 min. Cells were then incubated in PBS containing 1% skim milk for 1 h. Alexa Fluor 488 (ThermoFisher, A20181)-conjugated anti-ASC (Millipore, 04-147), Alexa Fluor 532 (ThermoFisher, A20182)-conjugated anti-AIM2 (Abcam, ab119791), Alexa Fluor 568 (ThermoFisher, A20184)-conjugated anti-pyrin (Abcam, ab195975), Alexa Fluor 647 (ThermoFisher, A20186)-conjugated anti-ZBP1 (AdipoGen, AG-20B-0010), Alexa Fluor 568 (ThermoFisher, A20184)-conjugated anti-RIPK3 (ProSci, 2283) or Alexa Fluor 647 (ThermoFisher, A20186)-conjugated anti-caspase-8 (Enzo, 1G12) were made according to the manufacturer’s instructions, and the coverslips were incubated with indicated antibodies (1:100) for 1 h. Cells were counterstained with DAPI mounting medium (Invitrogen, P36931). Images were acquired by confocal laser scanning microscopy (LSM780; Carl Zeiss) using 63× Apochromat objective. Cytokine analysis Cytokines were detected by using multiplex ELISA (Millipore, MCYTOMAG-70K), IL-18 ELISA (Invitrogen, BMS618-3) or IFN-β ELISA (BioLegend, 439408) according to the manufacturer’s instructions. Statistical analysis GraphPad Prism 8.0 software was used for data analysis. Data are presented as mean ± s.e.m. Statistical significance was determined by t -tests (two-tailed) for two groups or one-way ANOVA with Dunnett’s multiple comparisons test for more groups. Survival analysis was performed using the log-rank (Mantel–Cox) test. P values less than 0.05 were considered statistically significant where, * P < 0.05, ** P < 0.01, *** P < 0.001 and **** P < 0.0001. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper. Data availability The datasets generated and analysed during the current study are contained within the manuscript and the accompanying extended data figures. Source data are provided with this paper.

St. Jude Children's Research Hospital immunologists have identified how immune sensors in infected cells organize and launch a multi-faceted innate immune response to infections with live viruses and bacteria. The discovery appears today in Nature. The findings offer a new paradigm for understanding the functional and regulatory role that inflammasome sensors and cell death complexes play in infections. The work also highlights new therapeutic targets for treatment of diseases such as cancer and inflammatory autoimmune disorders that are associated with abnormal inflammasome sensor activation. Inflammasomes are protein complexes that form in infected cells or cells that sense damage. The complexes include sensors that recognize different viruses, bacteria and other pathogens or danger signals. Inflammasomes drive inflammatory signaling. Those signals activate inflammatory cell death pathways and eliminate the infection but can also contribute to pathological inflammation. Previous research has focused on inflammasomes working alone. "This new work builds on our quest to identify inflammasome regulation," said corresponding author Thirumala-Devi Kanneganti, Ph.D., of the St. Jude Department of Immunology. "Our study highlights how inflammasomes and multiple cell death components can and do work together in a mega-protein complex called the PANoptosome to activate the innate immune response and unleash PANoptosis." The Kanneganti laboratory showed that regulatory and molecular interactions among three inflammasome sensors, in concert with cell death proteins, drive formation of a mega-cell death complex called a PANoptosome. Rather than regulating one type of inflammatory programmed cell death, PANoptosomes control three—pyroptosis, apoptosis and necroptosis, referred to as PANoptosis. Investigators also determined that the AIM2 inflammasome sensor served as a master regulator of PANoptosome assembly in response to infections with herpes simplex virus 1 and the Francisella novicida bacterium. AIM2 also proved essential for helping mice survive the infections. "The findings address a central question in the fields of innate immunity, cell death and inflammasome biology," Kanneganti said. From inflammasomes to PANoptosomes The findings build on previous research from Kanneganti's laboratory, which is a pioneer in the field. Kanneganti identified one of the first inflammasome sensors and helped to establish inflammasome research. Researchers in the field have focused on how individual inflammasome sensors detect invading pathogens or other threats. Inflammasomes were historically thought to respond by activating one inflammatory cell death pathway. The Kanneganti laboratory has a longstanding interest in understanding regulation of inflammasomes and have identified redundancies among cell death pathways. In 2016, the researchers reported for the first time that influenza infections activated molecules in all three cell death pathways. The scientists called the process PANoptosis. Investigators also determined that a single innate immune sensor called ZBP1 regulated PANoptosis in flu-infected cells. This study laid the foundation for the development of the PANoptosis research area. PANoptosomes today Now, Kanneganti's group has identified AIM2 as the master regulator of a new PANoptosome. First author SangJoon Lee, Ph.D., a postdoctoral fellow in the Kanneganti laboratory, used immunoprecipitation, microscopy and other techniques to show that AIM2, other inflammasome sensors Pyrin and ZBP1, and cell death molecules were part of this AIM2-PANoptosome. The PANoptosome drove inflammatory cell death. "This was critical evidence that the inflammasome sensors and molecules from multiple cell death pathways are in the same complex and highlighted the PANoptosome's role in protecting the host during live pathogenic infections," Lee said. Live pathogens broadcast their presence more widely to the immune system, which helps to explain why the infections trigger PANoptosome assembly and a more robust immune response. Pathogens can also carry proteins that prevent activation of specific cell death pathways. PANoptosis provides an immune system workaround to protect the host. "Our working hypothesis is that while the sensors involved may vary, most infections will induce formation of these unique innate immune complexes called PANoptosomes to unleash inflammatory cell death, PANoptosis," Kanneganti said.

10.1038/s41586-021-03875-8

Medicine

Study identifies neural connections that regulate prosocial and selfish behavior in mice

Diego Scheggia et al, Reciprocal cortico-amygdala connections regulate prosocial and selfish choices in mice, Nature Neuroscience (2022). DOI: 10.1038/s41593-022-01179-2 Journal information: Nature Neuroscience

https://dx.doi.org/10.1038/s41593-022-01179-2

https://medicalxpress.com/news/2022-11-neural-prosocial-selfish-behavior-mice.html

Abstract Decisions that favor one’s own interest versus the interest of another individual depend on context and the relationships between individuals. The neurobiology underlying selfish choices or choices that benefit others is not understood. We developed a two-choice social decision-making task in which mice can decide whether to share a reward with their conspecifics. Preference for altruistic choices was modulated by familiarity, sex, social contact, hunger, hierarchical status and emotional state matching. Fiber photometry recordings and chemogenetic manipulations demonstrated that basolateral amygdala (BLA) neurons are involved in the establishment of prosocial decisions. In particular, BLA neurons projecting to the prelimbic (PL) region of the prefrontal cortex mediated the development of a preference for altruistic choices, whereas PL projections to the BLA modulated self-interest motives for decision-making. This provides a neurobiological model of altruistic and selfish choices with relevance to pathologies associated with dysfunctions in social decision-making. Main Many decisions are made in the context of social interactions. These decisions require the integration of different cognitive processes and behaviors, which allow an individual to understand and interact with others 1 . The psychological conflict between self-interest and the interest of others (especially when this involves a personal cost) is a key element of social decisions 2 . From an evolutionary perspective, altruism likely evolved to promote survival through actions associated with kin selection, parental care and reciprocity 3 , 4 , 5 . Increasing evidence suggests that nonhuman animals, including rodents, engage in prosocial behaviors that resemble altruism; rats help conspecifics that are in need 6 , have been harmed 7 or are seeking food 8 and reciprocate previously received help 9 , and mice display consolatory 10 and collaborative 11 behaviors. Current research paradigms can detect animals’ cooperative propensity to help or to prevent pain in others 4 , 12 . However, the social factors and neurobiological determinants of whether an animal engages in altruism or self-interest are unclear. Mammals live in social groups with dominant and subordinate members, which determine a hierarchy that can affect multiple behaviors 13 and represent an important variable in social relationships and prosocial behaviors 14 . Moreover, socially close individuals share more easily subjective affective states of another through emotional contagion 14 . How social interaction and relationships might change decisions affecting selves and others among members of a group has been overlooked. In nonhuman primates, activity in a prefrontal–amygdala network contributes to social decision-making (SDM) 15 . Further, the same structures are involved in social interactions 16 and social transmission in rodents 17 . In particular, the basolateral amygdala (BLA) has a central position in the neurobiological circuit for our abilities of choosing among options that differ in rewards and costs 18 . Yet, our understanding of its role in decisions involving altruism is still limited. Here we devised a SDM task for mice, modeled on the human game-theoretical paradigm known as the ‘dictator game’ (ref. 19 ), in which a ‘dictator’ (that is, the actor) decides whether to share food with a ‘recipient’ (that is, the observer), who is a passive player. We found that the majority of male mice, but not females, displayed a preference for sharing food with familiar, but not unfamiliar, conspecifics. Substantial individual differences in altruistic choices originated from the hierarchy status of each individual. Chemogenetic silencing of either the BLA or its projections to the prefrontal cortex (PFC) abolished the development of altruistic choices, whereas PFC inputs to the BLA modulated self-interest motives for decision-making. Results Mice choose altruistic actions over selfish decisions To develop a SDM for mice, we expanded a standard operant cage with an adjacent compartment, separated by a metal mesh. This compartment hosted a ‘recipient’ that would receive food rewards depending on the ‘dictator’s’ choice (hereafter referred to as the ‘actor’). The recipient was a passive player with a chance to receive a food reward from a magazine depending on the actor’s choice. The actor was presented with a two-choice decision-making paradigm in which nose poking resulted in either food rewards for themselves only (selfish choice) or for both themselves and the recipient (altruistic choice; Fig. 1a ). We compared this condition against a control group of actor mice without recipients. The structure of the task was identical between these two conditions (‘with recipient’ and ‘no recipient’). Thus, any differences in the response could be attributed to the influence of the recipient. Fig. 1: Mice prefer altruistic over selfish decisions. a , Experimental design of the SDM. b , Decision preference score in mice tested in the SDM with a recipient (orange) or no recipient (gray) (two-way repeated-measures (RM) ANOVA, group (with recipient, no recipient) × time (days 1–5): F (4, 116) = 2.771, P = 0.0305; the decision preference scores were found to fit a normal distribution across 5 days of testing (D’Agostino and Pearson normality test, ‘with recipient’: min K 2 = 3.122, P = 0.225, n = 16; ‘no recipient’: min K 2 = 0.944, P = 0.623, n = 15). Inset, altruistic responses on left ( n = 9) and right ( n = 7) nose pokes on day 1 (two-tailed unpaired t -test: t = 3.37, degrees of freedom (d.f.) = 14, P = 0.0046) and day 5 ( t = 0.79, d.f. = 14, P = 0.4419). c , Number of nose pokes with a recipient ( n = 16) and no recipient ( n = 15; two-way RM ANOVA, group (with recipient, no recipient) × response (nose poke 1, nose poke 2): F (1, 58) = 6.877, P = 0.0111). d , Change of preference in an additional session with a recipient (R→R, n = 10) or with a toy (R→T, n = 10) (two-tailed unpaired t -test: t = 2.24, d.f. = 18, P = 0.0374). e , Left, the total number of mice grouped by preference and sex. Right, altruistic and selfish preferences in males and females. f , Data distribution of decision preference score in altruistic and selfish mice. g , Cumulative frequency distribution of decision preference scores ( n = 52). h , Left, social exploration of altruistic (orange, n = 8) and selfish (blue, n = 10) actors toward their recipients during SDM days 1 and 5 (two-way ANOVA, group (altruistic, selfish): F (1, 32) = 16.29, P = 0.0003). Right, schematic of the testing chambers. i , Social exploration of recipients toward altruistic (orange, n = 6) or selfish (blue, n = 7) actors during SDM days 1 and 5 (two-way ANOVA, group (altruistic, selfish): F (1, 11) = 0.16, P = 0.6902;). j , Correlation between social exploration on day 1 and preference for altruistic choices on day 5 (linear regression: r = 0.4890, P = 0.039, n = 18 pairs). k , Left, decision preference scores in mice tested with a metal mesh (orange, n = 10), a transparent partition (light blue, n = 8) or an opaque partition (gray, n = 8) (two-way RM ANOVA, group (metal mesh, transparent partition, opaque partition) × time (days 1–5): F (8, 100) = 2.037, P = 0.0494). Right, individual curves representing decision preference score. * P < 0.05, ** P < 0.01. NS, not significant. Values are expressed as mean ± s.e.m. Full size image Male and female 3- to 6-month-old littermates were housed in same-sex pairs for at least 2 weeks before the start of testing. The mice were tested for 5 days until they reached a stable performance for 3 consecutive days. At the group level, actor mice with recipients preferred to share food rewards (altruistic choices) more frequently than not (selfish choices). They exhibited a positive decision preference score compared with that of mice in the ‘no recipient’ condition, which did not display any choice preference (Fig. 1b and Extended Data Fig. 1a ). The location of the nose poke that was associated with altruistic or selfish responses did not modify the preference for altruistic choices (Fig. 1b ). At the end of the training, the mice showed an increased number of altruistic over selfish responses when a recipient was present (Fig. 1c and Extended Data Fig. 1b ), whereas the mice in the ‘no recipient’ condition chose similarly between the two nose pokes (Fig. 1c ). Following the last session (day 5), we replaced the recipient mice with an inanimate object (‘with a toy’; Fig. 1a ) and tested whether any changes in the actors’ preference could be detected in the absence of social motivation. With an inanimate object, the actors decreased their preference (both altruistic and selfish) compared to their behavior in the presence of a recipient (Fig. 1d ). These results confirmed that the preference for altruistic or selfish choices was contingent on the presence of a conspecific. We observed marked individual differences in the responses of the mice over days. We found that 11 of the 16 mice showed an increase in altruistic responses, more frequently than could be explained by chance (Extended Data Fig. 1c,d ). The remaining five mice showed a decrease in altruistic responses (Extended Data Fig. 1c,d ). Over the course of training, mice developed a strong preference for one of the two choices. Differences between altruistic and nonaltruistic mice appeared from day 2 of testing, even though the mice exhibited both choices at the beginning of the training (Extended Data Fig. 1b ). The animals displayed clear preferences starting from day 3 of testing (Extended Data Fig. 1b,c ). Indeed, trial-by-trial analyses of responses on day 5 showed that both altruistic and selfish mice displayed a negligible number of nonpreferred choices (Extended Data Fig. 1e,f ). We next analyzed the preference for altruistic or selfish choices in a larger group of animals ( n = 52 actor–recipient pairs). We replicated the SDM task several times in naive and virus-injected mice (for later chemogenetic experiments) and confirmed similar results to our initial findings (Fig. 1e,f ), with the majority of mice displaying a preference for altruistic choices (Fig. 1e–g ). Overall, the relative frequency of mice not showing an altruistic preference was about 33% (Fig. 1e–g ), but this percentage was higher in females. Thus, we analyzed pairs of males and females separately. At the group level, males displayed a preference to allocate food rewards to their recipient (Extended Data Fig. 2a ) and only one of eight males preferred selfish over altruistic responses (Extended Data Fig. 2b ). By contrast, the females did not show an overall choice preference (Extended Data Fig. 2a ), with half of the females displaying a preference for altruistic choices and half preferring selfish choices (Extended Data Fig. 2b ). Therefore, compared to the performance of sex-matched actors that performed the task in the absence of a recipient, only the group of males showed a preference for altruistic responses (Extended Data Fig. 2c,d ). We measured the time spent on social exploration in both groups of mice and found that altruistic actors spent more time than selfish actors exploring their recipients (Fig. 1h ). This was evident from the first session onward. By contrast, we did not observe any differences in social exploration by the recipients (Fig. 1i ). Notably, social exploration of the actor mice during the first day of testing was positively correlated with altruistic responses on the last day, at which point the actors display a consistent preference (Fig. 1j ). We replaced the metal mesh with a transparent partition that prevented social contact but allowed the passage of visual, auditory and olfactory stimuli or an opaque Plexiglas partition that allowed only auditory and olfactory stimuli (Fig. 1k ). Actor mice tested with either partition showed fewer altruistic responses compared to mice who were tested with a metal mesh that allowed social contact. However, an analysis of individual performances revealed that mice tested with the transparent partition established a clear preference for one of the two options (Fig. 1k ), whereas mice tested with the opaque partition performed randomly (Fig. 1k ). These findings suggest that, although mice can use social visual cues to establish their decision preferences, social contact is a determinant of developing an altruistic bias. We then asked whether the recipients’ food-seeking behavior could modulate decision processes by quantifying the number of entries of the recipient’s head into the food magazine. Recipient mice that received rewards from altruistic mice showed more head entries after training than before training (Extended Data Fig. 1g ). However, at the beginning of the training, food-seeking behavior did not differ between recipients tested with altruistic or selfish mice. To understand how the recipients’ hunger state could motivate altruistic behavior, we tested actor mice with sated or food-restricted recipients following the training in the SDM task (Extended Data Fig. 1h ). Actor mice tested with sated recipients decreased their altruistic choices compared to the previous session with food-restricted recipients. In a separate cohort of mice, actors that had been trained in the SDM task with sated recipients (Extended Data Fig. 1i ) had a reduced altruistic preference compared to actors trained with food-restricted recipients. These results suggest that the hunger state of the recipients is an important factor in the actors’ decision to share a food reinforcement. Finally, we tested whether sharing food with recipients could motivate a change in decision preference. The actors were trained to trigger one of the two nose pokes, which both delivered the same food reward to themselves. After the mice displayed a stable preference for one nose poke, a recipient mouse was introduced, and nose poking into the nonpreferred hole resulted in the delivery of rewards to both mice, whereas nose poking into the preferred hole delivered rewards only to the actor (Extended Data Fig. 3a ). The location of the recipient compartment did not bias actor mouse preferences during baseline training (Extended Data Fig. 3b ). At the group level, actor mice displayed a positive change from their baseline preference across days, which suggested that some mice shifted their responses to share food rewards with their recipients (Extended Data Fig. 3c ). Although there were individual differences, the majority of mice switched their preference (‘altruistic’ 8 of 13 mice; Extended Data Fig. 3c–e ). On the day following the last session, we replaced the recipient mouse with an inanimate object and found a decrease in preference compared with that expressed when the recipient was present (Extended Data Fig. 3f ). These results suggest that mice learned or were willing to change their behaviors to share a food reward with their conspecifics. Actor mice take altruistic actions under costly conditions Next, we increased the cost of altruistic decisions by reinforcing the responses at a fixed ratio of 2 (FR2; Fig. 2a ). Under this condition, two nose pokes were required to receive food together with the recipient, whereas only one poke was necessary for selfish responses (FR1; Fig. 2a ). We tested only those mice that had previously demonstrated a preference for altruistic responses after 5 days (Extended Data Fig. 2 ). We similarly tested mice in the ‘no recipient’ condition, in which their natural preference was set to FR2, whereas the other nose poke option was maintained at FR1 (Fig. 2a ). Fig. 2: Mice are willing to take altruistic decisions under costly situations. a , Experimental design of the SDM with a different FR schedule. b , Left, the number of nose pokes on FR1 versus FR2, FR4 and FR6 in male ( n = 7) and female ( n = 4) actors and actors tested without a recipient ( n = 5) (between groups: two-way RM ANOVA, group (with recipient males, with recipient females, no recipient) × response (FR2, FR4, FR6): F (10, 52) = 4.25, P = 0.0002; within groups: two-way RM ANOVA, group (with recipient males, with recipient females, no recipient) × response (FR2, FR4, FR6): F (4, 26) = 4.48, P = 0.0069). Right, the number of nose pokes on SDM day 5 (one-way ANOVA: F (2, 13) = 0.67, P = 0.5270). c , Decision preference scores with FR2, FR4 and FR6, compared to FR1, in mice tested with a recipient (male, n = 7; female, n = 4) and without a recipient ( n = 5) (two-way ANOVA, group (with recipient males, with recipient females, no recipient) × response (FR2, FR4, FR6): F (4, 26) = 3.55, P = 0.0193), * P = 0.0265 (FR4) and P = 0.0678 (FR6) for males versus no recipient and ## P = 0.0010 versus females. d , Actors’ change in altruistic choices during devaluation test in the conditions with rewards to recipients only ( n = 6) and no reward ( n = 8; two-tailed paired t -test: t = 2.28, d.f. = 12, P = 0.0410) and number of nose pokes during valued and devalued sessions (two-way RM ANOVA, session type (valued, devalued): F (1, 14) = 43.07, P < 0.0001). e , f , Following SDM training, altruistic choices did not result in a concurrent reward for the actor. Shown are the percentage of selfish choices ( e ) (two-way RM ANOVA, group (selfish, altruistic) × time (sessions 1–2): F (1, 11) = 4.90, P = 0.0488) and the number of altruistic choices ( f ) over 120 min of SDM in mice grouped by selfish ( n = 7) or altruistic ( n = 6) prefere n ce (inset, percentage of altruistic choices in the first 40 min/total number of altruistic choices; two-tailed unpaired t -test: t = 8.17, d.f. = 10, P = 0.0001). g , Selfish choices in the SDM ( n = 13), without a concurrent reward (as in e and f ), following satiety-induced reward devaluation compared to a valued session (two-tailed paired t -test: t = 5.41, d.f. = 12, P = 0.0002) and differences between altruistic ( n = 6) and selfish ( n = 7) mice (two-way RM ANOVA, group (selfish, altruistic) × time (sessions 1–2): F (1, 11) = 6.37, P = 0.0282). * P < 0.05, ** P < 0.01, *** P < 0.001. NS, not significant. Values are expressed as mean ± s.e.m. Full size image Both males and females showed more altruistic responses over selfish responses, even when an additional effort was required (Fig. 2b,c and Extended Data Fig. 4 ). Moreover, male FR2 responses were higher than those performed by mice tested without a recipient (Extended Data Fig. 4 ). This difference was not confounded by the baseline number of nose poke responses (Fig. 2b ). When the effort necessary to perform an altruistic action was further increased to FR4, males showed more altruistic responses than females or mice without recipients (Fig. 2b,c ). Here females did not show a preference between the two responses and mice without a recipient switched their preference to nose poke reinforced at FR1. When the altruistic responses were reinforced to FR6, the females switched their preference to the nose poke that delivered food rewards more easily (FR1), whereas males continued to prefer altruistic responses (Fig. 2b,c and Extended Data Fig. 4 ). Male mice only switched their preference to selfish responses at FR8 (Extended Data Fig. 4 ). These results suggest that male mice are more willing to share food rewards under more costly conditions. To dissect the social motivation to make an altruistic decision from the motivation to collect a food reward, we introduced a satiety-induced reward devaluation. After training in the SDM, mice were brought to satiety using reward pellets and tested in a session that did not provide reward pellets for the actors. We tested one condition in which neither the actor nor the recipient received rewards (no reward), whereas another group of actor mice was tested under conditions in which the actors did not receive any reinforcements but were able to allocate rewards to the recipient (reward to the recipient only). Both groups of mice displayed reward devaluation, as indicated by a decrease in nose poke responses (Fig. 2d ) compared with the previous session without prefeeding. However, the mice increased their preference for altruistic responses when allocation to a recipient was possible, whereas mice that did not receive rewards and could not allocate rewards to the recipient did not modify their preference (Fig. 2d ). We then tested whether actors would give food rewards to the recipients even if they did not receive concurrent reinforcement themselves. We first trained mice in the SDM and then modified the paradigm such that one nose poke resulted in food rewards for themselves only (selfish choice) and the other to the recipient only (altruistic choice; Fig. 2e ). We tested mice in a longer session (120 min) to observe the effects of satiety on their choices. Although both groups of mice displayed a high percentage of selfish choices, this preference was reduced in altruistic mice (Fig. 2e ). Moreover, while altruistic mice completed most of the altruistic choices in the first part of the session, selfish mice decided to give rewards later in the session (Fig. 2f ), likely due to satiety. Consistent with our previous experiment (Fig. 2d ), after satiety-induced reward devaluation both groups of mice decreased their selfish choices (Fig. 2g ); however, over sessions, the decrease was greater in altruistic than in selfish mice (Fig. 2g ). Altogether, these results suggest that mice were willing to help their conspecifics, even in the absence of a food reward for themselves. Familiarity facilitates altruistic choices Familiarity between individuals is known to amplify prosocial behaviors 6 , 10 . In a new cohort of mice, we, therefore, tested the actions of actors in response to unfamiliar recipients that were housed in different cages. Actors tested with unfamiliar recipients showed an opposite pattern of choices compared with actor mice tested with familiar recipients (Fig. 3a ). Both males and females made fewer altruistic responses in the presence of an unfamiliar recipient compared to mice tested in the presence of cage mates (Fig. 3b ). Analyses of individual responses showed that, in the unfamiliar recipient condition, 9 of 15 mice displayed a preference for selfish responses (Fig. 3c,d ), whereas only 3 mice acted altruistically. Three mice did not show any preference (Fig. 3c,d ). Moreover, altruistic mice tested with unfamiliar recipients, when challenged with increased FR for the altruistic choices, showed a rapid change in their preference (Fig. 3d ). Thus, actor mice paired with noncage mates acted more selfishly than actors paired with cage mates. Fig. 3: Mice display more selfish choices with unfamiliar recipients. a , Decision preference score in the 5 days of SDM in mice tested with familiar (orange; n = 13 (7 males, 6 females)) or unfamiliar (green; n = 15 (10 males, 5 females)) recipients (two-way RM ANOVA, group (familiar recipients, unfamiliar recipients) × time (days 1–5): F (4, 104) = 2.707, P = 0.0342). b , Number of nose poke responses in the conditions with familiar (black border, n = 13) and unfamiliar (green border, n = 15) recipients (two-way RM ANOVA, group (familiar, unfamiliar) × response (altruistic, selfish): F (1, 52) = 12.03, P = 0.0011). c , Individual decision preference score in the SDM in mice tested with familiar or unfamiliar recipients and the number of mice tested with unfamiliar recipients (chi-square test: χ 2 = 5.99, P = 0.0143). Mice were assigned as altruistic (orange), selfish (blue) or no preference (gray) using a one-sample t -test compared to chance (50%, red line). d , Individual decision preference score in mice tested with unfamiliar recipients ( n = 15) grouped by preference and change from preference (expressed in percent; n = 12) in the SDM with altruistic choices reinforced on FR2 and FR4. * P < 0.05, ** P < 0.01, *** P < 0.001. NS, not significant. Values are expressed as mean ± s.e.m. Full size image Social dominance differentiates altruistic preference Social hierarchies influence social relationships 14 . To determine whether the hierarchical relationship between animals within the same cage could influence altruistic propensity, we assessed the social rank of each mouse using the tube test 20 (Fig. 4a ) and then analyzed the relationships of 39 actor–recipient pairs. In all cages, the relationships between mice were transitive and linear ( α is dominant over β , β is dominant over γ , and γ is dominant over δ , with α dominant over all the others). Of mice that performed the SDM task as actors and were dominant in the tube test, the majority preferred altruistic choices (13 of 20; Fig. 4b ); in the group of selfish mice, only 7 of 19 were dominant. At the group level, dominant actor mice displayed a higher decision preference score, suggesting more altruistic choices, compared to subordinate actor mice (Fig. 4c and Extended Data Fig. 5a–c ). We quantified this difference by calculating David’s score (DS) for each mouse, which is a ranking method based on the outcomes of agonistic interactions between group members (higher values indicate higher dominance). Actor mice that preferred selfish choices had a lower DS—and thus lower social rank—than their recipient conspecifics (Extended Data Fig. 5d,f ). By contrast, DS did not differ between actor mice preferring altruistic choices and their recipients (Extended Data Fig. 5e,g ). Fig. 4: Social dominance hierarchy modulates preference for altruistic choices. a , After the SDM daily session, mice were tested on the tube test (at least 1 h after SDM), to measure the hierarchical relationship of animals within the same cage. Actor and recipient mice were tested pairwise and using a round-robin design. b , Number of altruistic or selfish actor mice (A) that were dominant (red) or subordinate (gray) compared to the recipient (R) in the tube test ( n = 39). c , Decision preference score of actor mice that were dominant or subordinate in the tube test compared to their recipient (two-way ANOVA: F (4, 148) = 3.46, P = 0.097; dominant, n = 20; subordinate, n = 19). d , Individual decision preference score in the SDM of dominant actor mice grouped by altruistic or selfish preference ( n = 20). e , f , Social dominance (normalized DS) quantified based on the number and directionality of interactions in the tube test in actor mice that were dominant compared to their recipient, grouped by altruistic (two-tailed paired t -test: t = 5.01, d.f. = 23,97, P < 0.0001; n = 13 pairs) ( e ) and selfish (two-tailed paired t -test: t = 2.27, d.f. = 6,87, P = 0.0576; n = 7 pairs) ( f ) preference. Inset, number of losses by dominant altruistic and selfish actor mice in the tube test (two-tailed paired t -test: t = 2.45, d.f. = 18, P = 0.0244). g , Individual decision preference score in the SDM of dominant actor mice grouped by altruistic or selfish preference ( n = 19). h , i , Normalized DS in actor mice that were subordinate compared to their recipient, grouped by altruistic (two-tailed paired t -test: t = 7.66, d.f. = 11,39, P < 0.0001; n = 7 pairs) ( h ) and selfish (two-tailed paired t -test: t = 8.6, d.f. = 21,67, P < 0.0001; n = 12 pairs) ( i ) preference. Inset, number of losses by subordinate altruistic and selfish actor mice in the tube test (two-tailed paired t -test: t = 1.65, d.f. = 17, P = 0.1154). * P < 0.05, ** P < 0.01, *** P < 0.001. NS, not significant. Values are expressed as mean ± s.e.m. Full size image To explore this variability, we then considered the individual differences between dominant and subordinate actors’ choices and analyzed the animals’ dominance in the social hierarchy in relation to their altruistic or selfish preferences (Fig. 4d,g) . Dominant actors with altruistic preferences had a higher DS than their recipients (Fig. 4e ). By contrast, although dominant actors expressing selfish preferences had a higher rank, they did not display a significant increase in DS compared to their recipients (Fig. 4f ). Furthermore, selfish mice suffered more losses in the tube test than altruistic mice (Fig. 4f ). These results suggest that dominant mice that developed a selfish preference were in competition with their recipient for the same rank. We did not find any differences in DS among subordinate actors that could differentiate mice with altruistic versus selfish preferences (Fig. 4h,i ). Finally, we analyzed the performance in the SDM task by grouping mice based on their social rank. Mice in the α rank made more altruistic choices compared to mice in the β and γ ranks (Extended Data Fig. 5h ) but not compared to mice in the lowest rank ( δ ), which showed similar altruistic preferences (Extended Data Fig. 5h ). These results suggest that a mouse’s willingness to share food rewards with their conspecifics is motivated by in-group dynamics involving the social status of the members. Altruistic choices are linked to emotional state matching The motivation to help others can be related to an individual’s sensitivity to others’ emotional states 21 . Thus, we assessed whether increased altruistic choices in familiar dominant mice could relate to increased affective-state matching between individuals. We used an observational fear conditioning paradigm (Extended Data Fig. 5i ) 22 in which the actor mouse was allowed to observe the recipient mouse (demonstrator) receiving repetitive foot shocks. Freezing behavior, reflecting the observational fear induced by social transmission, was higher in altruistic mice than in selfish mice (Extended Data Fig. 5i ), and observational fear learning scores correlated with social dominance (Extended Data Fig. 5j ). Both groups of mice spent a similar amount of time in exploration of their conspecific demonstrator (Extended Data Fig. 5i ). These results indicate that dominant mice show more emotional contagion, and this may influence SDM. BLA silencing abolishes preference for altruistic choices We found that emotional contagion and social dominance influence altruistic choices. The encoding of information needed for observational learning depends on the BLA 17 , 22 , and in nonhuman primates, the activity of BLA neurons mirrors the value of the reward for self and others 23 . We found that SDM training activated the BLA: following SDM training, mice tested with a recipient mouse had more c-Fos-positive neurons in the BLA than control mice tested without a recipient (Fig. 5a ). Fig. 5: BLA neuronal silencing abolishes the preference for altruistic choices. a , Representative images and bar graph quantification of c-Fos expression in mice following the last day of the SDM task and the number of c-Fos-positive cells in mice tested with or without a recipient ( n = 42 sections from seven animals, three independent experiments; two-tailed unpaired t -test: t = 2.13, d.f. = 40, P = 0.0394). Scale bars (applicable to all micrographs), 50 μm. b , Male mice were bilaterally injected in the BLA with AAV-CamKIIa-mCherry (control, orange) or AAV-CamKIIa-hM4D-mCherry (hM4D, fuchsia). Representative image of a coronal section of BLA. CeA = central amygdala, LA = lateral amygdala, BLA = basolater amygdala, BMA = basomedial amygdala, BLV = basolateral amygdala—ventral part. c , Thirty minutes before the daily SDM session, control and hM4D mice received an intraperitoneal (i.p.) injection of CNO. As a control, we also tested hM4D animals that received vehicle. As we did not observe differences, we pooled the control animals together (two-tailed unpaired t -test: t = 0.927, d.f. = 8, P = 0.3810). d , Left, decision preference score in the 5 days of SDM in control ( n = 9) and hM4D ( n = 10) mice (two-way RM ANOVA, group (control, hM4D) × time (days 1–5): F (12, 140) = 1.981, P = 0.0301; one-sample t -test compared to chance (0.0): control: t = 3.146, d.f. = 44, P = 0.0030; hM4D: t = 1.730, d.f. = 49, P = 0.0899). Right, individual decision preference score in SDM of control and hM4D mice. e , Average decision preference score across 5 days of SDM (two-tailed paired t -test: t = 2.175, d.f. = 17, P = 0.0440) and number of control ( n = 9) and hM4D ( n = 10) mice displaying preference for altruistic or selfish choices. f , Number of altruistic and selfish choices in control (two-way RM ANOVA, choice (altruistic, selfish) × time (days 1–5): F (4, 64) = 5.0, P = 0.0013, n = 9) and hM4D (choice (altruistic, selfish) × time (days 1–5): F (4, 80) = 1.5, P = 0.2024, n = 10) mice over 5 days of the SDM task. g , Representation of altruistic and selfish choices at the end of the training in the SDM task (day 5) in control (left) and hM4D (right) mice. * P < 0.05, *** P < 0.001. Values are expressed as mean ± s.e.m. Full size image To determine whether BLA activation has a causal role in the establishment of altruistic preferences, we evaluated the effects of silencing BLA glutamatergic neurons (the majority of BLA neurons 24 ) during the SDM task using a chemogenetic approach. We injected a virus carrying the inhibitory designer receptor exclusively activated by designer drugs (DREADD) receptor hM4Di (AAV-CaMKIIa-hM4Di-mCherry) or a control virus (AAV-CaMKIIa-mCherry) into the BLA (Fig. 5b and Extended Data Fig. 6a,b ). Both hM4Di mice and control mice injected with AAV-CaMKIIa-mCherry received clozapine- N -oxide (CNO) 30 min before testing (Fig. 5c ). Control mice displayed a preference for altruistic choices (as expected), but BLA-silenced mice did not (Fig. 5d,e ). Indeed, control mice showed more altruistic choices than selfish choices, whereas BLA-silenced mice showed no differences in altruistic versus selfish choices (Fig. 5f,g ). Individual performance showed that 6 of 10 BLA-silenced mice preferred selfish choices, while one did not exhibit a preference (Fig. 5e ). In contrast, the majority of control mice (7/9) preferred altruistic choices (Fig. 5e ), similar to naive animals (Fig. 1b ). Similarly, BLA-silenced mice showed reduced interest in social exploration over testing days (Extended Data Fig. 6c ). BLA silencing did not affect the number of responses or the latency to make a choice and did not produce motor impairments (Extended Data Fig. 6d–f ). Emotional contagion and social hierarchy influenced the preference for altruistic or selfish choices (Fig. 4 and Extended Data Fig. 5i ). Therefore, we explored the involvement of these factors in the reduced preference for altruistic choice induced by BLA silencing. BLA-silenced mice froze less during observational fear conditioning than control mice (Extended Data Fig. 6g–i ), suggesting reduced social transmission of fear to BLA-silenced mice, consistent with data from a previous study 22 . In primates, amygdala lesions produce divergent effects on social dominance 25 . To better understand whether BLA activity can be linked to the representation of social ranks, we chemogenetically silenced the BLA before the tube test. BLA silencing reduced ranks in the tube test starting from 1 to 2 h following CNO injection (Extended Data Fig. 7a ), although this effect was not observed in the highest-ranking mice (Extended Data Fig. 7b ). In a different cohort of mice, tested in the SDM, a higher number of BLA-silenced actor mice compared to control mice were subordinate to their recipient mouse (Extended Data Fig. 7c,d ). Indeed, BLA-silenced actor mice had a lower DS than control actor mice, suggesting reduced dominance (Extended Data Fig. 7e–i ). Furthermore, BLA-silenced mice that had a lower rank than their recipient made more selfish choices than dominant BLA-silenced mice (Extended Data Fig. 7j ), in agreement with our previous results showing selfish preferences among subordinate actors (Fig. 4d ). Altogether, consistent with our findings linking altruistic decision preference with social dominance, these experiments provide initial evidence of the BLA as an information crossroads of social dominance, emotional contagion and SDM. The BLA is required to develop an altruistic preference To monitor BLA neural activity during all decisional processes leading to the development of altruistic versus selfish choices, we performed fiber photometry recordings during the SDM task (Fig. 6a and Extended Data Fig. 8a ). We injected virus carrying a genetically encoded fluorescent calcium indicator, AAV-CaMKIIa-GCaMP6f, into the BLA. Beginning on the third day of testing (learning phase), we found increased neural activity time-locked to the nose poke response after both altruistic and selfish choices (Fig. 6c,f ) compared to the baseline. Moreover, in this phase, BLA neural activity was higher in altruistic mice than in selfish mice after altruistic choices (Fig. 6c ), but not after selfish choices (Fig. 6f ). There was no such difference in BLA neural activity on the first day of the SDM task (Fig. 6b,e ). Moreover, BLA neural activity in altruistic mice was higher in the learning phase than on the first day of testing after altruistic (Fig. 6h ) but not selfish (Fig. 6j ) choices. When the task was fully acquired, neural activity after nose poke responses was lower than at baseline (Fig. 6d,g ). In particular, neural activity after altruistic choices was lower in selfish than in altruistic mice (Fig. 6d,g ). Accordingly, neural activity was decreased during the last day of SDM compared to the learning phase after altruistic choices (Fig. 6h,i ) but not selfish choices (Fig. 6j,k ). To examine this difference, we analyzed c-Fos expression following the last day of SDM. The percentage of c-Fos-positive GABAergic interneurons (GAD67-positive cells) was small; nevertheless, selfish mice had more c-Fos-positive GABAergic cells than altruistic mice (Extended Data Fig. 8b ). By contrast, the number of c-Fos-positive GAD67-negative cells did not differ between altruistic and selfish mice and was higher than in mice tested without a recipient (Extended Data Fig. 8b ). These data, combined with the results obtained using fiber photometry, suggest that inhibition of neural activity by activation of GABAergic cells following altruistic choices may be stronger in selfish mice than in altruistic mice. Fig. 6: The BLA is required for the learning of altruistic choices. a , Virus encoding GCaMP6f (AAV-CaMKIIa-GCaMP6f) in the BLA for fiber photometry. Scale bar, 250 μm. b – d , GCaMP6f fluorescence changes in the BLA of altruistic and selfish actor mice in response to an altruistic nose poke during the first day of testing (‘start’; two-way RM ANOVA, group (altruistic, selfish) × time (min): F (1, 101) = 0.86, P = 0.8248; n = 42 trials from six mice) ( b ), the ‘learning’ phase (two-way RM ANOVA, group × time: F (101, 14948) = 3.01, P < 0.00001; group: F (1, 148) =5.45, P = 0.0208; n = 151 trials) ( c ) and the last day of testing in the SDM (‘acquired’; two-way RM ANOVA, group × time: F (101, 68882) = 2.46, P < 0.00001; n = 684 trials) ( d ). e – g , GCaMP6f fluorescence changes in the BLA of altruistic and selfish actor mice in response to a selfish nose poke during the first day of testing (two-way RM ANOVA, group (altruistic, selfish) × time (min): F (101, 3939) = 0.97, P = 0.5541; n = 42 trials from six mice) ( e ), the learning phase (two-way RM ANOVA, group × time, F (101, 56964) = 2.87, P < 0.00001; n = 144 trials) ( f ) and the last day of testing in the SDM (two-way RM ANOVA, group × time: F (101, 68882) = 4.25, P < 0.00001; group: F (1, 564) = 5.63, P = 0.0179; n = 566 trials) ( g ). h , i , Area under the curve (AUC) after altruistic choices (0–5 s) at different periods of the SDM task in altruistic ( h ; one-way ANOVA: F (2, 680) = 51.17, P < 0.0001, first day n = 25, learning phase n = 117, last day n = 551) and selfish ( i ; one-way ANOVA: F (2, 179) = 5.62, P = 0.0043, first day n = 16, learning phase n = 33, last day n = 133) actor mice. j , k , AUC after selfish choices in altruistic ( j ; one-way ANOVA: F (2, 137) = 0.07, P = 0.9434, first day n = 28, learning phase n = 48, last day n = 64) and selfish ( k ; one-way ANOVA: F (2, 607) = 0.55, P = 0.5741, first day n = 13, learning phase n = 95, last day n = 502) actor mice. l , Male mice were bilaterally injected in the BLA with AAV-CamKIIa-mCherry (control, n = 7) or AAV-CamKIIa-hM4D-mCherry (hM4D, n = 7). Both groups received CNO on testing days 2 and 3, 30 min before a SDM session with familiar recipients. Shown is the number of altruistic choices in control and hM4D mice (two-way RM ANOVA, group (control, hM4D) × time (days 2–3): F (1, 12) = 5.44, P = 0.0378). m , Decision preference score in the 5 days of SDM in control ( n = 7) and hM4D ( n = 7) mice (two-way RM ANOVA, group (control, hM4D) × time (days 1–5): F (4, 48) = 2.719, P = 0.0404; one-sample t -test compared to chance (0.0): control: t = 10.81, d.f. = 34, P < 0.0001; hM4D: t = 3.17, d.f. = 34, P = 0.0032). n , Number of altruistic choices on days 2 to 3 of the SDM in control and hM4D mice (mixed model analysis; day 2: group (control, hM4D) × time (min): F (42, 267) = 0.42, P = 0.9994; day 3: F (73, 685) = 2.85, P < 0.0001). In the box plots in h – k center, median; box, quartiles; whiskers, minimum and maximum. All other values are expressed as mean ± s.e.m. * P < 0.05, ** P < 0.01, *** P < 0.001. NS, not significant. Full size image Finally, to test whether the increased BLA activity in the learning phase underlies the establishment of altruistic preferences (Fig. 6c ), we silenced BLA neuronal activity only during the learning phase (days 2–3; Fig. 6l ) of the SDM test. This reduced the number of altruistic choices compared to control mice (Fig. 6l ) on test day 3, but not day 2. This effect was long-lasting as these mice displayed reduced decision preference scores also in the absence of CNO administration in the following 2 days (Fig. 6m ). To investigate the effects of BLA silencing on decision processing, we analyzed the latency to response (that is, the time between one choice and the following one). BLA silencing increased the latency to make altruistic, but not selfish, choices (Extended Data Fig. 9a ). We then monitored the number of altruistic choices over time, and we found that BLA-silenced mice displayed increased time to make altruistic choices on test day 3, but not day 2 (Fig. 6n and Extended Data Fig. 9b ). This suggests that BLA-silenced mice shared food rewards only later in the session, unlike controls. This effect disappeared with additional training (Extended Data Fig. 9b ). When we tested mice 1 week later, BLA-silenced mice still made fewer altruistic choices than control mice (Extended Data Fig. 9c,d ). Collectively, these results indicate that the BLA is differently activated by decisions to share or not share a positive reinforcement in mice that prefer altruistic choices versus mice that prefer selfish choices. In particular, our data point to a crucial role of the BLA in the establishment of altruistic preference. Distinct roles of BLA–PFC connections in altruistic choices BLA inputs and outputs mediate different types of learning 26 and support circuits involved in the valence processing of environmental stimuli 27 . PFC subregions are both major targets of the BLA and a major source of inputs 26 . We targeted the prelimbic (PL) region of the PFC, corresponding to primate Brodmann area A32 (ref. 28 ), which supports goal-directed behavior 29 . We injected a retrogradely transported canine adenovirus-2 expressing Cre recombinase (CAV2-Cre) into the PL and injected the BLA with an rAAV carrying a Cre-dependent hM4D(Gi)DREADD receptor and mCherry (hM4D BLA→PL; Fig. 7a ). With this combination, we achieved DREADD(Gi)-mCherry expression exclusively in BLA neurons projecting to the PL. We used the same approach to study PL neurons projecting to the BLA (hM4D PL→BLA; Fig. 7a ). CNO was injected 30 min before each session. Fig. 7: Chemogenetic silencing of the BLA–PFC reciprocal connection has a different impact on altruistic choices. a , Schematic showing viral injection and projection areas and example images of coronal sections of BLA and PL. Mice received virus encoding Cre-dependent hM4D receptor in the BLA and CAV2-Cre in the PL or Cre-dependent hM4D receptor in the PL and CAV2-Cre in the BLA. With this combination, we achieved DREADD expression exclusively in BLA neurons projecting to the PL (hM4D BLA→PL) and vice versa (hM4D PL→BLA). CeA, central amygdala; M2, secondary motor cortex. b , Decision preference score in the 5 days of SDM in control CNO (orange, n = 10), hM4D BLA→PL (purple, n = 11) and hM4D PL→BLA (light blue, n = 9) mice (two-way RM ANOVA, group (control CNO, hM4D BLA→PL, hM4D PL→BLA) × time (days 1–5): F (8, 108) = 2.03, P = 0.0493). c , The number of mice displaying preference for altruistic or selfish choices. Mice were assigned as altruistic (orange), selfish (blue) or no preference (gray) by analyzing decision preference scores using a one-sample t -test compared to chance. d , Individual decision preference score in the SDM of control CNO and hM4D BLA CNO mice. e – g , Representation of altruistic and selfish choices at the end of the training in the SDM task (day 5). h , Number of altruistic and selfish choices in control CNO (two-way RM ANOVA, choice (altruistic, selfish) × time (days 1–5): F (4, 72) = 3.6, P = 0.0088, n = 10), hM4D BLA→PL (choice (altruistic, selfish) × time (days 1–5): F (4, 64) = 2.6, P = 0.0401, n = 11) and hM4D PL→BLA (choice (altruistic, selfish) × time (days 1–5): F (4, 80) = 0.69, P = 0.5981, n = 9) mice over 5 days of the SDM task. i , Learning index representing the preference development in control CNO ( n = 10), hM4D BLA ( n = 10), hM4D BLA→PL ( n = 11) and hM4D PL→BLA ( n = 9) mice (two-way RM ANOVA, group × time: F (12, 140) = 1.91, P = 0.0376). * P < 0.05, ** P < 0.01. NS, not significant. Values are expressed as mean ± s.e.m. Full size image Silencing BLA→PL projections abolished the preference for altruistic choices (Fig. 7b ), similar to the effect of silencing the entire BLA (Fig. 6 ). While the majority of control mice (7 of 10) showed a preference for altruistic choices (Fig. 7c ), after silencing BLA→PL projections, the preference was equally distributed between selfish and altruistic choices and three of these mice did not display any preference (Fig. 7c,d ). In line with these results, control mice made more altruistic than selfish choices, which was not observed after silencing of BLA→PL projections (Fig. 7e,f,h ). By contrast, silencing PL→BLA projections produced a negative decision preference score (Fig. 7b ); the majority of mice expressed a selfish preference (6 of 9; Fig. 7c ) and made more selfish than altruistic choices (Fig. 7e,g,h ). Finally, to quantify the efficiency of preference development regardless of its value (positive or negative decision score), we calculated a learning index. We found that, similar to BLA silencing, silencing BLA→PL projections resulted in a lower learning index compared to that seen in control mice or after silencing PL→BLA projections (Fig. 7i ). This suggests that silencing BLA→PL projections slows the development of choice preferences, consistent with the finding of increased BLA neural activity during the learning phase of the task. Collectively, these data suggest that reciprocal PFC–BLA connections have distinct roles in the establishment of altruistic or selfish choices (Fig. 8a ). Fig. 8: A brain circuit and factors involved in altruistic choices. a , Schematic model of the involvement of the BLA–PFC reciprocal connections in the SDM. Green circle, normal; red cross, impaired; blue arrow, selfish; orange arrow, altruistic. b , Influence of multiple factors on the SDM task ( n = 146 mice). Average decision preference score (stable performance for the last 3 days) is shown when considering the impact of multiple factors (from 1 to 5 factors simultaneously; one-way ANOVA, Welch’s test: F (17, 67.28) = 9.03, P < 0.0001). * P < 0.05. Blue indicates a higher preference for altruistic choices and red indicates a higher preference for selfish choices. Box plots: center, median; box, quartiles; whiskers, minimum and maximum. Full size image Discussion In this study, we showed that mice intentionally engage in choices that will favor another conspecific or only themselves. Divergence in SDM originates from different interindividual factors, including familiarity, sex, social contact, physiological need (that is, hunger), hierarchical status and emotional state matching. Notably, prosocial actions, even if they required more effort or had no direct benefit to the actor mouse, were more generally observed toward familiar, hungry males with the highest hierarchical distance to the actor. We showed that BLA neuronal activity is needed to develop a preference for altruistic behaviors in the SDM task. More specifically, BLA→PFC projections guide the establishment of altruistic actions, whereas PFC→BLA projections exert control over selfish decisions. We developed an operant task to explore how basic decision-making systems operate within a socially interactive environment. Most experimental studies of decision-making have examined behaviors with clearly defined probabilities and outcomes, such as choosing between food rewards. In our task, mice chose between two actions that yielded either a reward only to themselves or to both themselves and a partner placed in an adjacent compartment. To our knowledge, no previous studies in mice have included such complexity of social interactions 30 , even though many important decisions are made in the context of social interactions with others and change based on social feedback. Thus, our SDM paradigm provides an approach to examining distinct and complex social behaviors in mice. Our results are consistent with recent studies showing complex prosocial behaviors in rats, such as preference for mutual rewards 8 , 31 , helping behaviors 6 and the avoidance of harming others 7 . Rodents show social behaviors that suggest they can act to increase mutual benefits 6 , 10 , 11 . Few studies have demonstrated prosocial behaviors reminiscent of altruism constructs in mice. In the SDM task, mice learned to make altruistic choices although conditioned responses were reinforced by a positive outcome for the actor and the recipient; however, they displayed an interest in sharing food with their hungry conspecifics even in the absence of concurrent positive reinforcement or any explicit return favors from their actions (that is, without evident self-benefit), a critical factor that defines ‘altruistic’ choices 32 . Moreover, when mice were presented with the opportunity to stop making altruistic choices by exerting less effort to obtain a food reward only for themselves, they continued to display a preference for sharing food with their companions. Altogether, these findings identify social values that determine the development of choices that will favor another conspecific. The establishment of social choices is complex and heterogeneous, depending on several factors. Indeed, our results reveal that when considering components such as familiarity, sex, dominance, hunger state, social contacts and emotional state matching individually, their effects on selfish versus altruistic choices were considerably heterogeneous. By contrast, the combination of all these factors revealed cumulative effects that reduced the heterogeneity and better separated subpopulations of mice that showed altruistic choices (Fig. 8b ). Dominance hierarchy contributed to the preference for altruistic or selfish choice. Social status can guide behavior and motivation in a social group, including in humans 33 . Our finding that most mice that displayed a preference for selfish over altruistic choices were subordinate to their recipient and belonged to an intermediate rank could reflect competition for food, as dominant members might benefit from easier access to food and dictate priorities to access resources 20 . Similarly, in nonhuman primates 14 and rats 34 , prosocial responses are more often directed from dominant toward subordinate members. Thus, dominant individuals may behave in ways that benefit others to advertise their dominance. On the contrary, animals acted selfishly because they were in direct competition for both upward and downward ranks. In summary, our task generated distinct behavioral responses that could address several complex aspects of SDM triggered by interpersonal interactions. Studies in rodents under several conditions, such as risk-taking 18 , punishments 35 and threats 36 , have revealed a critical role of the BLA in integrating reward-related information and costs to guide decision-making. Our results expanded this role to the social domain. The BLA integrates cue-response associations with motivational and emotional inputs, updating the value of these associations through connections with the PFC 37 . In nonhuman primates, the synchronization of neural activity between the BLA and PFC is important for the establishment of other-regarding preferences 15 . Accordingly, silencing of PFC→BLA neuronal projections produced a bias in the development of a preference toward selfish choices. This result supports the hypothesis that the dorsolateral part of the PFC in humans is involved in modulation of the relative impact of self-interest impulses on decision-making 38 . The preferential connection between the BLA and cortical structures, such as the PFC, also has an important modulatory effect on social behavior and the transmission of social cues 16 , 17 . Indeed, similarly to a previous study 22 , the BLA silencing reduces social fear learning, which was correlated with preferences in the SDM. Thus, the establishment of a preference toward altruistic or selfish choices could be at least partially related to empathy-like capacity in mice. Altogether, the effects observed following neuronal silencing of the BLA inputs and outputs indicate its involvement in the social value of reward for self and for others. Previous studies have provided evidence pointing to the involvement of the PFC in plastic modulation of the social hierarchy 20 . However, it has been unclear whether PFC–BLA reciprocal connections are involved in the social hierarchy. We found that the effects of BLA→PFC projections on altruistic choices were more similar to overall BLA silencing than the effects of PFC→BLA projections, which were more relevant to selfish choices. The hierarchy was correlated to altruistic but not selfish preference, and silencing of the BLA was associated with rank changes down the hierarchy, suggesting that BLA→PFC projections could also be more relevant for regulating hierarchy. In agreement with this, PFC→BLA projections are not involved in the modulation of hierarchical dominance 39 . Thus, whether the PFC and BLA distinctly modulate social status or bottom-up BLA→PFC connections are a key circuit involved in the social hierarchy is an intriguing topic for future studies. Altruistic behaviors were learned in our task through positive reinforcements. Nevertheless, we cannot rule out the possibility that some degree of innate or impulsive altruism may have assisted during the initial learning process 12 . In support of this possibility, mice displaying a preference for altruistic choices were more interested in social exploration than mice that preferred selfish choices. Kin selection has been suggested as a biological explanation of the motivation to act in an altruistic manner 12 . However, in a laboratory setting, animals do not face such selection pressures. Therefore, other explanations should be considered, such as the emotional engagement between the actor and recipient. Mice can discriminate 40 and share 41 the affective state of their conspecifics. Indeed, mice that expressed a preference for altruistic choices displayed more empathy-like behaviors. The food-seeking behavior of the recipient could trigger an emotional transfer between mice, which may motivate altruism. Moreover, familiarity can amplify the empathic response 42 . Consistent with this, emotional contagion was linked with the preference for altruistic choices, suggesting that affective state matching could motivate altruism in mice. Together with empathy, social motivation could represent another explanation for altruism. The social value can guide how social animals interact with others and adapt behavior and actions in response to them, influencing social-value-based decisions 43 . Rodents are social animals that display preferences for social closeness and avoid social isolation 44 , which can have rewarding properties 45 . Thus, well-being conferred by sharing a positive experience may also have a social value for the actor. Altogether, our results suggest that altruistic choices in mice are motivated by a positive connection favoring prosocial behavior. In summary, we developed a task enabling the detection of interindividual propensities in mice for altruistic or selfish choices and the factors modulating them (that is, hierarchy, familiarity, sex, hunger state and social contact). We started to elucidate the neurobiology of such social decisions, revealing the BLA and BLA–PFC reciprocal connections as critical substrates for the establishment of altruistic choices. These results could have important implications for psychiatric, psychological and neurodevelopmental conditions associated with disruptions in SDM. Methods Mice All procedures were approved by the Italian Ministry of Health (permits: 107/2015-PR, 749/2017-PR and 191/2020-PR) and the local Animal Use Committee and were conducted in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and the European Community Council Directives. Routine veterinary care and animal maintenance were provided by dedicated and trained personnel. Two- to 6-month-old male and female C57BL/6J animals were used. Distinct cohorts of naive mice were used for each experiment. Two to four animals were housed per cage in a climate-controlled facility (temperature, 22 ± 2 °C; humidity, 45–65%), with ad libitum access to food and water throughout and with a 12-h light–12-h dark cycle (7 pm to 7 am schedule). Experiments were run during the light phase (between 10 am and 5 pm). All mice were handled on alternate days during the week preceding the first behavioral testing. Behavioral paradigm SDM task Experimental setup Experiments were conducted in a standard operant chamber (actor’s compartment, length of 24 cm × width of 20 cm × height of 18.5 cm; ENV-307W-CT, Med Associates) fused with a custom-made small triangle-shaped chamber that hosted the recipient (length of 18 cm × width of 14 cm × height of 18.5 cm). The separation wall between the compartments (operant chamber and recipient chamber) was replaced by a metal mesh with 1-cm holes that allowed social exploration and nose-to-nose interaction. The actor’s compartment was equipped with two nose poke holes and a food magazine between them, for delivery of food rewards (14 mg; Test Diet, 5-TUL). The recipient’s compartment presented only a food magazine connected to a food dispenser. The setup was placed inside a sound-attenuating cubicle (ENV-022V, Med Associates) homogeneously and dimly lit (6 ± 1 lux) to minimize gradients in light, temperature, sound and other environmental conditions that could produce a side preference. All tasks were controlled by custom scripts written in MED-PC V (Med Associates). A digital camera (Imaging Source, DMK 22AUC03 monochrome) was placed on top of the setup to record the test using a behavioral tracking system (Anymaze 6.2, Stoelting). Task design The testing subjects, the ‘actors’, were tested in the following three different conditions: (1) with a recipient, in which a cage mate was placed in the adjacent compartment and acted as the recipient; (2) no recipient, in which the compartment of the recipient was empty; and (3) with toy, in which the recipient was replaced with an inanimate object. The actor (Fig. 1a ) was determined based on the decision to receive a food reward for himself (selfish choice) or to allocate the reward also to his companion (altruistic choice), ‘the recipient’. Both choices were reinforced on FR1 such that poking into the left or right nose poke resulted in one food reward delivery. Altruistic and selfish responses were counterbalanced between left and right nose pokes across mice. After one nose poke, an intertrial interval of 5 s occurred. The recipient was a passive player and only received food rewards upon actor choices. In the no-recipient condition, the adjacent compartment was empty, while in the toy condition, an inanimate black object was placed in the recipient compartment. The task design was identical across the condition with a recipient and served as a control for pellet delivery sounds and for potential secondary effects of reinforcement. The actor and recipient were mildly food restricted to 90% of their free-feeding body weights to promote food-seeking behavior and were housed together for at least 2 weeks before the experiment. In the condition with unfamiliar recipients, the actor and recipient were never housed together. The actors were tested for 5 days, in 40-min sessions, with (with a recipient) or without (no recipient) a partner, with an inanimate object or with an opaque or transparent partition dividing the recipient and actor, depending on the testing condition. Actors were always paired with the same recipient throughout the same experiment. In the toy condition, actors were tested for 5 days with a partner and the day following the last session (day 5), the recipient was replaced by an inanimate object (day 6). In the conditions with the opaque or transparent partition between the actor and recipient, actor mice were tested for 5 days. The opaque partition did not allow visual cues and social exploration/interaction, while the transparent partition allowed visual cues but not social exploration/interaction. FR schedules To test whether mice made voluntary choices to benefit others under costly conditions, we tested mice using an increasing FR schedule for altruistic decisions from FR2 to FR8. In this condition, the number of operant responses required to dispense food to the recipient is increased each day (from two to eight). Selfish responses remained on FR1 throughout the experiment. In the ‘no-recipient’ condition, for each actor, the preferred nose poke was reinforced using the increasing FR schedule and the other nose poke was kept on FR1. SDM task without concurrent reward Mice were trained in the SDM task and then the paradigm was modified such that one nose poke resulted in food rewards for only the actor (selfish choice) with the other to the recipient only (altruistic choice). Mice were tested in longer sessions (120 min) to observe possible effects of satiety on their choices. Satiety-induced reward devaluation Mice were tested for 5 days in the SDM, and following the last session, actors were singly housed for 2 h and the reward outcome was devalued by prefeeding them to satiety, giving free access to reward pellets in their cage. Then, mice were transferred to the operant chamber and tested in a nonreinforced session. Sated actor mice were tested in two different experiments. In experiment 1, two groups of mice were tested: in one condition, both the actor and recipient did not receive rewards (no reward) and in the other, actor mice did not receive any reinforcement, but they could still allocate food rewards to their recipients (reward to the recipient only). In experiment 2, mice were tested in the ‘SDM without concurrent reward’ after altruistic choices. SDM task with sated recipients Both actor and recipient mice were food restricted as described above. Two hours before each session of the SDM task, recipient mice were separated from their cage mate actors and the reward outcome was devalued by prefeeding them to satiety by giving them free access to reward pellets in their cage. Then, both actor and recipient mice were transferred to the operant chamber and tested in a reinforced session. In a different cohort of mice, we tested satiety-induced reward devaluation in the recipient mice after standard training in the SDM task for 5 days. In this condition, one group of actor mice was tested with food-restricted recipients and one group was tested with sated recipients following satiety-induced reward devaluation. Analyses The number of nose poke responses was counted by software (MED-PC V, Med Associates). To quantify individual preferences of altruistic over selfish responses or left and right nose pokes, we calculated a decision preference score as follows: (number of altruistic responses − number of selfish responses)/total number of responses. Video images were analyzed a posteriori for scoring exploratory behavior using Anymaze 6.2 (Stoelting) and Boris 46 . We measured the time spent by the recipient and the actor in social exploration in the area (highlighted in red, Fig. 1h ) in the proximity of the adjacent compartment, where they could explore each other. Tube tests were performed as described in a previous study 47 . We used a transparent Plexiglas tube (length of 30 cm, inside diameter of 3 cm). For habituation, the tube was placed inside the cage for three consecutive days. After habituation, mice were trained to run inside the tube. Each mouse was released at alternating ends of the tube and was allowed to run through the tube. We used a plastic stick to guide the mouse to the end of the tube if needed. Each animal was given ten training trials on two consecutive days. For the test, two mice were simultaneously released into the opposite ends of the tube and care was taken to ensure that they met in the middle of the tube. The first mouse that retreated and placed its two rear paws outside the tube was recorded as the ‘loser’ of the trial with the other mouse recorded as the ‘winner’. Between each trial, the tube was cleaned with 75% ethanol. Mice were tested pairwise using a round-robin tournament, on daily sessions. Each pair of cage mates was tested in consecutive trials, alternating the starting side of the tube. The test was performed until all the ranks were stable for at least four continuous daily trials. To assign each animal social rank, we used the normalized DS for dominance. The score was calculated from the individual proportion of wins and losses in all the trials, in relation to the wins and losses of its opponents, as reported in a previous study 48 . We then normalized the score to be between 0 and n − 1 (where n is the number of subjects in each cage), using the following formula: $${{{\mathrm{Normalized}}}}\,{{{\mathrm{DS}}}} = \frac{1}{{{{{n}}}}}\left( {{{{\mathrm{DS}}}} + \frac{{n\left( {n - 1} \right)}}{2}} \right)$$ In hM4D-expressing animals, the tube test was performed in a different cohort of mice with or without the SDM task. The training in the tube for habituation was performed before the SDM task, and then, the tube test started and the SDM task started on the same day. The tube test was performed at least 1 h after the SDM task (with CNO injection). In hM4D-expressing animals that did not perform the SDM task, after reaching stable ranking, mice received CNO or vehicle and were tested at different time points following injection (1–2, 6–8 and 24 h). For BLA silencing, one mouse received CNO and the other cage mates received vehicle. In control cages, all the animals received vehicle. Observational fear conditioning The apparatus consisted of two identical and adjacent fear conditioning chambers (Ugo Basile, 24 × 20 × 30 cm 3 ) separated by a transparent Plexiglas partition. Olfactory and auditory cues could be transmitted between the chambers. A demonstrator mouse (previously a recipient in the SDM task) and an observer (previously an actor in the SDM task) were individually placed in the two chambers and allowed to explore the chambers for 5 min (baseline). Then, a 2-s foot shock (0.7 mA) was delivered every 10 s for 4 min to the demonstrator mouse using behavior tracking software (Anymaze 6.2, Stoelting). The same pairs of mice tested in the SDM task were used. Based on a previous study 28 , we used 10-s intervals for foot shocks and a 4-min training. At the end of the procedure, the mice were returned to their home cage. Stereotaxic surgeries Viral vectors AAV5-CamKIIa-mCherry (114469, titer ≥7 × 10 12 viral genomes (vg) ml −1 ), AAV5-CamKIIa-hM4D(Gi)-mCherry (50477, titer ≥3 × 10 12 vg ml −1 ), AAV5-Syn-DIO-hM4D(Gi)-mCherry (44362, titer ≥7 × 10 12 vg ml −1 ) and AAV1.CamKII.GCaMP6f.WPRE.SV40 (100834, titer ≥1 × 10 13 vg ml −1 ) were purchased from Addgene. CAV2 equipped with Cre recombinase (titer ≥2.5 × 10 11 vg ml −1 ) was purchased from the Institute of Molecular Genetics in Montpellier CNRS, France. Surgical procedures C57BL/6J mice were naive and 2 months old at the time of surgery. All mice were anesthetized with a mix of isoflurane/oxygen 2%/1.5% by inhalation and mounted into a stereotaxic frame (Stoelting) linked to a digital micromanipulator. Brain coordinates of viral injection were chosen in accordance with the mouse brain atlas as follows 49 : (1) BLA: AP, −1.7 mm; ML, ±3 mm; DV, −4.5 mm and (2) PL: AP, 2.0 mm; ML, ±0.25 mm; DV, −2.4 mm. The volume of AAVs injected was 400 nl for hM4D and 150 nl for CAV2-Cre, per hemisphere. We infused the virus through a 10-μl Hamilton syringe. After infusion, the pipette was kept in place for 5 min. After virus injection, mice were allowed 4 weeks to recover and for the viral transgenes to adequately express before behavioral experiments. For fiber photometry, a glass micropipette connected to a 10-μl Hamilton syringe was lowered into the BLA (DV, −4.75 from the skull) and 300 nl of virus (AAV1.CamKII.GCaMP6f.WPRE.SV40) was injected (0.1 μl min −1 ) using a syringe pump (Harvard Apparatus). After infusion, the pipette was kept in place for an additional 10 min and then slowly withdrawn. A multimode fiberoptic cannula (200-μm core, 0.5 NA, ~6 mm, Thorlabs) was implanted 0.15 mm above the injection site (DV, −4.60 from the skull). The implant was secured to the skull with MetaBond and dental cement. Mice were housed in pairs immediately after surgery and were allowed to recover for at least 4 weeks after surgery for virus expression before the experiment began. Fiber photometry recordings To assess the activity of BLA neurons during the SDM task, the fluorescence signal emitted by GCaMP6f-expressing neurons was recorded using fiber photometry 50 . A signal processor (RZP5, Tucker Davis Technologies) was used to control two light sources (465-nm LED, CLED_465; 405-nm LED, CLED_405, Doric Lenses), which were modulated at 211 and 539 Hz, respectively. The two wavelengths were combined by a fluorescence minicube (Doric Lenses) and transmitted through an optical patch cable (Doric Lenses) to the mouse head implant. Emitted fluorescence was collected by the same patch cable, delivered back to the same minicube through a 525-nm filter and sent to a photoreceiver (Femtowatt Silicon Photoreceiver, DC-750 Hz; Newport). Real-time signals were acquired, lowpass filtered (6 Hz) and demodulated with Synapse Essentials software (Tucker Davis Technologies). Med-PC system (Med-PC V Software Suite, Med Associates) generated TTLs to time-stamp specific events (that is, nose pokes, food receptacle entries and pellet deliveries). Data were extracted from TDT files and analyzed using custom MATLAB scripts. Demodulated datastreams were filtered at 1,017 Hz, divided into discrete trials by aligning with TTL representing a trial (that is, a nose poke resulting in a pellet delivery) and binned into 100-ms bins. Z -scores were calculated for each nose poke-related signal by taking the mean divided by the s.d. of a 5-s baseline period preceding each nose poke. Area under the curve was calculated for the 5 s following the nose poke. The prelearning phase was defined as the first experimental day when mice underwent the task for the first time. If no nose pokes were observed during the session, prelearning was considered as the first day in which mice were performing at least one nose poke for each side. The learning phase was defined as the day in which a mouse performed more than 20 pokes throughout the session. Postlearning included the last two experimental days, in which altruistic or selfish behavior was consistent with this observation. For the postlearning phase, data were averaged across the 2 days. Quantification of c-Fos-positive cells Actor mice were tested in the SDM task with or without recipient mice for 5 days. On the last day of training, actor mice were killed 90 min after the session and brains were collected and processed for immunohistochemical detection of Fos protein. All cells were counted bilaterally from two to six coronal sections of BLA for each mouse. Drugs For hM4D activation, we used i.p. administration of clozapine- N -oxide dihydrochloride (CNO, HB6149 Hello Bio) dissolved in physiological saline (0.9% NaCl) at a dose of 3 mg kg −1 in a volume of 10 ml kg −1 , 30 min before the behavioral experiments. All mice (control and hM4D) received i.p. CNO injections. Tissue-slice preparation and immunohistochemistry Mice were transcardially perfused with 40 ml of 0.1 M PBS and then with cold paraformaldehyde (PFA, 4% in PBS). The brain was removed from the skull and postfixed in 4% PFA in PBS for 1 h at 4 °C. The brain was sliced into 40-μm coronal sections using a Vibratome 1000 Plus Sectioning System (3M). Brain slices were permeabilized in 0.3% Triton X-100 in PBS (0.3% T-PBS) for 1 h at room temperature while shaking. After permeabilization, brain slices were blocked with 0.1% Triton X-100 in PBS (0.1% T-PBS) supplemented with 10% normal goat serum (NGS) for 2 h at room temperature, while shaking. After permeabilization and blocking, to examine c-Fos expression, slices were incubated with anti-c-Fos and anti-GAD67 antibodies in 0.1% T-PBS supplemented with 3% NGS for 3 overnight at 4 °C, while shaking. The appropriate Alexa Fluor-conjugated secondary antibodies in 0.1% T-PBS with 3% NGS were applied for 2 h at room temperature followed by nuclei staining with the fluorescent dye 4′,6-diamidino-2-phenylindole (1:50,000 in PBS; Thermo Fisher Scientific). Labeling in the BLA and in the PFC was visualized with a confocal microscope (Zeiss) with a ×20 objective. Specifically, a z -stack of 1-µm steps was taken and then analyzed using Fiji (ImageJ) software. To detect the viral expression of hM4D in BLA- and PFC-projecting neurons, brain slices were stained with rabbit anti-DsRed. To visualize the viral expression of hM4D and GCaMP6f and fiber placement, BLA- and PFC-containing brain slices were acquired with a Nanozoomer S60 (Hamamatsu), using constant settings, or an Axiovert 200M microscope (Zeiss). Antibodies For immunohistochemistry analyses, the following primary antibodies were used: rabbit anti-c-Fos (sc-7202, Santa Cruz Biotechnology; dilution: 1:1,000), mouse anti-GAD67 (MAB5406, Sigma-Aldrich; dilution: 1:800) and rabbit anti-DsRed (632496, Takara; dilution: 1:1,000). The following secondary antibodies were used: goat anti-rabbit-Alexa488 (A-11034, Invitrogen; dilution: 1:1,000) and goat anti-mouse-Alexa647 (A-21235, Invitrogen; dilution: 1:1,000). Statistics and reproducibility No statistical methods were used to predetermine sample sizes but sample size was selected based on previous experience and on estimation from related studies 6 , 7 , 8 , 17 . Animals were randomly assigned to control and manipulation groups. Experimenters were not blinded during data acquisition, but all analyses were performed with blinding to the experimental conditions as stated in the Methods . One mouse was excluded from data collection because it showed little motivation to engage in nose poke responses (fewer than 10 total pokes), two were excluded because viral expression patterns were not appropriate (outside the target region) and two were excluded due to fiber misplacement. Statistical analyses and figure plotting were performed using Prism version 9 (GraphPad). Data are reported as mean ± s.e.m. plots or box plots. In box plots, the central mark indicates the median and the bottom and top edges of the box indicate the maximum and the minimum. Statistical methods used in this study include two-way RM ANOVA and one-way ANOVA followed by Bonferroni correction and Wilcoxon matched-pairs signed-rank test. Single-variable comparisons were made using two-tailed paired and unpaired t -tests. Mice were assigned to altruistic or selfish groups using one-sample t -tests compared to chance (50%). The accepted value for significance was P < 0.05. Sample sizes and statistical tests are reported in the figure legends. Data distribution was tested using the D’Agostino and Pearson normality test. The experiments reported in this work were repeated independently at least two to four times. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability All source data used to generate the figures are available at . Code availability Custom-written analysis code is available at .

Humans and animals can exhibit a wide variety of behaviors when interacting with their peers. These include both prosocial behaviors, voluntary actions aimed at supporting or helping others, and selfish or opportunistic behaviors, which prioritize one's own needs or desires. Several past neuroscience studies have investigated the neural processes involved in social connection, particularly in terms of the brain regions and neurotransmitters associated with bonding or intimacy. However, so far very little is known about the brain mechanisms underlying selfish or altruistic actions. Researchers at the Italian Institute of Technology (IIT) and University of Milan have recently carried out a study aimed at filling this gap in the literature, by examining the activation and firing of neurons in the mice brain as the animals engaged in prosocial or selfish behaviors. Their paper, published in Nature Neuroscience, identifies a series of prosocial or selfish behavior-specific reciprocal connections between neurons in the prefrontal cortex and amygdala, two brain regions associated with complex behavioral planning and emotional regulation, respectively. "Over the past years, I felt a loss of the common sense of collectiveness, even before the pandemic, that contributed to self-centered concern and a disregard of others," Diego Scheggia, one of the researchers who carried out the study, told MedicalXpress. "Of course, recent physical distancing and quarantine policies further disrupted our daily social interactions and resulted in increased social isolation. The goal of my recent study was to understand the social factors and neurobiological determinants of altruism and self-interest." To examine the neural mechanisms underpinning altruistic and selfish behaviors, Scheggia and his colleagues first designed a new experimental task that would encourage these different types of behaviors in animals. This social decision-making task offered animals two simple choices: sharing a reward with their peers or not. The researchers then used this task to perform an experiment on several adult mice. "The task we used was modeled on the human game-theoretical paradigm known as the 'dictator game' in which a 'dictator' decides whether to share a reward with a conspecific," Scheggia explained. "In this task, we measured the neural activity of the basolateral nucleus of the amygdala that show substantial differences between prosocial and selfish subjects." Magnification of the basolateral amygdala. In red the neurons activated during the behavioral task "dictator game." Credit: Scheggia et al. Essentially, Scheggia and his colleagues found that the mice's decision to share their food reward with other mice depended on several factors. These factors included the mice's level of familiarity with their peers, their sex, previous social contacts, levels of hunger, hierarchical status and their emotional attunement. Subsequently, the researchers used chemogenetics, techniques that entail the use of synthetic drugs to manipulate specific brain pathways, to silence the activity of some neurons. This allowed them to determine whether silencing specific neurons automatically resulted in more prosocial or selfish behaviors. "We observed that silencing the neurons in the prefrontal cortex that are connected to the basolateral amygdala modulated choices guided by self-interest," Scheggia said. The results gathered by this team of researchers highlight the role of neurons in the basolateral amygdala (BLA) in promoting prosocial decisions. Specifically, they suggest that BLA neurons projecting on the prelimbic (PL) part of the prefrontal cortex mediate the development of a preference for altruistic choices in mice. In contrast, the preference for selfish behaviors appeared to be mediated by the projection of neurons in the PL prefrontal cortex to the BLA brain region. In the future, the findings of this study could pave the way for interesting new discoveries about the impact of interactions between neurons in the amygdala and prefrontal cortex on the nature of animal behaviors. The work by Scheggia and his colleagues also outlines a possible neurobiological model of altruistic and selfish choices, which could help to better understand the issues with social decision-making associated with some psychiatric disorders. "As a result of the pandemic, more selfish and antisocial behavior have risen, thus threatening to multiply psychiatric diagnoses in the next few years," Scheggia added. "Unraveling the nature of brain mechanisms underlying social decision-making could represent a crucial step to suggest novel treatments for social dysfunctions and antisocial behavior that occur in several psychiatric conditions and neurodegenerative disorders. Building on the findings of our recent study, we are now exploring the role of previous experience and emotional memories on prosocial and selfish choices."

10.1038/s41593-022-01179-2

Earth

Study suggests Australians can be sustainable without sacrificing lifestyle or economy

Steve Hatfield-Dodds et al. Australia is 'free to choose' economic growth and falling environmental pressures, Nature (2015). DOI: 10.1038/nature16065 Journal information: Nature

http://dx.doi.org/10.1038/nature16065

https://phys.org/news/2015-11-australians-sustainable-sacrificing-lifestyle-economy.html

Abstract Over two centuries of economic growth have put undeniable pressure on the ecological systems that underpin human well-being. While it is agreed that these pressures are increasing, views divide on how they may be alleviated. Some suggest technological advances will automatically keep us from transgressing key environmental thresholds; others that policy reform can reconcile economic and ecological goals; while a third school argues that only a fundamental shift in societal values can keep human demands within the Earth’s ecological limits. Here we use novel integrated analysis of the energy–water–food nexus, rural land use (including biodiversity), material flows and climate change to explore whether mounting ecological pressures in Australia can be reversed, while the population grows and living standards improve. We show that, in the right circumstances, economic and environmental outcomes can be decoupled. Although economic growth is strong across all scenarios, environmental performance varies widely: pressures are projected to more than double, stabilize or fall markedly by 2050. However, we find no evidence that decoupling will occur automatically. Nor do we find that a shift in societal values is required. Rather, extensions of current policies that mobilize technology and incentivize reduced pressure account for the majority of differences in environmental performance. Our results show that Australia can make great progress towards sustainable prosperity, if it chooses to do so. Main Our analysis uses a new integrated multi-model framework developed for the Australian National Outlook 1 . Australia is globally relevant: a major exporter of energy, mineral and agricultural products, with high per capita income, greenhouse gas emissions, water extractions, and habitat loss. The framework assesses energy–water–food interactions (and links to ecosystem services) in the context of climate change 2 , and uses more than 20 scenarios to explore a diverse range of factors shaping future Australian economic and environmental outcomes 1 , 2 . Interacting national trends and policies include energy and resource efficiency, agricultural productivity, consumption and working hours, and new land-sector markets for energy feed-stocks and ecosystem services (carbon sequestration and biodiversity conservation). These are modelled against four levels of national and global greenhouse gas emissions reduction effort (from no abatement to very strong abatement), and associated global climate trajectories (see Extended Data Fig. 9 ). As well as assessing the range of scenario outcomes, we identify the relative contributions of different types of choices. ‘Collective choices’ are defined as decisions that can only be implemented by groups of actors, and then constrain or empower ‘individual choices’ (particularly through changing rules and institutions). For example, individual choices about whether to drive or catch a train to work are strongly shaped by prior collective choices about transport infrastructure. The framework accounts for detailed interactions across sectors and spatial scales. The focal scale is national (the continent of Australia), accounting for key processes at higher (global) and lower (sub-national) spatial scales. This cross-domain integrated approach is needed because partial assessments may not account for constraints or adverse impacts that would undermine an otherwise ‘sustainable’ trajectory 3 , 4 , 5 , 6 , 7 , 8 . The projections and indicators are fully consistent with the international System of National Accounts 9 . We provide more details in the Supplementary Methods (section ‘Overview of modelling framework and scenarios’) and results for more than 60 national and global indicators in the Supplementary Data . Novel aspects of the analysis include assessing the potential for markets for ecosystem services to supply carbon sequestration and habitat restoration (and implications for agricultural output 7 and extinction risk) 10 , 11 ; assessing future water stress rather than simple volume of water extracted 2 , 12 ; exploring material extractions and environmental footprints 13 ; and integrating these elements with established models for analysing energy, greenhouse gas emissions and economic performance 2 , 14 , 15 , 16 , 17 . We are not aware of any other future-looking modelling that integrates this range of issues and indicators ( Supplementary Methods , ‘Overview of modelling framework and scenarios’). Economic and physical decoupling is possible We find that substantial economic and physical decoupling is possible 18 . Economically, Australia can achieve strong economic growth to 2050, indicated by rising gross domestic product (GDP) and gross national income (GNI) per capita, in scenarios where environmental pressures fall or are stable. Physically, we find the services derived from natural resources (energy ( Extended Data Fig. 2 ), water ( Extended Data Fig. 3 ), food ( Extended Data Fig. 4 )) can increase, while associated environmental pressures ease (greenhouse emissions ( Extended Data Fig. 6 ), water stress ( Extended Data Fig. 3 ), native habitat loss ( Extended Data Fig. 5 )). Importantly, these projected decouplings do not involve a reduction in the value of Australia’s heavy industry ( Extended Data Fig. 1g ), or outsourcing its environmental footprint to other nations 13 , 19 . Instead energy- and material-intensive sectors are projected to increase their share of economic activity, even in scenarios with the strongest global abatement efforts 1 , 2 . In all scenarios, Australia’s economy and living standards are projected to grow strongly (see Extended Data Fig. 1 ). As shown in Fig. 1 , the value of economic activity (GDP) is projected to rise tenfold over the 80 years to 2050, driven by a 2.9-fold increase in population ( Extended Data Fig. 8 ) and a 3.2–3.6-fold increase in GDP per capita (all values are in real 2010 Australian dollars, adjusted for inflation). National income (GNI) grows at a similar rate as GDP, with GNI per capita increasing by 58–82% from 2010 to 2050. Around two-thirds of the range of outcomes is explained by choices about working hours and consumption rather than environmental constraints. Average incomes rise by up to 66% if average working hours decline another 11% over the next four decades, in line with recent trends, and rise by 75% or more if there is no decline in working hours. The remaining income differential is accounted for by different assumptions and outcomes on resource efficiency, new land markets, agricultural productivity, and national and global abatement efforts. Figure 1: Economic activity (GDP) and national income (GNI) continue to rise strongly in all scenarios. Projections for 20 scenarios. GDP measures the market value of goods and services produced. GNI here measures payments to national residents from domestic production (as foreign production is not modelled). All values are in real 2010 Australian dollars, adjusted for inflation; one trillion is defined as 1 × 10 12 . Neither GDP or GNI is adjusted for changes in asset values, such as depreciation or the depletion of stocks of natural resources, and so do not measure pure income. More information on models and scenarios is provided in Supplementary Methods , ‘Overview of modelling framework and scenarios’. Sources: Supplementary Data worksheets 1a and 1c. PowerPoint slide Full size image Net greenhouse emissions show a clear decoupling from the growing economy, falling to zero or lower in some scenarios by 2040 (top row of Fig. 2 ). Australian emissions per capita could fall below the global average by 2050, from four times the global average today ( Extended Data Figs 6b and 9f ). One-third to one-half of Australia’s projected emissions reductions are achieved through biosequestration from large areas of new carbon plantings (29–59 Mha in 2050, see Extended Data Fig. 5 ). The remainder is achieved by reducing the emissions- and resource-intensity of the economy. If there is a strong or very strong abatement effort, domestic emissions could fall by up to 33%, even as GDP grows more than 150%; and energy emissions could fall by up to 29% while energy use grows by 55–120%. Similarly, the total mass of fossil fuels, metals, non-metallic minerals and biomass 20 Australia uses is projected to decrease by 36% by 2050 in scenarios with very strong abatement and improved resource efficiency ( Extended Data Fig. 1h ). In other scenarios, total resource use is projected to increase by 69% 13 . Figure 2: Decoupling of emissions, water stress, and native habitat from the supply of energy, water and food, respectively, for 18–21 scenarios, 1970–2050. Each panel shows the scenario trajectories for a key indicator of resource use or environmental pressure. The shaded areas indicate scenarios in which environmental pressure decreases from current levels (in the left-hand panel), with the same scenarios shaded in the right hand panel of each row. Models and scenarios are described in Supplementary Methods , ‘Overview of modelling framework and scenarios’, and information on performance of multiple pressures across scenarios is provided in Supplementary Methods , ‘Analysis of multiple pressures across scenarios’. Sources: Supplementary Data worksheets 6a, 2a, 3e, 3a, 5h and 4d. PowerPoint slide Full size image National water extractions (by all sectors) are projected to increase by up to 101% by 2050. However, up to half (32–56%) of this water demand can be met by desalinisation in coastal cities and water recycling for industrial use. Water stress, indicated by rain-fed water use in water-limited catchments 12 , 21 , improves or is stable in 7 of 18 scenarios (and is sensitive to governance of new carbon and biodiversity plantings, as noted below). Pressures on biodiversity can also be reduced alongside economic growth and increased agricultural activity—resulting in increased native habitat and agricultural output volumes (including protein) in many scenarios 22 (bottom row of Fig. 2 ). Settings that give weight to biodiversity restoration could see mixed local native species plantings make up 36–47% of all carbon plantings in 2050 (against only 5% under a carbon-focused approach), increasing native habitat by up to 25% (37 Mha) in Australia’s intensive use zone, and reversing the long-term trend. With strong abatement incentives, we find 11 Mha of habitat could be restored without large government outlays, reducing climate-related extinction risk by 7–9% (assessed for RCP 4.5 climate) 1 . However, these carbon and biodiversity plantings would reduce surface water flows, which could exacerbate pressures on river-based ecosystems in water-limited catchments (middle row of Fig. 2 ). Integrated governance is needed to properly balance their interceptions with competing extractive uses 23 ( Supplementary Methods , ‘Analysis of multiple pressures across scenarios’). Existing Australian governance arrangements cap extractions from water-limited catchments around current levels. The requirement to hold a water licence for new plantings embeds the price of water licences in these governance arrangements, as discussed below. Alternative governance assumptions could further restrain plantings, better safeguarding river health, but forgoing up to 0.5 Gt (5%) of cumulative national carbon sequestration by 2050. Overall, two-thirds of the scenarios assessed (13 of 18) show improvement in at least one environmental indicator, but only three scenarios (all involving strong or very strong abatement and new land markets) show improvement or stable performance in all three environmental indicators, reflecting the tensions between reducing water stress and restoring terrestrial native habitat, and the importance of integrated governance (see Supplementary Methods Fig. 6 and Supplementary Methods , ‘Analysis of multiple pressures across scenarios’). Policies to ease pressures extend established options The scenario assumptions that result in reduced environmental pressures are all continuations of existing trends, combined with greater uptake of energy and water efficiency, and a shift towards stronger global and national greenhouse gas abatement ( Supplementary Methods , ‘Overview of modelling framework and scenarios’). Policy settings reflect market-based approaches that are already in place in Australia or other countries. Greenhouse gas abatement is modelled as a uniform global broad-based carbon price, representing a variety of potential real-world mixes of regulation, standards, grants, taxes, or cap-and-trade arrangements. The carbon price in 2015 is US$15 (moderate scenario), US$30 (strong) and US$50 (very strong) per tonne of CO 2 emissions, and increases by around 4.5% per year in real terms (above inflation) to 2050. This drives a 90% reduction in the emissions intensity of Australian electricity from 2010 to 2050 in the stronger abatement scenarios (eliminating coal-fired electricity without carbon capture and storage before 2035 under the highest carbon price). Wholesale generation prices are 61–106% higher in 2050, and household electricity prices are 11–12% higher (strong) or 32% higher (very strong), compared to the no-abatement scenarios. However, affordability changes very little, owing to higher household incomes (in all scenarios) and higher energy efficiency in scenarios with higher prices 17 . Payments to Australian landholders for biosequestration are 15% below the global carbon price, with the forgone carbon revenue applied to increasing the share of native habitat plantings from 4–5% to 36–46% of total area in 2050. The resulting biodiversity ‘top up payments’ account for 22–30% of payments to habitat plantings in these scenarios over the decade to 2050, complementing carbon income. (These payments should be interpreted as a one-off payment for implementing a conservation covenant, for the area of new habitat added in that period.) On water, we find that interceptions from new plantings result in increased water stress in many of the very strong abatement scenarios (which have the highest levels of new plantings). We find the profitability of carbon plantings is not sensitive to water licence prices: a doubling results in just a 4% reduction in the area of new plantings in water-limited catchments. Limiting the area of plantings to avoid this increased water stress would require a 200% increase in the water licence price (increasing the asset value of licences to existing owners). Policy choices are crucial, not changes in values These results provide insights into the contested relationship between economic growth and environmental sustainability 24 , complementing historical analyses 18 , 25 , 26 , 27 ( Supplementary Methods , ‘Competing views on the prospects for sustainability’). A ‘technological optimist’ view considers market-driven technological advances will ensure that growth does not transgress key environmental thresholds 28 , 29 , 30 . Others suggest that institutional reform and new policies could achieve necessary changes within established values and paradigms 25 , 31 , 32 , 33 , noting that environmental damage may occur during the long lags between problem identification and policy responses 18 , 25 , 34 , 35 , 36 . A third ‘communitarian limits’ view argues that sustainability will require a fundamental shift in societal values, often involving a rejection of economic growth 37 , 38 , or a shift from consumerism to a values-based commitment to living within ecological limits 39 . We find that decoupling economic growth from environmental pressure before 2050 would not require a change in societal values, but is not automatic—contrary to both the communitarian limits and technological optimist positions. It is not projected to occur under existing trends, and requires, in our scenarios, collective choices to increase global and national abatement efforts. The analysis explores potential behavioural change in several ways. The modelling simulates bottom-up individual choices on working hours and consumption that shape production and consumption as incomes rise (income elasticity) and relative prices change (price elasticity). These choices interact with different assumptions about policy settings (reflecting collective choices), such as incentives for greenhouse gas abatement, and about bottom-up trends, such as the uptake of energy and water efficiency. None of the scenarios assume a new social or environmental ethic. In particular, increasing Australia’s abatement effort in line with emissions reductions by other countries would be consistent with Australian public opinion 40 and assessments of Australia’s national interest 41 , 42 , 43 in limiting the rise in average global temperature to 2 °C 5 , 7 , 32 , 44 , and so is not interpreted as implying a change in values. Rather, the analysis reflects how goal-oriented human behaviour can change with circumstances (including new information, or changes in the actions of others), without requiring any change in underlying goals and values. We find collective policy choices are crucial, explaining 46–94% of differences in environmental performance and resource use across the scenarios examined (see Extended Data Fig. 7 and Supplementary Methods , ‘Assessing the contributions of individual and collective choices’). Consistent with the institutional reform approach 25 , 32 , 45 , 46 , we find top-down collective choices are particularly important in shaping ‘public good’ outcomes—accounting for 83–94% of the differential in scenario outcomes for net greenhouse gas emissions, and 69–89% for greenhouse emissions excluding land sector sequestration. Bottom-up individual choices play a greater role when private and public benefits are aligned, such as when improved resource efficiency delivers financial savings. Individual choices account for up to half of the differential in scenario outcomes for energy use (33–47%) and non-agricultural water consumption (16–53%). Giving value to natural assets can build new advantage Economic analysis of climate change mitigation typically finds that limiting emissions involves near-term costs, but can yield net benefits over the long term (well after 2050) through avoided climate impacts 5 , 32 , 41 , 44 . Near-term co-benefits such as improved air quality and human health are also identified 47 , 48 . However, our analysis identifies additional near-term economic benefits for nations with a comparative advantage in ecosystem services, particularly carbon sequestration from reforestation. For these nations, stronger action to improve resource efficiency and environmental performance could unlock new sources of economic opportunity and growth, boosting near-term income while protecting natural assets essential to long-term well-being. Figure 3 compares national income and net emissions outcomes in 2030 and 2050 for 18 scenarios. All seven stronger abatement scenarios (blue and purple) with land sector markets have better economic performance to 2050 than those with moderate abatement (green scenarios). National income (GNI) in 2050 in these scenarios is up to 6% higher than under existing trends (see quadrant 1). These win-win outcomes occur because carbon sequestration becomes more profitable than beef and other agricultural production across large areas of Australia (up to 58 Mha, or 70% of the intensive-use zone), in a world taking stronger action to reduce emissions. Stronger abatement incentives also promote electrification and the use of biofuels in road transport, reducing oil imports. These economic gains outweigh the costs of more stringent national emissions targets, as well as the impacts of lower global demand for (and value added from) Australia’s emissions-intensive exports, relative to moderate national and global abatement (see Supplementary Methods , ‘Calculations for Fig. 3 and assessment of potential economic performance with different levels of global and national action to reduce greenhouse emissions’ and Extended Data Fig. 1i ). Figure 3: Comparing living standards and emission outcomes across multiple scenarios. Differences in national income (GNI) and net greenhouse gas emissions in 2030 and 2050, relative to existing trends. Calculations based on 18 scenarios. Emissions, water stress and native habitat all improve or are stable in three scenarios, combining step change energy efficiency with very strong abatement (L1XI)—marked as (a)—or strong abatement (M3XI) (b), or trend energy efficiency with strong abatement (M3XR) (c). Differences shown are relative to existing trends (M2XR) controlling for working hours and consumption trends. Scenario assumptions and notation (such as M2XR) described in the text and Supplementary Methods , ‘Calculations for Figure 3 and assessment of potential economic performance with different levels of global and national action to reduce greenhouse emissions’. Extended Data Fig. 6e shows time paths for each scenario from 2015 to 2050. Source: Supplementary Data worksheet 6e; see Extended Data Figs 1c and 6a . PowerPoint slide Full size image Across the scenarios explored, we find land-sector markets are needed to exploit these shifts in comparative advantage. Quadrant 4 reflects missed opportunities, including the scenario where very strong abatement action without land-sector markets leads to the worst relative economic performance (solid purple circle). Other scenarios in this quadrant involve transitions: pathways where emissions reductions generate net costs around 2030, but net benefits by 2050, relative to existing trends (see Extended Data Fig. 6e for time paths). Quadrant 2 shows the scenarios in which there is no global or national action to reduce emissions, reflecting a decline from current modest abatement efforts. Here, national income in 2050 is projected to be 5–7% higher than for existing trends, while emissions are projected to be 35–51% higher. These scenarios illustrate the classic ‘unsustainable development’ trade-off, where higher near-term living standards are achieved at the cost of increased risks and future damage to the Earth’s natural capital and life-support systems 5 , 46 . Adverse environmental feedbacks might see these scenarios shift towards quadrant 3 after 2050, combining worse economic performance and higher emissions. Limitations of the current modelling framework suggest that the analysis is likely to overstate the relative economic performance of the no-action scenarios (orange) and understate that of the very strong abatement scenarios (purple), because it does not fully account for all potentially significant climate impacts 1 , 2 . Making progress towards sustainable prosperity In summary, we find that Australia could materially ease environmental pressures while enjoying strong economic growth. Many of the 20 scenarios we explored would represent substantial progress towards sustainable prosperity 46 . Australia could begin to repair past damage; restoring significant areas of native habitat and achieving negative emissions (net sequestration) of greenhouse gasses. But none of these scenarios would guarantee sustainability, or eliminate future threats to Australia’s natural capital and the Earth’s life-support systems 6 , 46 . Instead, each implies a different portfolio of risks and opportunities, which we have not fully modelled beyond 2050. For example, new native habitat established before 2050 could provide a permanent flow of biodiversity benefits and other ecosystem services, while the flow of carbon sequestration provided will peak and eventually decline to zero, drawing attention to challenges and opportunities beyond our modelling horizon, such as the possibility of using carbon plantations to generate negative emission bioenergy with carbon capture and storage 49 . Reducing environmental pressures will not require a shift in societal values, but neither will technology deliver it automatically. Collective choices and public policy settings have a crucial contribution, and well-designed markets can boost national income by exploiting new areas of comparative advantage in some circumstances. However, these scenarios may present new longer-term risks and opportunities, and the synergies and trade-offs involved will be influenced by global circumstances. We also find an important threshold effect: moderate global action to reduce greenhouse emissions may diminish Australia’s traditional comparative advantage (particularly in fossil fuel-based sectors) without creating new areas of advantage; while stronger global action that places tangible value on emissions reductions could create new opportunities for creating value, providing win-win economic and environmental benefits relative to existing trends. While Australia could dramatically reduce environmental pressures across a wide range of global contexts, the economic costs of doing so will be smaller (and benefits larger) in global settings that support the stable functioning of key Earth systems, including through promoting clean energy. As these global circumstances emerge, Australia’s opportunities will multiply. Sustainable prosperity is possible, but not predestined. Australia is free to choose.

A sustainable Australia is possible – but we have to choose it. That's the finding of a paper published today in Nature. The paper is the result of a larger project to deliver the first Australian National Outlook report, more than two years in the making, which CSIRO is also releasing today. As part of this analysis we looked at whether achieving sustainability will require a shift in our values, such as rejecting consumerism. We also looked at the contributions of choices made by individuals (such as consuming less water or energy) and of choices made collectively by society (such as policies to reduce greenhouse gas emissions). We found that collective policy choices are crucial, and that Australia could make great progress to sustainability without any changes in social values. Competing views Few topics generate more heat, and less light, than debates over economic growth and sustainability. At one end of the spectrum, "technological optimists" suggest that the marvellous invisible hand will take care of everything, with market-driven improvements in technology automatically protecting essential natural resources while also improving living standards. Unfortunately, there is no real evidence to back this, particularly in protecting unpriced natural resources such as ocean fisheries, or the services provided by a stable climate. Instead the evidence suggests we are already crossing important planetary boundaries. Other the other end of the spectrum, people argue that achieving sustainability will require a rejection of economic growth, or a shift in values away from consumerism and towards a more ecologically attuned lifestyles. We refer to this group as advocating "communitarian limits". A third "institutional reform" approach argues that policy reform can reconcile economic and ecological goals – and is attacked from one side as anti-business alarmism, and from the other as indulging in pro-growth greenwash. Income up, environmental pressures down My colleagues and I have spent much of the past two years developing a new framework to explore how Australia can decouple economic growth from multiple environmental pressures – including greenhouse emissions, water stress, and the loss of native habitat. We use nine linked models to assess interactions between energy, water and food (and links to ecosystem services) in the context of climate change. The project provides projections for more than 20 scenarios, exploring different potential trends for consumption and working hours; energy and resource efficiency; agricultural productivity; new land-sector markets for energy feedstocks and ecosystem services; national and global abatement efforts, climate, and global economic growth. While our major focus is on Australia, at the national scale, we also model what might happen globally, and at more detailed state and local scales within Australia. We find economic growth and environmental impacts can be decoupled − in the right circumstances. National income per person increases by 12-15% per decade from now to 2050, while the value of economic activity almost triples. In stark contrast to income, which rises across all scenarios, environmental performance varies widely. Key environmental indicators such as greenhouse gas emissions, water stress, and native habitat and biodiversity are projected to more than double, stabilise, or fall across different scenarios to 2050. As shown in the chart below, we find that energy rises in all scenarios, but that greenhouse emissions can fall at the same time – with the right choices and technologies. Water use can also rise without increasing extractions from already stressed catchments. Food output (here indicated by protein) can increase, while native habitat is restored. Many of the 20 scenarios explored would represent substantial progress towards sustainable prosperity. Indeed, we find that Australia could begin to repair past damage: restoring significant areas of native habitat and achieving negative emissions (net sequestration) of greenhouse gasses. Growth of what? We use the normal definition of economic growth as measured by increase in Gross Domestic Product (GDP) – the value of goods and services produced in an economy – consistent with the national accounts framework. Some authors use a different definition, most notably Herman Daly a leading advocate for a steady state economy. Daly defines growth as an increase in physical economic scale, such as resource extraction, and goes on to argue that indefinite (material) economic growth is not possible. While this may be true, for his definition, it can be confusing for people that do not realise he is not referring to GDP growth. Indeed, Daly recently acknowledged that economic (GDP) growth is possible with finite resources and steady material throughput. These definitions matter: we project growth (GDP - measured in real dollars, adjusted for inflation) increases by more than 160% in scenarios where domestic material extractions and throughput (measured in tonnes) decreases by around 40%. Choosing a sustainable future But here is the real crunch: we find these substantial steps toward sustainability could build on policy approaches that are already in place in Australia or other countries. This implies Australia could make enormous progress towards a more sustainable future without a major change in what we value. We can be confident that a values shift is not required to achieve these outcomes – at least before 2050 – because none of the scenarios we modelled assume change in values or a new social or environmental ethic. Instead, we show that people will make choices to change their behaviour to make the best of particular policy settings. These choices shape production and consumption. For instance, we consider increasing Australia's climate effort in line with other countries would be consistent with Australian public opinion and assessments of Australia's national interest in limiting the rise in average global temperature to 2°C. So we do not interpret this as implying a change in values. But we find collective choices are crucial. For example, individual choices about whether to drive or catch a train to work are strongly shaped by prior collective choices about transport infrastructure. Collective choices are often, but not always implemented through changes in government policy, legislation, and programs. We find collective choices explain around 50-90% of differences in environmental performance and resource use across the scenarios we model. Consistent with the institutional reform approach, we find top-down collective choices are particularly important in shaping "public good" outcomes – accounting for at least 83% of the difference between scenarios for greenhouse gas emissions. Bottom-up individual choices play a greater role when private and public benefits are aligned. For instance individual choices account for up to half of the difference between scenarios for energy use (33–47%) and non-agricultural water consumption (16–53%). While individual choices are important, we find decisions we make as a society are likely to shape Australia's future sustainability more than the decisions we make as businesses and households. Sustainable prosperity is possible, but not predestined. Australia is free to choose.

10.1038/nature16065

Medicine

Study shows rate of extreme inbreeding in the U.K. and possible health impacts of it

Loic Yengo et al. Extreme inbreeding in a European ancestry sample from the contemporary UK population, Nature Communications (2019). DOI: 10.1038/s41467-019-11724-6 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-019-11724-6

https://medicalxpress.com/news/2019-09-extreme-inbreeding-uk-health-impacts.html

Abstract In most human societies, there are taboos and laws banning mating between first- and second-degree relatives, but actual prevalence and effects on health and fitness are poorly quantified. Here, we leverage a large observational study of ~450,000 participants of European ancestry from the UK Biobank (UKB) to quantify extreme inbreeding (EI) and its consequences. We use genotyped SNPs to detect large runs of homozygosity (ROH) and call EI when >10% of an individual’s genome comprise ROHs. We estimate a prevalence of EI of ~0.03%, i.e., ~1/3652. EI cases have phenotypic means between 0.3 and 0.7 standard deviation below the population mean for 7 traits, including stature and cognitive ability, consistent with inbreeding depression estimated from individuals with low levels of inbreeding. Our study provides DNA-based quantification of the prevalence of EI in a European ancestry sample from the UK and measures its effects on health and fitness traits. Introduction Mating between close relatives, that is inbreeding, is reported in many species to yield deleterious outcomes, such as reduced fertility 1 , 2 , 3 , 4 , stature 2 , 4 , 5 , 6 , 7 , 8 , 9 , 10 and lifespan 2 . In humans, consanguineous mating leads to higher childhood mortality 3 , 11 , 12 and to adverse effects on traits such as lung function 4 , 10 , 13 and cognitive ability 4 , 8 , 10 , 13 . Because of its detrimental consequences, also referred to as inbreeding depression, a number of species have developed inbreeding avoidance mechanisms to limit its effects 14 . In humans, inbreeding avoidance mechanisms, include cultural and religious taboos on incest, and laws explicitly forbidding certain types of mating. For instance, the Sexual Offences Act (2003) in the UK specifically forbids mating between first-degree (parent–offspring or fullsibs (FS), i.e., coefficient of relationship of 0.5) and second-degree (halfsibs (HS), grandparent–grandchild, avuncular or double-first cousins, i.e., coefficient of relationship of 0.25) relatives; and also forbids mating between step and relatives when one of the family members is below 18 years old. Cultural, legal, religious and health-related constraints strongly weigh on the ability to observe, and therefore study the causes and consequences of inbreeding between first- and second-degree relatives, hereafter referred to as extreme inbreeding (EI). A number of previous studies have attempted to quantify the prevalence and incidence of EI 15 , 16 , 17 , 18 , 19 , 20 . However, as underlined by van den Berghe 21 , the estimates which they produced are questionable given the “disinclination of family members to report incest when it occurs, and the countervailing bias of many scholars and crusaders to magnify the problem on which they build their career”. Add to these limitations, the relatively small size of these studies (often <1000 participants) and the discrepancies between them with respect to the definition of EI, as some of these studies included mating between step-relatives 21 . Here, we leverage a large observational study of ~450,000 participants, the UK Biobank (UKB), to quantify EI and its consequences in contemporary European descents from the UK population. We compare our estimates with the prevalence of police-recorded cases of incest offences reported in the Crime Survey for England and Wales (CSEW) between April 2002 and March 2017. We also characterise the distribution of runs of homozygosity (ROHs) in EI cases and assess its consistency with theoretical predictions. Finally, we characterise the phenotypic consequences of EI on a number of health-related traits measured in UKB participants. Results Prevalence of EI in European descents from the UKB We previously identified 22 456,426 individuals of European ancestry among the 487,409 UKB participants who have been genotyped. Ancestry was called in our previous study using projected principal components analysis based on known ancestry and whole-genome sequence data from 2504 participants of the 1000 Genomes Project 23 (Methods). Given that 12 participants had retracted consent, we only analysed 456,414 UKB participants in the present study. We used 301,412 quality-controlled genotyped single-nucleotide polymorphisms (SNPs) to call ROHs using the PLINK software (Methods). As in previous studies 4 , 8 , 10 , ROHs were defined as homozygous >1.5 Mb long genomic segments (Methods). We then estimated for each study participant the percentage of their autosome comprising ROHs as a measure of inbreeding. Such inbreeding measure, hereafter denoted F ROH , is a well-established predictor of pedigree inbreeding 24 , 25 . Following guidelines from the American College of Medical Genetics and Genomics (ACMG) 26 , 27 , EI was called for individuals with F ROH > 0.1. The use of both F ROH as a measure of inbreeding and of this threshold are recommended by the ACMG for detecting suspected consanguinity between parents. We thus identified 125 unrelated participants (65 males and 60 females) whose genomes are consistent with their parents being first- or second-degree relatives. That represents a prevalence of EI ~0.03%, i.e., ~1/3652 (95% confidence interval—CI 95% : [1/4428–1/3106]). As a sensitivity analysis, and consistent with theory predicting much longer ROHs under EI, we re-estimated the prevalence of EI considering only ROHs > 2 Mb or >5 Mb long. Using these alternative definitions of ROH, also recommended in the ACMG guidelines, we detected 115 (prevalence of ~1/3969; CI 95% : [1/4857–1/3355]) and 98 (prevalence of ~1/4658; CI 95% : [1/5807–1/3887]) cases of EI, respectively. We also estimated the prevalence of EI using allele-frequency based inbreeding measures or using ROHs detected on both autosome and X-chromosomes of female participants. (Supplementary Table 1 ). Given that the latter estimates of the prevalence of EI are not statistically distinct (paired t test: p > 0.05) from our first estimate based on ROHs > 1.5 Mb, we will hereafter only consider ROHs > 1.5 Mb. We then compared our estimate of the prevalence of EI with the prevalence of incest offences reported in the CSEW between April 2002 and March 2017. That survey reports a total of 11,196 cases of police-recorded incest offences over this time period (URLs). Relative to the population of England and Wales, which varied from 52,602,200 to 58,744,600 between those years (URLs), this represents a prevalence ranging from ~1/5247 (CI 95% : [1/5346–1/5151]) to 1/4699 (CI 95% : [1/4787–1/4612]). The latter estimate is of the same order of magnitude as our estimated prevalence of EI in the UKB although these two estimates are based on different time periods (births 1938–1967 in the UKB vs. reports 2002–2017 in the CSEW). We then compared the mean years of birth among EI cases with the rest of UKB participants and found no statistical difference ( p = 0.11). That suggests that the prevalence of EI is relatively unchanged over time although mean inbreeding coefficients have significantly decreased over the years (correlation between year of birth and F ROH : r = −0.01%; Pearson’s correlation test p = 5.5 × 10 −14 ). However, it is important to note that the prevalence of EI and that of police-recorded incest offences cannot naïvely nor strictly be compared because (i) only an unknown but likely small fraction of incest cases are reported to the police, (ii) not all cases of incest would result in viable offspring as observed in this study, and (iii) viable offspring with severe cognitive impairment due to inbreeding are unlikely to enrol themselves as participants in the UKB. Fry et al. 28 previously reported that the UKB is not representative of the entire UK population, as it notably, includes healthier and more educated participants than the average population. Such an ascertainment on traits which are negatively correlated with inbreeding (e.g., educational attainment (EA) or height 28 ), may lead the prevalence of EI in the UKB to be an underestimation of the actual prevalence of EI in the UK population. As a consequence, our estimate of prevalence of EI is likely conservative, although the magnitude of the underestimation is difficult to predict as it depends on many other unknown factors which might differ between UKB participants and the general population. Deconvolution of underlying mating types We next estimated the proportion of EI cases born from mating between first-degree relatives (mating type 1; MT1) vs. second-degree relatives (MT2) using a threshold-based approach based on F ROH . To determine an optimal threshold, we simulated inbreeding under MT1 and MT2 using phased genotypes from 972 unrelated UKB participants (Methods, Supplementary Table 2 ). These 972 UKB participants are the offspring from 972 independent parent–offspring (PO) trios identified in the UKB 29 . Over ~20,000 simulation replicates (one replicate is one simulated EI case) we found that F ROH as a predictor of underlying mating type (MT1 vs. MT2) yields an area under the receiver operating characteristic curve (AUC) of ~0.97 and that using F ROH > 0.17 as a threshold yields optimal sensitivity and specificity both >0.92 (Fig. 1 ). Using this threshold, we therefore identified 54/125 (i.e., ~43.2%) EI UKB cases whose parents are most likely first-degree relatives. It is worth noting that complex inbreeding loops between second degree-relatives may also lead to extreme values of F ROH . However, mating between first-degree relatives remains a more parsimonious explanation of the empirical observations, in particular in a population of European ancestry where such complex inbreeding loops are uncommon. Fig. 1 Predictive performances of FROH to discriminate different types of inbreeding: mating type 1 (MT1: parent-offspring or fullsibs mating), mating type 2 (MT2: halfsibs, avuncular, grandparent-grandchild or double-first cousins mating) or mating type 3 (MT3: first-cousins mating). Panels a and c correspond to the comparison of MT1 and MT2, while panels b and d correspond to the comparison of MT1 and MT2 on the one hand and MT3 on the other hand. Predictive statistics assessed are the area under the receiver characteristics operating curve (AUC), the sensitivity to detect MT1 over MT2 (true positive rate) and specificity to distinguish MT1 from MT2 (true negative rate). FROH>0.17 yields a sensitivity and specificity >0.92 to discriminate MT1 from MT2; and FROH>0.087 yields a sensitivity of ~0.94 and a specificity of ~0.79 to discriminate MT1 or MT2 from MT3 Full size image We further attempted to quantify the proportion of MT1 born from PO vs. FS mating ( π PO / FS ). Given that the theoretical expectation of F ROH is 0.25 both under PO and FS mating, we found, as expected, that F ROH alone cannot discriminate PO from FS in our simulations (AUC of ~0.5). However, F ROH being proportional to the cumulative length of ROHs across the genome also implies that the same value of F ROH could reflect either fewer larger ROHs, or more smaller ROHs. Therefore, we investigated if the numbers of ROHs detected ( N ROH ) under PO or FS mating are different and if so can discriminate those two types of mating. We found on average over ~20,000 simulation replicates that N ROH ~45 ROHs are detected in offspring of FS mating as compared to N ROH ~38 ROHs detected in offspring of PO mating (Table 1 ). Moreover, ROHs detected in offspring of PO mating were on average ~2.7 Mb longer than ROHs detected under FS mating (Table 1 ). Consistent with these observations, we found that N ROH as a predictor of mating type yields a discriminative AUC of ~0.81, with the optimal threshold of >41 yielding a sensitivity of ~0.77 and specificity of ~0.69. Using that threshold we predict that 24/54 (i.e., π PO / FS ~44.4%; CI 95% : [31.2–57.7%]) EI cases with F ROH > 0.17 are likely offspring of parent–offspring mating. We also considered an alternative approach that aims at directly estimating the proportion of EI cases born from PO vs. FS mating from modelling the length distribution of ROHs (Methods). We applied this method to 2244 ROHs segments detected in 54 EI cases with F ROH > 0.17 and estimated that π PO / FS ~67.6% (CI 95% : [45.2–90.1%]). To confirm this finding, we analysed the distribution of F ROH from X-chromosome ROHs (hereafter denoted F ROH-X ) in 26 female EI cases with F ROH > 0.17. This analysis is justified by the fact that the theoretical expectation of F ROH-X equals 0.5 under PO mating vs. 0.25 under FS mating. We first stratified these 26 female EI cases into two groups (Group 1 and Group 2) depending on whether the likelihood of their autosomal segments lengths is larger under PO mating or under FS mating. More specifically, Group 1 ( N = 10) and Group 2 ( N = 16) contain female EI cases predicted to be offspring of FS and PO, respectively (Supplementary Fig. 1 ) from the length distribution of autosomal ROHs. The mean F ROH-X in Group 2 is 0.53 (CI 95% : [0.41–0.65]), consistent with PO mating, while the mean F ROH-X in Group 1 is 0.34 (CI 95% : [0.19–0.49]), which is consistent with FS mating, although standard errors are large. Altogether, we found that between 44.4 and 67.6% of EI cases with F ROH > 0.17 are likely offspring of PO mating. Table 1 Mean number and length of runs of homozygosity (ROHs) detected in participants from the UK Biobank (UKB), including extreme inbreeding (EI) cases (defined as F ROH > 0.1) and unrelated EI controls (defined as F ROH < 0.01). We also report the mean and length of ROHs in simulated data under various mating types Full size table We simulated inbreeding between first-cousins (hereafter denoted MT3) in order to quantify the ability of the F ROH > 0.1 threshold recommended by the ACMG guidelines to discriminate MT1 or MT2 from MT3. We recall here that the coefficient of relationship between first-cousins is 0.125, and therefore the expected inbreeding coefficient of their offspring is E[ F ROH ] = 0.5 × 0.125 = 0.0625. Also, MT3 is legal in most countries and thus more common in the population. We found over ~20,000 simulation replicates that F ROH yields an AUC of ~0.95, and that using F ROH > 0.1 as a threshold yields a sensitivity of ~0.94 and a specificity of ~0.79 to discriminate MT1 or MT2 from MT3 (Fig. 1 ). This, therefore, suggests that ~8/125 EI cases identified (i.e., ~6.4%) in this study could in fact be offspring of first-cousins mating. Hill and Weir 30 derived that the theoretical standard deviation of inbreeding coeffcients of offspring of first-cousins is ~0.024. Therefore, assuming under MT3 that F ROH is normally distributed with mean 0.0625 and standard deviation 0.024, follows that the probability of F ROH > 0.1 equals ~5.9%, which is consistent with our simulations. Distribution of ROH in EI cases As expected, we found that EI cases harboured significantly more and significantly longer ROHs than EI controls ( F ROH < 0.01) in the population (Table 1 ). On average, we detected N ROH ~33.6 ROHs in EI cases vs. ~4.9 ROHs in EI controls. The mean length of ROHs was L ROH ~14.8 Mb in EI cases vs. ~2.1 Mb in EI controls. Both mean numbers and mean lengths of ROHs detected are consistent with our simulations of EI (mean N ROH ~33.6 and L ROH ~14.0; Table 1 ). We represent in Fig. 2 the histogram of ROHs length in EI cases, and report in Fig. 3 , a few examples of very large ROHs (>100 Mb) covering ~50% of an entire chromosome. We also report X-chromosome ROHs detected 54/125 female EI cases in Supplementary Fig. 2 . Fig. 2 Histogram of the lengths of 4,196 runs of homozygosity (ROHs) detected in 125 EI cases ( F ROH > 0.1). Each length was subtracted 1.5 Mb (i.e., minimum length used to detect ROHs) before mixture distribution was fitted. A 84:16 mixture of two exponential distributions with means ~15.7 Mb (rate = 1/15.7 ~0.06) and ~0.72 Mb (rate = 1/0.72 ~1.4), respectively was found to best fit the observed length distribution (dotted line) Full size image Fig. 3 Chromosomal and positional distribution of runs of homozygosity (ROHs) detected in 125 EI cases ( F ROH > 0.1). Each row, with possibly multiple segments, represents a unique participant. Segments are groups by autosomal chromosomes from chromosome 1 (bottom of each panel) to chromosome 22 (top of each panel) ROHs are grouped in 6 length categories: between 1.5 and 5 Mb ( a ), between 5 and 10 Mb ( b ), between 10 and 20 Mb ( c ), between 20 and 50 Mb ( d ), between 50 and 100 Mb ( e ), and above 100 Mb ( f ). f also show inbreeding coefficients of individuals harbouring the largest ROHs Full size image Previous theoretical studies have often considered the length of genomic segments homozygous by descent (HBD) to follow an exponential distribution 31 , 32 . These studies generally relied on specific assumptions regarding recombination map functions, like Haldane or Kosambi map functions, which yield tractable algebraical simplifications. However, empirical evidence supporting these assumptions remains limited. Moreover, some of these simplifying assumptions like that of independence between the lengths and the numbers of HBD segments have also been criticised 33 . Here, we used an empirical approach to estimate the length distribution of ROHs segments detected in EI cases using a mixture of exponential distributions. Given that only ROHs larger than 1.5 Mb were detected, we modelled the distribution of lengths minus 1.5 Mb and not directly the length distribution, which would better fit a mixture of truncated exponential distributions (Methods). Mixtures of exponential distributions represent a flexible family of probability distributions, from which the exponential distribution is a special case. We selected the number of mixture components best fitting the data using the Bayesian Information Criterion (BIC). To calibrate our inference, we first estimated the length distribution of >282,635 simulated true HBD segments under various mating types. Our simulations are based on observed recombination maps from the 1000 Genomes Project 23 , and therefore do not make additional assumptions regarding recombination rates (Methods). We found for all simulated mating types that BIC selects two mixture components, which suggests that the single exponential distribution is likely too simple to characterise the length distribution of HBD segments. Of note, mixtures of two exponential distributions also yield a better fit than gamma distributions that have previously also been proposed 1 . Similarly, we estimated the length distribution of >99,794 ROHs detected in our simulated data. We found consistently that the length distribution of simulated ROHs is also well characterised by a mixture of two exponential distributions. We report in Table 2 , the parameters of the mixture distributions estimated from true HBD segments and from ROHs. We then estimated the length distribution of the 4196 ROHs detected over all EI cases. We found this distribution to fit a 84:16 mixture of exponential distributions with means ~15.7 Mb (larger component) and ~0.7 Mb (smaller component), respectively (Table 2 ; Fig. 2 ). Overall, our findings suggest that the length distribution of HBD segments and ROHs can be well approximated with a mixture of two exponential distributions. Table 2 Parameters of mixtures of exponential distributions estimated from observed length distributions of homozygous-by-descent (HBD) genomic segments and runs of homozygosity (ROH) Full size table Another observation in our simulations was that that the mean number of ROHs detected in an individual was larger than the number of true HBD segments simulated. This somewhat counterintuitive observation is explained by the fact that HBD were defined as segments identical-by-descent (from parents to offspring), while ROHs were re-estimated from the genotypes of simulated offspring. As a consequence, although simulated offspring of matings between unrelated parents have exactly zero HBD segments, they still harbour ROHs > 1.5 Mb given that their chromosomes were sampled from 972 existing UKB participants. Despite not being closely related (genomic relationship (GRM) < 0.05), these 972 UKB participants are still likely to have a distant common ancestor (>25 generations ago), which would lead to detection of ROHs > 1.5 Mb in their (simulated) offspring. We found that simulated offspring of matings between unrelated parents had on average 4.8 ROHs > 1.5 Mb (Table 1 ). If we subtract that number (i.e., 4.8 ROHs) from the mean number of ROHs detected under simulated inbred matings (Tables 1 and 2 ), we now find very consistent mean numbers of ROHs and HBD segments per individual. More specifically, for each simulated inbred mating we find, after this correction, 32.5 HBD vs. 33.3 ROH for PO mating, 41.6 HBD vs. 40.4 ROH for fullsibs mating, 20.8 HBD vs. 20.2 ROH for HSs mating, 25.2 HBD vs. 23.5 ROH for avuncular mating, 20.8 vs. 20.1 ROH for grandparent–grandchild mating, 29.8 vs. 26.8 ROH for double-first cousin mating and 14.9 HBD vs. 13.3 ROH for first-cousin mating. Phenotypic consequences of EI We quantified the consequences of EI on multiple traits measured in the UKB. We first analysed ten control traits with prior evidence of inbreeding depression 4 , 8 , 10 , 13 . Those ten traits are height, hip-to-waist ratio (HWR), handgrip strength (HGS; average of left and right hand), lung function measured as the peak expiratory flow (PEF), visual acuity (VA), auditory acuity (AA), number of years of education (EA), fluid intelligence score (FIS), cognitive function measured as the mean time to correctly identify matches (MTCIM) and fertility measured as the number of children (NCh). We performed linear regressions of these traits on the EI status adjusted for age at recruitment, recruitment centre (treated as a categorical factor), sex, year of birth (treated as a continuous variable), genotyping batch (treated as a factor), socioeconomic status measured by the Townsend deprivation index and population structure measured by ten genetic principal components estimated from HM3 SNPs. As expected, we found that EI cases had a reduced mean in these ten traits as compared to EI controls. More specifically, we found phenotypic means in EI cases to be between 0.3 and 0.7 standard deviation below the population mean (Table 3 ). Note, that under normality assumptions, between ~25 and ~40% of the population has a phenotype below 0.7 and 0.3 standard deviations below the mean, respectively. Despite the small sample size of 125 EI cases, the reduction was statistically significant (Wald-test p < 0.5/10 = 0.005) for 7 out the 10 traits (Table 3 ). We also specifically estimated the inbreeding load (often denoted B ), which represents the number of loci with deleterious alleles that would cause one death on average if made homozygous 3 . As previously recommended 34 , we estimated B using Poisson regression of the number of children engendered onto F ROH . Poisson regression was performed using a logarithmic link function as also previously recommended 34 and adjusted for the same covariates listed above. For this analysis, we used the entire distribution of F ROH , (i.e., includes both EI cases and EI controls) and found an estimate of B ~1.46 (CI 95% : [0.87–2.05]; Wald-test p = 1.3 × 10 −6 ; Table 3 ). The effect of inbreeding on fertility of the resulting inbred offspring, that we have quantified here, has been previously detected in humans 35 . However, the latter study did not provide an estimate of inbreeding load that can be directly compared with ours. Nonetheless, we found that our estimate is consistent with estimates of inbreeding load on survival of offspring from inbred mating in humans 3 , 36 and other species 34 , 37 , although these are different traits. Table 3 Association between extreme inbreeding (EI) and multiple traits measured in UK Biobank participants (125 EI cases vs. 345,276 EI controls) Full size table We then assessed whether the observed reduction in these ten traits was consistent with inbreeding depression quantified within EI controls. Under the assumption that inbreeding depression results only from directional dominance effects of deleterious alleles or heterozygote advantage (overdominance), phenotypes are expected to decline linearly with increased inbreeding. However, if epistasis contributes to inbreeding depression 38 , 39 or if causal variants for inbreeding depression are rarer 1 , a nonlinear relationship could be observed in particular for large inbreeding coefficients. To test this hypothesis we first estimated inbreeding depression in 345,276 EI controls unrelated with each other and unrelated with the 125 EI cases. For each of the 10 control traits, we then compared the phenotypic mean in the 125 EI cases, with a linear prediction based on the estimate of inbreeding depression in EI controls. For this analysis inbreeding depression was also estimated using an alternative inbreeding measure ( F UNI ), which we previously showed to be more powerful for detecting inbreeding depression 4 . The latter analysis did not reveal a significant deviation from the linear prediction (Wald-test p > 0.005) regardless of the inbreeding measure used, which therefore underlines that the observed phenotypic reduction in EI cases is consistent with inbreeding depression observed within EI controls (Fig. 4 ). This also suggests that causal variants contributing to inbreeding depression in those traits are likely well-tagged (i.e., correlated) by common variants in the population. However, we acknowledge that the estimate of inbreeding depression from the EI cases present in the UKB might be too low if, as seems plausible, they are a relatively healthy sample from the population of all EI cases in the UK 28 . Fig. 4 Phenotypic reduction (in trait standard deviation; SD) observed in 125 extreme inbreeding (EI: F ROH > 0.1) cases compared to 345,276 unrelated EI controls ( F ROH < 0.01). Observed means for EI cases and controls are reported in Table 3 . Phenotypic reduction was assessed for ten traits: auditory acuity (AA), fluid intelligence score (FIS), peak expiratory volume (PEF), hip-to-waist ratio (HWR), visual acuity (VA), height, cognitive ability measured as the mean time to correctly identify matches (MTCIM), handgrip strength (HGS), number of children (NCh) and educational attainment (EA) measured as the number of years of education. Traits were adjusted for age at recruitment, sex, recruitment centre, year of birth, genotyping batch, socioeconomic status measured by the Townsend deprivation index and population structure measured by 10 genetic principal components estimated from HM3 SNPs. Inbreeding depression was estimated within unrelated EI controls using two inbreeding measures: F UNI and F ROH . Resulting estimates were used to linearly predict the reduction in EI cases. Vertical bars around predictions corresponds to 99.5% confidence interval as the significance was defined here at p < 0.05/10 Full size image We next analysed the number of diseases diagnosed in an individual as an overall measure of health (Methods). We used overdispersed Poisson regression to estimate the relative risk (RR) of being diagnosed with at least one disease in EI cases as compared to EI controls. We found a RR of ~1.44 (Wald-test p = 3.6 × 10 −5 ; Table 3 ). To minimise potential biases due to partial or differential disease reporting between UKB participants, we re-estimated RR in individuals with at least one disease diagnosed. This analysis included only 110 of the 125 EI cases identified and similarly showed a reduced but still significant RR ~1.34 (Wald-test p = 4.4 × 10 −4 ; Table 3 ). In summary, we confirm that EI produces offspring with reduced stature (height), cognitive function (EA, FIS, and MTCIM), AA, muscular fitness (HGS), and lung function (PEF), consistent with a linear decline in these traits as inbreeding increases. We also provide additional evidence that offspring resulting from EI have increased risk for developing any type of disease. Social context of EI cases We tested the association between EI and the Townsend depression index, which quantifies the level of socioeconomic deprivation in areas where UKB participants live. We found significant evidence that EI is enriched in more socioeconomically deprived area (odds ratio: 1.22; CI 95% : [1.16–1.29]; Wald-test p = 2.6 × 10 −13 ), consistent with a previous study 13 , which reported association between F ROH and the same index in the UKB. We further investigated the social contexts in which EI arose. For that we compared different characteristics of the parents of EI cases with that of the parents of EI controls. We found that 14.5% (i.e., 18/124, 1 missing value) of EI cases vs. 1.5% of controls reported to be adopted as a child (Fisher exact test p = 7.3 × 10 −13 ). Given the significance of this difference we therefore focused all subsequent comparisons in nonadopted participants (106 EI cases vs. 339,241 EI controls) in order to minimise biases due to differential reporting of parental traits. Previous studies 40 have suggested that low EA of parents could be a cause of inbreeding in the population. Given that EA of parents of UKB participants has not been measured, we therefore tested this hypothesis by comparing mean genetic predictors of EA in UKB participants between EI cases and EI controls. Note that mean genetic predictor of EA is an estimate of the parental average for this trait. We found no statistical evidence that the mean genetic predictor of EA in EI cases deviate from that of EI controls ( t test p = 0.538; Table 3 ). In fact, the mean genetic predictor of EA in EI cases approximately equals the median of the EA genetic predictor distribution in EI controls, which highlights that EI cases are not outliers on this scale. Besides EA, we then used overdispersed Poisson regression to compare the number of diseases (Online method) reported in parents of EI cases vs. parents of controls, which we used as another proxy for socioeconomic status of parents. We found no significant evidence that parents of EI cases are enriched for comorbidities as compared to parents of EI controls (RR ~0.96; Wald-test p = 0.507; Table 3 ). However, this observation must be interpreted with caution as it may simply reflect that EI cases observed in the UKB may be from more healthier background as compared to EI in the general population. Although additional information on parents of UKB participants was available (i.e., age of parents or age when parent died), missing values rates were often too large (>50%) among EI cases to draw reliable inference. Finally, we investigated if EI cases were geographically clustered, but found no significant association between EI and birth location (North coordinate: Wald-test p = 0.15; East-coordinate: Wald-test p = 0.08). Note that the absence of geographical clustering that we report only applies to these extreme events and could also reflect lack of statistical power as we still observed variance in mean F ROH between different geographical areas of the UK. Altogether, although we observed that EI is more prevalent in more socially deprived areas of the UK, our results point to an absence of evidence that social and geographical stratification of parents contribute to the prevalence of EI in the population. Discussion In this study, we estimated a prevalence of EI of ~1/3652 in individuals of European ancestry born in the UK between 1938 and 1967. Importantly, our estimate of the UK prevalence of EI is likely downwardly biased partly because of the ascertainment of UKB participants, who are on average healthier and more educated than the rest of the UK population 28 . It also worth mentioning that our estimate only accounts for mating between close relatives that have led to viable offspring. Altogether, our findings suggest that the prevalence of EI in the population is small and that very large observational studies are required to quantify it accurately. We aimed in this study to quantify EI as it can routinely be detected in clinical screenings if genotypes are available. Therefore, we followed guidelines from the American College of Medical Genetics and Genomics, which recommend the use of both F ROH and a threshold at 0.1. Nevertheless, we acknowledge that ACMG guidelines may be suboptimal with respect to detection of EI and that other approaches could have been in implemented 41 , 42 . We found in our simulations that a threshold 0.1 may in fact be too conservative, while using a threshold of ~0.08 is optimal with respect to specificity and sensitivity to detect EI (Fig. 1 ). In addition, our study has addressed theoretical questions regarding the distribution of genomic segments homozygous-by-descent, which are classically approximated using long ROHs. Indeed, we explored how the distribution of long ROHs can be utilised to infer mating types underlying EI. Although we only applied threshold-based methods, we found that such simple approaches perform quite well in our simulations (AUC > 0.95). However, it is worth mentioning that previous studies have addressed a similar question using more elaborate models. For example, Druet and Gautier 42 introduced a model-based approach which assumes individual genomes to be a mosaic of HBD and non-HBD segments, and allows HBD segments to originate from different ancestors at different time points. The aim of their method is therefore to estimate simultanesouly the age and the HBD status of genomic segments. Note that knowing the age of an HBD segments directly informs the likelihood of certain mating types. One similarity between Druet and Gautier’s approach and ours, is that we both assumed the distribution of HBD segments to follow a mixture of exponential distributions. However, our approach relies on observed ROHs, which we have assumed to be HBD, whereas Druet and Gautier models HBD segments as unobserved states of a hidden Markov chain. Consequently, their inference is likely more robust to biases from ROHs calling, which often requires arbitrary choices to be made (e.g., minimum length of ROHs, minimum distance between ROHs and number of occasional heterozygotes allowed). On the other hand, the Druet and Gautier method relies on the assumption that the length of HBD segments follows an exponential distribution as a consequence of assuming a constant recombination rate. Our study provides a simulation-based (using observed genetic maps) and an empirical quantification of the length distribution of long genomic segments identical-by-descent, which we found to best fit a mixture of two exponential distributions. Therefore, our results confirm that the assumption of constant recombination rate is inappropriate for describing segments length distribution 33 , and we show that mixtures of exponential distributions provide a mathematically tractable framework to accommodate arbitrary recombination maps. We note that Druet and Gautier acknowledged that violation of the assumption of a constant recombination rate across the genome could limit the interpretation of their model parameters. We showed in this study that the reduction in measured values of multiple complex fitness-related traits resulting from EI is consistent with inbreeding depression estimated within EI controls, who still harbour ROHs in their genome 43 . If inbreeding depression in EI controls is well estimated then the latter finding would suggest that gene × gene or gene × environment interactions contribute little to inbreeding depression in the traits analysed and also that variants causal of inbreeding depression in these traits are well tagged (i.e., correlated) by common SNPs. However, because of ascertainment of UKB participants who are on average healthier and more educated than the general population 28 , estimates of inbreeding depression in UKB participants may also be underestimated. Moreover, Curik et al. 44 showed using computer simulations that the absence or presence of a nonlinear relationship between inbreeding and traits should be interpreted with caution in particular when inbreeding depression is estimated using an inbreeding measure which only partially reflects realised autozygosity, as is the case for F ROH . Lastly, we attempted to quantify the contribution of social contexts to the prevalence of EI. Despite the sparsity of parental information for EI cases, we found no evidence that EI is more prevalent in health-deprived families nor that low education contributes to increase the likelihood of EI in the population. In conclusion, our study provides an objective quantification of EI in the UK population and shed lights on its causes and phenotypic consequences. Methods SNP genotyping We used genotyped and imputed allele counts at 16,652,994 SNPs imputed to the Haplotype Reference Consortium 45 imputation reference panel, in 487,409 participants of the UKB 29 , 46 . Extensive description of data can be found here 26 . We restricted our analysis to 456,414 participants of European ancestry identified using projected principal components based on sequenced participants of the 1000 genomes projects with known ancestry 26 . This subset of the UKB contains 348,502 conventionally unrelated participants, i.e., whose estimated pairwise SNP-based GRM < 0.05, estimated using 1,124,803 common (minor allele frequency (MAF) ≥ 1%) HapMap3 47 SNPs using GCTA (v1.9) 48 . The North West Multi-Centre Research Ethics Committee (MREC) approved the study and all participants in the UKB study analysed here provided written informed consent. Polgenic predictor of EA We used estimated SNP effects from the Lee et al. 49 GWAS of EA to calculate polygenic score predicting EA. HM3 SNP effects were re-estimated after excluding data from the UKB. Marginal SNP effects were then transformed into conditional SNP effects using the LD-pred method 50 assuming all SNPs to be causal. The latter analysis used genotypes at HM3 imputed SNPs of ~300,000 unrelated UKB participants as linkage disequilibrium reference panel. ROH detection ROH were called using only 301,412 SNPs genotyped in 456,414 UKB participants of European descent. These SNPs were filtered on missingness rate (missingness < 1%), MAF > 5% and Hardy–Weinberg equilibrium test p value > 0.0001. As in previous studies 4 , 8 , 10 , we used the following PLINK (versions 1.07 and 1.9) 51 , 52 command to call ROH: --maf 0.05 --homozyg --homozyg-density 50 --homozyg-gap 1000 --homozyg-kb 1500 --homozyg-snp 50 --homozyg-window-het 1 --homozyg-window-missing 5 --homozyg-window-snp 50. That command detects ROHs at least 1.5 Mb long, at least 1 Mb apart from one another, containing at least 50 SNPs, and such that SNPs overlapping ROH can have at most 5 missing values and 1 occasional heterozygote. Once ROHs detected, we calculate an inbreeding measure F ROH for each individual by dividing the cumulated length of ROH in Mb by an estimate of the length of the human autosome, i.e., ~2881 Mb under genome build hg19. Note that this estimate of autosome length may vary between genome builds, and therefore may impact the number of individuals detected above a given threshold. Simulation of EI To simulate EI we used 972 independent (GRM < 0.05) trios (both parents and one offspring) out of 1066 identified in the UKB 29 . We used the same set of 301,412 genotyped and quality-controlled SNPs as to call ROH to phase haplotypes using SHAPEIT 2 with the following options: --duohmm -W 5 -T 10 and using genetic maps from the 1000 Genomes (1KG) Project phase 3 (hg19, see URL) 23 . We considered eight different mating types (pedigrees): mating between unrelated individuals (i.e., any pair among the unrelated 972 samples), between first-cousins, between double-first cousins, between grandchildren and grandparents, between uncles/aunts and nieces/nephews, between HSs, between fullsibs and between parents and offspring. Nonetheless, we describe here the case of PO mating. First, we sample a random pair of individuals (denoted P 1 and P 2 ) out of 972 × 971/2 = 471,906 possible pairs. We then create recombined chromosomes from haplotypes of P 1 and P 2 . For all genetic intervals defined in the 1KG genetic maps, we sample the Bernoulli distributed indicator of the presence of a recombination breakpoint with probability equal to 0.01 × genetic distance of the interval in Morgan(s). Once the recombined chromosomes of the offspring O of P 1 and P 2 are simulated, we then repeat this procedure to simulate an offspring resulting from mating of O with one of the parent, i.e., P 1 or P 2 . To then mimic real data, which contain genotyping errors, we also add a random number of errors to the simulated genotypes. The number of errors is sampled from a Poisson distribution with a mean corresponding to the mean number of genotyping errors estimated, for each chromosome, from comparing genotypes of 168 twin pairs (Supplementary Table 2 ). We found overall a genotyping error at quality controlled SNPs ~4.5 × 10 −4 , which is orders of magnitude larger than the rate of new somatic mutation, which was previously estimated around ~2.8 × 10 −7 in human fibroblasts 53 . Therefore, somatic mutation would have a negligible effect on ROH calling given the set of parameters that we used. Association with phenotypes measured in the UKB We used GCTA with the --ibc command to estimate for each UKB participants the correlation between uniting gametes 48 . That statistic denoted F UNI (also known as “Fhat3”) is an estimate of inbreeding using allele frequencies in the current population and was previously shown to be more powerful to detect ID 4 . We nonetheless condidered F ROH as a reference inbreeding measure in this study in accordance with the ACMG guidelines. We tested the association between inbreeding measures ( F ROH and F UNI ) and traits using linear regression adjusted for age at recruitment (UKB field 21022–0.0), sex, assessment centre (UKB field 54–0.0), genotyping chip and batch, year of birth (UKB field 34–0.0), socioeconomical status measured by the Townsend deprivation index (UKB field 189–0.0) and 10 genetic principal components calculated using PLINK 2.0. Analyses were performed in 345,276 unrelated EI controls ( F ROH < 0.01). Traits were pre-adjusted and inverse normal transformed and phenotypic values larger than >4 standard deviations were excluded. UKB identifiers for tested traits are: height (UKB field 50-0.0), hip-to-waist ratio (HWR: ratio of UKB field 49-0.0 over UKB field 48-0.0), HGS (average of UKB fields 46-0.0 and 47-0.0), lung function measured as the PEF (UKB field 3064-0.0), VA measured on log MAR scale (VA: average between UKB field 5201-0.0 and UKB field 5208-0.0), auditory acuity measured as te speech reception threshold (AA: average between UKB field 20,019-0.0 and UKB field 20,021-0.0), number of years of education (EA), fluid intelligence score (FIS: UKB field 20,016-0.0), cognitive function measured as the mean time to correctly identify matches (MTCIM: UKB field 20,023-0.0) and fertility measured as the number of children (NCh: for males UKB field 2405-0.0 and for females UKB field 2734-0.0). To test the association between number of diseases diagnosed and inbreeding, we used overdispersed Poisson regression implemented in R 3.2.0 ( glm function with option family = “quasipoisson”). Number of diseases diagnosed was estimated as the number International Classification of Diseases, Tenth Revision (ICD10) codes reported for UKB participants. We also analysed reported illnesses in fathers and mothers of UKB participants (UKB fields 20,107 and 20,110, respectively) as measure of health deprivation in the family. Illnesses of parents were reported among 12 groups of diseases (URLs). We created for each participant a count of diseases in both parents. Analysis were adjusted for adoption status (UKB field 1767) and missing values on the parental diseases were excluded. Length distribution of ROHs We estimated the length distribution of ROHs using a mixture of exponential distributions with a number of components from 1 to 10. Given that only ROHs larger than 1.5 Mb are detected, we therefore analysed lengths of ROHs in Mb minus the minimum threshold (as in Fig. 2 ). This choice is justified by following property of exponential distributions. If X follows an exponential distribution of rate λ , then Y = X |X > s , i.e., the truncated distribution of X with values larger than a given threshold s , is such that (Y- s ) also follows an exponential distribution with the same rate ( λ ) as X . Estimation of mixture distribution was performed using the R package Renext . Model selection was performed using BIC criterion. Discriminate PO vs. FS mating from ROH length distribution Given a collection of autosomal ROH segments lengths, we developed a method for estimating the proportion π PO / FS of these segments resulting from PO vs. FS mating. We denote f PO and f FS as the probability density functions of (ROHs) segments length under PO and FS, respectively. We assume that the length distribution of the set of ROHs used for inference is a mixture of f PO and f FS and we denote π PO / FS as the mixture proportion. We also assumed f PO and f FS to be known so that the parameter of interest, i.e., that we want to estimate, is π PO / FS . The log-likelihood l ( x ; π PO / FS ) of one segment of length x can be written as $${l\left( {x;\pi _{PO/FS}} \right) = {\mathrm{log}}\left[ {\pi _{PO/FS}f_{PO}(x) + \left( {1 - \pi _{PO/FS}} \right)f_{FS}(x)} \right] = {\mathrm{log}}\left[ {\pi _{PO/FS}\left( {f_{PO}(x) - f_{FS}(x)} \right) + f_{FS}(x)} \right]}.$$ (1) From that we can write the Fisher information as $${\Bbb E}\left[ { - \frac{{\partial ^2l\left( {X;\pi _{PO/FS}} \right)}}{{\partial \pi _{PO/FS^2}}}} \right] = {\Bbb E}\left[ {\left( {\frac{{f_{PO}(X) - f_{FS}(X)}}{{\pi _{PO/FS}\left( {f_{PO}(X) - f_{FS}(X)} \right) + f_{FS}(X)}}} \right)^2} \right],$$ (2) where X ’s probability log density function is l ( x ; π PO / FS ). Therefore the asymptotic variance of the maximum likelihood estimator \(\hat \pi _{PO/FS}\) of π PO / FS would be $${\mathrm{var}}[\hat \pi _{PO/FS}] \approx \frac{1}{N} \times \left\{ {{\Bbb E}\left[ {\left( {\frac{{f_{PO}(X) - f_{FS}(X)}}{{\pi _{PO/FS}\left( {f_{PO}(X) - f_{FS}(X)} \right) + f_{FS}(X)}}} \right)^2} \right]} \right\}^{ - 1}.$$ (3) where N is the number of segments used to estimate π PO / FS . We use parameters from Table 2 to characterise f PO and f FS . Each of these two distributions were approximated using mixtures of two exponential distributions which parameters were estimated from >648,125 simulated ROHs under PO and FS. Conditional on f PO and f FS , estimating π PO / FS is therefore a straightforward univariate optimisation problem. We used Eq. ( 3 ) to quantify the standard error of \(\hat \pi _{PO/FS}\) . The expectation in Eq. ( 3 ) was approximated using one million Monte Carlo simulations conditional on \(\hat \pi _{PO/FS}\) , f PO and f FS . URLs For Crime Survey for England and Wales, see (Table 8; Offence code 23, total number of cases is 11,196). For Population sizes in England and Wales, see . For Educational attainment in the UK from 2011 Census, see . For Genetic maps from the 1,000 Genomes Project, see ftp://ngs.sanger.ac.uk/production/samtools/genetic-map.tgz . For UK Biobank groups of diseases affecting parents, see . Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability This study makes use of genotype and phenotype data from the UK Biobank data under project 12505. UKB data can be accessed upon request once a research project has been submitted and approved by the UKB committee. We also provide data source for generating all figures. Code availability We provide the R script that we used to simulate genotypes of EI cases individuals from a collection of phased haplotypes in SHAPEIT2 format at . We also provide R scripts for generating all figures.

A team of researchers has found a way to gauge the rate of extreme inbreeding (EI) in the U.K. and its possible health repercussions. In their paper published in the journal Nature Communications, the group describes their study of data from the U.K. Biobank and what they found. The researchers began their study by noting that not a lot of research has been done surrounding EI, which they define as reproduction between people that are closely related, such as siblings or aunts or uncles. They note that EI is considered taboo in most societies, and is very often outlawed. This has led to limited information on the topic. To learn more about EI in the U.K., the researchers turned to the U.K. Biobank, which contains information from approximately 450,000 voluntary participants, all of whom have European ancestry. In sifting through the data in the Biobank, the researchers looked at genetic information, specifically, for large runs of homozygosity—an indicator of close family ties between parents. They report that they found 125 cases of individuals who they believed were the product of inbreeding—a rate of one in 3,652. That number differs significantly from police incest reports, which show a rate of one in 5,247. The researchers then looked at the health histories of those individuals and compared them with people in society at large. They report that they found that such individuals were at a slightly higher risk of a variety of health effects. They were on average slightly shorter, were less smart, and were less able to reproduce. They also were more likely to have lung function problems and were more likely to contract diseases than the average person. The researchers acknowledge that their dataset might have been somewhat limited—people who volunteer to be tested and have their data added to the Biobank tend to be wealthy, healthy and more highly educated. That could have skewed the results. If so, the researchers suggest it likely skews low, because those with more health problems due to inbreeding would not volunteer to as participants. They conclude by claiming that their study backs up claims of the ill health effects of inbreeding.

10.1038/s41467-019-11724-6

Biology

Quick test for potential probiotic in seawater may reveal health of corals

Ruriko Kitamura et al, Specific Detection of Coral-Associated Ruegeria, a Potential Probiotic Bacterium, in Corals and Subtropical Seawater, Marine Biotechnology (2021). DOI: 10.1007/s10126-021-10047-2

http://dx.doi.org/10.1007/s10126-021-10047-2

https://phys.org/news/2021-10-quick-potential-probiotic-seawater-reveal.html

Abstract Coral microbial flora has been attracting attention because of their potential to protect corals from environmental stresses or pathogens. Although coral-associated bacteria are considered to be acquired from seawater, little is known about the relationships between microbial composition in corals and its surrounding seawater. Here, we tested several methods to identify coral-associated bacteria in coral and its surrounding seawater to detect specific types of Ruegeria species, some of which exhibit growth inhibition activities against the coral pathogen Vibrio coralliilyticus . We first isolated coral-associated bacteria from the reef-building coral Galaxea fascicularis collected at Sesoko Island, Okinawa, Japan, via random colony picking, which showed the existence of varieties of bacteria including Ruegeria species. Using newly constructed primers for colony PCR, several Ruegeria species were successfully isolated from G. fascicularis and seawater. We further investigated the seawater microbiome in association with the distance from coral reefs. By seasonal sampling, it was suggested that the seawater microbiome is more affected by seasonality than the distance from coral reefs. These methods and results may contribute to investigating and understanding the relationships between the presence of corals and microbial diversity in seawater, in addition to the efficient isolation of specific bacterial species from coral or its surrounding seawater. Access provided by MPDL Services gGmbH c/o Max Planck Digital Library Working on a manuscript? Avoid the common mistakes Introduction Coral reefs are considered to play an essential role in maintaining marine biodiversity, including fish (Descombes 2015 ; Jones et al. 2004 ) and bacteria involved in the biogeochemical cycle of sulfur (Raina et al. 2009 ). Since corals are endangered (Carpenter 2008 ; Dietzel et al. 2021 ), extensive research has been conducted to understand the mechanisms of maintaining coral health. In recent years, coral microbiota have attracted attention from the viewpoint of relevance to coral health (Bourne et al. 2016 ; Peixoto et al. 2017 ; Peixoto 2021 ; Reshef et al. 2006 ; Rosenberg et al. 2007 ). It is well known that the microbiota in diseased coral changes compared to the healthy state (Meyer et al. 2019 ; Rosales et al. 2019 ; Sweet et al. 2019 ). Coral microbiota are also affected by bleaching (Pootakham 2018 ) or environmental stresses (Maher et al. 2019 ; Vega Thurber 2009 ), including heat stress (Pootakham 2019 ) and the combined stress of heat and acidification (Grottoli 2018 ). Conversely, the composition of microbial flora is considered to be related to coral resilience to environmental stresses (Grottoli et al. 2018 ; Ziegler et al. 2017 ). In fact, microbial mixtures extracted from corals have been shown to protect the reef-building coral Pocillopora damicornis from bleaching induced by heat stress as well as exposure to a coral pathogen, Vibrio coralliilyticus (Rosado 2019 ), suggesting that the composition of coral microbiota can assist in maintaining the healthy status of corals. Hence, monitoring coral microbiota is considered to be effective to predict risks of environmental or pathological coral bleaching. Coral-associated bacteria are considered to be acquired from seawater (Bernasconi et al. 2019 ). Previously, Glasl and colleagues suggested that the composition of the microbial flora in seawater around coral could be used as an environmental indicator (Glasl et al. 2019 ). However, whether the distribution of specific types of bacteria in seawater is related to the existence or status of coral reefs is not known. Recently, it has been suggested that specific types of bacteria can contribute to protecting coral health. For example, carotenoid-producing Muricauda sp. isolated from the reef-building coral Galaxea fascicularis has been shown to recover photosynthetic activities of the coral symbiont zooxanthellae under heat and higher-light stress (Motone et al. 2020 ), suggesting a role of Flavobacteriia in protecting corals from environmental stress. It has also been suggested that several bacterial species inhibit the growth of pathogens. Various bacterial species isolated from corals have been shown to inhibit the in vitro growth of V . coralliilyticus (Miura et al. 2019 ; Raina 2016 ; Tang et al. 2020 ), which is considered to be one of the coral pathogens (Ben-Haim et al. 2003 ). We have previously isolated several G . fascicularis -associated Ruegeria spp. that inhibit the in vitro growth of V . coralliilyticus P1 strain (Miura et al. 2019 ). Ruegeria spp. are known to play essential roles in marine ecosystems by supplying vitamin B 12 to phytoplankton and contributing to marine carbon and sulfur cycles (Durham 2015 ). Hence, Ruegeria spp. are not only potentially probiotic toward corals but also considered to be important bacterial lineages to protect marine ecosystems. Although Ruegeria spp. are known to be widely distributed in seawater (Sonnenschein 2017 ), their distribution and functions associated with corals are not well understood. In this study, we evaluated several methods, including classical techniques for isolation of coral-associated bacteria, direct colony polymerase chain reaction (colony PCR), real-time PCR using previously constructed Ruegeria -specific primer sets (Kitamura et al. 2020 ), in addition to the 16S metabarcoding analysis of seawater, to survey the distribution of Ruegeria spp. inhabiting the interior and exterior of corals. We investigated the distribution of coral-associated Ruegeria spp. near Sesoko Island, Okinawa, Japan, which harbor a variety of stony and soft corals (Loya et al. 2001 ), including G . fascicularis and Acropora tenuis (Yuyama et al. 2012 ). Although G . fascicularis is a relatively minor player among reef-building corals found around Sesoko Island, it is easily detectable and identifiable in the field. Here, we isolated several Ruegeria spp. from G . fascicularis , and tested methods to detect Ruegeria spp. in seawater from around Sesoko Island. Furthermore, we investigated microbial flora in seawater related to seasonality and distance from coral reefs. Materials and Methods Strains and Media Escherichia coli BW25113 ( rrnB DElacZ4787 hdR514 DE ( araBAD ) 567 DE ( rhaBAD ) 568 rph-1 ) was cultured in LB medium (Nacalai Tesque Inc., Kyoto, Japan). Vibrio coralliilyticus P1 was purchased from the Belgian Coordinated Collections of Microorganisms ( ). V . coralliilyticus P1 and other bacteria isolated in this study were cultured using either Difco™ Marine Agar 2216 (MA medium; BD Biosciences, Franklin Lakes, NJ, USA) or Difco™ Marine Broth 2216 (MB medium; BD Biosciences). Isolation of Coral-Associated Bacteria from G. fascicularis Three small coral colonies of G . fascicularis were collected from around Sesoko Island, Okinawa, on January 30, 2019. Corals were handled with gloved hands, avoiding touching the mouth of the coral, for all of the following procedures. The colonies were rinsed in filter-sterilized 40 g/L sea salt (Sigma-Aldrich, St. Louis, MO, USA) solution in dH 2 O and carefully separated into polyps using nippers. Three polyps from each colony (i.e., a total of nine polyps) were used for the direct culture of bacteria. Another three polyps from each colony were used to prepare glycerol stocks. Direct cell culture of bacteria was performed following previously reported procedures (Miura et al. 2019 ) with minor modifications. Polyps were inserted into 1.5-mL tubes, and sea salt solution remaining on the surface of polyps was removed by centrifugation at 100 × g for 10 s. Polyps were transferred to new 1.5-mL tubes using tweezers and centrifuged at 8,000 × g for 2 min to remove coral skeleton from the coral body. After removing the coral skeleton using tweezers, coral bodies were mixed with 200–500 µL of sea salt solution using a 100–1,000-µL pipette tip. Fifty microliters of the mixture was plated on MA plates and statically cultured at 25°C for 2–4 days. After incubation, growing bacterial colonies were randomly picked and 16S rRNA gene sequencing was conducted for species identification. Bacteria annotated as Ruegeria spp. were further used for the evaluation of Ruegeria -specific PCR amplification primers. For the preparation of glycerol stocks, coral body exfoliated from the skeleton was mixed with 325 µL of MB medium and 175 µL of sterilized 50% glycerol and then stored at −80°C until use. Glycerol stocks were thawed at room temperature. After mixing with a sterilized pipette tip, 100 µL of glycerol stock solution was plated onto MA plates and statically cultured at 25°C for 2–4 days. Developed colonies were randomly picked and recultured on fresh MA plates at 25°C for 2–3 days. A portion of each randomly picked colony was used for PCR to identify Ruegeria spp. using Ruegeria -specific primers. Selection and Evaluation of Primer Sets for Colony PCR A primer set for easy detection of Ruegeria spp. colonies was selected from a previously reported pool of 55 PCR primer candidates (Kitamura et al. 2020 ). A primer set that amplifies a ~400-bp fragment, with each primer 18–24 bp in length, Tm values of 55–65°C, and GC content of 40%–60%, was selected using FastPCR software v.6.6 (Primer Digital Ltd., Helsinki, Finland). The obtained primers, namely, r102F (5′-ATCTTTCACTTCGGTGACTCGGTG-3′) and r109R (5′-AGCGTCGTCGGGTAGAACCA-3′), were used for colony PCR. Colony PCR and Gel Electrophoresis Colony PCR of picked colonies mentioned in the previous section was performed using ExTaq DNA polymerase (Takara Bio Inc., Otsu, Japan) on an S1000 thermal cycler (Bio-Rad, Hercules, CA, USA) according to the manufacturer’s instructions. A total of 10 µL (for each colony) of PCR mixture was prepared with 1 µL 10 × ExTaq buffer, 0.8 µL dNTP mix, 1 µL of 3.2 µM of each primer, 6.1 µL autoclaved Milli-Q water, and 0.1 µL ExTaq DNA polymerase. PCR was conducted using an initial denaturation temperature of 94°C for 5 min, followed by 30 cycles of amplification (denaturation at 94°C for 30 s, annealing at 55°C for 30 s, and elongation at 72°C for 25 s), with a final extension step at 72°C for 10 min, followed by cooling to 4°C. The size of the PCR products was evaluated via gel electrophoresis using 2.5% AgaroseXP (Nippon Gene Co., Ltd., Tokyo, Japan) with a 20-bp DNA Ladder (Dye Plus, Takara). PCR-positive colonies were further cultured, and genomic DNA was extracted for species identification as described below. Simple Genome Extraction Colonies were inoculated into 1.5-mL tubes containing 400 µL MB medium and then cultured using a shaker incubator (RLS-25R-3; Sanki Seiki Co., Ltd., Osaka, Japan) at 25°C, with 300 rpm for 2–4 days. After culturing, the tubes were centrifuged at 20,400 × g for 5 min and the supernatants were discarded. One-hundred microliters of autoclaved Milli-Q water was mixed with cell pellets, and the mixture was incubated at 95°C for 5 min and then set on ice for > 1 min. The tubes were again incubated at 95°C for 5 min and set on ice for > 1 min. The tubes were then centrifuged at 13,000 × g for 1 min, and the supernatants were used as genomic isolates. The concentration of the DNA preparation was measured using a Biospec-nano microvolume spectrophotometer (Shimadzu, Kyoto, Japan). The DNA solution was stored at 4°C until use. PCR Amplification and Analysis of 16S rDNA Sequence For species identification, 16S rDNA was amplified and analyzed as previously described (Miura et al. 2019 ). DNA amplification was performed using primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) (Lane 1991 ) and 1492R (5′-TACCTTGTTACGACTT-3′) (Frank et al. 2008 ) using KOD FX Neo polymerase (TOYOBO Co., Ltd., Osaka, Japan) on an S1000 Thermal cycling platform, according to manufacturer’s instructions. A total of 50 µL of reaction solution per sample was prepared with 25 µL 2 × PCR buffer, 10 µL of 2 mM dNTPs, 1.5 µL of 10 µM of each primer, 200 ng of extracted genomic DNA, 1 µL KOD FX Neo polymerase, and autoclaved Milli-Q water. PCR was conducted using an initial denaturation temperature of 94°C for 5 min, followed by 30 cycles of amplification (denaturation at 98°C for 10 s, annealing at 51°C for 30 s, and elongation at 68°C for 1 min), and then cooled to 4°C. After amplification, 10 µL of each reaction solution was reserved for gel electrophoresis using 2% agarose (Nacalai) with a 1-kb DNA marker ladder (BRG-1 kb-0205; WATSON Bio Lab, Kobe, Japan). The remaining 40 µL of the reaction solution was purified using Monarch PCR and DNA Cleanup Kit (New England Biolabs, Ipswich, MA, USA). The purified PCR products were sequenced by Eurofins Genomics Inc. (Tokyo, Japan) using 27F and 1492R primers. The data were analyzed using SnapGene Viewer software v.4.1.8 (GSL Biotech LLC, Chicago, IL, USA) and annotated using the Nucleotide Blast tool against the GenBank nucleotide (nr/nt) database ( ). Seawater Collection Seawater was collected at three locations: location 1: coral reef area (26.6271°N, 127.8605°E), location 2: near the coral reef area (26.6225°N, 127.8614°E), and location 3: open sea (26.6368°N, 127.8481°E) area. Sampling was performed three times on January 30, June 27, and November 12, 2019, using plastic bags. Seawater was collected from 1 to 3 m below the water surface. Isolation of Bacteria from Seawater One-hundred microliters of the sampled seawater or seawater diluted 10- to 1,000-fold with Milli-Q water was directly plated on MA medium and cultured at 25°C for 6 days. After culturing, colonies were randomly selected from plates and recultured on MA medium. Preparation of Environmental DNA from Seawater Environmental DNA was extracted as previously reported (Kitamura et al. 2020 ). Seawater (1 L) was filtered through Sterivex-GP 0.22-µm filters (Millipore, Burlington, MA, USA) using tubing pumps (62–90 mL/min #1–7580-12 or 30–280 mL/min #1–9028-11; AS ONE, Osaka, Japan). Sterivex was filled with RNAlater™ Stabilization Solution (Thermo Fisher Scientific–Invitrogen, LaFayette, CO, USA) and stored at − 80°C. For DNA extraction, RNAlater™ Stabilization Solution was thawed and removed from Sterivex by centrifuging at 4°C at 5,000 × g for 1 min. Sterivex was then filled with a mixture of 220 µL PBS, 200 µL buffer AL (DNeasy Blood and Tissue Kit, QIAGEN, Hilden, Germany), 180 µL buffer ATL (DNeasy Blood and Tissue Kit), and 20 µL proteinase K (DNeasy Blood and Tissue Kit); sealed; and incubated at 56°C for 3 h in a shaker incubator (RLS-25R-3, Sanki Seiki Co., Ltd.). The inner solution was collected by centrifuging Sterivex in a 50 mL tube at 5,000 × g for 1 min, transferred to a 1.5-mL tube, and mixed with 200 µL ethanol. The mixture was purified using a column (DNeasy Blood and Tissue Kit) and washed according to the manufacturer’s instructions. DNA was eluted with 100 µL RNAlater™ Stabilization Solution and stored at 4°C. DNA concentration was measured using Qubit™ 4 Fluorometer (Thermo Fisher Scientific). qPCR Analysis A qPCR analysis of environmental DNA was conducted as previously reported (Kitamura et al. 2020 ), using TB Green™ Premix Ex Taq™ II (Tli RNaseH Plus; Takara) and Thermal Cycler Dice TP800 (Takara) with either a 16S universal primer set: U357′F (5′-GAGGCAGCAGTGGGGAAT-3′) and U515′F (5′-TGGCACGGAGTTAGCCGG-3′) or a Ruegeria -specific primer set: r38F (5′-GACCGGTCCAGAGATGGATCTT-3′) and r30R (5′-GCTGGCAACTAAGGATGTGG-3′) or r445F (5′-TTTCACTTCGGTGACTCGGT-3′) and r446R (5′-ACTAAGGATGTGGGTTGCGC-3′) (Kitamura et al. 2020 ). A total of 25 µL of reaction solution per sample was prepared with 8.5 µL RNase-free water, 1 µL each of 10 µM primers, 2 µL of extracted genomic DNA solution, and 12.5 µL TB Green™ Premix Ex Taq™ II. PCR was conducted using an initial denaturation temperature of 95°C for 30 s, followed by 40 cycles of amplification (denaturation at 94°C for 5 s and then annealing and elongation at 60°C for 30 s), and melting curve analysis was conducted. All experiments were repeated three times, and sample differences were tested statistically using Student’s t -test. When p < 0.05, the difference was determined as statistically significant. After analysis, PCR products were purified using Monarch PCR and DNA Cleanup Kits and the purified PCR products were sequenced by Eurofins Genomics Inc. using r38F, r30R, r445F, or r446R primers. The data were analyzed using SnapGene Viewer software and annotated using the Nucleotide Blast tool against the GenBank nucleotide collection (nr/nt) database. Amplicon Sequencing of 16S rRNA Gene PCR Products Paired-end (2 × 300 bp) DNA sequencing of the 16S rRNA V3/V4 region was conducted by Bioengineering Lab. Co., Ltd. (Sagamihara, Japan), using the Illumina MiSeq platform (San Diego, CA, USA). A library was prepared by two-step tailed PCR using primers 1st-341f_MIX (5′-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-NNNNN-CCTACGGGNGGCWGCAG-3′) and 1st-805r_MIX (5′-GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT-NNNNN-GACTACHVGGGTATCTAATCC-3′), then 2ndF (5′-AATGATACGGCGACCACCGACATCTACAC-Index2-ACACTCTTTCCCTACACGACGC-3′) and 2ndR (5′-CAAGCAGAAGACGGCATACGAGAT-Index1-GTGACTGGAGTTCAGACGTGTG-3′). Amplified DNA was purified using AMPure XP kits (Beckman Coulter, Inc., Brea, CA, USA) after each amplification step. Data Analysis for Abundance and Composition of Ruegeria spp. Raw FASTQ files were processed using VSEARCH v.2.4.3 (Rognes et al. 2016 ) and Cutadapt v.1.8 (Martin 2011 ) on Ubuntu 16.04 LTS. First, the reads 1 and 2 of the MiSeq output were merged using the vsearch –fastq_mergepairs command. The merged reads were subjected to trimming of sequence regions corresponding to the forward and reverse primer sites using Cutadapt. Next, a quality filtering step was conducted by discarding reads with more than one expected sequencing error using the vsearch –fastx_filter command, and the filtered reads were subjected to a chimera check step with the use of the vsearch –uchime_denovo command. To extract reads derived from Ruegeria spp. and other potential Ruegeria bacteria closely related to known Ruegeria isolates, DNA sequences with 97% identity or higher with any type strain of Ruegeria were selected using vsearch –usearch_global command. Selected reads were assigned to phylotypes using 16S RNA gene sequences of all 18 type strains of Ruegeria (accession nos. JQ807219 for R . arenilitoris , D88526 for R . atlantica , D88527 for R . meteori , HQ171439 for R . conchae , MH023307 for R . denitrificans , GU057915 for R . faecimaris , HQ852038 for R . halocynthiae , FR832879 for R . intermedia , U77644 for R . lacuscaerulensis , HE860713 for R . litorea , FJ872535 for R . marina , KP726356 for R . marisrubri , HE860710 for R . mediterranea , AF098491 for R . pomeroyi , KP726355 for R . profundi , D88523 for R . gelatinovorans , NR_118639 for R . meonggei , NR_170429 for R . sediminis ). Data Analysis for Total Microbial Community Raw FASTQ files were analyzed using QIIME2 (v.2020.8) on Ubuntu 20.04.1 LTS. For forward and reverse reads, 10–300 or 10–250 bp, respectively, were selected and used for denoising and processed using Divisive Amplicon Denoising Algorithm 2 (Callahan et al. 2016 ) on QIIME2 and representative sequences were determined. A database file was prepared from Greengenes OTUs (16S) 13_8 (most recent), which was downloaded from the QIIME homepage ( ) and used for annotation. Further analysis was performed using either QIIME2 or QIIME2R employing R studio v.1.3.1093 (R v.4.0.4) with the packages devtools, ggplot2, rgl, Rtools, qiime2R, phyloseq (McMurdie and Holmes 2013 ), and MicrobeR. Alpha diversity analysis was conducted using the Shannon index. The Kruskal–Wallis test was conducted to determine whether microbiome alpha diversity from different months or locations was significantly different from others. Beta diversity analysis was conducted using unweighted UniFrac with pairwise permutational multivariate analysis of variance (PERMANOVA) to evaluate diversity between samples. Phylogenetic Analysis For sequences of which 16S rRNA partial sequences were annotated, phylogenic analysis was conducted using MEGA v.10.0.5 software (Kumar et al. 2018 ), with the neighbor-joining algorithm. For NGS data, the R packages qiime2R and phyloseq were used for phylogenic analysis. Availability of Bacteria Isolated in This Study Ruegeria sp. strain okinawa_3_1_9, which was isolated from G . fascicularis , was deposited with the NBRC (National Institute of Technology and Evaluation) with the accession number NBRC 115114. Data Availability The raw sequencing data obtained via NGS analysis are available in the DDBJ Sequence Read Archive under the accession numbers DRX267149–DRX267157. Results Detection of Ruegeria spp. in Colonies Obtained from Seawater Around Sesoko Island Using a Novel Primer Set for Colony PCR By random colony picking, varieties of bacteria that were annotated as Ruegeria spp., Vibrio spp., and other bacteria such as Muricauda spp. were isolated from G . fascicularis (Table 1 ). To test detectability of the newly designed Ruegeria -specific primer set, r102F-r109R, genomes of the six Ruegeria spp., besides E . coli and V . coralliilyticus as negative controls, were used (Fig. 1 A). The primer set r102F-r109R successfully amplified the targeted region from six Ruegeria spp. genomes that contain sequences partially identical with 16S rRNA of R . arenilitoris strain G-M8 (GenBank accession no. NR_109635) (Okinawa_3_2_23, Okinawa 3_2_25, and Okinawa_3_1_9) and R . conchae strain TW15 (GenBank accession no. NR_109062) (Okinawa_1_3_7 and Okinawa_1_4_4) (Fig. 1 B). The primer set r102F-r109R failed to amplify genomes of Ruegeria sp. strain Okinawa_3_2_26, which contain partially identical with 16S rRNA sequence of R . mediterranea strain M17 (GenBank accession no. NR_125557), besides E . coli and V . coralliilyticus (Fig. 1 A, B), suggesting that the primer set can detect certain clades of Ruegeria spp. Table S1 and Figure S1 illustrate the sequence identity of primers r102F and r109R to 16S rRNA sequences of Ruegeria spp. Sequence identity of primers r102F and r109R with 16S rRNA sequence of Okinawa_3_2_26 was ~ 67% and 70%, respectively. Conversely, the sequence identity of primers r102F and r109R to 16S rRNA sequence of Ruegeria spp. was > 83%, suggesting the sequence identity-dependent detectability of 16S rRNA by primer set r102F-r109R. Table 1 Summary of bacteria isolated from G . fascicularis by random colony picking Full size table Fig. 1 Detection of Ruegeria spp. isolated from G . fascicularis . A Validation of primers for colony PCR using Ruegeria spp. isolated from G . fascicularis . “Asterisk” indicates amplification of target (399 bp). B Phylogenetic tree analysis of coral-associated bacteria in association with strains detected using the primer set r102F-r109R. The evolutionary history was inferred using the neighbor-joining method (Saitou and Nei 1987 ). An optimal tree with a sum of branch length of 0.80901725 is shown. The percentages of replicate trees in which the associated taxa clustered together in the bootstrap test (1,000 replicates) are shown next to the branches (Felsenstein 1985 ). The tree is drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the maximum composite likelihood method (Tamura et al. 2004 ) and are in the units of the number of base substitutions per site. All ambiguous positions were removed for each sequence pair (pairwise deletion option). There were a total of 1,567 positions in the final dataset. Evolutionary analyses were conducted using MEGA v.10.0.5 (Kumar et al. 2018 ) Full size image To determine whether primer set r102F-r109R could be used for easy detection of Ruegeria spp., colonies of bacteria derived from G . fascicularis were randomly picked and used for colony PCR. As the results, 14 of 303 colonies yielded amplification of target bands. Due to 16S rRNA sequencing of each colony, nine and five isolates were identified to have identical partial 16S rRNA sequence with R . arenilitoris strain G-M8 and R . atlantica strain NBRC 15792, respectively, demonstrating the validity of primer set r102F-r109R for specific detection of coral-associated Ruegeria spp. via colony PCR. Detection of Ruegeria spp. Colonies in Seawater Around Sesoko Island Next, bacterial colonies obtained from seawater were used for colony PCR to determine whether the distance from coral reefs affected the detection of Ruegeria spp. in seawater. For bacterial colonies obtained from seawater collected in locations 1, 2, and 3 (Fig. 2 A, B), positive rates of colony PCR were 0%, 9.4%, and 6.3%, respectively (Fig. 2 C), showing a lower positive rate for locations near coral reef areas (p < 0.05 by Fisher’s exact test for the number of colony PCR-positive samples and the total number of colonies tested, as shown in Fig. 2 C). Fig. 2 Detection of Ruegeria spp. from seawater. A Map of seawater sampling locations. Location 1: coral reef area (26.6271°N, 127.8605°E), location 2: near the coral reef area (26.622528°N, 127.861417°E), and location 3: open sea (26.63675°N, 127.848083°E). A shapefile for Okinawa Prefecture was downloaded from National Land Numerical Information, Japan ( ). A map was prepared using maptools in R v.4.0.4. B Representative figures of colonies formed on a culture plate. Location 1: colonies obtained by plating 1 × seawater collected at location 1 in June 2019. Location 2: colonies obtained by plating 1 × seawater collected at location 2 in June 2019. Location 3: colonies obtained by plating 1 × seawater collected at location 3 in June 2019. Single colonies were randomly picked from plates. C The number of positive colonies obtained via colony PCR of bacteria collected from seawater Full size image Seasonal and Location-Dependent Composition of Ruegeria in Seawater Of the three samples collected in January, June, and November, the amount of environmental DNA extracted from 1 L of seawater was highest in June at all locations (Table 2 ). Real-time PCR analysis using 16S universal primers U357′F and U515′R revealed no significant differences in bacteria detected between three locations in January or June (Fig. 3 A). In November, samples collected from location 1 exhibited significantly lower Ct values than did samples collected in other locations, suggesting a higher abundance of detectable bacterial genomes in location 1 in November (Fig. 3 A). When Ruegeria- specific primer sets r38F-r30R (Fig. 3 B) and r445F-r446R (Fig. 3 C) were used, samples collected at location 1 showed significantly lower Ct values than did samples collected at location 3 in January, showing a higher abundance of Ruegeria spp. in location 1 in January. In June, the results using primer sets r38F-r30R (Fig. 3 B) and r445F-r446R (Fig. 3 C) were slightly different. When primer set r38F-r30R was used, location 1 showed lower Ct values than did location 2 (Fig. 3 B). When primer set r445F-r446R was used, location 1 exhibited lower Ct values than did location 2, and location 3 presented lower Ct values than did location 1 (Fig. 3 C). Sequences amplified by primer sets r38F-r30R and r445F-r446R showed a partially mixed sequence waveform (Fig. S2 ). The former sequence was annotated with 100% identity as 16S rRNA sequences of Ruegeria sp. strain A9 (GenBank accession number MK100386), a close relative of an R . profundi -type strain ZGT108 (98.6% identity between each 16S rRNA gene). The latter was identical with all 18 type strains of Ruegeria . These results suggested that the existence of different compositions of Ruegeria spp. at locations 1, 2, and 3. Table 2 Amount of environmental DNA extracted from seawater Full size table Fig. 3 Real-time PCR analysis of environmental DNA extracted from seawater using 16S universal and Ruegeria -specific primers. A Specific detection of bacterial DNA by 16S universal primers U357′F and U515′R. B Specific detection of Ruegeria spp. DNA by Ruegeria -specific primers r38F and r30R. C Specific detection of the Ruegeria spp. DNA by Ruegeria -specific primers r445F and r446R. The sample name indicates the sampling location and collection month. n = 3, data represents means ± SD. “Asterisk” indicates a statistically significant difference ( p < 0.05 as determined using Student’s t -test) Full size image To confirm the distribution and abundance of bacterial taxa, including the genus Ruegeria in seawater samples, DNA sequencing of 16S rRNA gene V3-V4 regions was conducted using the Illumina MiSeq platform. Sequences with 97% identity or higher with 16S rRNA genes of any Ruegeria type strain were found in all seawater samples (Fig. 4 A). The relative abundance of the sequences among samples was consistent and inversely correlated with the Ct values, as shown in Fig. 3 , indicating the validity of our Ruegeria detection method on the basis of real-time PCR. Fig. 4 Abundance and distribution of Ruegeria spp. and closely related bacteria dependent on season and location. A Abundance of bacteria belonging to the genus Ruegeria and closely related bacteria. B Heatmap illustrating the composition of bacterial phylotypes of the genus Ruegeria and closely related bacteria. The representative sequences of the phylotypes were those of the 16S rRNA gene of the type strains of Ruegeria Full size image Each sequence of Ruegeria -related bacteria was assigned to each phylotype whose representative sequence was a 16S rRNA gene of a Ruegeria type strain (Fig. 4 B). The composition of the Ruegeria phylotypes was actually different mainly among seasons and less among locations. For example, an outbreak of the phylotype of R . marisrubri was found in samples collected in June. This contributed to a higher abundance of Ruegeria -related bacteria in that month (Fig. 4 A). There were also relatively stable phylotypes across seasons, such as the ones of R . profundi or R . sediminis . Overview of Microbiota Composition in Environmental DNA Determined via NGS Analysis Table S2 shows the sequence counts for each sample obtained via NGS analysis. Analysis was performed with 604 representative sequences (Table S3 ) obtained. Figure 5 shows the bacterial composition of representative sequences for each sample. In all samples, Proteobacteria and Cyanobacteria were the most abundant phyla (Fig. 5 A). More specifically, Alphaproteobacteria, Cynechoccophycideae, Gammaproteobacteria, and Flavobacteriia were the most abundant classes of bacteria, besides Chloroplast (Fig. 5 B). The alpha diversity of microbiomes collected in different months (Fig. 5 C) or locations (Fig. 5 D) was not significantly different from each other (Table S4 ), indicating that microbial diversities within each sample were comparable. Fig. 5 Bacterial composition of environmental DNA identified via NGS analysis. A Composition of bacteria divided by biological phylum. B Composition of bacteria divided by biological class. The sample name indicates the sampling location and collection month. Boxplot of alpha diversity index (Shannon) on month ( C ) and sampling location ( D ) (see also Table S4 ). Quartiles were calculated using a comprehensive median Full size image Seasonal and Location-Dependent Detection of Specific 16S rRNA Sequences To test whether the composition of overall bacterial 16S rRNA sequences was dependent on seasonality or location of sample collection, principal coordinate analysis (PCoA) was calculated on an unweighted UniFrac distance matrix. Samples collected in the same month formed clusters, suggesting that the composition of 16S rRNA sequences was dependent on the season rather than on the sampling location (Fig. 6 A). The beta diversity of samples collected in June and November compared with those in January (Fig. 6 B), samples collected in January and November compared with those in June (Fig. 6 C), and samples collected in January and June compared with those in November (Fig. 6 D) were not significantly different (Table S5 ), showing comparable variation in microbial communities between samples collected in different months. Fig. 6 Seasonal and location-dependent distribution of bacteria in seawater. A Principal coordinate analysis (PCoA) of NGS data dependent on sampling location. PCoA was calculated on Unifrac distances using phyloseq in R v.4.0.4. Dotted circles show samples collected in the same month. B – D Beta diversity analysis using unweighted UniFrac measure among different months. P values were estimated using PERMANOVA (see also Table S5 ). PERMANOVA was calculated with the 999 Monte Carlo permutation and Benjamini–Hochberg correction. Boxplots show distance to January ( B ), June ( C ), and November ( D ). Quartiles were calculated using the comprehensive median Full size image To overview whether there were major specific bacterial lineages detected in specific months, phylogenic tree analysis was conducted for the top 100 abundant OTUs (Fig. 7 A). When focused on the phylum Proteobacteria (Fig. 7 B), bacteria in the class Alteromonadaceae were found to be specific to samples collected in June. Fig. 7 Phylogenetic tree analysis. A Phylogenetic tree of the 100 most abundant OTUs. B Phylogenetic tree of the phylum Proteobacteria extracted from A . Phylogenetic analysis was conducted using QIIME2 pipeline phylogeny raxml-rapid-bootstrap option with 1000 bootstrap replicates. The phylogenetic tree was prepared using phyloseq and qiime2R in R v.4.0.4 Full size image Discussion This study proposes the use of newly prepared Ruegeria -specific primers r102F and r109R for the convenient detection of Ruegeria spp. in coral or seawater via colony PCR. Bacteria annotated to certain Ruegeria spp., including R . arenilitoris and R . chonchae , that were previously isolated from G . fascicularis (Miura et al. 2019 ) were successfully detected either from coral or seawater using the primers for PCR amplification. It was also suggested that R . mediterranea was not suitable for detection with the primer set used because of the low sequence identity between the primers and target sequence (Fig. 1 ). These results indicated that the specific primer set for colony PCR enables easy and rapid isolation of specific bacterial species, facilitating isolation of bacteria of interest. Focusing on the culturable bacteria identified in this study, the abundance of Ruegeria spp. in coral was higher than in seawater, suggesting that some Ruegeria spp. may be concentrated in coral. Since Ruegeria spp. are considered to contribute to carbon and sulfur cycling (Durham et al. 2015 ) besides their potential role as probiotics (Miura et al. 2019 ), the role of Ruegeria spp. in the coral holobiont would be important in future research. Further studies are needed to determine differences in the abundance of Ruegeria spp. in corals and seawater. Whether the abundance of Ruegeria spp. in seawater is related to the presence of corals could not be determined via environmental DNA analysis, because it is likely to be affected by the amount and variability of all bacterial species in seawater that appear to change significantly over time (Fig. 3 ). However, the lower proportion of colony PCR-positive colonies obtained from coral reef areas than that from the open sea (Fig. 2 C) may indicate that the abundance of Ruegeria spp. is not related to or proportional to the presence of coral. However, this study demonstrated the effectiveness of real-time and colony PCR for investigating sparsely populated microorganisms. By determining the exact DNA copy number of target bacteria in seawater or coral, it would be more possible to determine the abundance of specific bacteria in coral than in the surrounding seawater in the future. Interestingly, real-time PCR analysis (Fig. 3 ) and amplicon sequence analysis (Fig. 4 ) showed similar trends in the distribution of Ruegeria spp. across seasons, including a higher abundance of Ruegeria spp. in June. Since 16S rRNA sequences of the Ruegeria spp. listed in Fig. 4 B shows high identity with the primers used for real-time PCR analysis and the amplified sequences were highly preserved across Ruegeria spp., it is possible that the primers used for real-time PCR amplified the 16S rRNA sequences of these Ruegeria spp. shown in Fig. 4 B. Slightly mixed sequence waves of the amplified sequences (Fig. S2 ) may support such speculation. Constructing other primer sets specifically targeting the 16S rRNA sequence diversities of Ruegeria spp. would be needed for further study. Because amplicon sequence data for Ruegeria spp. are not sufficiently registered and the definition of the strain is not well specified, a more detailed analysis regarding the distribution of Ruegeria spp. using NGS data requires detailed research involving this species. Our overview of the metabarcoding analysis revealed season-dependent changes in the composition of bacterial species (Figs. 5 A, 6 A, and 7 A), as reported previously (Glasl et al. 2019 ). More specifically, we found a class of bacteria, Alteromonadaceae, to be abundant in samples collected in June. Alteromonadaceae are known to be one of the common coral-associated bacteria (Sunagawa 2009 ; Ziegler et al. 2017 ). The role of Alteromonadaceae in coral is controversial; they have been considered to be pathogenic (Gignoux-Wolfsohn et al. 2017 ; Liang 2017 ; Sunagawa et al. 2009 ), whereas Dungan and colleagues reported in a preprint that some species of bacteria belonging to the class exhibit radical scavenging ability and could be used as probiotics (Dungan et al. 2020 ). Previously, it has been reported that Glaciecola spp., which belong to Alteromonadaceae, show higher abundance and growth rates in elevated seawater temperatures (von Scheibner et al. 2017 ), whereas in the low temperature range (0–8°C). Whether any species of bacteria in the class Alteromonadaceae can be used as an indicator for elevated water temperatures needs further investigation. Interestingly, we also identified mitochondrial DNA from A . tenuis , a major reef-building coral in the sampling locations, from NGS data (DDBJ accession numbers DRX267149–DRX267157). Although this was unexpected, it has been reported that environmental DNA collected from seawater can be used to survey coral abundance (Nichols and Marko 2019 ; Shinzato et al. 2018 ; West 2020 ). In future research, combining coral abundance with microbial composition in seawater can provide insights into the relationship between microbial diversity and the presence of corals. To conclude, this study demonstrated the existence of various bacteria including Ruegeria spp. in the reef-building coral G . fascicularis and in the surrounding seawater. It was also suggested that specific classes of bacteria in seawater may be used as indicators for seasonality or higher water temperature. These results would be useful for investigating and understanding relationships between the presence of corals and the diversity of microbial flora in seawater, besides the efficient isolation of specific bacterial species from coral or seawater. Availability of Data and Material The data and materials that support the findings of this study are available on request from the corresponding author.

The multi-institute research team developed a method to quickly and noninvasively test for bacteria species known to benefit coral—they test the seawater near the coral. Their approach, first published online on July 18 in Marine Biotechnology, was featured on the cover of the journal's August print edition. "Some coral-associated bacteria are thought to protect coral as probiotics, and in order to develop coral probiotics, it is important to collect such bacteria for further study," said paper author Natsuko Miura, assistant professor in the Graduate School of Life and Environmental Sciences at Osaka Prefecture University. She also noted that detecting and identifying bacteria is key to checking the balance of the coral's bacterial flora, the collective balance of bacteria that exist on any organism. "Previously, researchers suggested that the composition of the microbial flora in seawater around coral could be used as an environmental indicator," Miura said, explaining that the challenge is determining exactly how the distribution of bacteria in seawater reflects the coral health. "To better understand this, we collect and analyze bacteria from coral. But, in previous collection methods, coral-associated bacteria were obtained by destroying the coral body. We thought it was problematic to destroy the coral in order to protect it." The researchers set out to develop a quick detection method to identify beneficial bacteria in seawater samples taken from around coral reefs. They focused on Galaxea fascicularis, a reef-building coral found off the coast of Sesoko Island in Okinawa, Japan. Species of Ruegeria bacteria are thought to offer probiotic benefits to this coral, such as B12 vitamin production and protection against some pathogens, according to Miura. In general, to identify specific species, researchers use two fragments of defined genetic code called primers to target and isolate desired longer genetic sequences. Once isolated, the sequence can be copied and amplified in quantities large enough to match to individual species. Frequently, potential primers are proposed based on various genetic factors but require verification. Seawater around the corals was collected using plastic containers for bacterial isolation and microbiota investigation. Credit: H.Yamashiro, The University of the Ryukyus In addition to newly developed primers, Miura and her team hypothesized that the physical characteristics of two primers from a set of 55 proposed in 2020 that were needed to verify the seawater contents during this development period, would reliably identify Ruegeria species in samples of seawater and coral from around Sesoko Island in just a few hours. They were right. "Our results indicate that this specific primer set enables easy and rapid isolation of specific bacterial species, facilitating isolation of bacteria of interest," Miura said, adding that they found higher concentrations of the Ruegeria species in coral than in seawater, suggesting that the bacteria may be concentrated in coral. "While further studies are needed to determine differences in the abundance of bacteria in corals and seawater, the newly developed method makes it easier to quickly check if seawater around the coral obtains beneficial coral-related bacteria without destroying the coral." The researchers are now supplying samples of the identified bacteria species to public repositories with the hope that others will be able to expand and accelerate the work to better understand and use coral probiotics, according to Miura. "In the future, checking for bacteria in the seawater around the coral will allow us to quickly check the health of the coral," Miura said.

10.1007/s10126-021-10047-2

Space

Elusive origin of stellar geysers revealed by 3-D simulations

Yan-Fei Jiang et al, Outbursts of luminous blue variable stars from variations in the helium opacity, Nature (2018). DOI: 10.1038/s41586-018-0525-0 Journal information: Nature

http://dx.doi.org/10.1038/s41586-018-0525-0

https://phys.org/news/2018-09-elusive-stellar-geysers-revealed-d.html

Abstract Luminous blue variables are massive, evolved stars that exhibit large variations in luminosity and size on timescales from months to years, with high associated rates of mass loss 1 , 2 , 3 , 4 , 5 . In addition to this on-going variability, these stars exhibit outburst phases, during which their size increases and as a result their effective temperature decreases, typically to about 9,000 kelvin 3 , 6 . Outbursts are believed to be caused by the radiation force on the cooler, more opaque, outer layers of the star balancing or even exceeding the force of gravity, although the exact mechanisms are unknown and cannot be determined using one-dimensional, spherically symmetric models of stars because such models cannot determine the physical processes that occur in this regime 7 . Here we report three-dimensional simulations of massive, radiation-dominated stars, which show that helium opacity has an important role in triggering outbursts and setting the observed effective temperature during outbursts of about 9,000 kelvin. It probably also triggers the episodic mass loss at rates of 10 −7 to 10 −5 solar masses per year. The peak in helium opacity is evident in our three-dimensional simulations only because the density and temperature of the stellar envelope (the outer part of the star near the photosphere) need to be determined self-consistently with convection, which cannot be done in one-dimensional models that assume spherical symmetry. The simulations reproduce observations of long-timescale variability, and predict that convection causes irregular oscillations in the radii of the stars and variations in brightness of 10–30 per cent on a typical timescale of a few days. The amplitudes of these short-timescale variations are predicted to be even larger for cooler stars (in the outburst phase). This short-timescale variability should be observable with high-cadence observations. Main We used 60 million CPU hours on the supercomputer Mira, awarded by the Argonne Leadership Computing Facility for the INCITE programme, as well as computational resources from NASA and NERSC (National Energy Research Scientific Computing Center), to solve the three-dimensional radiation hydrodynamic equations 8 and follow the physically realistic evolution of the stellar envelopes of luminous blue variables (LBVs; see Methods for the simulation set-up). These simulations take as inputs the fixed mass of the core of the star and its luminosity from the bottom (inner) boundary of the simulation area, and determine self-consistently the structure of the envelope, its effective temperature and the rate of mass loss. We performed three simulations: the first (which we refer to as T9L6.2) assumes an initial core mass of M i = 80 M ʘ , where M ʘ is the mass of the Sun, a luminosity relative to that of the Sun of log( L / L ʘ ) = 6.2 and an effective temperature of T eff = 9,000 K; the second (T19L6.4) assumes the same initial mass, M i = 80 M ʘ , log( L / L ʘ ) = 6.4 and T eff = 19,000 K; and the third (T19L6) assumes M i = 35 M ʘ , log( L / L ʘ ) = 6.0 and T eff = 19,000 K. Although the three simulations are based on one-dimensional models for stars with different initial masses and at different evolutionary stages (see Methods ), they share properties such as density and Eddington ratio (the ratio of radiation to gravitational acceleration) at the peak in opacity due to iron (at around 180,000 K). The main difference between our three simulations is the pressure scale height at the iron opacity peak, which results in different total stellar masses and optical depths above the convective region, and hence different surface temperatures 9 . The locations of the three models in the Hertzsprung–Russell diagram and the stellar-evolution tracks determined from the one-dimensional models are shown in Fig. 1 . The histories of spherically averaged radial profiles of density, turbulent velocity, radiation temperature and opacity for run T9L6.2 are shown in Fig. 2 . The envelope is convectively unstable at the iron opacity peak 7 , 9 , 10 , as indicated by the density increasing with radius around that region in the initial hydrostatic structure. Convection takes about 10 dynamical times (about 43 h; where one dynamical time is the time required for the radiation-supported sound waves to travel one pressure scale height in the convective region) to destroy the density inversion (density inversion means that the density increases rather than decreases with radius), which causes high-density clumps to rise, expand and cool to a temperature of less than about 6 × 10 4 K. Because the density of these cooled clumps is much greater than that of the unclumped gas at the same temperature before the onset of convection, a strong helium opacity peak appears (Fig. 2 , bottom). The local radiation acceleration after the helium opacity peak has formed is ten times larger than the gravitational acceleration, which causes a large fraction of the envelope to expand markedly, with most of the gas above that region blown away. The mass flux of unbound gas (positive total energy 11 ) leaving our simulation domain can reach an instantaneous rate of around 0.05 M ʘ yr −1 . After 400 h, the envelope settles to a steady state(Fig. 2 , right). Convection is still operating at around 80.90 R ʘ (where R ʘ is the radius of the Sun), with a second peak in helium opacity at around 200 R ʘ . Convection also causes envelope oscillations (that is, oscillations in the size or radius of the star) with a typical timescale of a day. The time-averaged location of the photosphere, at which the total Rosseland optical depth to the outer boundary of the simulation box is unity, is at 342.8 R ʘ , as indicated by the dashed blue lines in Fig. 2 . At this location, the average radiation temperature is 9.06 × 10 3 K. The mass-averaged turbulent velocity is only 1% of sound speed deep in the envelope, but becomes supersonic near the photosphere, causing strong shocks and large temperature and density fluctuations near the photosphere. During each oscillation (evident in both the density and the turbulent velocity in the right panels of Fig. 2 ), part of the mass becomes unbound, with a mass loss rate of about 5 × 10 −6 M ʘ yr −1 . This simulation naturally produces a massive star with luminosity, effective temperature and mass loss consistent with an LBV during an outburst. The traditionally used one-dimensional models cannot capture these properties because of two physical processes that can occur only in the three-dimensional simulations: supersonic turbulent motions that provide effective turbulent pressure support to maintain the extended envelope, and radiative and convective energy transport through the turbulent regions around the opacity peaks. In addition, conversion from radiation to kinetic energy is not well captured by traditional mixing-length theory in one-dimensional models 9 . Fig. 1: Hertzsprung–Russell diagram for LBVs. The shaded areas represent the locations in the diagram where LBVs (grey circles) are most commonly found 5 , 6 : the diagonal band is the S Dor instability strip (LBVs in quiescence). The vertical band represents LBVs in outburst. Dotted lines indicate observed excursions from quiescence to outburst. The solid red, green and blue lines correspond to main-sequence stellar evolution tracks determined from one-dimensional models with different initial masses as indicated: MESA 21 (modules for experiments in stellar astrophysics), GENEC 22 (Geneva evolution code) and STERN 23 , respectively. Our three simulated stars are indicated by coloured polygons. The dashed black line is the Humphreys–Davidson limit 1 . Full size image Fig. 2: Evolution of spherically averaged radial profiles for run T9L6.2. The left and right columns break at t = 426 h to separate the initial transition and the steady-state structures. From top to bottom, the colour scales indicate density ρ (scaled by the fudicial density ρ 0 = 3.6 × 10 −9 g cm −3 ), turbulent flow velocity v (scaled by the local isothermal sound speed c g , which takes the value c g = 1.05 × 10 6 cm s −1 at the photosphere), radiation temperature T r (scaled by the fudicial temperature T 0 = 1.67 × 10 5 K) and opacity κ (scaled by the fudicial opacity κ 0 = 0.34 cm 2 g −1 ). The dashed blue lines indicate where the time-averaged optical depth to the outer boundary of the simulation domain is unity at the steady state. The iron opacity peak is evident in the bottom two panels as the region of larger κ at smaller R , where R is the radial distance from the centre of the star. Full size image Run T19L6.4 results in a similar evolution history and similar turbulent structures to run T9L6.2, with one snapshot shown in Fig. 3 . Owing to a smaller pressure scale height and a smaller optical depth across the typical convective element, the gas rising as a result of convection experiences a much smaller temperature change. This results in a much lower value of the opacity at the helium peak compared with run T9L6.2, and thus a smaller total optical depth above the iron-opacity-peak region. Although the luminosity for run T19L6.4 is slightly larger than for run T9L6.2, the less substantial helium opacity peak places the time-averaged location of the photosphere at a smaller radius of 102 R ʘ , with a higher effective temperature of 1.87 × 10 4 K. This finding confirms that without the helium opacity peak the star will not undergo an outburst and shift into the constant-temperature strip in the Hertzsprung–Russell diagram (see Fig. 1 ). The presence of a smaller helium opacity peak results in a substantial reduction in the amplitude of the envelope oscillation and a lower associated mass loss rate of around 1 × 10 −6 M ʘ yr −1 . Fig. 3: A snapshot of the three-dimensional density and radiation energy density from run T19L6.4. The radial range in this plot covers 14.8 R ʘ (the bottom of the envelope) to 102 R ʘ (the photosphere). Convection develops at the bottom of the envelope owing to the iron opacity peak at 44.6 R ʘ (indicated by the region with density of around 5 × 10 −8 g cm −3 , where small-scale turbulence is observed). The photosphere has large-scale plumes, which cause strong variations in the radiation temperature at the photosphere across the surface of the star. A video of the evolution of this simulation is provided in Supplementary Video 1 . Full size image Finally, run T19L6 has very similar properties to T19L6.4, in particular a comparable value of the pressure scale height at the iron opacity peak. However, this model is calculated for a smaller core mass and a lower luminosity. At the steady state, the envelope has an effective temperature of 1.89 × 10 4 K, a time-averaged photosphere radius of 63.7 R ʘ and an episodic mass loss rate associated with envelope oscillations of only around 5 × 10 −7 M ʘ yr −1 . This result confirms that when the iron opacity peak is in a region with a small pressure scale height the effective temperature remains too hot for the helium opacity to become important, and the star stays closer to the S Dor instability strip in the Hertzsprung–Russell diagram (see Fig. 1 ). Our simulations predict that LBVs undergoing outbursts should exhibit irregular variability with typical timescales of days. In particular, we expect the variability pattern to be different for massive stars in the S Dor instability strip and during outburst (see Fig. 4 ). For massive stars with effective temperatures near 9 × 10 3 K, a substantial helium opacity peak exists in the envelope and causes large-amplitude oscillations. The predicted stellar brightness then varies by a factor of roughly 1.5−2 in a day at the steady state (Fig. 4 , top). For stars with hotter effective temperatures of nearly 1.9 × 10 4 K and a weaker helium opacity peak, the variability at the steady state has a much smaller amplitude. However, the luminosity can still vary by about 20% on timescales of a week to a few weeks, which corresponds to the thermal timescale of the envelope above the iron-opacity-peak region. This kind of variability has been seen in recent high-cadence observations of massive stars 12 , 13 , 14 , and the correlation between variability and effective temperature can be tested with future observations. The envelope is loosely bound and dominated by turbulent convection (Fig. 2 ), so the oscillation at the stellar surface is chaotic. However, there are moments in the evolution of the envelope when the majority of the photosphere is falling back onto the core, as suggested by the integrated luminosity in Fig. 4 . This finding can potentially explain the time-dependent behaviour characterized by P Cygni and inverse P Cygni profiles of some LBVs 13 , 15 . Fig. 4: Evolution of the total luminosity measured from the outer boundary of the simulation box. Top, T9L6.2 ( M i = 80 M ʘ ); middle, T19L6.4 ( M i = 80 M ʘ ); bottom, T19L6 ( M i = 35 M ʘ ). The vertical dashed red lines indicate the time by which the density inversion in the initial conditions has been removed owing to convection. The effective temperature and average luminosity at the steady state are labelled for each simulation. Full size image The mass loss rates that we obtain from our steady-state simulations are broadly consistent with inferred mass loss rates for the quiescent and outburst phases of LBVs 5 , 15 . We find that the physical mechanism that is responsible for driving the mass loss in LBVs is the interaction of their large radiative flux with opacity peaks that appear in their optically thick envelopes as they expand and cool. Importantly, although the iron opacity peak is strongly metallicity dependent, as long as the turbulent stellar envelope cools to low enough temperatures, the helium and hydrogen opacity peaks will always cause large Eddington factors. This mode of mass loss may therefore be less sensitive to metallicity than are line-driven winds 16 . Traditionally, mass loss due to radiation force on the ultraviolet lines in the optically thin region of the stellar envelope is thought to be the dominant mechanism for winds in these massive stars 17 , 18 , although other models are probably important for outbursts 3 , 19 . Our work suggests that it is important to study these mechanisms with the turbulent envelope as found by our simulations. The simulations also suggest possible paths for the transition between the S Dor instability strip and the outburst phase in the Hertzsprung–Russell diagram, as indicated by the dotted black lines for the observed LBVs in Fig. 1 . When a star expands due to nuclear evolution and a substantial helium opacity peak appears, the star will undergo the first outburst. The amount of mass initially above the iron-opacity-peak region can sustain the associated mass loss rate for only around 10 years; the thermal timescale below the convective region in the envelope is also about 10 years. The star may lose a substantial fraction of the mass above the iron-opacity-peak region via the wind before it has time to adjust to a new structure to keep this large mass loss rate. This will reduce the total optical depth above the iron opacity peak and increase the effective temperature. When the helium opacity peak is reduced substantially, these stars will return to the S Dor instability strip. As the iron opacity peak moves to a deeper region of the envelope on the thermal timescale, this process can repeat. Alternatively, if the massive star is in a binary system, as has been suggested for some LBVs 20 , and the companion deposits mass on the surface of the star, then the additional mass will probably be ejected by the massive star, as we found in our initial evolutions for each numerical simulation. This mass ejection could be a trigger of the giant eruptions observed for some LBVs. The detailed properties of this process will need to be studied with future calculations. Methods Typical LBVs have luminosities of 6 × 10 5 L ʘ to about 4 × 10 6 L ʘ , and effective temperatures either hotter than roughly 2 × 10 4 K in quiescence or around 9 × 10 3 K during outburst (Fig. 1 ). We use the stellar evolution code MESA 7 , 24 , 25 , 26 to evolve solar-metallicity stars with initial masses of M i = 35 M ʘ and M i = 80 M ʘ to the ranges of luminosity and temperature shown in the Hertzsprung–Russell diagram 9 in Fig. 1 . Although these one-dimensional evolution models for stars in this regime are very uncertain, as illustrated by the widely varying end points of the tracks in Fig. 1 , they do provide a useful first approximation of the physical conditions in the radiation-dominated envelopes of these stars. One important feature of this region of luminosity–temperature space is the presence of the iron opacity peak at temperatures of T ≈ 1.8 × 10 5 K due to lines of iron-group elements. The opacity of the iron peak is often a factor of a few larger than that from free-electron scattering and can cause the local radiation acceleration to exceed the gravitational acceleration 7 . In this situation, the envelope is unstable to convection at the location of the iron opacity peak 7 , 27 , with the properties of convection and the structure of the envelope depending crucially on how deep within the star the iron opacity peak occurs 9 . The opacity can in principle increase further—to values 100 times that of electron scattering—when the temperature drops below (1–4) × 10 4 K. This arises from helium and hydrogen recombination, but only when the density exceeds about 10 −9 g cm −3 , a value that is much larger than what is realized in one-dimensional hydrostatic structures around this temperature range. We take the typical luminosity, density and gravity at the location of the iron opacity peak in our one-dimensional models and construct envelopes in hydrostatic and thermal equilibrium in a spherical polar geometry as the initial conditions for our three-dimensional simulations. The whole simulation box initially covers the temperature range 10 4 –10 6 K. The bottom boundary of the simulation box has a density and temperature that are orders of magnitude larger than the values at the iron opacity peak and that remain approximately constant. Both OPAL Rosseland and Planck mean opacity tables 28 are included in the simulations to capture the momentum and thermal coupling between the radiation field and the gas 29 . The outer boundary of the simulation box is at least three times the photosphere radius for all of the simulations. Code availability The code that we used to do the simulations, Athena++ with the radiative transfer module, is available from the corresponding author on request. Data availability The simulation data are available from the corresponding author on request.

Astrophysicists finally have an explanation for the violent mood swings of some of the biggest, brightest and rarest stars in the universe. The stars, called luminous blue variables, periodically erupt in dazzling outbursts nicknamed "stellar geysers." These powerful eruptions launch entire planets' worth of material into space in a matter of days. The cause of this instability, however, has remained a mystery for decades. Now, new 3-D simulations by a team of astrophysicists suggest that turbulent motion in the outer layers of a massive star creates dense clumps of stellar material. These clumps catch the star's intense light like a solar sail, erupting material into space. After jettisoning enough mass, the star calms down until its outer layers re-form and the cycle begins anew, the astrophysicists report online September 26 in Nature. Identifying the cause of the stellar geysers is significant because every extremely massive star probably spends part of its life as a luminous blue variable, says study co-author Matteo Cantiello, an associate research scientist at the Center for Computational Astrophysics at the Flatiron Institute in New York City. A simulation of the turbulent gas that envelops a star 80 times the mass of the sun. Intense light from within the star pushes against dense pockets of helium-rich material in the star's outer layers, launching the material spaceward. The colors represent the density of the gas, with lighter colors signifying denser regions. Credit: Joseph Insley/Argonne Leadership Computing Facility "This finding represents an important step forward in understanding the life and death of the biggest stars in the universe," says Cantiello. "These massive stars, despite their small number, largely determine the evolution of galaxies through their stellar winds and supernova explosions. And when they die, they leave behind black holes." Luminous blue variables, or LBVs, are exceedingly rare, with only around a dozen spotted in and around the Milky Way galaxy. The gargantuan stars can exceed 100 times the mass of the sun and approach the theoretical limit of how massive stars can get. LBVs are also exceptionally radiant: The brightest ones shine with more than 1 million times the luminosity of the sun. That light pushes matter spaceward because absorption and re-emission of a photon by an atom results in a net outward shove. The tug of war between extreme gravity pulling material in and extreme luminosity pushing it out is responsible for the trademark outbursts of LBVs, scientists believe. The absorption of a photon by an atom, however, requires that electrons be bound in orbits around the atom's nucleus. In the deepest, hottest layers of a star, matter behaves as a plasma with electrons untethered from atoms. In the cooler outer layers, electrons begin rejoining their atoms and can therefore absorb photons again. Previously proposed explanations for the outbursts predicted that elements such as helium in the outer layers could absorb enough photons to overcome gravity and fly into space as an outburst. But simple, one-dimensional calculations didn't back up this hypothesis: The outer layers didn't seem sufficiently dense to catch enough light to overpower gravity. A simulation of the turbulent gas that envelops a star 80 times the mass of the sun. Intense light from within the star pushes against dense pockets of helium-rich material in the star's outer layers, launching the material spaceward. The solid colors denote radiation intensity, with bluer colors representing regions of larger intensity. The translucent purplish colors represent the gas density, with lighter colors denoting denser regions. Credit: Joseph Insley/Argonne Leadership Computing Facility Those simple calculations, however, didn't capture the full picture of the complex dynamics within a colossal star. Cantiello, along with Yan-Fei Jiang of the Kavli Institute for Theoretical Physics at the University of California, Santa Barbara, and colleagues took a more realistic approach. The researchers created a detailed, three-dimensional computer simulation of how matter, heat and light flow and interact within supersize stars. The calculations involved required more than 60 million computer processor hours to solve. In the simulation, the average density of the outer layers was too low for material to go flying—just as the one-dimensional calculations predicted. However, the new calculations revealed that convection and mixing in the outer layers resulted in some regions being denser than others, with some clumps opaque enough to be launched into space by the star's light. Such eruptions occur over timescales ranging from days to weeks as the star churns and its brightness fluctuates. The team estimates that such stars can shed around 10 billion trillion metric tons of material each year, roughly double the Earth's mass. The researchers plan to improve the accuracy of their simulations by incorporating other effects such as the star's rotation, which can make launching material into space easier near the star's fast-spinning equator than near the almost stationary poles. (This effect is the reason NASA launches its rockets from Florida and California rather than Maine or Alaska.) Improving the fidelity of star simulations is crucial to achieving astrophysical insights, Cantiello says. The move from simple, single-dimensional calculations to full 3-D simulations requires more computational muscle and more complex physics, but the results are well worth the trouble. "We had to implement all of these physics to see, with our own eyes, that this process—that we didn't expect to be important—would turn out to be key to understanding these violent eruptions and the evolution of these massive stars," he says.

10.1038/s41586-018-0525-0

Biology

Learning how to battle harmful algae blooms

Drahomíra Faktorová et al, Genetic tool development in marine protists: emerging model organisms for experimental cell biology, Nature Methods (2020). DOI: 10.1038/s41592-020-0796-x Journal information: Nature Methods

http://dx.doi.org/10.1038/s41592-020-0796-x

https://phys.org/news/2020-07-algae-blooms.html

Abstract Diverse microbial ecosystems underpin life in the sea. Among these microbes are many unicellular eukaryotes that span the diversity of the eukaryotic tree of life. However, genetic tractability has been limited to a few species, which do not represent eukaryotic diversity or environmentally relevant taxa. Here, we report on the development of genetic tools in a range of protists primarily from marine environments. We present evidence for foreign DNA delivery and expression in 13 species never before transformed and for advancement of tools for eight other species, as well as potential reasons for why transformation of yet another 17 species tested was not achieved. Our resource in genetic manipulation will provide insights into the ancestral eukaryotic lifeforms, general eukaryote cell biology, protein diversification and the evolution of cellular pathways. Main The ocean represents the largest continuous planetary ecosystem, hosting an enormous variety of organisms, which include microscopic biota such as unicellular eukaryotes (protists). Despite their small size, protists play key roles in marine biogeochemical cycles and harbor tremendous evolutionary diversity 1 , 2 . Notwithstanding their significance for understanding the evolution of life on Earth and their role in marine food webs, as well as driving biogeochemical cycles to maintain habitability, little is known about their cell biology including reproduction, metabolism and signaling 3 . Most of the biological knowledge available is based on comparison of proteins from cultured species to homologs in genetically tractable model taxa 4 , 5 , 6 , 7 . A main impediment to understanding the cell biology of these diverse eukaryotes is that protocols for genetic modification are only available for a small number of species 8 , 9 that represent neither the most ecologically relevant protists nor the breadth of eukaryotic diversity. The development of genetic tools requires reliable information about gene organization and regulation of the emergent model species. Over the last decade, genome 4 , 5 , 6 and transcriptome sequencing initiatives 7 have resulted in nearly 120 million unigenes being identified in protists 10 , which facilitates the developments of genetic tools used for model species. Insights from these studies enabled the phylogenetically informed approach 7 for selecting and developing key marine protists into model systems in the Environmental Model Systems (EMS) Project presented herein. Forty-one research groups took part in the EMS Project, a collaborative effort resulting in the development of genetic tools that significantly expand the number of eukaryotic lineages that can be manipulated, and that encompass multiple ecologically important marine protists. Here, we summarize detailed methodological achievements and analyze results to provide a synthetic ‘transformation roadmap’ for creating new microeukaryotic model systems. Although the organisms reported here are diverse, the paths to overcome difficulties share similarities, highlighting the importance of building a well-connected community to overcome technical challenges and accelerate the development of genetic tools. The 13 emerging model species presented herein, and the collective set of genetic tools from the overall collaborative project, will not only extend our knowledge of marine cell biology, evolution and functional biodiversity, but also serve as platforms to advance protistan biotechnology. Results Overview of taxa in the EMS initiative Taxa were selected from multiple eukaryotic supergroups 1 , 7 to maximize the potential of cellular biology and to evaluate the numerous unigenes with unknown functions found in marine protists (Fig. 1 ). Before the EMS initiative, reproducible transformation of marine protists was limited to only a few species such as Thalassiosira pseudonana , Phaeodactylum tricornutum and Ostreococcus tauri (Supplementary Table 1 ). The EMS initiative included 39 species, specifically, 6 archaeplastids, 2 haptophytes, 2 rhizarians, 9 stramenopiles, 12 alveolates, 4 discobans and 4 opisthokonts (Fig. 1 ). Most of these taxa were isolated from coastal habitats, the focus area of several culture collections 7 . More than 50% of the selected species are considered photoautotrophs, with another 35% divided between heterotrophic osmotrophs and phagotrophs, the remainder being predatory mixotrophs. Almost 20% of the chosen species are symbionts and/or parasites of marine plants or animals, 5% are associated with detritus and several are responsible for harmful algal blooms (Supplementary Table 2 ). Fig. 1: Phylogenetic relationships and transformation status of marine protists. A schematic view of the eukaryotic tree of life with effigies of main representatives. Color-coordinated species we have attempted to genetically modify are listed below. Current transformability status is schematized in circles indicating: DNA delivered and shown to be expressed (yellow, for details see text and Table 1 ); DNA delivered, but no expression seen (gray) and no successful transformation achieved despite efforts (blue). The details of transformation of species that belong to ‘DNA delivered’ and ‘Not achieved yet’ categories are described in Supplementary Table 5 . mRNA, messenger RNA; FITC–dextran, fluorescein isothiocyanate (FITC)-conjugated dextran. Full size image While some transformation systems for protists have been developed in the past 8 , 9 , 11 , the challenge for this initiative was to develop genetic tools for species that not only require different cultivation conditions but are also phenotypically diverse. It should be noted that not all main lineages were explored. For example, amoebozoans did not feature in this aquatic-focused initiative, in part because they tend to be most important in soils, at least based on current knowledge, and manipulation systems exist for members of this eukaryotic supergroup, such as Dictyostelium discoideum 12 . The overall EMS initiative outcomes are summarized in Fig. 1 and Table 1 . We provide detailed protocols for 13 taxa, for which no transformation systems have been previously reported (category A) and eight taxa, for which existing protocols 9 , 11 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 were advanced (category B; Figs. 2 , 3 and 4 , Table 1 , Supplementary Tables 1 – 5 and Methods ). We also review an already published EMS transformation protocol 22 in one species (category C), and we discuss unsuccessful transformation attempts for 17 additional taxa (Fig. 1 and Methods ). Finally, we synthesize our findings in a roadmap for the development of transformation systems in protists (Fig. 5 ). Table 1 Parameters used for successful transformation as shown in Figs. 2 , 3 and 4 Full size table Fig. 2: Epifluorescence micrographs of transformed marine protists. Representative images of transformants and wild-type cell lines of ten selected protist species. Colored boxes behind species names refer to phylogenetic supergroup assignments given in Fig. 1 . Representative data of at least two independent experiments are shown. The fluorescent images show the expression of individual fluorescent marker genes introduced via transformation for all organisms shown, except in the case of A. amoebiformis . For this, red depicts the natural autofluorescence of photosynthetic pigments in the cell, while the additional green spheres in the transformant fluorescence panel shows introduced GFP fluorescence (see Supplementary Fig. 15c for a trace of these different regions in the cell). Scale bars are as follows: 10 µm for A. amoebiformis , T. pseudonana , A. limacinum , B. saltans , N. gruberi , A. whisleri and S. rosetta ; 15 µm for P. marinus ; 20 µm for F. cylindrus and 100 µm for P. multiseries . Full size image Fig. 3: Various methods were used to demonstrate successful transformation in different archaeplastid species: luminescence and fluorescence. a – c , Luminescence ( a , b ) and fluorescence (by FACS and epifluorescence microscopy) ( c ) were used to verify expression of introduced constructs in three archaeplastids: O. lucimarinus ( a ), B. prasinos ( b ) and M. commoda ( c ). For the latter, red in the image depicts the natural autofluorescence of photosynthetic pigments in the cell, while green shows introduced eGFP fluorescence and blue shows the DAPI-stained nucleus; the overlay shows colocalization of eGFP and nucleus signals. See Supplementary Fig. 15d for a trace of these different regions in the cell. NS, not significant; trans., transformed. Representative data of at least two independent experiments are shown. For a detailed figure description see Supplementary Notes 2 . Source data Full size image Fig. 4: Various methods were used to demonstrate successful transformation in different species: RT–PCR, western blot and sequencing. a – j , Western blot, RT–PCR or sequencing (in case of Cas9-induced excision by CRISPR) were used to verify expression of introduced constructs in one haptophyte: I. galbana ( a ), one rhizarian— A. amoebiformis ( b ), two stramenopiles— F. cylindrus ( c ) and P. tricornutum ( d ), three alveolates— K. veneficum ( e ), P. marinus ( f ) and A. carterae ( g ), two discobans— B. saltans ( h ) and D. papillatum ( i ) and one opisthokont— A. whisleri ( j ). Note that nptII/neo is used synonymously with amino 3′-glycosyl phosphotransferase gene ( aph (3′)) conferring resistance to kanamycin and neomycin. Representative data of at least two independent experiments are shown. For a detailed figure description see Supplementary Notes 2 . Source data Full size image Fig. 5: ‘Transformation roadmap’ for the creation of genetically tractable protists. a , Vector design and construction for microeukaryotes of interest and a natural community. b , Transformation approaches. Different symbols represent methods (for example chemical, physical or biological) for introducing DNA/RNA/protein into a living cell. c , Protocol. Key methodological steps for successful transformation are listed in an abbreviated form (for particular examples, see Table 1 and text). Full size image Archaeplastids Prasinophytes are important marine green algae distributed from polar to tropical regions 23 . They form a sister group to chlorophyte algae, and together, these two groups branch adjacent to land plants, collectively comprising the Viridiplantae, which are part of the Archaeplastida 1 , 23 (Fig. 1 ). Genome sequences are available for the picoprasinophytes (<3 µm cell diameter) tested herein, specifically, Micromonas commoda , M. pusilla , Ostreococcus lucimarinus and Bathycoccus prasinos . As part of the EMS initiative, we report on genetic tools for Bathycoccus , a scaled, nonmotile genus, and Micromonas , a motile, naked genus with larger genomes than Bathycoccus and Ostreococcus 22 . We also report on genetic tools for Tetraselmis striata and O. lucimarinus . The latter was transformed based on an adapted homologous recombination system for O. tauri 24 , 25 . O. lucimarinus (RCC802) and B. prasinos (RCC4222) were transformed using protocols adapted from O. tauri 24 , 25 . Briefly, using electroporation for transfer of exogenous genes, O. lucimarinus was transformed using a DNA fragment encoding the O. tauri high-affinity phosphate transporter ( HAPT ) gene fused to a luciferase gene and a kanamycin selection marker (Table 1 and Supplementary Table 3 ), which resulted in transient luciferase expression 24 h after electroporation (Table 1 and Fig. 3a ). After 2 weeks of growth in low-melting agarose plates containing G418 (1 mg ml −1 ), 480 colonies were obtained, picked and grown in artificial seawater with the antibiotic neomycin. Of these, 76 displayed luminescence ≥2.5-fold above background (80 relative luminescence units (RLU)), with widely variable levels (200–31,020 RLU), likely reflecting either variations in the site of integration and/or the number of integrated genes (Fig. 3a , Supplementary Fig. 1 and Methods ). The O. tauri construct did not work in B. prasinos , while the use of the B. prasinos histone H4 and HAPT sequences in an otherwise identical construct and conditions was successful. Although luciferase expression was not detected 24 h after electroporation, 48 G418-resistant colonies were obtained 2 weeks later, 20 being luminescent when grown in liquid medium. Analysis of 14 resistant transformants revealed that the luciferase sequence was integrated into the genome of five luminescent clones, and one nonluminescent clone (Fig. 3b and Methods ), suggesting that the chromatin context at integration sites in the latter was not favorable to luciferase expression. Although transformation methods successful for Bathycoccus and Ostreococcus failed in Micromonas , Lonza nucleofection was successful with M. commoda (CCMP2709) (Table 1 and Fig. 3c ) using two different codon-optimized plasmids, one encoding the luciferase gene (NanoLuc, Promega) flanked by an exogenous promoter and terminator sequence from the 5′ and 3′ untranslated regions (UTRs) of histone H3 in Micromonas polaris (CCMP2099), and the other encoding an enhanced green fluorescent protein ( eGFP ) gene flanked by endogenous promoter and terminator sequences from ribosomal protein S9 (Supplementary Table 5 ). Sensitivities to antibiotics were established (Supplementary Table 3 ). Constructs did not include a selectable marker, as we aimed to introduce and express foreign DNA while developing conditions suitable for transfection that supported robust growth in this cell wall-lacking protist (Table 1 ). Transformants revealed a significantly higher level of eGFP fluorescence than wild-type cells, with 1.3% of the population showing fluorescence per cell 45-fold higher than both the nontransformed portion of the culture and the wild-type cells (Fig. 3c and Methods ). Additionally, the RLU was 1,500-fold higher than controls when using the luciferase-bearing construct, such that multiple experiments with both plasmids confirmed expression of exogenous genes in M. commoda . T. striata (KAS-836) was transformed using microprojectile bombardment (Supplementary Fig. 2a ). Two selectable marker genes were tested, consisting of a putative promoter and 5′ UTR sequences from the T. striata actin gene and either the coding sequences of the Streptoalloteichus hindustanus bleomycin gene (conferring resistance to zeocin) or the Streptomyces hygroscopicus bar gene (conferring resistance to glufosinate) (Table 1 , Supplementary Fig. 2a and Methods ). The terminator sequence was obtained from the T. striata glyceraldehyde-3-phosphate dehydrogenase gene. Linearized plasmids were coated on gold particles and introduced into T. striata cells by using the PDS-1000/He Particle Delivery System (Bio-Rad). Transformants were successfully selected on half-strength f/2 at 50% salinity agar plates containing either 150 μg ml −1 zeocin or 150 μg ml −1 glufosinate. Haptophytes (incertae sedis) Haptophytes are a group of photosynthetic protists that are abundant in marine environments and include the principal calcifying lineage, the coccolithophores. Genome sequences are available for Emiliania huxleyi 6 and Chrysochromulina tobin 26 , and there is one report of nuclear transformation of a calcifying coccolithophore species 27 but transformation of E. huxleyi , the most prominent coccolithophore, has not been achieved yet 27 . Here, as part of the EMS initiative, a stable nuclear transformation system was developed for Isochrysis galbana , a species that lacks coccoliths, but represents an important feedstock for shellfish aquaculture 28 . I. galbana (CCMP1323) was transformed by biolistic bombardment with the pIgNAT vector, which contains nourseothricin (NTC) N -acetyltransferase ( NAT ), (for nourseothricin resistance) driven by the promoter and terminator of Hsp70 from E. huxleyi (CCMP1516). Twenty-four hours after bombardment, cells were transferred to liquid f/2 medium at 50% salinity containing 80 µg ml −1 NTC and left to grow for 2–3 weeks to select for transformants (Table 1 ). The presence of NAT in NTC-resistant cells was verified by PCR and PCR with reverse transcription (RT–PCR) (Fig. 4a , Supplementary Fig. 2b and Methods ) and the sequence was verified. To confirm NTC resistance was a stable phenotype, cells were subcultured every 2–4 weeks at progressively higher NTC concentrations (up to 150 µg ml −1 ) in the above-mentioned media. Cells remained resistant to NTC for approximately 6 months, as confirmed by PCR screening to identify the presence of the NAT gene. Rhizarians Rhizarians include diverse nonphotosynthetic protists, as well as the photosynthetic chlorarachniophytes that acquired a plastid via secondary endosymbiosis of a green alga 4 . Uniquely, they represent an intermediate stage of the endosymbiotic process, since their plastids still harbor a relict nucleus (nucleomorph). Here, we report on an advanced transformation protocol for the chlorarachniophyte Amorphochlora (Lotharella) amoebiformis for which low-efficiency transient transformation has previously been achieved using particle bombardment 14 . A. amoebiformis (CCMP2058) cells were resuspended in 100 µl of Gene Pulse Electroporation Buffer (Bio-Rad) with 20–50 µg of the reporter plasmid encoding eGFP-RubisCO fusion protein under the control of the native rbcS1 promoter and subjected to electroporation (Table 1 ). Cells were immediately transferred to fresh ESM medium and incubated for 24 h. Transformation efficiency was estimated by the fraction of cells expressing eGFP, resulting in 0.03–0.1% efficiency, as enumerated by microscopy, showing an efficiency up to 1,000-fold higher than in the previous study 14 (Table 1 ). Stable transformants were generated by manual isolation using a micropipette, and a transformed line has maintained eGFP fluorescence for at least 10 months without antibiotic selection (Figs. 2 and 4b and Methods ). Stramenopiles Stramenopiles are a diverse lineage harboring important photoautotrophic, mixotrophic (combining photosynthetic and phagotrophic nutrition) and heterotrophic taxa. As the most studied class in this lineage, diatoms (Bacillariophyceae) were early targets for the development of reverse genetics tool 11 , 29 . Diatoms are estimated to contribute approximately 20% of annual carbon fixation 30 and, like several other algal lineages, are used in bioengineering applications and biofuels 31 . Although other cold-adapted eukaryotes have, to our knowledge, yet to be transformed, here we present a protocol for the Antarctic diatom Fragilariopsis cylindrus 32 . A transformation protocol has also been developed herein for Pseudo-nitzschia multiseries , a toxin-producing diatom 33 . Here we also present work for nondiatom stramenopiles, including a transformation protocol for the eustigmatophyte Nannochloropsis oceanica , and an alternative protocol for the labyrinthulomycete Aurantiochytrium limacinum 20 , both of which are used for biotechnological applications. Furthermore, we report on advances for CRISPR/Cas-driven gene knockouts in Phaeodactylum tricornutum 8 , 13 and a more efficient bacterial conjugation system for Thalassiosira pseudonana 13 . Microparticle bombardment was used on F. cylindrus (CCMP1102) that was grown, processed and maintained at 4 °C in 24 h light. Exponential phase cells were harvested onto a 1.2 µm membrane filter that was then placed on an 1.5% agar Aquil plate for bombardment with beads coated with a plasmid containing zeocin resistance and eGFP , both controlled by an endogenous fucoxanthin chlorophyll a/c binding protein (FCP) promoter and terminator (Table 1 , Supplementary Table 3 and Methods ) 34 . Transformation was performed using 0.7 µm tungsten particles and the biolistic particle delivery system PDS-1000/He (Bio-Rad). Rupture disks for 1,350 and 1,550 pounds per square inch (psi) gave the highest colony numbers with efficiencies of 20.7 colony forming units (c.f.u.) per 10 8 cells and 30 c.f.u. per 10 8 cells, respectively. Following bombardment, the filter was turned upside down and left to recover for 24 h on the plate, then cells were rinsed from the plate/filter and spread across five 0.8% agar Aquil plates with 100 µg ml −1 zeocin. Colonies appeared 3–5 weeks later. PCR on genomic DNA showed that 100 and 60% of colonies screened positive for the bleomycin gene ( ShBle ) for zeocin resistance and the gene encoding eGFP, respectively. As confirmed by fluorescence-activated cell sorting (FACS) and microscopy, eGFP was localized to the cytosol and was distinguishable from plastid autofluorescence (Fig. 2 ). Additional confirmation by PCR and RT–PCR (Fig. 4c ) revealed that the ShBle and eGFP genes were present in the genomes of transformants after multiple transfers (>10) 2 years later, indicating long-term stability. Bacterial conjugation methods were improved in T. pseudonana (CCMP1335) using the silaffin precursor TpSil3p (Table 1 and Methods ) as the target gene. TpSil3p was fused to eGFP flanked by an FCP promoter and terminator, cloned into a pTpPuc3 episomal backbone and transformed into mobilization plasmid-containing EPI300 E. coli cells (Lucigen). The donor cells were grown in super optimal broth with catabolite repression (SOC) medium at 37 °C until OD 600 of 0.3–0.4, centrifuged and resuspended in 267 μl SOC medium. Next, 200 μl of donor cells were mixed with T. pseudonana cells, cocultured on predried 1% agar plates, dark incubated at 30 °C for 90 min, then at 18 °C in constant light for 4 h, followed by selection in 0.25% agar plates containing 100 µg ml −1 NTC. Colonies were observed after 2 weeks, inoculated into 300 μl L1 medium and supplemented with 200 µg ml −1 NTC to reduce the number of false positives. Positive transformants were identified by colony PCR screening (Supplementary Fig. 3 ) and epifluorescence microscopy (Fig. 2 ). The diatom P. multiseries (15093C) and other members of this genus form buoyant linear chains with overlapping cell tips during active growth, and were unconducive to punctate colony formation on agar, where their growth is generally poor. To address this challenge, a low-gelation-temperature agarose seawater medium (LGTA) was developed to facilitate growth, antibiotic selection and cell recovery. P. multiseries exhibited growth inhibition at relatively low concentrations under NTC, formaldehyde and zeocin (Supplementary Table 3 ). Biolistic transformation of two other P. species had been demonstrated at low efficiency 35 . To complement this approach and explore potentially higher efficiency methods for transformation with diatom episomal plasmids, we modified the existing conjugation-based method 13 . The published conjugation protocol was modified to enhance P. multiseries postconjugation viability by reducing SOC content. An episomal version of the Pm_actP_egfp_actT expression cassette was transfected into E. coli EPI300+pTAMOB and used for conjugation (Table 1 and Methods ). After 48 h in L1 medium, cells were plated in LGTA and eGFP-positive cells were observed 7 d later (Fig. 2 ). PCR revealed the presence of plasmids in all eGFP-positive colonies (Supplementary Fig. 4 ). Similarly, conjugation with the episome pPtPUC3 (bleomycin selection marker)-containing bacterial donors was followed under zeocin selection (200 μg ml −1 ). After 7 d, only viable cells (based on bright chlorophyll fluorescence) contained the episome, as confirmed by PCR. Propagation of transformants after the first medium transfer (under selection) has so far been unsuccessful. Stable transformation of A. limacinum (ATCC MYA-1381) was achieved by knock-in of a resistance cassette composed of ShBle driven by 1.3 kb promoter and 1.0 kb terminator regions of the endogenous glyceraldehyde-3-phosphate dehydrogenase gene carried in a pUC19-based plasmid (18GZG) along with the native 18S ribosomal RNA gene, and by knock-in of a similar construct containing a eGFP:ShBle fusion (Supplementary Fig. 5 ). Approximately 1 × 10 8 cells were electroporated, adapting the electroporation protocol used for Schizochytrium 36 . The highest transformation efficiency was achieved using 1 µg of linearized 18GZG plasmid with two pulses, resulting in a time constant of ~5 ms (Table 1 and Methods ). Expression of the fusion protein was confirmed by both the zeocin-resistance phenotype and the detection of eGFP (Fig. 2 ). Six 18GZG transformants derived from uncut and linearized plasmids were examined in detail. All maintained antibiotic resistance throughout 13 serial transfers, first in selective, then subsequently in nonselective media and then again in selective medium. Integration of the plasmid into the genome was confirmed by PCR as well as by Southern blots using a digoxigenin-labeled ShBle gene probe, showing that four transformants had integrations by single homologous recombination, while in two transformants additional copies of the antibiotic resistance cassette were integrated by nonhomologous recombination elsewhere in the genome (Supplementary Fig. 5 ). Electroporation of N. oceanica (CCMP1779) was optimized based on observation of cells treated with fluorescein-conjugated 2,000 kDa dextran and subsequent survival (Table 1 and Methods ). A sorbitol concentration of 800 mM and electroporation at between 5 and 9 kV cm −1 resulted in highest cell recovery. These conditions were used during introduction of plasmids containing the gene for the blue fluorescent reporter mTagBFP2 under the control of cytomegalovirus ( CMV ), the cauliflower mosaic virus 35S , or the VCP1 promoter previously described from Nannochloropsis sp. 37 . Transient expression of blue fluorescence (compared to cells electroporated simultaneously under the same conditions without plasmid) appeared within 2 h, lasted for at least 24 h and disappeared by 48 h in subsets of cells electroporated with mTagBFP2 under the control of CMV (Supplementary Fig. 6 ). The transient transformation was more effective when a linearized plasmid was used compared to a circular plasmid (Table 1 ). VCP1 did not induce blue fluorescence with a circular plasmid, while 35S gave inconsistent results with either circularized or linearized plasmids. For P. tricornutum (CCAP1055/1), we adapted the CRISPR/Cas9 system 8 for multiplexed targeted mutagenesis. Bacterial conjugation 13 was used to deliver an episome that contained a Cas9 cassette and two single-guide RNA (sgRNA) expression cassettes designed to excise a 38 basepair-long domain from the coding region of a nuclear-encoded, chloroplastic glutamate synthase ( Phatr3_J24739 ) and introduce an in-frame stop codon after strand ligation (Table 1 and Methods ). The GoldenGate assembly was used to clone two expression cassettes carrying sgRNAs into a P. tricornutum episome that contained a Cas9–2A-ShBle expression cassette and the centromeric region CenArsHis (Supplementary Fig. 7 ). After their addition to a P. tricornutum culture, plates were incubated in a growth chamber under standard growth conditions for 2 d and transformed P. tricornutum colonies began to appear after 2 weeks. Only colonies maintaining Cas9–2A-ShBle sequence on the delivered episome were able to grow on selection plates because Cas9 and ShBle were transcriptionally fused by the 2A peptide 38 (Supplementary Fig. 7 ). Gel electrophoresis migration and sequencing of the genomic target loci confirmed the 38 bp-long excision and premature stop codon (Fig. 4d ). Alveolates This species-rich and diverse group comprises ciliates, apicomplexans and dinoflagellates (Fig. 1 ). As a link between apicomplexan parasites and dinoflagellate algae, perkinsids are key for understanding the evolution of parasitism, and also have potential biomedical applications 17 . Techniques currently exist for transformation of only a small number of ciliates, perkinsids and apicomplexans 39 . Here, we present a transformation protocol for Karlodinium veneficum (CCMP1975), a phagotrophic mixotroph that produces fish-killing karlotoxins 40 . Experiments were also performed on Oxyrrhis marina (CCMP 1788/CCMP 1795), a basal-branching phagotroph that lacks photosynthetic plastids and Crypthecodinium cohnii (CCMP 316), a heterotroph used in food supplements. For both of these taxa, evidence of DNA delivery was achieved (Table 1 , Supplementary Results , Supplementary Fig. 15 and Methods ), a goal recently achieved for C. cohnii using electroporation 19 . Additionally, we report on improved transformation systems for Perkinsus marinus (PRA240) and Amphidinium carterae (CCMP1314) chloroplast, published recently as part of the EMS initiative 15 . K. veneficum (CCMP1975) was transformed based on electroporation and cloning the selectable marker gene aminoglycoside 3′-phosphotransferase ( nptII/neo ; note that nptII/neo is used synonymously with amino 3′-glycosyl phosphotransferase gene conferring resistance to kanamycin, neomycin, paromomycin, ribostamycin, butirosin and gentamicin B) into the backbone of the dinoflagellate-specific expression vector DinoIII-neo 41 , which confers resistance to neomycin and kanamycin (Table 1 ). In brief, DinoIII-neo was linearized and electroporated using the Nucleofector optimization pulse codes, buffer SF/Solution I (Lonza), and 2 μg μl −1 of linearized DinoIII-neo. Electroporated cells were selected under 150 μg ml −1 kanamycin 3 d postelectroporation. Fresh seawater with kanamycin was added every 2 weeks to the cultures and new subcultures were inoculated monthly. After 3 months, DNA and RNA were isolated from the resistant cultures as previously reported 42 and cDNA was synthesized using random hexamers. Out of 16 transformations, two cell lines (CA-137, DS-138) showed stable growth under kanamycin selection. CA-137 developed dense cultures after 3 months, and the resistance gene was detected in both DNA and RNA by nested PCR and RT–PCR, respectively (Fig. 4e , Supplementary Fig. 8 and Methods ). We improved the transformation protocol 16 , 17 of P. marinus , a pathogen of marine mollusks, fish and amphibians 43 (Supplementary Table 5 ). We coexpressed two genes and efficiently selected transient and stable transformants using FACS (Table 1 , Figs. 2 and 4f , Supplementary Fig. 9 and Methods ). In addition, we established the integration profile of ectopic DNA once introduced into the P. marinus genome. We did not see evidence of integration through homologous recombination and observed a propensity for plasmid fragmentation and integration within transposable elements sites. An optimized alternative protocol for transformation using glass bead abrasion was also developed. Two versions of the previously published Moe gene promoter 16 were tested. Whereas the 1.0 kb promoter version induced expression after 2 or 3 d, the truncated version (0.5 kb) took 7 d for expression to be detected. Resistance genes to zeocin, blasticidin and puromycin have all been shown to confer resistance to transformed P. marinus ; however, selection regimes are still relatively slow and inefficient, indicating further room for improvement 17 . We also report a vector for the transformation of the A. carterae chloroplast, a photosynthetic dinoflagellate. A. carterae , like other dinoflagellates with a peridinin-containing chloroplast, contains a fragmented chloroplast genome made up of multiple plasmid-like minicircles 40 . The previous transformation protocols made use of this to introduce two vectors based on the psbA minicircle 15 . Here, we show that other minicircles are also suitable for use as vectors. We created an artificial minicircle, using the atpB minicircle as a backbone, but replacing the atpB gene with a codon-optimized chloramphenicol acetyltransferase (Table 1 and Methods ). This circular vector was introduced by biolistics to A. carterae (Supplementary Fig. 10a ). Following selection with chloramphenicol, we were able to detect transcription of the chloramphenicol acetyltransferase gene via RT–PCR (Fig. 4g ). This result suggests that all 20 or so minicircles in the dinoflagellate chloroplast genome would be suitable for use as artificial minicircles, thus providing a large pool of potential vectors. Discobans This diverse group, recently split into Discoba and Metamonada 44 , includes heterotrophs, photoautotrophs and predatory mixotrophs, as well as parasites. Discobans include parasitic kinetoplastids with clinical significance, such as Trypanosoma brucei , T. cruzi and Leishmania spp., for which efficient transformation protocols are available 45 . However, such protocols are missing for aquatic species. Here, we describe available transformation protocols for the kinetoplastid Bodo saltans and the heterolobosean Naegleria gruberi . The former was isolated from a lake, but identical 18S rRNA gene sequences have been reported from marine environments 46 . The latter is a freshwater protist that represents a model organism for closely related marine heterolobosean amoebas. Furthermore, we provide advanced methods that build on previous EMS results 18 for the diplonemid Diplonema papillatum . B. saltans (ATCC 30904) was transformed with a plasmid containing a cassette designed to fuse an endogenous EF1-α gene with eGFP for C-terminal tagging. This cassette includes downstream of eGFP , a B. saltans tubulin intergenic region followed by the selectable marker nptII/neo gene, conferring resistance to neomycin. EF1-α genes exist in tandem repeats. The homologous regions that flank the cassette were chosen as targets for inducing homology-directed repair; however, they target only one copy of the gene. As transcription in B. saltans is polycistronic 46 , insertion of the tubulin intergenic region into the plasmid is essential for polyadenylation of the EF1-α/GFP fusion and trans -splicing of the nptII/neo gene (Supplementary Table 5 ). Selection of transfected cells began with 2 µg ml −1 of neomycin added 24 h after electroporation, and this concentration was gradually increased over 2 weeks to 5 µg ml −1 (Table 1 and Methods ). Cells were washed and subcultured into fresh selection medium every 4 d, and neomycin-resistant cells emerged 7–9 d postelectroporation. The eGFP signal was detected 2 d postelectroporation, albeit with low intensity. This may be due to the inefficient translation of eGFP since it has not been codon-optimized for B. saltans (Fig. 2 ). Genotyping analysis 9 months posttransfection confirmed the presence of the nptII/neo gene and at least partial plasmid sequence (Fig. 4h and Supplementary Fig. 10b ). However, plasmid integration into the B. saltans genome through homologous recombination is still unconfirmed. This suggests either off-target plasmid integration or episomal maintenance. For N. gruberi (ATCC 30224) two plasmids were designed. The first one carried the hygromycin B resistance gene ( hph ) with an actin promoter and terminator, along with an HA-tagged eGFP driven by the ubiquitin promoter and terminator. The second plasmid carried the nptII/neo gene instead. For each individual circular plasmid, 4 μg was electroporated (Table 1 and Methods ). About 48 h after electroporation, dead cells were removed from the suspension and viable cells were washed with PBS. Afterward, 300 μg ml −1 of hygromycin B or 700 μg ml −1 of neomycin was added to the fresh media. One to 4 weeks later, resistant clones were recovered and expression of eGFP and/or hygromycin was confirmed by western blotting (Supplementary Fig. 11 ). Expression of eGFP was observed by epifluorescence microscopy (Fig. 2 and Supplementary Fig. 11 ) with ~80% of transformants maintaining hygromycin B or neomycin resistance in addition to expressing eGFP. D. papillatum (ATCC 50162) was transformed by electroporation using 3 μg of a SwaI -linearized fragment (cut from p57-V5+NeoR plasmid) containing the V5-tagged nptII/neo gene flanked by partial regulatory sequences derived from the hexokinase gene of the kinetoplastid Blastocrithidia (strain p57) (Table 1 and Methods ) using a published protocol 18 . About 18 h after electroporation, 75 μg ml −1 G418 was added to the medium and after 2 weeks, seven neomycin-resistant clones were recovered. Transcription of nptII/neo was verified in four clones by RT–PCR (Supplementary Fig. 12 ) and the expression of the tagged nptII/neo protein was confirmed in two clones by western blotting using the α-V5 antibody (Fig. 4i ). Opisthokonts The opisthokont clade Holozoa includes animals and their closest unicellular relatives choanoflagellates, filastereans, ichthyosporeans and corallochytreans. The establishment of genetic tools in nonmetazoan holozoans promises to help illuminate the cellular and genetic foundations of animal multicellularity 47 . Genomic and transcriptomic data are available for multiple representatives characterized by diverse cell morphologies, some of which can even form multicellular structures 46 . Until recently, only transient transformations had been achieved for some opistokonts such as the filasterean Capsaspora owczarzaki 48 , the ichthyosporean Creolimax fragrantissima 49 and the choanoflagellate Salpingoeca rosetta 21 . Through the EMS initiative, we report on evidence for transient transformation of the ichthyosporean Abeoforma whisleri , isolated from the digestive tract of mussels, and review a recently published stable transformation protocol for S. rosetta achieved by using the selectable puromycin N -acetyl-transferase gene (Fig. 2 ) 22 . All A. whisleri life stages are highly sensitive to a variety of methods for transformation. However, we developed a 4D-nucleofection-based protocol using 16-well strips, wherein PBS-washed cells were resuspended in 20 μl of buffer P3 (Lonza) containing 40 μg of carrier plasmid (empty pUC19) and 1–5 μg of the reporter plasmid ( A. whisleri H2B fused to mVenus fluorescent protein, mVFP ) (Table 1 and Methods ), and subjected to code EN-138 (Lonza). Immediately after the pulse, cells were recovered by adding 80 μl of marine broth (Gibco) before plating in 12-well culture plates previously filled with 1 ml marine broth. After 24 h, ~1% of the culture was transformed based on the fraction of cells expressing mVFP in the nucleus (Figs. 2 and 4j ). Microbial eukaryotes in natural planktonic communities Model organisms are typically selected based on criteria such as relative ease of isolation and asexual cultivation in the laboratory; however, these attributes may not correlate with the capacity for uptake and expression of the exogenous DNA. We explored whether natural marine planktonic pico- and nanoeukaryote communities would take up DNA in a culture-independent setting. Microbial plankton from natural seawater was concentrated and electroporated with plasmids containing mTagBFP2 under the control of CMV or 35S promoters ( Supplementary Results and Methods ). In most trials, blue fluorescent cells were rare if detected at all (compared to control samples). However, in one natural community tested, a photosynthetic picoeukaryote population exhibited up to 50% of cells with transient expression of blue fluorescence when the CMV promoter was used (Supplementary Fig. 13 ). This suggests it might be possible to selectively culture eukaryotic microorganisms based on capacity to express exogenous DNA. Discussion The collaborative effort by the EMS initiative facilitated identification and optimization of the steps required to create new protist model systems, which culminated in the synthetic transformation roadmap (Fig. 5 ). Our genetic manipulation systems for aquatic (largely marine) protists will enable deeper insights into their cell biology, with potentially valuable outcomes for aquatic sciences, evolutionary studies, nanotechnology, biotechnology, medicine and pharmacology. Successes and failures with selectable markers, transformation conditions and reporters were qualitatively compared across species (Supplementary Tables 3 and 4 , Table 1 , Figs. 2 – 4 and Methods ). For some of the selected species, the first step was to identify cultivation conditions for robust growth in the laboratory to either generate high cell densities or large culture volumes for obtaining sufficient biomass required for a variety of molecular biology experiments. Unlike established microbial model species, cultivation of marine protists can be challenging, especially under axenic conditions or for predatory taxa that require cocultivation with their prey. Nevertheless, 13 out of 35 species were rendered axenic before the development of transformation protocols. For the remaining species, we were unable to remove bacteria and therefore had to make sure that transformation signals were coming from the targeted protist rather than contaminants (Supplementary Table 2 ). Subsequent steps included the identification of suitable antibiotics and their corresponding selectable markers (Table 1 and Supplementary Table 3 ), conditions for introducing exogenous DNA (Table 1 and Supplementary Table 4 ) and selection of promoter and terminator sequences for designing transformation vectors (Table 1 , Methods , Supplementary Table 5 and Supplementary Notes 1 ). As exemplified in the model systems provided herein (Table 1 and Figs. 2 – 4 ), a variety of methods were used to test whether exogenous DNA was integrated into the genome or maintained as a plasmid, and whether the introduced genes were expressed. Approaches to show the former included inverse PCR, Southern blotting and whole genome sequencing, whereas approaches to demonstrate the latter included various combinations of PCR, RT–PCR, western blotting, epifluorescence microscopy, FACS, antibody-based methods and/or growth assays in the presence of antibiotics to confirm transcription and translation of introduced selection and reporter genes (for example, eGFP , YFP , mCherry ). For fluorescent markers, it was first ensured that the wild-type, or manipulated controls cells, had no signals conflicting with the marker (Figs. 2 and 3c ), an important step because photosynthetic protists contain chlorophyll and other autofluorescent pigments. Overall transformation outcomes for each species were parsed into three groups according to the level of success or lack thereof (A, first transformation protocol for a given species; B, advanced protocol based on previous work and C, published protocol based on the EMS initiative) and are discussed according to their phylogenetic position (Fig. 1 ). Our studies did not result in a universally applicable protocol because transformability and a range of other key conditions varied greatly across taxa and approaches, such as intrinsic features of the genome and differences in cellular structure and morphology. In general, electroporation proved to be the most common method for introducing exogenous DNA stably into cells. This approach was used for naked cells and protoplasts, yet frequently also worked, albeit with lower efficiency, on cells protected by cell walls. Linearized plasmids were most effective for delivery, and 5′ and 3′ UTR-containing promotors of highly expressed endogenous genes provided the strongest expression of selective reporters and markers. If successful, teams usually continued with fluorescence-based methods. Furthermore, large amounts of carrier DNA usually facilitated successful initial transformations (for example, M. commoda , A. whisleri ) or improved existing protocols ( S. rosetta 21 ). We also provide the contact details of all coauthors who are assigned to particular species (Supplementary Table 6 ). Some lineages were difficult to transform, especially dinoflagellates and coccolithophores. Here, even if DNA appeared to be delivered (Supplementary Table 5 ), expression of the transformed genes could not be confirmed. Examples include the dinoflagellates C. cohnii , Symbiodinium microadriaticum and the coccolithophore E. huxleyi . Thus, at least these three species need concerted future efforts. The combination of results presented herein together with previously published protocols from the EMS initiative 50 significantly expands the segment of extant eukaryotic diversity amenable to reverse genetics approaches. Out of the 39 microbial eukaryotes selected for the initiative, exogenous DNA was delivered and expressed in more than 50% of them. The transformation systems enable us to shed light on the function of species-specific genes, which likely reflect key adaptations to specific niches in dynamic ocean habitats. Methods Studied species and used transformation methods For the full list of vector sequences and maps see Supplementary Notes 1 and for detailed description of Figs. 3 and 4 see Supplementary Note 2 . Antibiotic concentrations effective for selection of transformants can be found in Supplementary Table 3 , the details of the transformation methods applied to this study in Supplementary Table 4 and contact details for individual laboratories in Supplementary Table 6 . Full list of protists (including details of culture collection) and links to the complete step-by-step transformation protocols and published vector sequences are listed in Supplementary Table 5 . The protocols.io links listed in Table 1 and Supplementary Table 5 are summarized in Supplementary Tables 7 and 8 . Reporting Summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The data that support the findings of this study are available from the corresponding authors as well as the other authors upon request (for the contacts see Supplementary Table 6 ). Source data for Figs. 3 and 4 and Supplementary Figs. 9b ,c, 11a and 12b,c are available online. Change history 15 April 2020 A Correction to this paper has been published:

Throughout the world's oceans in global nutrient cycles, food chains, and climate, as well as increasingly in human-made industrial processes, a diverse set of planktonic microbes, such as algae, play an integral role. For nearly all of these planktonic microbes, however, little is known about them genetically beyond a few marker sequences, while their morphology, biological interactions, metabolism, and ecological significance remain a mystery. Algae produce half of the oxygen in earth's atmosphere and some forms of algae are used in industrial applications—such as producing high omega-3 fatty acids for baby formula or being used for biofuels—so there are many reasons a better understanding of algae could be beneficial. There is another side to algae, however, as some species can create harmful algal blooms (HABs), and those have been the focus of the research of University of Delaware's Kathryn Coyne. Advancing the study of microbes To help advance the understanding of the cellular instructions that underpin microbial life in the sea, Coyne joined more than 100 scientists from institutions around the globe to publish a compilation of methods, or protocols, for laboratory experiments that will help scientists gain a better understanding of the genetic underpinnings of marine algae as a resource article in the journal Nature Methods. The work was funded by the Gordon and Betty Moore Foundation's Marine Microbiology Initiative. For her contribution to the collaboration, Coyne worked specifically with Heterosigma akashiwo, a species of algae that can produce HABs. One of the mysteries about H. akashiwo is that while some strains produce toxins that can kill fish, other strains are non-toxic. "We don't have a clear understanding of what kind of toxin they produce. We just know that when there are blooms of this algae in some areas of the world, they are associated with massive fish kills," said Coyne, an associate professor of marine biosciences in UD's College of Earth, Ocean, and Environment (CEOE), and director of the Delaware Sea Grant program. "We also don't know why some strains produce toxins, or what stimulates this toxin production." Scientists often use genome manipulation to better understand how microbes respond to the environment or to identify genes that may be involved in a specific response, like production of toxins. Unlike other algal species that serve as models for genome manipulations, however, H. akashiwo doesn't have a cell wall, instead having only a thin membrane that holds the cell shape. Coyne explained that having a cell wall can be an impediment to genome manipulations and that these kinds of experiments usually entail some effort initially just to remove the cell wall or make it more porous. Because H. akashiwo lacked a cell wall, Coyne and her research team proposed that genome manipulation might be more straightforward with this species, and were able to demonstrate that using a couple of gene manipulation methods that have been successful on other model species. "We created a piece of genetic material that could be introduced into Heterosigma cells that would make them resistant to a specific antibiotic," said Coyne. "If we were successful, we would be able to grow them in this antibiotic and cells that had incorporated the resistance gene would survive." Coyne worked with Deepak Nanjappa, a postdoctoral researcher in her lab who is also an author on the paper, as well as Pam Green and her lab members, Vinay Nagarajan, a postdoctoral researcher, and Monica Accerbi, a research associate in Green's lab at the Delaware Biotechnology Institute (DBI). Together, they tried a handful of methods and optimized those that were successful for Heterosigma. One method in particular was replicated successfully several times, showing that they were able to produce a genetically modified strain of Heterosigma. Using this approach, scientists can now probe the genome of Heterosigma akashiwo to gain a better understanding of how this species responds to environmental cues, or what genes are responsible for its toxicity. One of the aims of the project was to make all of the methods developed freely available so that scientists can take that information and use it in their own research. "The Moore Foundation funded this project with the expectation that all of the methods developed during this research would be published," said Coyne. "Nothing is proprietary for this project, so we can share any of the protocols that we developed for Heterosigma." Immobilizing algicidal bacteria In addition, Coyne had another paper published in the scientific journal, Harmful Algae, that detailed her work with Yanfei Wang, a doctoral student in CEOE, studying the algicidal bacterium Shewanella and how it could be used to remediate HABs. Shewanella, which is an algicidal bacterium that has been isolated from the Delaware Inland Bays, is being developed as a biological control for HABs. It secretes water-soluble compounds that inhibit the growth of dinoflagellates, single-celled organisms that often produce toxins and contribute to HABs. Other research on this species of Shewanella shows that it has no negative effects on the growth of other species of algae, or on fish or shellfish. Since it was isolated from local waters, it may be considered an "environmentally neutral" solution to controlling HABs. In order to use Shewanella in the natural environment to control HABs, there first needs to be a method to safely deploy the bacterium in areas that are at risk for HABs. To move this HAB control solution closer to reality, Coyne and Wang immobilized Shewanella into several porous materials. Funded by Delaware Sea Grant, this research determined how well each material retained the bacteria over time, and whether the immobilized form of Shewanella was effective at controlling the growth of dinoflagellates. Unlike other HAB control approaches, such as application of toxic chemicals like copper sulfate, the advantage of using immobilized algicidal bacteria is the potential for continuous control of HABs without the need for frequent reapplication. The immobilized bacteria can also be removed when it is no longer needed. This research found that an alginate hydrogel was the most successful of the porous materials used in the study, and had the best retention of Shewanella cells. This research also showed that Shewanella cells immobilized in alginate beads were as effective as free bacteria in controlling the growth of the harmful species while at the same time having no negative impacts on a non-harmful control species. Overall, the study suggests that immobilized Shewanella may be used as an environmentally friendly approach to prevent or mitigate the blooms of harmful dinoflagellates and provides insight and directions for future studies.

10.1038/s41592-020-0796-x

Chemistry

Damage-reporting and self-healing skin-like polymeric coatings

Subin Yoon et al, Mechanochromic and thermally reprocessable thermosets for autonomic damage reporting and self-healing coatings, NPG Asia Materials (2022). DOI: 10.1038/s41427-022-00406-3

https://dx.doi.org/10.1038/s41427-022-00406-3

https://phys.org/news/2022-08-damage-reporting-self-healing-skin-like-polymeric-coatings.html

Abstract Autonomous polymers that report damage prior to loss of function and simultaneously self-heal are highly relevant for preventing catastrophic failures and extending the lifetimes of materials. Here, we demonstrate mechanochromic and thermally reprocessable thermosets that can be used for autonomic damage reporting and self-healing coatings. A mechanochromic molecule, spiropyran (SP), is covalently incorporated into thermoreversible Diels–Alder (DA) cross-linking networks. Mechanical activation of SPs in DA networks is confirmed by computational simulations and mechanical testing. The damaged areas of the polymers change colour, emit fluorescence signals, and completely recover after heat treatment. Because of the thermoreversible covalent networks, these polymers can be recycled up to fifteen times without degrading their mechanical, damage-reporting, or self-healing properties. Our autonomic material systems provide a new way to enhance the lifespans and reliabilities of thermosetting coatings, which also expands the range for practical applications of force-induced chemical reactions in polymers. Introduction Thermosets are chemically cross-linked polymer networks formed by irreversible chemical reactions of soft solids or viscous liquid prepolymers. Owing to their superior mechanical properties, high thermal stabilities, and outstanding chemical resistance, thermosets are used in a wide range of applications, including coatings, adhesives, composites, electronic packaging, etc. 1 , 2 . As a key element in the modern plastic industry, thermosets comprise 20% of total polymer production today, with worldwide annual production of more than 65 million tons 3 . Although permanent cross-links endow thermosets with desirable properties, their high stability precludes repair, reshaping, and reprocessing of thermosets even at high temperatures; hence, they are among the most difficult materials to recycle 4 . With increased interest in sustainability and environmental responsibility, dynamic and exchangeable chemical reactions have been investigated for developing reprocessable thermosets known as covalent adaptable networks (CANs) 5 , 6 , 7 , 8 . This type of polymer behaves like a classical thermoset under ambient conditions, but the network topology can be rearranged repeatedly in response to external stimuli. Since every reaction rate is dependent on temperature, thermal treatment is a universal stimulus used to trigger or control dynamic chemical reactions 9 , 10 , 11 . In particular, thermoreversible Diels–Alder (DA) reactions have been employed for syntheses of CANs due to their relatively fast kinetics and catalyst-free mild reaction conditions 12 , 13 . The DA reaction involves a [4 + 2] cycloaddition between an electron-rich diene and an electron-poor dienophile to form a stable cyclohexene derivative. The use of furan as the diene and maleimide as the dienophile constitutes an excellent combination for CANs since this DA reaction has relatively low coupling and high decoupling temperatures 14 . Specifically, multifunctional maleimides and furans have been used as cross-linking agents with several furan- and maleimide-containing polymers, respectively 15 , 16 . Therefore, we chose to use DA chemistry as a thermoreversible cross-linking tool for self-healing and reprocessable thermosets. Autonomous polymers that report damage prior to loss of function and simultaneously self-heal are highly relevant for preventing catastrophic failures and extending the lifetimes of materials, thus ultimately keeping us on the path to sustainability 17 , 18 . In particular, damage visualization has been achieved by incorporating mechanochromic molecules that change optical properties in response to mechanical stresses into polymeric materials. Spiropyran (SP) is a well-studied mechanochromic molecule that undergoes a force-induced reversible 6π electrocyclic ring-opening reaction to produce a merocyanine (MC) form 19 . Different strategies have been used to synthesize SP-linked mechanochromic polymers: Homo and block copolymers, including polyacrylates 20 , 21 and polystyrenes 22 , 23 , were prepared by using SP as a bifunctional initiator via atom transfer radical polymerization (ATRP). Dihydroxy-terminated SPs were incorporated into polyurethanes 24 , 25 , polyesters 26 , 27 and polycarbonates 28 through step-growth or ring-opening polymerization. SP-linked silicone rubbers were fabricated using bis-alkene functionalized SP as a cross-linker 29 , 30 . However, there has been no attempt to integrate SP molecules into CANs, especially those prepared with DA chemistry. Since some DA isomers undergo force-induced retro-DA (rDA) reactions 31 , SP-incorporated DA networks would be useful for studying how two different force-sensitive molecules can be mechanically activated in a single polymeric system. Here, we demonstrate self-healing and reprocessable thermosets that indicate damage with optical changes involving mechanochromic SPs and thermoreversible DA chemistry (Scheme 1 ). Linear random copolymers containing furfuryl functionalities are synthesized from bifunctional SP initiators through ATRP. We discuss in detail how polymerization kinetics and comonomer compositions are influenced by reaction conditions. Then, tris-maleimide cross-linkers are mixed with SP-linked linear polymers to prepare a thermally reversible cross-linked network. When the thermoset is mechanically deformed or damaged, the affected areas change colour and emit fluorescence during the SP to MC transition. Upon thermal treatment, the coloured MC reverts to the colourless SP, and reversible DA–rDA reactions reorganize the polymer network, which results in self-healing and reprocessing. Mechanical activation of SPs in DA polymeric networks is investigated with both ab initio steered molecular dynamics simulations and mechanical testing. We also evaluate the self-healing abilities and thermomechanical properties of mechanochromic thermosets after multiple recycling processes. Scheme 1 Mechanochromic, self-healing and reprocessable thermosets based on SP mechanophores and Diels–Alder adducts. Full size image Materials and methods Materials Unless otherwise states, all reagents were purchased from commercial source and used as received. Lauryl methacrylate (LMA) (96%) and furfuryl methacrylate (FMA) (97%) were passed through a column of basic alumina before use to remove the inhibitors. Copper(I) bromide (98%) was purified by stirring with glacial acetic acid, rinsed with ethanol, and dried under vacuum. Syntheses of mechanically active spiropyran (SP), control spiropyran (Ctrl) and cross-linkers SP, Ctrl and tris-maleimide cross-linkers were prepared according to literature procedures or with some modification. Details of their syntheses and characterization data are provided in the Supplementary material. SP-linked poly(LMA- co -FMA) ( P ) and Ctrl-linked poly(LMA- co -FMA) ( C M3 ) Nine types of SP-linked copolymers ( P ) containing SP mechanophores near the chain midpoint were synthesized by changing the molecular weights and FMA molar contents. Ctrl-linked copolymer ( C M3 ) was also prepared, and it had a chemical composition and molecular weight similar to those of M3 . The synthetic conditions, molecular weight, molar content of FMA incorporated, and glass transition temperature ( T g ) are summarized for each sample in Table 1 . The ranges of molecular weights for L , M and H were 13 k–19 k, 34 k–38 k and 67 k–77 k, respectively. The ranges of FMA molar contents for 1 , 2 and 3 were 2.6–3.0 mol%, 7.1–9.1 mol% and 19–29 mol%, respectively. Table 1 SP-linked or Ctrl-linear copolymers (P or C M3 ) synthesized in this study. Full size table The polymerization procedure used for H3 proceeded as follows. LMA (2.8 ml, 9.6 mmol, 320 equiv.), FMA (0.29 ml, 1.9 mmol, 64 equiv.), SP or Ctrl (20 mg, 0.03 mmol, 1 equiv.), CuBr (4.3 mg, 0.03 mmol, 1 equiv.), and copper wire (3 cm) were accurately weighed and transferred to a 50 ml Schlenk flask. After several vacuum and Ar purging cycles, toluene (9.3 ml) and PMDETA (6.3 μl, 0.03 mmol, 1 equiv.) were added to the flask. To remove oxygen, we used the freeze−thaw method three times. The polymerization was started by heating the solution to 80 °C under an argon atmosphere. Aliquots were periodically taken with a syringe to determine the reaction conversion and monitor the change in molecular weight. Upon completion of the reaction, the flask was opened to air, and the viscous solution was diluted with THF. The resulting solution was passed through a column of aluminium oxide to remove the copper catalyst, and then the produced was precipitated by addition to methanol. Similar polymerization procedures were adopted for synthesizing different SP- or Ctrl-linked copolymers. Detailed information on each sample is included in the Supplementary material. Cross-linked polymer using the Diels–Alder reaction ( xP and xC M3 ) By mixing P or C M3 with different amounts of tris-maleimide cross-linker, cross-linked polymers ( xP or xC M3 ) were prepared. The method used for preparation of xL1 with a furan-to-maleimide molar ratio of 1 proceeded as follows. L1 (0.5 g, 0.037 mmol, moles of furan in the polymer = 0.056 mmol) and tris(2-maleimidoethyl)amine (7.3 mg, 0.019 mmol, moles of maleimide in the cross-linker = 0.056 mmol) were dissolved in chloroform (10 wt%) in a 5 ml vial. The mixture was stirred for 1 h and poured onto a Teflon plate. The solution was dried in a vacuum oven and further cured at 50 °C for 12 h, resulting in cross-linked polymers. The dried materials were moulded between parallel stainless-steel plates via compression moulding. A pressure of 4 tons and a temperature of 140 °C were applied to the plates for 30 min. The final sheet was cut into bar-shaped specimens (15 × 5 × 0.7 mm). Similar procedures were adopted to prepare other cross-linked polymers (from xL2 to xH3 and xC ). Simulation methods We carried out ab initio steered molecular dynamics (AISMD) simulations for five different molecules: three single mechanophore molecules (SP, endo -DA adduct, and exo -DA adduct) and two compound molecules (SP- endo -DA compound and SP- exo -DA compound) (Fig. S1a ). A force ranging from 0.7 to 3.5 nN was exerted on the molecule for 10 ps at 300 K. We turned on the external force at the beginning of the simulations and maintained the magnitude of the external force during the simulations. We conducted independent AISMD simulations with different applied external forces to investigate the effects of the external forces. In the simulations, constant forces were exerted on two atoms (A1 and A2 in Fig. S1b ) to which the polymer chains would be connected. Each force was set to be directed towards a fixed point (P1 and P2 in Fig. S1b ). The fixed points were located on the line that passed through the two atoms (A1 and A2). The distance between the point and the atom (A1-P1 and A2-P2) was five times longer than the distance between the atoms (A1-A2). The simulations were performed with the canonical ensemble (NVT) using a Langevin thermostat with a friction constant of 5 × 10 12 s −1 . We employed TeraChem software with the B3LYP functional and 6–31G* basis set 32 . Ten simulations with different initial configurations and velocities were conducted for each kind of molecule. Note that the magnitude of the external force was fixed during each simulation trajectory. We found from our simulations that only two types of bond cleavage reaction occurred within our simulation times: scission of a CO bond of SP and two CC bonds of the DA adduct (each bond is highlighted in Fig. S1a ). We calculated the bond lengths of the CO and CC bonds ( l CO and l CC , respectively) to examine the bond cleavage reactions. For CC bonds, the lengths of two CC bonds were averaged. A bond cleavage reaction was defined when the distance between two atoms was stretched to more than 2 Å. For the CC bonds of DA adducts, we decided that bond cleavage occurred when both CC bonds broke. For both CO and CC bonds, their bond lengths fluctuated within a range 1 to 1.5 Å before they were broken; therefore, 2 Å seems to be arbitrary but should be sufficiently large to ensure that bond cleavage had occurred. Since CO bond breaking of SP constitutes a ring-opening reaction, the l CO after bond cleavage was ~4 Å (inset of Fig. 1a ). For the CC bond, however, the DA adduct split into two molecules after bond cleavage. Therefore, l CC kept increasing and varied from 4 to 12 Å (inset of Fig. 1b ) during the simulations. Fig. 1: Ab initio steered molecular dynamics simulations of SP and DA adduct mechanophores. Fractions of bond cleavage reactions for a the CO bond of SP ( Φ CO ) and b the CC bonds of the DA adduct ( Φ CC ) for a given external force strength ( F ext ). The inset shows the lengths ( l CO and l CC ) of those bonds averaged over the last 400 fs of the simulation. Full size image Characterization 1 H NMR and 13 C NMR spectra were recorded on a Bruker Avance III 400 MHz NMR spectrometer with CDCl 3 as the solvent. Spectra were referenced to the residual chloroform solvent peaks ( 1 H NMR: 7.24 ppm, 13 C NMR: 77.23 ppm). Gel permeation chromatography (GPC) measurements were performed on the polymer samples with an Agilent 1260 Infinity II equipped with a 1260 Infinity II refractive index detector (RID) and two PLgel 10 μm MIXED-B columns with a prefilter. Tetrahydrofuran (inhibitor free, HPLC grade, Tedia) was used as the eluent with a flow rate of 1 ml/min. The molecular weights were calibrated by using monodisperse polystyrene standards. Fourier transform infrared (FT-IR) spectra were recorded using a Thermo Fisher Scientific Nicolet 380 spectrometer. The spectra were recorded over the range 4000–650 cm −1 . Optical and fluorescence images were acquired with a Leica DMI6000B camera. Differential scanning calorimetry (DSC) analyses were performed with a Perkin Elmer DSC 4000. The samples were measured over two heating runs from −80 °C to 200 °C with a heating rate of 10 °C/min under a nitrogen atmosphere. Dynamic mechanical analyses (DMA) were performed on a Q800 system from TA Instruments in tension mode and with an amplitude of 0.1% at a frequency of 1 Hz. Storage and loss moduli were measured as a function of temperature from −50 °C to 120 °C at a heating rate of 5 °C/min. Uniaxial tensile tests were also performed on the Q800 system from TA Instruments. Bar-shaped specimens (15 × 5 × 0.7 mm) were subjected to a preload of 0.01 N at a strain rate of 0.05 s −1 . At least three samples were tested. Stress relaxation tests were performed on an MCR 302 instrument from Anton Paar in the 25 mm parallel plate geometry. A normal force of 0.5 N was used in compression mode, and a strain of 1% was applied to the material. Relaxation moduli, G( t ), were measured over time at constant temperature (100, 110, 120, 130 and 140 °C). Results and discussion Ab initio steered molecular dynamics simulations Ab initio steered molecular dynamics (AISMD) simulations were initially used to investigate the mechanochemical reactivities of multiple mechanoresponsive residues, SP and DA adducts, on an atomistic scale (Fig. 1 , Movie S1 ). AISMD is a useful approach for analysing the nonequilibrium reaction dynamics of mechanically induced chemical reactions 33 , 34 , 35 , 36 , 37 . To reduce computational costs, we prepared artificial molecules with two mechanoresponsive residues connected directly to each other and compared the reaction dynamics of the compound molecules with those of the single mechanophore residues. In addition, the effects of DA adduct stereochemistries, endo and exo isomers, on reaction dynamics were examined for both the compounds and single residues. From 10 simulations with different initial configurations and velocities for each molecule, we calculated the probability of the bond cleavage reaction by counting the number of trajectories in which a cleavage reaction occurred. Φ CO and Φ CC denote the fractions of cleavage reactions for the CO bond in SPs and for both CC bonds in DA adducts, respectively. The lengths ( l CO and l CC ) of those bonds were also averaged over the last 400 fs of the total simulation time of 10 ps. Consistent with previous results for constrained geometries simulated by external force (CoGEF) calculations by Klein et al. 38 , cleavage of the CO bond in a single SP residue occurred with a relatively small external force (1.1 nN) compared to those for reactions of the CC bonds in the DA adduct (3.1 nN for the endo -DA adduct and 3.2 nN for the exo -DA adduct). In compound molecules with multiple mechanophores, however, no CO bond was broken with the same force strength, and all ten trajectories for a single SP exhibited cleavage of the CO bond. A force of at least 1.5 nN was required to completely break the CO bonds (Φ CO = 1.0) in the SP- exo -DA and SP- endo -DA compounds. Cleavage of the CC bonds in DA adducts occurred only when the external force exceeded 2.5 nN. Therefore, we anticipate that SP is mechanically activated prior to DA adduct cleavage within a certain range of external forces, resulting in changes in optical properties due to the SP-to-MC transition. Upon application of a high external force sufficient for mechanical activation of both SP and DA, interesting behaviour was observed for a specific stereoisomeric compound, SP- exo -DA. For example, the CO bond of the SP- exo -DA compound was not cleaved in 20% of the trajectories (Φ CO = 0.8) at F ext = 2.8 nN. For the trajectories in which the CO bond did not break even with F ext = 2.8 nN, we found that breakage of the CC bonds in DA residues had already occurred; the CO bond did not break within our simulation times. On the other hand, in the trajectories where the CO bond broke at F ext = 2.8 nN, cleavage of the CC bond and the CO bond occurred almost simultaneously. In the 20% of trajectories with no CO bond cleavage, the external force might not have been transmitted efficiently to the CO bond of the SP residue due to early cleavage of the CC bonds. This behaviour was not observed for the SP- endo -DA compound. We suggest that the sequence of reactions for mechanophores is dependent on the molecular structure, including the stereochemical structures of complex polymer networks with multiple mechanophores. SP-linked linear copolymers Encouraged by the simulation results, we first used ATRP to synthesize SP-linked copolymers comprising lauryl methacrylates (LMA, M 1 ) and furfuryl methacrylates (FMA, M 2 ). The monomer concentration affected the polymerization kinetics and dispersity ( Ð ) (Fig. 2a ). Upon increasing the monomer concentration, polymerization was accelerated (Fig. S2a ), but the kinetics deviated from first-order behaviour, and Ð was increased (Table S1 ). We optimized the total monomer concentration ([M tot ] 0 = [M 1 ] 0 + [M 2 ] 0 ) by adjusting the amount of solvent. In contrast, the LMA-to-FMA ratio ([M 1 ] 0 :[M 2 ] 0 ) did not influence the polymerization behaviour: the living nature and the rate of polymerization were maintained (Fig. 2b , Fig. S2b ). The monomer composition of the synthesized copolymer was calculated with 1 H NMR spectroscopy by using the integrated areas of the signals at 4.9 ppm (-OCH 2 - protons of FMA) and 3.9 ppm (-OCH 2 - protons of LMA). Since FMA tends to have a reactivity ratio higher than those of alkyl methacrylates 39 , 40 , more FMA units were incorporated into the copolymer from the monomer feed except when the [M 1 ] 0 :[M 2 ] 0 was 30:1 (Table 1 ). Nine types of SP-linked copolymers ( P ) were synthesized, and their NMR, GPC and DSC data are included in the Supplementary material (Figs. S3 – S11 ). Although the FMA content was increased up to 23 mol%, the glass transition temperature ( T g ) of the polymer was not changed significantly, as summarized in Table 1 and Fig. S12 . By increasing the molecular weights of the polymers, higher T g values were obtained. However, all the linear copolymers had T g values lower than −46 °C. Fig. 2: Living radical polymerization kinetics. Kinetic plots of monomer loss vs . time measured by a varying the initial total monomer concentration ([M tot ] 0 ) at a fixed comonomer ratio ([M 1 ] 0 :[M 2 ] 0 = 5:1) and b varying the comonomer ratio ([M 1 ] 0 :[M 2 ] 0 ) at [M tot ] 0 = 0.93 M ( k p app ~ (1.3 ± 0.1) × 10 –4 s −1 , R 2 ~ 0.99). k p app is the apparent polymerization rate, which was calculated with a first-order kinetic equation. Full size image Thermoreversible SP-linked CANs SP-linked CANs were prepared by dissolving P and cross-linkers in organic solvents, followed by solution casting on a heated glass plate. Formation of the DA adducts was confirmed by their FT-IR spectra (Fig. 3a ). After mixing with tris-maleimide cross-linkers, characteristic peaks were observed, including those for the C-H bonds of C=C (700 cm −1 ) and C=O carbonyl groups (1700 cm −1 ) in the maleimide rings ( L3 + cross-linker before and after DA) 41 . When the DA reaction was completed ( xL3 ), new absorption peaks appeared for C–O–C ether stretching vibrations (1070 cm −1 ) and C=C double bond stretching vibrations (1774 cm −1 ) of the DA adducts. Fig. 3: Formation of thermoreversible SP-linked CANs. a FT-IR spectra of SP-linked linear copolymers ( L3 , black solid line), L3 with cross-linkers immediately after mixing (before DA reaction, red solid line), and after the DA reaction ( xL3 ) was completed (blue solid line). b DSC results for xL3 obtained by varying the molar ratio of the cross-linkers. c Reversible solubility changes of SP-linked CANs in toluene caused by varying the temperature. Full size image DA adducts were thermally cleaved through the rDA reaction, the reverse reaction that formed the initial linear copolymers at temperatures above 100 °C 42 , 43 . When we measured the DSC thermogram of xL3 , two endothermic peaks were observed during the first heating cycle (Fig. 3b ). These peaks were centred at ~120 °C and 150 °C, indicating rDA reactions of the thermally less stable endo and more stable exo adducts. The heat of the endothermic reaction (Δ H ) increased with the amount of cross-linker used, and the value was constant at 24.0 J/g until the molar ratio of cross-linker to P reached 1.0. Therefore, we prepared fully cross-linked samples by adding one equivalent of maleimide derivative per furan moiety. Solubility tests were conducted to confirm the thermoreversible cross-linking and retro cross-linking reactions (Fig. 3c ). The cross-linked polymers were insoluble in toluene at room temperature but dissolved at 110 °C since both linear polymers and cross-linkers are soluble in toluene. This is because the rDA reaction occurred above 100 °C according to our DSC result. When the temperature was maintained at 50 °C for 12 h, the DA reaction proceeded, and cross-linked polymers precipitated from the solution. This solubility change with temperature was reversible. Autonomic damage-reporting and self-healing coatings Initially, we used continuous irradiation with ultrasound to look for mechanical activation of SP in linear copolymers in solution. A difunctional control SP (Ctrl) was prepared that could not transfer mechanical forces from polymer chains to the spiro C–O bonds (Fig. S13a ). Linear copolymers with similar compositions and molecular weights were prepared from both the SP and Ctrl functional groups. Only the SP-linked polymers ( M3 ) exhibited a visible pink colour after ultrasonication, indicating mechanochemical ring opening of the SP. Ctrl-linked polymers ( C M3 ) did not change their colours after application of force, but photochromic behaviour was maintained (Fig. S13b, c ). Cross-linked polymers ( xM3 and xC M3 ) were prepared from M3 and C M3 by mixing them with tris-maleimide cross-linkers and coating them on glass plates to demonstrate force-induced colour changes in solids. When we scratched each sample with a blunt object, only the xM3 samples showed a purple colour at the damaged area (Fig. 4a ). Irradiation with 365 nm UV light changed the colours of both xM3 and xC M3 , indicating that SP and Ctrl were photochromic. Mechanical activation of SP was also confirmed with fluorescence microscopy (Fig. 4b ). Strong red fluorescence signals were detected at the scratched areas, suggesting that the SP-to-MC transition was activated by force. The self-healing test was conducted above 150 °C, at which temperature the rDA reaction proceeded for the exo adduct. After thermal treatment at 160 °C for 1 min, the damaged regions self-healed, and the MC reverted to SP: the purple colour and red fluorescence disappeared. The damage-reporting and self-healing behaviours were repeated multiple times with the same specimen, which makes them especially suitable for use with coatings and related applications. For these kinds of materials, scratch-healing tests provided useful information on the healing efficiency, and the area of the damaged surface area was measured before and after the healing process 44 , 45 . Using nine types of cross-linked samples (from xL1 to xH3 ), we evaluated the healing efficiencies and measured the times required for recovery from damage. In most cases, the sample was self-healed by over 90% within a minute, and this was not dependent on either molecular weight or FMA composition. This is because all of the linear polymers had T g values well below the temperature required for the rDA reaction. Fig. 4: Autonomic damage-reporting and self-healing capabilities of xM3 samples. a Digital images of SP-linked and Ctrl-linked cross-linked polymers ( xM3 and xC M3 ) in the pristine, scratched, and UV-activated states (scale bar = 2 mm). b Optical and fluorescence images of scratched samples and self-healed samples (after thermal treatment at 160 °C for 1 min). c Mechanochromic and self-healing coatings on diverse substrates (glass, steel and wood) (scale bar = 5 mm). Full size image Our mechanochromic and self-healing polymers were coated on diverse substrates, including glass, steel, and wood, by using solvent casting (Fig. 4c ). To avoid any wetting issues, we chose proper solvents depending on the substrates. Once the materials were fully cured, the coated areas were scratched using tweezers. The scratched regions turned purple and completely healed within a minute after thermal treatment (Movie S2 ). We anticipate that our polymers can be applied in versatile coating materials that require self-reporting and self-healing capabilities. Thermomechanical properties and reprocessability We analysed the thermomechanical properties of xP by using dynamic mechanical analyses (DMA) in tension mode (Fig. 5a, b ). With increasing FMA molar content (from 1 to 3 ), the storage moduli ( E ′) in the glassy and rubbery regions increased while the height of the loss tangent (tan δ ) decreased, implying that the material became more elastic due to the increased number of cross-linking sites (Fig. 5a , Fig. S14 ). T g values, as determined from the peak of tan δ , for xH1 (0 °C), xH2 (23 °C) and xH3 (50 °C) increased as well. Below the temperature required for the rDA reaction (120 °C), a single glass transition was observed that became broader with higher FMA content, implying an increase in the inhomogeneity of the matrix. We could not find any trends for E ′ and tan δ when the molecular weight of xP was changed with constant FMA molar content (Fig. 5b , Fig. S15 ). For instance, x1 and x2 had higher moduli with lower molecular weights, while the reverse trend was observed for x3 . The peak heights and shapes of tan δ were similar, but T g decreased with increasing molecular weight. Fig. 5: Thermomechanical and tensile properties of xP samples. Dynamic mechanical analyses of a xH and b x2 to determine the effects of molecular weight and FMA molar content. E ′ refers to the storage modulus, and tan δ refers to the loss tangent. c Normalized relaxation modulus (G( t )/G 0 ) for xH3 as a function of time at constant temperature. The relaxation times ( τ *) were defined when G( t )/G 0 was equal to 1/e (dashed lines). d Arrhenius plots obtained from τ * at different temperatures. The activation energy ( E a ) for the rDA reaction was calculated from a linear fit. e , f Representative engineering stress ( σ ) and strain ( ε ) curves for xP samples. Full size image The stress relaxation rates of xH3 were measured as a function of temperature to determine the activation energy ( E a ) of the rDA reaction (Fig. 5c ). The relaxation time at each temperature was obtained by using Maxwell’s model, and the value of E a was obtained with the Arrhenius equation (see the Supplementary material for equations). The value of E a was calculated as 105 kJ mol −1 , which is somewhat above the range of values previously reported (88–95 kJ mol −1 ) for rDA reactions run within a similar temperature range (Fig. 5d ) 46 . This increase might be caused by the difficult diffusion of bulky tris-maleimide cross-linkers in viscous polymeric media. The engineering stress ( σ ) and strain ( ε ) responses of xP were characterized under the application of uniaxial tension. Increasing the FMA molar content with similar molecular weights increased the brittleness of the materials, which is consistent with the DMA results: the Young’s modulus ( E ) increased and the elongation at break ( ε max ) decreased with increasing FMA molar content (Fig. 5e , Fig. S16 ). The tensile strength ( σ TS ) values for x1 and x2 were similar, but the highest value was observed for x3 . Changes in the molecular weight with a fixed FMA content also affected the tensile properties of xP (Fig. 5f , Fig. S17 ). We anticipate that there is a range of molecular weights that would show optimal mechanical properties based on the amount of FMA. Unexpectedly, we did not observe colour changes under tension for any of the xP samples, which is inconsistent with the behaviour of other SP-linked polymeric systems. Our polymers can be reprocessed several times with compression moulding, confirming the robust nature of Diels–Alder chemistry. We performed up to fifteen recycling processes with xH2 samples and measured E , σ TS and ε max with tensile testing (Fig. 6 ). With a series of t tests, we found that the differences in the stress–strain curves for each sample were not statistically significant. Furthermore, there was no difference in E , σ TS and ε max for the recycled and pristine samples at the 95% confidence level. Specifically, fifteen recycled xH2 samples ( E = 343 ± 49 MPa, σ TS = 72.7 ± 6.7 MPa, ε max = 27.6 ± 5.5%) exhibited mechanical properties that were experimentally identical to those of the pristine sample ( E = 321 ± 42 MPa, σ TS = 63.2 ± 3.8 MPa, ε max = 24.6 ± 3.0%). These specimens retained their mechanochromic and self-healing capabilities (Fig. S18 ). Fig. 6: Mechanical properties of pristine and recycled xH2 . a Representative engineering stress ( σ ) and strain ( ε ) curves for virgin, 5x recycled, 10x recycled, and 15x recycled xH2 specimens. b Elastic modulus ( E ), tensile strength ( σ TS ), and elongation at break ( ε max ) for xH2 as a function of recycling runs. Reported values and error bars represent the average and one standard deviation, respectively. Full size image Conclusions We demonstrated self-healing and thermally reprocessable thermosetting polymers that change colour and emit fluorescence in response to mechanical stimuli. With ab initio steered molecular dynamics simulations, we confirmed mechanical activation of spiropyrans (SPs) in thermoreversible and force-sensitive Diels–Alder (DA) networks. Nine cross-linked DA networks containing SP were prepared by changing the molecular weights and the cross-linking densities of linear polymers. These polymers exhibited purple colours and red fluorescence after compression or scratching. The damaged areas were self-healed, and the colour had disappeared within a minute after thermal treatment. In addition, the cross-linked polymers showed reversible solubility changes as well as thermal reprocessability. With up to fifteen reprocessing cycles, the thermosets retained their mechanical, damage-reporting, and self-healing properties. We envision that our autonomic polymers can be applied as universal coatings to enhance the lifespan and reliability of products by warning of damage with colour and then undergoing self-healing.

Skin-like polymeric coatings are applied to the surfaces of automobiles, ships, and buildings to protect them from the external environment. As it is difficult to determine whether the currently used coatings are already damaged or not, these non-reusable coatings must be regularly replaced, leading to a large amount of waste generation and high disposal costs. The Korea Institute of Science and Technology (KIST) announced that Dr. Tae Ann Kim's team at the Soft Hybrid Materials Research Center has developed a polymeric coating wherein the damaged area changes color, enabling immediate detection and high temperature self-healing. Existing studies on damage-reporting and self-healing polymeric coatings involve the use of extremely small capsules containing functional agents. However, these capsules cannot be used again for subsequent damage detection and self-healing if broken. The KIST research team has developed a thermoset polymer that can recover its original chemical structure after being disrupted by an external stimulus, thereby allowing this material to self-report damage and self-heal multiple times. In this study, a mechanochromic molecule, which changes color when an external force is applied due to a specific bond cleavage, and a thermoset polymer containing a molecule that can be separated and re-formed by temperature were synthesized. When a force is applied to a mechanochromic molecule, a certain bond is broken, thus changing into a form that can exhibit color. The damaged part of the synthesized polymeric coating changed to purple. When a temperature of 100°C or higher was applied, the material became processable and was physically healed and became colorless. The research team used molecular dynamics simulations to predict and confirm that only certain desired chemical bonds are selectively cleaved when a mechanical force is applied to yield a colored structure; the functionality was implemented by synthesizing the actual coating agent. Mechanochromic and self-healing coatings on diverse substrates. Credit: Korea Institute of Science and Technology (KIST) The novel multifunctional polymeric coating developed herein can be extensively used in automotive, marine, defense, timber, railway, highway, and aerospace industries, and can significantly contribute toward the reduction of industrial waste. In addition, it can be used as an artificial skin for robots, such as humanoids, since its functionality is similar to that of skin and it does not require an external energy source. Dr. Tae Ann Kim of KIST said, "This study reports a method for the simultaneous realization of damage detection and self-healing technology without any external agents such as capsules." He added, "However, even if repeated self-healing is possible, it cannot be used permanently. Therefore, additional research is underway to transition materials that have reached their lifespan into materials that are harmless to the environment or convert them into a re-cyclable form."

10.1038/s41427-022-00406-3

Medicine

Loss of a gene can be compensated by another gene

"Genetic compensation induced by deleterious mutations but not gene knockdowns." Nature; 13 July, 2015 (DOI: 10.1038/nature14580 Journal information: Nature

http://dx.doi.org/10.1038/nature14580

https://medicalxpress.com/news/2015-07-loss-gene-compensated.html

Abstract Cells sense their environment and adapt to it by fine-tuning their transcriptome. Wired into this network of gene expression control are mechanisms to compensate for gene dosage. The increasing use of reverse genetics in zebrafish, and other model systems, has revealed profound differences between the phenotypes caused by genetic mutations and those caused by gene knockdowns at many loci 1 , 2 , 3 , an observation previously reported in mouse and Arabidopsis 4 , 5 , 6 , 7 . To identify the reasons underlying the phenotypic differences between mutants and knockdowns, we generated mutations in zebrafish egfl7 , an endothelial extracellular matrix gene of therapeutic interest, as well as in vegfaa . Here we show that egfl7 mutants do not show any obvious phenotypes while animals injected with egfl7 morpholino (morphants) exhibit severe vascular defects. We further observe that egfl7 mutants are less sensitive than their wild-type siblings to Egfl7 knockdown, arguing against residual protein function in the mutants or significant off-target effects of the morpholinos when used at a moderate dose. Comparing egfl7 mutant and morphant proteomes and transcriptomes, we identify a set of proteins and genes that are upregulated in mutants but not in morphants. Among them are extracellular matrix genes that can rescue egfl7 morphants, indicating that they could be compensating for the loss of Egfl7 function in the phenotypically wild-type egfl7 mutants. Moreover, egfl7 CRISPR interference, which obstructs transcript elongation and causes severe vascular defects, does not cause the upregulation of these genes. Similarly, vegfaa mutants but not morphants show an upregulation of vegfab . Taken together, these data reveal the activation of a compensatory network to buffer against deleterious mutations, which was not observed after translational or transcriptional knockdown. Main Interfering with a gene’s function is a widely used strategy to decipher its role. Several different approaches have been developed over the years to achieve this goal. Yet, despite having the same goal of functional inactivation, different strategies, namely knockdown (via antisense) and knockout (via genetic inactivation), often lead to different phenotypes. These discrepancies could be caused by off-target effects of the knockdown reagents, the generation and use of hypomorphic alleles, or other and more fundamental reasons. The recent development of new genome engineering techniques, such as TAL effector nucleases (TALENs) and clustered regularly interspaced short palindromic repeats (CRISPRs), is allowing the facile generation of mutations and has revived concerns over the lack of specificity of knockdown reagents 1 , 2 , 3 . In several cases, toxicity due to off-target effects, induction of p53 (also known as tp53 ) transcription, interferon response, engagement of toll-like receptors and/or saturation of the RNA interference machinery can lead to phenotypes unrelated to the silencing of the target gene 8 , 9 . To investigate further whether toxicity effects are the main reason for the differences between genetic mutation and gene knockdown phenotypes, we analysed the EGF-like-domain, multiple 7 ( egfl7 ) gene. The egfl7 gene is a good candidate to address this question because of the lack of obvious phenotypes in the mouse mutants 10 , 11 and the severe vascular tube formation defects observed in knockdown experiments in zebrafish, frogs and human cells 12 , 13 , 14 . We first generated egfl7 mutants using TALENs 15 targeting exon 3, which encodes part of the EMI domain ( Fig. 1a and Extended Data Fig. 1 ). This domain precedes other domains critical for Egfl7 activity, including EGF domains and the leucine–valine-rich carboxy (C) terminus ( Fig. 1a ) 16 . We identified several deletion alleles including a Δ3 and a Δ4 ( Fig. 1b ). The egfl7 Δ3 (hereafter egfl7 s980 ) allele encodes a protein that lacks a non-conserved proline at position 50 (p.P50del) while the egfl7 Δ4 allele (hereafter egfl7 s981 ) is predicted to encode a truncated polypeptide containing a stretch of 29 incorrect amino acids starting with a Gln to Leu substitution at position 49 (p.Gln49Leufs*30) ( Fig. 1b ). To investigate the severity of these mutant alleles, we first examined egfl7 transcript levels by quantitative PCR (qPCR). The premature stop codon in egfl7 s981 led to a decrease of approximately 50% in transcript levels compared with wild-type (WT) and egfl7 s980 mutant embryos, indicating an increased messenger RNA (mRNA) degradation rate ( Extended Data Fig. 2a ). To characterize the different egfl7 mutant alleles further, we cloned the egfl7 WT, s980 and s981 complementary DNAs (cDNAs) in a mammalian expression vector and transfected HUVEC cells. Unlike Egfl7 WT and Egfl7 s980 , the Egfl7 s981 protein was mostly absent in the medium or the cells, suggesting that this truncated polypeptide is rapidly degraded and/or poorly translated and secreted ( Extended Data Fig. 2b ). Altogether, these data indicate that egfl7 s981 is a severe mutant allele, possibly even a null. Figure 1: Generation of zebrafish egfl7 mutant alleles and sporadic brain haemorrhage in mutant larvae. a , Top: egfl7 consists of 9 exons and mir126b is embedded in intron 6. The protein is encoded by exons 2–9 (grey boxes). a , Bottom: Egfl7, 277 amino acids (aa) long, contains a signal peptide (blue), an EMI domain (yellow), an EGF domain that contains a Delta-Serrate-LAG-2 (DSL) motif (orange) and a Ca 2+ -binding EGF domain (red). b , Top: the egfl7 s980 lesion (Δ3) leads to the deletion of proline at position 50 (p.P50del). b , Bottom: the egfl7 s981 allele (Δ4) encodes a truncated 77-amino-acid-long polypeptide (p.Gln49Leufs*30) that contains a signal peptide (blue) and a partial EMI domain (yellow) followed by a frameshift leading to a premature stop codon. c , d , Brightfield micrographs of 72 hpf WT and egfl7 s981/s981 larvae in lateral and ventral views. Arrows point to area of haemorrhage. e , f , Confocal micrographs of 72 hpf Tg(kdrl:HRAS-mCherry) WT and egfl7 s981/s981 larvae in lateral and dorsal views. PowerPoint slide Full size image To analyse Egfl7 function during vascular development, the egfl7 s981 mutant fish were crossed into the Tg(kdrl:HRAS-mCherry) 17 and Tg(kdrl:GFP) 18 backgrounds. We also developed a robust method based on high-resolution melt analysis to identify the different genotypes ( Extended Data Fig. 1b ). Surprisingly, no differences in gross morphology were evident between egfl7 WT and mutant animals. However, a sporadic onset of brain haemorrhage was observed in fewer than 5% of the mutant animals at 72 hours post-fertilization (hpf) ( Fig. 1c, d ). Besides the haemorrhagic foci, no obvious abnormalities were detected in vasculogenesis, angiogenesis or circulation in any region of the brain or the rest of the body ( Fig. 1e, f and Extended Data Fig. 3 ). Moreover, egfl7 s981 mutant animals survive to become fertile adults. In summary, while egfl7 morphants exhibit severe vascular defects 12 , egfl7 mutants exhibit very mild, if any, phenotypes. This discrepancy between mutant and morphant phenotypes could be explained by several reasons including morpholino (MO) off-target effects. We thus sought to assess the specificity and toxicity of egfl7 MO. First, to evaluate the effectiveness of the egfl7 MO, we engineered the egfl7 locus through the co-injection of TALENs and a single-stranded DNA (ssDNA) donor encoding a Myc-tag ( Extended Data Fig. 4 ), and generated a stable transgenic line. We then injected Egfl7 Myc-tag embryos with 1 ng of egfl7 MO and analysed protein levels by western blot at 24 hpf. The relative expression of Egfl7 Myc-tag was reduced by approximately 80% in the morphants compared with uninjected embryos, revealing the ability of the MO to inhibit egfl7 mRNA translation. A widely recognized MO off-target effect is the transcriptional activation of p53 (ref. 9 ). We thus measured p53 expression by qPCR and observed no significant difference between embryos injected with 1 ng of MO and uninjected embryos. However, p53 expression was clearly induced in embryos injected with 2 or 4 ng of MO ( Extended Data Fig. 5 ). We next reasoned that if the egfl7 MO did not induce off-target effects, it should not cause defects in egfl7 null mutants. Thus, we injected embryos obtained from egfl7 s981/+ incrosses with 1 ng of egfl7 MO. We subsequently selected and genotyped 32 embryos that showed a vascular phenotype, namely intersegmental vessel defects, reduced circulatory loop and/or pericardial oedema ( Fig. 2a ). Notably, we found that these embryos did not follow the Mendelian pattern observed in controls: 17 embryos were WT, 12 heterozygous and only 3 mutant ( Fig. 2b, c ), suggesting that the egfl7 mutants were less sensitive than WT to MO injections. Confocal micrographs of WT, heterozygous and mutant embryos injected with 1 ng of egfl7 MO ( Fig. 2a ) support this hypothesis. To investigate why some mutant embryos showed a phenotype when injected with egfl7 MO, we repeated this experiment using a lower MO dose (0.5 ng). This experiment resulted in a clear reduction of the number of mutants in the selected population (1 mutant, 20 heterozygous and 21 WT fish out of 42 selected for vascular abnormalities, P < 0.0001). In the same experiment, out of ten WT-looking embryos, eight were mutant and two heterozygous for egfl7 s981 ( P = 0.0003) (data not shown), supporting the observation that the mutant fish are less sensitive to egfl7 MO and indicating that the egfl7 MO has minimal off-target effects at these concentrations. Figure 2: Zebrafish egfl7 mutant embryos are less sensitive to egfl7 morpholino injections. a , Confocal micrographs of Tg(kdrl:GFP) WT, egfl7 s981/+ and egfl7 s981/s981 48 hpf embryos injected with 1 ng of egfl7 MO (AS -47 ) in lateral views. b , High-resolution melt analysis genotyping example of 32 embryos (from an egfl7 s981/+ incross) selected for vascular defects at 48 hpf, showing the melting curves of 17 egfl7 WT (green), 12 egfl7 s981/+ (blue) and 3 egfl7 s981/s981 (red) embryos. c , Genotype distribution (at 48 hpf) of egfl7 s981/+ incross progeny injected with 1 ng of egfl7 MO at the one-cell stage and subsequently selected for the vascular phenotypes (independent experiments (Exp) 1, 2 and 3) or randomly selected (Ctrl). The population of randomly selected embryos follows the expected Mendelian ratio, but the phenotype-selected populations show significant skewing towards egfl7 WT. P value represents two-tailed value for χ 2 test with two degrees of freedom; n = 32 embryos genotyped in each experiment. Note that the egfl7 s981/+ embryos are also underrepresented in the phenotype-selected populations (corresponding P values for experiments 1, 2 and 3 are 0.0033, 0.0066 and 0.032, respectively). PowerPoint slide Full size image To investigate the differences between the mutant and morphant phenotypes further, we used an alternative knockdown approach and took advantage of the recently developed CRISPR interference (CRISPRi) technology 19 to inhibit egfl7 transcript elongation. We designed two guide RNAs (gRNAs) to target the non-template strand of the 5′ untranslated region (UTR) and exon 2 of egfl7 as well as one gRNA targeting the template strand of exon 2 (negative control) ( Extended Data Fig. 6a ). The relative egfl7 expression levels were then quantified at 20 hpf using qPCR on pools of ten embryos injected with gRNAs and catalytically inactive (dead) CAS9 (dCAS9). Non-template gRNAs were able to inhibit egfl7 transcript levels by approximately 60% compared with uninjected or template gRNA-injected embryos ( Extended Data Fig. 6b ). Tg(kdrl:GFP) embryos injected with gRNAs and dCAS9 exhibited different degrees of vascular abnormalities at 48 hpf, including intersegmental vessel defects, reduced circulatory loop and pericardial oedema after non-template but not template gRNA injections ( Extended Data Fig. 6c ). Altogether, these data show that transcriptional or translational knockdown of egfl7 can lead to severe cardiovascular phenotypes while a severe genetic lesion does not. To identify molecules underlying the different phenotypes observed in mutants versus morphants, we performed mass spectrometry and RNA profiling analyses in egfl7 WT, homozygous mutant ( egfl7 s981 ) and morphant embryos at 24 hpf. We assessed the proteomes by 4 h ‘single shot’ liquid chromatography–tandem mass spectrometry (LC–MS/MS) and identified more than 6,000 proteins with high reproducibility ( r > 0.90 for biological and technical replicates between mutants and WT; Extended Data Fig. 7 ). To identify significant differences in individual protein expression, we used randomization-based false detection rate (FDR) estimation for multiple-testing correction and identified only one protein differentially expressed between mutants and WT ( Fig. 3a ). Strong upregulation (more than fivefold) was found for Emilin3a, suggesting its possible role in compensation. Additionally, we found no significant upregulation of Emilin3a in morphants compared with WT ( Fig. 3b ; Extended Data Fig. 8a ), further highlighting Emilin3a as a possible compensating protein. Moreover, RNA sequencing (RNA-seq) and qPCR analyses indicated that not only emilin3a but also emilin3b and emilin2a were upregulated in mutants but not in morphants or CRISPRi injected embryos ( Fig. 3c ; Extended Data Fig. 8b ). Interestingly, all these proteins contain an EMI domain, one of the key units of Egfl7 function 16 , and, like Egfl7, can regulate elastogenesis 20 , 21 . We then reasoned that if Emilins are able to functionally replace Egfl7, they might rescue egfl7 morphants. Embryos were injected with egfl7 MO or co-injected with egfl7 MO and egfl7 , egfl7 s981 , Emilin2 or Emilin3 mRNA and screened for circulatory loop defects at 48 hpf. Similarly to egfl7 mRNA, Emilin2 and Emilin3 mRNAs were both able to rescue the circulatory defects in a significant proportion of egfl7 morphants, while egfl7 s981 mRNA was not ( Fig. 4 ). These results support the hypothesis that the upregulation of emilin genes in egfl7 s981 mutants is at least partly responsible for their lack of phenotype. To test whether the transcriptional changes we identified in mutants but not morphants were a peculiarity of the egfl7 locus, we generated TALEN mutants for vegfaa (data not shown). Interestingly, qPCR analysis showed that vegfab , a paralogue of vegfaa , was upregulated in mutants but not morphants ( Extended Data Fig. 9a ). Additionally, blocking Vegfaa function using a dominant negative approach also failed to trigger vegfab upregulation, placing the signal triggering compensation upstream of protein function ( Extended Data Fig. 9b ). Figure 3: Emilin3a is upregulated in mutant but not in morphant or CRISPRi embryos. a , Volcano plot showing significantly dysregulated proteins between 24 hpf egfl7 WT and egfl7 s981 mutant embryos using label-free quantification. Emilin3a and Emilin3b are highlighted in blue. b , Morphants did not show a significant upregulation of Emilin3a in unbiased mass-spectrometry-based proteomics comparing mutants, WT and morphants. A two-sided t -test was used to assess P values and FDR was controlled by a randomization-based SAM approach. c , mRNA expression of emilin3a , emilin3b and emilin2a in egfl7 WT, mutant, morphant and CRISPRi (template and non-template strand) embryos at 20 hpf; qPCR data, pools of 20–30 embryos each, expression normalized to gapdh (WT expression set at 1 for each gene). The emilin genes were upregulated in egfl7 s981 mutants but not after translational or transcriptional inhibition. * P ≤ 0.05. PowerPoint slide Full size image Figure 4: Emilin2 and Emilin3 can rescue egfl7 morphants. a , Design of the rescue experiment. One nanogram of egfl7 MO was injected in WT embryos, alone or together with 400 pg of mRNA ( egfl7 WT, egfl7 Δ4 , Emilin2 or Emilin3 ). b , Injected embryos were sorted according to their circulatory loop phenotype into three classes: normal, slow and absent circulation. Injection of egfl7 MO resulted in 69% of embryos lacking circulation at 48 hpf. This percentage was reduced to 19% when co-injecting egfl7 mRNA, and to 37% and 35% when co-injecting Emilin2 and Emilin3 mRNA, respectively. In contrast, mRNA from the egfl7 Δ4 mutant allele did not rescue (80% of embryos lacked circulation). Uninjected siblings are shown for comparison (99% normal). Number at the bottom of each bar is the total number of embryos from two independent experiments. Error bars, s.e.m. for the ‘absent circulation’ class. P value represents two-tailed value for Fisher’s exact test. PowerPoint slide Full size image Concerns have been raised over the use of antisense reagents, including MOs, as they may cause off-target effects and lead to aberrant conclusions. This debate was recently revived by the generation of mutations in many genes whose function was previously studied using MOs; strikingly, a majority of the resulting mutants exhibit a different phenotype from the one reported for the corresponding morphants; in fact, most often the mutants exhibit no obvious phenotype 1 . In our study, we show that, at least for some genes, the phenotypic differences between mutants and morphants can be due to the activation of genetic compensation in the former but not the latter. We show here that the upregulation of Emilins can compensate for the loss of Egfl7, but anticipate that other genes are involved in this process. On the basis of our data, we propose two additional recommendations for using MOs: the first is to use doses that do not induce p53 expression as in many cases this induction indicates off-target effects; the second is to titrate the MO dose so that it does not cause additional phenotypes in a null mutant background, as such phenotypes would be due to non-specific effects. The mechanisms underlying the compensation observed in mutants but not in morphants are likely to be complex and so will their investigation. Interestingly, we observed no upregulation of the emilin genes in the Δ3 ( s980 ) allele, suggesting that a non-deleterious genomic lesion is not sufficient to trigger this response. On the other hand, we observed emilin gene upregulation in embryos injected with egfl7 TALENs, indicating that a deleterious mutation does not need to go through the germline to trigger this response. We also detected emilin gene upregulation in embryos carrying only one egfl7 s981 mutant allele (data not shown). This observation might explain the partial protection of heterozygous embryos from egfl7 MO injections ( Fig. 2 ). In summary, our data show that, for egfl7 , one can identify a dose of MO that has no effect in most egfl7 mutant embryos but causes clear vascular defects in WT, indicating that these morphant phenotypes are not due to off-target effects. Further, egfl7 mutants show no phenotypes but exhibit a clear upregulation of several members of the emilin gene family. These Emilin proteins share an important functional domain with Egfl7, and, probably with additional proteins, can compensate for the loss of Egfl7 function. Notably, a recent study of the Icelandic population identified individuals with homozygous loss-of-function mutations in EGFL7 (ref. 22 ), indicating that compensation for severe lesions at this locus might also be at play in humans. It will be interesting to determine whether the upregulation of EMILIN genes is also present in these individuals. Of course, detailed studies will be needed to determine whether such compensation is the reason for the phenotypic differences between mutants and morphants for other genes. More importantly, our study illustrates the power of comparing mutants and morphants to identify modifier genes, a goal that remains a major challenge in the field of human genetics. Methods No statistical methods were used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment, except for the data shown in Fig. 2 where the inherent design of the experiment includes a blinding component. Zebrafish handling All zebrafish husbandry was performed under standard conditions in accordance with institutional (UCSF and MPG) and national ethical and animal welfare guidelines. Confocal microscopy An LSM 700 confocal laser scanning microscope (Zeiss) was used for live imaging. Embryos and larvae were anaesthetized with a low dose of tricaine, placed in a glass-bottomed Petri dish (MatTek) with a layer of 1.2% low melt agarose and imaged using Plan-Apochromat 10×/0.45 and LCI Plan-Neofluar 25×/0.8 objective lenses. Vessel integrity and permeability were analysed using micro-angiography. Fluorescein isothiocyanate (FITC)-dextran, 2,000 kDa (Sigma) was injected into the posterior cardinal vein at 48 or 72 hpf and imaged after 10 min. Plasmids Total RNA extraction was performed using TRIZOL (Life Technologies) and used for cDNA synthesis using SuperScript second strand (Life Technologies). cDNAs encoding the Egfl7 WT, Egfl7 s980 and Egfl7 s981 proteins were PCR-amplified using whole embryo cDNA as template. PCR fragments were ligated into the mammalian expression vector pcDNA3.1 myc-HIS tag between EcoRI and XhoI. All constructs were verified by sequencing. pCMV-6 plasmids containing mouse elastin microfibril interfacer 2 ( Emilin2 ) or mouse elastin microfibril interfacer 3 ( Emilin3 ) cDNAs were purchased from Origene. pcDNA3.1 and pCMV-6 plasmids were respectively linearized using SmaI and AgeI and in vitro transcribed using the mMESSAGE mMACHINE T7 kit (Life Technologies). Cell culture and transfection Authenticated HEK293FT (human embryonic kidney) (Life Technologies, R700-07) and HUVEC (human umbilical vein endothelial cells) cells were cultured at 37 °C in 5% CO 2 , 95% air in appropriate medium containing 10% fetal bovine serum, 100 units per millilitre penicillin and 100 μg ml −1 streptomycin. HEK293FT cells were used for biochemical studies because they are easy to grow and transfect, and have been used widely for cell biology research. All cell lines are routinely tested for mycoplasma in our facilities, and only mycoplasma-free cells were used in this study. Cells were transfected with cDNAs in antibiotic-free medium 12–24 h before protein extraction, using FuGene HD (Roche) at a 3:1 ratio (µl:µg nucleic acid) and 0.18 µg DNA per square centimetre. Cells were lysed in RIPA buffer and extracellular proteins precipitated with TCA (final 20%). Samples were resuspended in Laemmli buffer before gel electrophoresis. Genome editing TALENs were designed targeting egfl7 ( Extended Data Fig. 1 ) using TALEN targeter 23 ( ) and constructed using Golden Gate assembly 15 . Zebrafish embryos were injected into the cell at the one-cell stage with 100 pg total TALEN RNA. Genome engineering was performed as previously described 24 . The C terminus egfl7 TALEN recognition sites are TGCTGGTAGACATCATC and TTGCAGTAGTGACTAGT. Between the binding sites is a 17-bp spacer with the ‘tag’ stop codon underlined (aggaaaac tag acgatc). A ssDNA oligonucleotide ( Supplementary Table 1 ) was designed to target the spacer sequence between the cutting sites. A Myc-tag sequence, flanked by XhoI restriction sites, was introduced in the centre of the oligonucleotide resulting in 25-base homology arms on the 5′ and 3′ ends. Polyacrylamide gel electrophoresis (PAGE)-purified oligonucleotides were obtained from Sigma. One-cell-stage embryos were injected with 100–200 pg TALEN mRNA and 75 pg ssDNA donor. Screening for founders was conducted using PCR followed by XhoI restriction enzyme digest and subsequently by sequencing. CRISPR interference gRNA and CAS9 plasmids 25 were purchased from Addgene. Dead Cas9 was generated using the zebrafish-codon-optimized WT CAS (pT3TS–nls–zCas9–nls; nls, nuclear localization signal) as a template. The D10A and H840A mutations were generated using the primers in Supplementary Table 1 . Site-directed mutagenesis was performed using PfuUltra Fusion HS (Agilent). The PCR reaction protocol was 95 °C for 1 min, then 18 cycles of 95 °C for 50 s, 60 °C for 50 s and 68 °C for 1 min per kilobase of plasmid length, then 68 °C for 7 min and 4 °C hold. Dpn1 (1 μl) was added to the PCR reaction and incubated at 37 °C for 1 h to digest parental DNA and transform into competent cells. CRISPR gRNAs were designed using CRISPR design ( ) (Zhang laboratory). Oligonucleotides were annealed in a thermo block at 90–95 °C for 5 min followed by a slow cooling to room temperature ( ∼ 20 °C) and cloned in gRNA plasmid between BsmbI sites. All constructs were verified by sequencing. To make nls–zCas9–nls RNA, the template DNA (pT3TS–nls–zCas9–nls) was linearized by XbaI digestion and purified using a QIAprep column (Qiagen). Capped nls–zCas9–nls RNA was synthesized using a mMESSAGE mMACHINE T3 kit (Life Technologies) and purified using an RNA Clean and Concentrator kit (Zymo Research). To make gRNA, the template DNA was linearized by BamHI digestion and purified using a QIAprep column. gRNA was generated by in vitro transcription using a T7 RNA polymerase MEGA short script T7 kit (Life Technologies). After in vitro transcription, the gRNA (approx 100 nucleotides long) was purified using RNA clean and concentrator (Zymo Research). dCAS9 mRNA (100–400 pg) and gRNA (50–100 pg) were co-injected in the cell at the one-cell stage and at least ten pooled embryos were used to evaluate the expression level of the targeted genes by qPCR. Initial experiments were performed with gRNAs targeting the tnnt2a gene. In general, a substantial increase in knockdown efficiency was observed when combining multiple guides, indicating a synergistic effect. The egfl7 gene was knocked down by using two to four gRNAs ( Supplementary Table 1 ). Microinjection of morpholinos The ATG morpholinos egfl7 (5′-CAGGTGTGTCTGACAGCAGAAAGAG-3′), vegfaa (5′-GTATCAAATAAACAACCAAGTTCAT-3′) and tp53 (5′-GCGCCATTGCTTTGCAAGAATTG-3′), were purchased from GeneTools and injected at the indicated amounts (0.5, 1, 2 or 4 ng for egfl7 , 2 ng for vegfaa and 1 ng for tp53 ). To identify the potential effects of p53 induction in egfl7 morphants, we compared the phenotype of embryos co-injected with egfl7 and p53 MO or egfl7 MO alone (1 ng for each MO), and did not detect any obvious differences. The experiments testing the egfl7 MO effect on egfl7 mutants were blind (injection into fertilized eggs from an incross of heterozygous fish followed by phenotyping and then genotyping). The egfl7 morphant rescue experiments were not blind. Sample sizes for these and other experiments were determined on the basis of previous experience. Genotyping Embryos or fin-clips were placed in PCR tubes, with 50 µl of elution buffer (10 mM Tris-Cl, pH 8.5) and 1 mg ml −1 proteinase K added to each well and then incubated at 55 °C for 2 h. The samples were then heated to 95 °C for 10 min to inactivate proteinase K. Primers were designed using primer3: . An Eco Real-Time PCR System (Illumina) was used for the PCR reactions and high-resolution melt analysis. DyNAmo SYBR green (Thermo Fisher Scientific) was used in these experiments. PCR reaction protocols were 95 °C for 15 s, then 40 cycles of 95 °C for 2 s, 60 °C for 2 s and 72 °C for 2 s. Following the PCR, a high-resolution melt curve was generated by collecting SYBR-green fluorescence data in the 65–95 °C range. The analyses were performed on normalized derivative plots. Electrophoresis Laemmli SDS–PAGE gels consisted of a 4–20% running gel and 3% stacking gel or Tricine SDS–PAGE 12%. For immunoblots, membranes were blocked with 5% non-fat milk and incubated at 4 °C overnight with mouse (SC-40) or rabbit (SC-789) anti-Myc antibody (Santa Cruz Biotechnology), or anti-α-tubulin (T9026, Sigma). Membranes were then rinsed, incubated for 1 h with horseradish-peroxidase-conjugated anti-rabbit immunoglobulin-G (IgG) or anti-mouse IgG (Santa Cruz Biotechnology), rinsed extensively, and labelled proteins were detected using the Clarity Western substrate (Biorad). Mass spectrometry Embryos ( egfl7 WT, egfl7 s981 mutants and morphants) at 20–24 hpf were lysed in 6 M urea and 2 M thiourea (in HEPES buffer pH = 8.5). The lysate was clarified by centrifugation and proteins were subjected to in-solution digestion. In brief, proteins were reduced (0.1 M DTT, 30 min at room temperature) and alkylated (55 mM IAA, 30 min at room temperature in the dark). Lys-C was added in a 1:100 enzyme:protein ratio and incubated for 3 h. Urea concentration was diluted to 2 M using 50 mM ammonium bicarbonate, and trypsin (Promega) was added in a 1:100 enzyme:protein ratio. After incubation for 18 h, generated peptides were de-salted using the Stop and Go Extraction tip technology before mass spectrometric analysis. All WT and egfl7 s981 mutant experiments were performed at least in technical and biological duplicates. For morphant embryos, we measured protein changes in technical triplicates after pooling more than 20 embryos, thus reducing biological variability. The instrumentation for LC–MS/MS analysis consisted of a nano LC 1000 (Proxeon, now Thermo Scientific) coupled via a nano-electrospray ionization source to a quadrupole-based benchtop QExactive Plus or QExactive mass spectrometer. Separation of peptides according to their hydrophobicities was achieved on a 50 cm in-house packed column (internal diameter 75 µm, C18 Beads (Dr. Maisch) diameter 1.8 µm) using a binary buffer system: (A) 0.1% formic acid in H 2 O and (B) 0.1% formic acid in 80% acetonitrile. A linear gradient within 220 min from 8% to 38% of B, followed by an exponential increase to 90% B and a re-equilibration step to 5% B within 20 min, was used for peptide elution. Mass spectra were acquired at a resolution of 70,000 (200 m / z ) using an AGC target of 1E6 and a maximal injection time of 20 ms. A top ten method was applied for subsequent acquisition of high-energy collision-induced dissociation (HCD) fragmentation MS/MS spectra of the ten most intense peaks. Resolution was set to 17,500 at 200 m / z and 5E5 ions (AGC target) were collected in the C-trap within a maximal injection time of 60 ms using an isolation window of 1.3 thomsons (Th) (1 Th = 1.036426 ×10 −8 kg C −1 ) 26 . Raw files were processed using MaxQuant 1.4.1.2 (ref. 27 ) and the implemented Andromeda search engine 28 . For peptide assignment, MS/MS HCD fragmentation spectra were correlated to the Uniprot Danio rerio database (2014). A list of common contaminants was included in the searches that were performed with tryptic specificity. Default settings were used for MS and MS/MS mass tolerances and peptide length. The FDR was set to 1% on protein and peptide levels and estimated by the implemented decoy algorithm. Oxidation of methionine residues and acetylation on the protein N-term were set as variable modifications, and carbamidomethyl at cysteine residues was defined as a fixed modification. The match-between-runs, label-free quantification and re-quantification were enabled. Statistical analysis and data visualization were done in the environment R. The package siggenes from Bioconductor was used to determine significance of proteome changes at a FDR cutoff of less than 0.05 (ref. 29 ). Note that the protein list was filtered for at least 50% quantification over all experiments. In label-free protein quantification, a common problem is that low abundance proteins are likely to be not quantifiable, leading to a right-shifted normal distribution. Thus, missing values were replaced along a Gaussian distribution using a log 2 downshift of 1 and a width of 0.4. Imputation was inspected by histograms to mimic a Gaussian distribution for the complete data set (columnwise) to avoid too high a frequency of low-intensity values. Significance was assessed by a two-sided t -test of log 2 intensity values. Note that to compare morphants with WT, we used technical replicates of pooled embryos against all experiments for WT embryos. Five hundred randomizations were used to estimate FDR, using a cutoff of 0.05 while s0 (the fudge factor) was defined as 0.1. Protein ratios were calculated by subtracting the average of the respective groups. Data are shown in Supplementary Table 2 . RNA profiling Total RNA from egfl7 WT, egfl7 s981 mutants and morphants at 24 hpf was prepared using TRIzol (Life Technologies). RNA profiling was performed by ZF-screens using an Illumina HiSeq 2500 ultra-high-throughput sequencing system. Data are shown in Supplementary Table 3 . qPCR An Eco Real-Time PCR System (Illumina) was used for qPCR experiments. Gene expression was normalized relative to gapdh . All reactions were performed in technical triplicates; the results represent biological triplicates (unless otherwise stated) including the s.e.m. Supplementary Table 1 shows the primers used for these experiments. Additional data Mature miR126 levels were quantified in WT and egfl7 s981/s981 embryos using the miRNA QRT–PCR Detection Kit (Agilent). No significant changes were observed at 48 or 72 hpf. Maternal zygotic egfl7 s981 mutant embryos were generated by incrossing homozygous mutant adults; they exhibited no additional phenotypes compared with zygotic mutant embryos. In addition, we observed no evidence of maternal egfl7 mRNA by RT–PCR. Mutant samples for proteomics, RNA-seq and qPCR analyses were MZ mutants.

New methods for modifying the genome are currently widely discussed: Using CRISPR/Cas for instance, scientists can remove parts of the genetic code of a gene, thereby knocking it out. Furthermore, there are ways to inhibit translation of a gene into a protein. Both methods have in common that they impede production of a protein and should therefore have comparable consequences for an organism. However, it has been shown that consequences can differ, after a gene is either knocked, out or only blocked. Scientist from the MPI for Heart and Lung Research in Bad Nauheim now found that additional genes compensate for a knocked out gene and either attenuate consequences or completely compensate deficits. The results suggest caution when interpreting data from molecular biological studies or developing gene therapies to treat various diseases. To analyse function of an unkown gene, scientists often extinguish the gene and investigate the consequences of this treatment for the organism. To do so, they cut DNA-fragments from the gene using enzymes deleting the genetic information for a functioning protein. Such method is called "Gene knockout". In contrast, in a "gene knockdown" scientists block protein production using particular substances, e.g. microRNAs. Recent studies, however, have shown that results may vary between knockout- and knockdown animals. Scientists from Didier Stainier's group at the Max Planck Institute for Heart and Lung Research have now identified the reason for this. The Bad Nauheim based researchers have investigated a gene called egfl7 in zebrafish. The gene is involved in the production of connective tissue in blood vessel walls, thereby stabilizing them. Doing so, egfl7 regulates blood vessel growth. Developmental biologists, however, are not sure, what happens in a fish organism, after the egfl7 gene has been deleted. "If the gene has been blocked in a knockdown, blood vessels do not develop normally", explains Andrea Rossi, together with Zacharias Kontarakis first author of the study. In contrast, if the gene itself is deleted by a genetic manipulation, blood vessel growth is not affected. In the beginning, Max Planck researchers excluded potential side effects of the knockdown substance being responsible for interference in vascular development. To this end, they injected the substance into fish larvae in which the egfl7 gene had already been deleted. However, the larvae almost developed normally. "Since the substance did not cause disturbances in blood vessel growth, we thought of a different mechanism: The gene loss could be compensated by another gene taking over the function", Kontarakis says. "Therefore, we were looking for rescue genes, which might have been produced in animals without a functional egfl7 gene." The researchers compared the mRNA molecules and proteins in fish with or without a functional egfl7 gene and detected several mRNAs and proteins being present in higher amounts in fish without egfl7. An example is emilin 3B. When "knockdown" animals are treated with emilin 3B after egfl7 has been blocked, blood vessels develop almost normally. "This tells us that emilin 3B can compensate for the loss of egfl7. In egfl7 knockout fish, emilin production is getting upregulated. This is not the case in knockdown fish", Stainier explains. As the next step, the group plans to analyse how genes "know" that another gene has been deleted and then compensate for the loss. Several researchers worldwide are trying to delete disease genes for therapeutic reasons. Before we establish such therapies, we have to fully understand the consequences the loss or blockade of a gene might have. "In addition, our study illustrates the power of comparing knockouts and knockdowns to identify modifier genes, a goal that remains a major challenge in the field of human genetics" says Stainier.

10.1038/nature14580

Medicine

New intestinal cancer treatment approach identified

Sandra Pflügler et al. IDO1+ Paneth cells promote immune escape of colorectal cancer, Communications Biology (2020). DOI: 10.1038/s42003-020-0989-y Journal information: Communications Biology

http://dx.doi.org/10.1038/s42003-020-0989-y

https://medicalxpress.com/news/2020-06-intestinal-cancer-treatment-approach.html

Abstract Tumors have evolved mechanisms to escape anti-tumor immunosurveillance. They limit humoral and cellular immune activities in the stroma and render tumors resistant to immunotherapy. Sensitizing tumor cells to immune attack is an important strategy to revert immunosuppression. However, the underlying mechanisms of immune escape are still poorly understood. Here we discover Indoleamine-2,3-dioxygenase-1 (IDO1) + Paneth cells in the stem cell niche of intestinal crypts and tumors, which promoted immune escape of colorectal cancer (CRC). Ido1 expression in Paneth cells was strictly Stat1 dependent. Loss of IDO1 + Paneth cells in murine intestinal adenomas with tumor cell-specific Stat1 deletion had profound effects on the intratumoral immune cell composition. Patient samples and TCGA expression data suggested corresponding cells in human colorectal tumors. Thus, our data uncovered an immune escape mechanism of CRC and identify IDO1 + Paneth cells as a target for immunotherapy. Introduction Colorectal cancer (CRC) is the third most common cancer worldwide and patients with metastases in distant organs have a 5-year survival rate below 13% 1 . Metastatic CRC is currently treated with several combinations of cytotoxic agents. They improved overall survival of patients, treated initially with fluoropyrimidine monotherapy, from 12 to 30 months 2 . However, chemotherapy reached its limits 3 , which fostered clinical trials for immunotherapies 4 . The importance of immunosurveillance in CRC is emphasized by the good prognostic value of CD3 + , CD8 + , and CD45RO + T-cell infiltration (Immunoscore) 5 , 6 , 7 . Unfortunately, immunotherapy with checkpoint inhibitors showed clinical benefits only in mismatch-repair-deficient CRC with high neo-antigen load. Durable responses in CRC with different etiologies remained scarce, which is due to immune escape mechanisms 8 . The transcription factor signal transducer and activator of transcription 1 (Stat1) is a key effector in tumor immunosurveillance mediated by natural killer (NK)- and T cells 9 , 10 . Consistently, Stat1 is part of the immunologic constant of rejection gene expression signature, which correlates with good prognosis of CRC 7 . In cancer cells, Stat1 inhibits proliferation and promotes apoptosis via induction of cyclin-dependent kinase inhibitors and pro-apoptotic proteins 9 . Stat1 also regulates the expression of tissue antigens and proteins of the antigen presentation machinery, which enhance the immunogenicity of tumors 11 . Therefore, it is generally considered that Stat1 expression and activation in immune cells and in cancer cells suppresses tumorigenesis. Type I and II interferon (IFN) are the major activators of canonical Stat1 signaling, which relies on Tyr701 phosphorylation (pY-STAT1) and mediates tumor suppressive effects of IFN 12 . However, the tumor cell-intrinsic role of Stat1 in CRC is not well defined. Lgr5 + stem cells at the bottom of intestinal crypts have been identified as possible precursor cells for CRC 13 . However, non-stem cells can also acquire tumor-initiating capacity 14 and Lgr5 + cancer stem cells are not essential for growth of primary tumors 15 . These cells are separated by Lysozyme + Paneth cells in the small intestine, which provide essential niche factors for stem cell proliferation and self-renewal 16 . In the colon, Lysozyme − deep crypt base secretory cells support Lgr5 + stem cells 16 but colonic Paneth cells can appear through epithelial metaplasia 17 . Paneth cells have been identified in intestinal tumors of Apc Min mice 18 and in sporadic CRC of humans, albeit at varying frequencies ranging from 0.2% to 39% 19 . However, intestinal tumors from familial adenomatous polyposis (FAP) patients with inherited Apc mutations harbored more than 90% of Paneth cells 20 . The role of Paneth cells is unclear but CRC developed predominantly in colonic mucosal tissue with Paneth cell metaplasia 21 and the presence of Paneth cell-containing adenomas increased the risk for synchronous CRC 19 . Therefore, Paneth cells might promote CRC formation. Here we identified a subset of Paneth cells that displayed Stat1-dependent expression of the immune checkpoint molecule IDO1. Loss of these cells in Stat1-deficient intestinal tumors of Apc Min mice resulted in reduced tumor load and increased infiltration of anti-tumor immune cells. Results Epithelial Stat1 promotes formation of intestinal tumors We used Apc Min mice 22 with conditional deletion of Stat1 in intestinal epithelial cells ( Stat1 ∆IEC Apc Min ) 23 , 24 to identify Stat1 functions in intestinal tumorigenesis. Deletion of Stat1 was confirmed by PCR (Supplementary Fig. 1a ), quantitative PCR (qPCR) of purified intestinal epithelial cells (Supplementary Fig. 1b ) and immunohistochemistry (IHC, Supplementary Fig. 1c ). Lamina propria cells of Stat1 ∆IEC Apc Min mice displayed STAT1 expression, which demonstrated specific ablation in intestinal epithelial cells (Supplementary Fig. 1c ). Goblet, enteroendocrine, Paneth, and proliferating cells in the intestinal crypts were present at normal numbers in Stat1 ∆IEC Apc Min mice (Supplementary Fig. 1d-h ). These data show that epithelial cell-specific deletion of Stat1 does not affect intestinal cell differentiation and crypt proliferation of Apc Min mice. Four-month-old Stat1 flox/flox Apc Min and Stat1 ∆IEC Apc Min mice were used to investigate epithelial cell-intrinsic functions of Stat1 in intestinal tumorigenesis. Tumor formation was reduced in Stat1 ∆IEC Apc Min male and female mice (Fig. 1a–c ). Angiogenesis (Supplementary Fig. 2a ), tumor cell proliferation, and apoptosis (Supplementary Fig. 2b ) were not affected but numbers of low-grade adenomas were increased (Fig. 1d ). These data show that epithelial cell-intrinsic Stat1 promotes the formation and progression of intestinal tumors in Apc Min mice. Fig. 1: Reduced intestinal tumor burden in Stat1 ∆IEC Apc Min mice. a – c Quantification of tumor load ( a ), number ( b ), and mean area ( c ) in Stat1 flox/flox Apc Min (7♀, 9♂) and Stat1 ∆IEC Apc Min (6♀, 13♂) mice. d Grading of tumors in Stat1 flox/flox Apc Min (115 tumors, 9 mice) and Stat1 ∆IEC Apc Min (68 tumors, 9 mice) mice. Low grade: p < 0.001; high grade: p < 0.001; carcinoma: p = 0.473. e IHC staining demonstrating STAT1 + tumor cells (arrowheads) in Stat1 flox/flox Apc Min colon tumors. STAT1 expression was not detectable in tumor cells of Stat1 ∆IEC Apc Min tumors but was increased in the stroma (arrow). Scale bars indicate 50 µm. f , g Quantification of STAT1 + tumor ( f ) and stroma ( g ) cells ( f : Stat1 flox/flox Apc Min si 12 tumors, 4 mice; colon 13 tumors, 4 mice; Stat1 ∆IEC Apc Min si 12 tumors, 4 mice; colon 12 tumors, 4 mice; g : 24 tumors of 4 mice each). h Venn diagram showing an overlap between human IRDS genes, IRDS genes downregulated in Stat1 ∆IEC Apc Min colon tumors, and genes that positively correlated with Stat1 expression in human CRC TCGA data. i Heat map of downregulated IRDS genes in Stat1 ∆IEC Apc Min colon tumors. j qPCR for Ifi44, Isg15, and Ido1 mRNA expression in colon tumors of Stat1 flox/flox Apc Min (5 mice) and Stat1 ∆IEC Apc Min (6 mice) mice (tumors from each mouse were pooled). k ELISA for kynurenine in supernatants of Stat1 flox/flox Apc Min (si: 7 tumors, 5 mice; colon: 7 tumors, 6 mice) and Stat1 ∆IEC Apc Min tumors (si: 7 tumors, 7 mice; colon: 7 tumors, 5 mice). l Quantification of iNOS + and Granzyme B + stroma cells ( Stat1 flox/flox Apc Min iNOS: 12 tumors, 4 mice; GZMB < 0.5 mm 2 : 9 tumors, 2 mice; >0.5 mm 2 : 21 tumors, 4 mice; Stat1 ∆IEC Apc Min iNOS: 14 tumors, 4 mice; GZMB < 0.5 mm 2 : 23 tumors, 5 mice; >0.5 mm 2 : 13 tumors, 3 mice). m , n FACS analysis of immune cells of Stat1 flox/flox Apc Min (pooled tumors of 5 mice in 4 experiments) and Stat1 ∆IEC Apc Min (pooled tumors of 4 mice in 4 experiments) colon tumors. si: small intestine. Bars represent mean ± SEM. Full size image Apc Min mice develop tumors mainly in the small intestine but also in the colon 25 . Similarly, we found tumors in the small intestine and the colon and stained them for STAT1 by IHC. STAT1 was detected in tumor and stroma cells of Stat1 flox/flox Apc Min mice. In contrast, STAT1 was not detectable in tumor cells of Stat1 ∆IEC Apc Min tumors demonstrating efficient conditional deletion (Fig. 1e–g ). However, Stat1 ∆IEC Apc Min tumors displayed a significant upregulation of STAT1 in the tumor stroma (Fig. 1e, g ). Numbers of STAT3- and activated pY-STAT3-positive cells were not changed in Stat1 ∆IEC Apc Min tumors (Supplementary Fig. 2c-e ). These data show that tumor cell-intrinsic Stat1 suppresses upregulation of Stat1 in the stroma of Apc Min tumors. Tumor cell-intrinsic Stat1 suppresses immune cell activation We have recently shown that enhanced anti-tumor immune cell activity is reflected by increased stromal Stat1 expression in azoxymethane-dextran sodium sulfate (AOM-DSS)-induced colorectal tumors 26 . Therefore, we investigated immune cell activation in Stat1 ∆IEC Apc Min tumors. As human intestinal tumors develop mainly in the colon, we performed RNA sequencing (RNA-seq) experiments with murine colon tumors (Supplementary Data 1 and 2 ). The analyses showed reduced expression of several IFN-stimulated genes (ISGs) in Stat1 ∆IEC Apc Min tumors (Supplementary Data 1 ). In particular, 24 out of 49 orthologs of human genes of the IFN-related gene signature for DNA damage (IRDS) 27 were downregulated in Stat1 ∆IEC Apc Min tumors (Fig. 1i–j and Supplementary Data 1 ). We analyzed The Cancer Genome Atlas (TCGA) data to evaluate whether these genes are also regulated by Stat1 in human CRC. A correlation analysis revealed 529 genes that are co-expressed with Stat1 in human CRC (Supplementary Table 1 ). Among them were 16 IRDS genes. A Venn diagram using (i) human IRDS genes 27 , (ii) orthologs of human IRDS genes, significantly downregulated in Stat1 ∆IEC Apc Min mouse tumors, and (iii) the 529 genes, identified to be co-expressed with Stat1 in the CRC TCGA dataset, revealed a substantial overlap. A signature of 13 genes was present in all three gene sets (Fig. 1h ). Gene Ontology (GO) term-enrichment analyses of deregulated genes in Stat1 ∆IEC Apc Min tumors and genes that are co-expressed with Stat1 in human CRC showed a substantial overlap of GO terms and revealed mainly pathways implicated in immunological processes (Supplementary Data 3 and 4 ). These data suggest that Stat1 target genes and Stat1-dependent regulation of immunological processes are similar in human and murine CRC. Among the most significantly downregulated modulators of immune responses in Stat1 ∆IEC Apc Min tumors was the enzyme Indoleamine-2,3-dioxygenase-1 (Ido1, Fig. 1j , Supplementary Fig. 3a , and Supplementary Data 1 ), which acts as an immune checkpoint 28 . Other immune checkpoints were not deregulated in Stat1 ∆IEC Apc Min tumors but expression of the T-cell activation marker CD28 was increased (Supplementary Fig. 3a ). IDO1 converts tryptophan into the immune-suppressive metabolite kynurenine. Consistently, levels of kynurenine were reduced in supernatants of Stat1 ∆IEC Apc Min tumors (Fig. 1k ). We analyzed expression of inducible nitric oxide synthase (iNOS) and the serine protease Granzyme B, because they are markers for activation of several immune cells such as macrophages, mature dendritic cells, cytotoxic T cells, or NK cells. IHC characterization of the stroma showed increased numbers of activated iNOS + immune cells (Fig. 1l ). Numbers of Granzyme B + cells were not significantly changed in large tumors but accumulated in small adenomas of Stat1 ∆IEC Apc Min mice (Fig. 1l and Supplementary Fig. 3b ). Fluorescence-activated cell sorting (FACS) analysis of T cells demonstrated increased numbers of CD3 + , CD4 + , and CD8 + immune cells (Fig. 1m ). However, the relative percentage of Granzyme B + cells among the CD8 + population was not changed (Fig. 1n and Supplementary Fig. 3c ). As kynurenine promotes differentiation of regulatory T cells (Tregs) 29 , we analyzed CD4 + CD25 + FoxP3 + cell infiltration by FACS. Treg numbers were significantly reduced in Stat1 ∆IEC Apc Min tumors (Fig. 1n and Supplementary Fig. 3c ). It has also been reported that kynurenine promotes β-catenin nuclear localization in intestinal cancer cells 30 , 31 but IHC staining revealed unchanged nuclear β-catenin levels in Stat1 ∆IEC Apc Min tumor cells (Supplementary Fig. 3d ). These data suggest that tumor cell-intrinsic Stat1 suppresses stroma immune cell activation in Apc Min tumors through Ido1. Stat1 promotes Ido1 expression in neoplastic Paneth cells We performed IHC and IF staining to assess downregulation of Ido1 in Stat1 ∆IEC Apc Min tumors at the cellular level. This analysis revealed specific IDO1 + tumor cells in Stat1 flox/flox Apc Min tumors that were absent in Stat1 ∆IEC Apc Min tumors (Fig. 2a, b, i ). It has been shown that commercially available IDO1 antibodies are unspecific and unable to detect IDO1 in western blottings 32 . To address this issue for IHC staining, we performed in-situ hybridization (ISH) experiments and verified loss of IDO1 + tumor cells at the RNA level (Fig. 2a ). The IDO1 + tumor cells were arranged in an alternating pattern with IDO1 − tumor cells in neoplastic adenoma sheets (Fig. 2a ). A similar arrangement was described for transformed Lgr5 + stem cells and Lysozyme + Paneth cells in adenoma sheets of Apc Min tumors, which resembles crypt organization of the small intestine 18 . This suggests that IDO1 + tumor cells are either related to stem cells or to Paneth cells. Co-expression of Lgr5 and Ido1 mRNA was barely detectable by ISH in Stat1 flox/flox Apc Min tumor cells (Fig. 2a ), indicating that IDO1 + cells are Paneth cells. However, the ISH signals for Lgr5 were weak and not clearly attributable to individual cells. Therefore, double immunofluorescence (IF) with Paneth cell markers was performed. These analyses revealed protein expression of Paneth cell markers Lysozyme and MMP7 in IDO1 + tumor cells (Fig. 2c, g, h ). More than 80% of IDO1 + tumor cells expressed Paneth markers (Fig. 2e, k ), indicating that Paneth cells are the major source for Ido1 expression in the neoplastic epithelium. Double-positive cells were absent in Stat1 ∆IEC Apc Min tumors (Fig. 2c, g ) but the overall numbers of Lysozyme + and MMP7 + Paneth cells were not reduced (Fig. 2d, j ). Moreover, about 50% of Paneth cells expressed IDO1 in Stat1 flox/flox Apc Min tumors (Fig. 2f, l ). These data demonstrate that Stat1 is required for the formation of IDO1 + Paneth cells in Apc Min tumors. IDO1 + Paneth cells are potential immunosuppressors and their absence might account for immunological changes in Stat1 ∆IEC Apc Min tumors. This assumption is challenged by the low number of IDO1 + Paneth cells in Stat1 flox/flox Apc Min tumors (Fig. 2b, i ). However, we identified a much higher percentage of IDO1 + Paneth cells in small and early Apc Min adenomas with almost 30% in the colon (Fig. 2m, n ). These data suggest that IDO1 + Paneth cells support immune escape during early stages of Apc Min -induced tumorigenesis. Fig. 2: Intestinal tumors of Stat1 ∆IEC Apc Min mice lack IDO1 + Paneth cells. a IHC staining for IDO1 (upper panel) and double ISH for Ido1 and Lgr5 mRNA (lower panel) in Stat1 flox/flox Apc Min and Stat1 ∆IEC Apc Min colon tumors. IDO1 + cells are indicated by arrowheads. Scale bars indicate 20 µm. b Quantification of IDO1 + tumor cells in IHC-stained tumors of Stat1 flox/flox Apc Min (si and colon: 13 tumors, 4 mice) and Stat1 ∆IEC Apc Min (si: 12 tumors, 5 mice; colon: 13 tumors, 4 mice) mice. c IF staining for IDO1 and Lysozyme (LYZ1) in Stat1 flox/flox Apc Min and Stat1 ∆IEC Apc Min colon tumors. Single fluorescent channels and composites are shown. Positive or double-positive cells are indicated by arrowheads. Scale bars indicate 50 or 10 µm (high-magnification insets). d Quantification of LYZ1 + tumor cells of Stat1 flox/flox Apc Min (si and colon: 12 tumors, 4 mice) and Stat1 ∆IEC Apc Min (si: 8 tumors, colon 12 tumors, 4 mice each) tumors. e , f Quantification of percentages of LYZ1 + IDO1 + tumor cells between IDO1 + ( e ) and LYZ1 + ( f ) tumor cells of Stat1 flox/flox Apc Min tumors. g IF staining for IDO1 and MMP7 in Stat1 flox/flox Apc Min and Stat1 ∆IEC Apc Min colon tumors. Single fluorescent channels and composites are shown. Positive or double-positive cells are indicated by arrowheads. Scale bars indicate 50 µm. h IF staining for IDO1, MMP7, and E-Cadherin in Stat1 flox/flox Apc Min tumors of the small intestine and colon. Triple-positive cells are indicated by arrowheads. Scale bars indicate 50 µm. i , j Quantification of IDO1 + ( i ) and MMP7 + ( j ) tumor cells in IF-stained tumors of Stat1 flox/flox Apc Min (si: 7 tumors, 3 mice; colon: 12 tumors, 3 mice) and Stat1 ∆IEC Apc Min (si: 9 tumors, 3 mice; colon: 11 tumors, 3 mice) mice. k , l Quantification of percentages of MMP7 + IDO1 + tumor cells between IDO1 + ( k ) and MMP7 + ( l ) tumor cells of Stat1 flox/flox Apc Min tumors. m , n IHC staining ( m ) and quantification ( n ) of IDO1 + Paneth cells in early and advanced Stat1 flox/flox Apc Min tumors (si and colon: 12 early adenomas, 5 mice each and 13 advanced tumors, 4 mice each). Scale bars indicate 50 µm. si: small intestine. n.d.: not detectable. Bars represent mean ± SEM. Full size image Ido1 is an ISG in human and murine tumor cells 33 , indicating a role of canonical Stat1 signaling in the formation of IDO1 + Paneth cells. IHC staining for canonical STAT1 activation detected <3% pY-STAT1 + tumor cells in Stat1 flox/flox Apc Min tumors (Supplementary Fig. 4a, b ). The pY-STAT1 + tumor cells appeared as cell clusters (Supplementary Fig. 4a ), which differed from the alternating arrangement of IDO1 + Paneth cells. pY-STAT1 was undetectable in tumor cells of Stat1 ∆IEC Apc Min tumors (Supplementary Fig. 4a, b ) but appeared upregulated in the stroma (Supplementary Fig. 4a, c ) similar to upregulation of total STAT1 (Fig. 1e, g ). We blunted type I IFN signaling by conditional deletion of Ifnar1 34 in Apc Min tumor cells. Tumor formation was not affected in the intestine of Ifnar1 ∆IEC Apc Min mice (Supplementary Fig. 4d-f ). Moreover, the number of IDO1 + Paneth cells was comparable in Ifnar1 flox/flox Apc Min and Ifnar1 ∆IEC Apc Min intestinal tumors (Supplementary Fig. 4g ). These data demonstrate that the formation of IDO1 + Paneth cells in tumors of Apc Min mice is independent of type I IFN signaling. We next investigated whether human CRC contain IDO1 + Paneth cells. Biopsies of 149 human T3 and T4 CRC that had not yet metastasized (Supplementary Table 2 ) were IHC-stained for STAT1 and IDO1 to compare staining patterns with Stat1 flox/flox Apc Min mouse tumors. STAT1 + and IDO1 + cancer cells were readily detectable in human CRC but unlike mouse tumors, IDO1 + cells did not show an alternating pattern with IDO1 − cells (Fig. 3a ). However, a TCGA-based correlation matrix of ISGs (IRDS genes) and marker genes for cell identities revealed a correlation between the expression of Stat1, Ido1, and Lysozyme. Lysozyme clustered with Ido1- and Stat1-regulated IRDS genes (Fig. 3b ). Moreover, we stained 14 early adenomas (5 from FAP patients) for neoplastic Lysozyme + IDO1 + Paneth cells. Lysozyme + and Lysozyme + IDO1 + Paneth cells were found in ten adenomas (four from FAP patients) and six adenomas (three from FAP patients), respectively. In particular, FAP adenomas showed a perinuclear signal for IDO1 in Paneth cells (Fig. 3c ). The relative contribution of Paneth cells to Ido1 expression was assessed by IF staining of the six adenomas harboring IDO1 + Paneth cells. About 50% of IDO1 + tumor cells were Lysozyme-positive, demonstrating a significant contribution of Paneth cells to Ido1 expression in the neoplastic epithelium (Fig. 3d ). In summary, these data suggest that IDO1 + Paneth cells are present in human CRC. Fig. 3: Evidence for Ido1-expressing Paneth cells in human CRC and early adenomas. a IHC staining for STAT1 and IDO1 on consecutive tissue microarray (TMA) sections of human CRC. A tubular structure with cancer cells showing co-expression of STAT1 and IDO1 is indicated by arrowheads in the lower images (scale bars 50 µm). b Correlation matrix of TCGA expression data for Stat1, Ido1, IRDS genes, and markers for Paneth, Goblet, enteroendocrine, and cancer stem cells in human CRC. c IF staining for Lysozyme (LYZ) and IDO1 of a human adenoma biopsy of a FAP patient. The perinuclear signal for IDO1 is indicated by an arrowhead in the right high-magnification image (scale bar 50 µm). d IF staining was used for quantification of IDO1 + non-Paneth (LYZ − ) and Paneth (LYZ + ) cells in the neoplastic epithelium of human adenomas. Full size image IDO1 + CRC cells promote immune escape Subcutaneous implantation of C57BL/6-derived MC38 cells into immunocompetent host mice is an established method for evaluation of pre-clinical immunotherapy approaches 35 . We transplanted green fluorescence protein (GFP)-labeled MC38 CRC cells to test whether deletion of Ido1 in neoplastic cells mimics immunologic consequences of IDO1 + Paneth cell ablation in Stat1 ∆IEC Apc Min tumors. Two independent MC38 ΔIdo1-GFP subclones (MC38 ΔIdo1-GFP-2 and MC38 ΔIdo1-GFP-6 ) with CRISPR/Cas9-mediated deletion of the Ido1 locus were generated. The presence of INDELs in MC38 ΔIdo1-GFP cells was verified by sequence analysis. Both clones contained an additional G in exon 6 of Ido1 , which is a common CRISPR/Cas9-mediated insertion, and resulted in a truncated IDO1 protein (MC38 ΔIdo1-GFP-2 cells are shown in Supplementary Fig. 5a, b ). Sequencing revealed also a bigger deletion of the genomic locus upstream of the sgRNA targeting site in MC38 ΔIdo1-GFP-2 cells, which might destabilize the Ido1 mRNA. Commercial antibodies are not suitable to detect a specific IDO1 protein by western blotting 32 . However, IFNγ stimulation induced Ido1 mRNA expression in MC38 wt-GFP cells but not in MC38 ΔIdo1-GFP-2 or MC38 ΔIdo1-GFP-6 cells (Fig. 4a and Supplementary Fig. 5c ). MC38 ΔIdo1-GFP cells displayed a reduced cumulative cell number (Supplementary Fig. 5d ), indicating an in-vitro proliferation defect similar to human cells 31 . In vivo, MC38 ΔIdo1-GFP cells formed smaller tumors than MC38 wt-GFP cells in immunocompetent C57BL/6 hosts (Fig. 4b, c ), which were strongly infiltrated with CD3 + T cells (MC38 ΔIdo1-GFP-2 cells are shown in Fig. 3d, e ). In contrast, growth of MC38 ΔIdo1-GFP-2 cells was not affected in immunocompromised NOD scid gamma (NSG) hosts, which lack mature T cells, B cells, and NK cells (Fig. 4f, g ). We performed transplantation experiments with mixtures of cells to evaluate protective effects acting in trans . MC38 ΔIdo1-GFP-6 cells were additionally labeled with dsRed to discriminate them from Ido1-proficient MC38 wt-GFP cells. A 1 : 1 mixture of MC38 wt-GFP /MC38 ΔIdo1-G/RFP-6 cells showed comparable growth to MC38 wt-GFP cells in immunocompetent C57BL/6 hosts, indicating that MC38 wt-GFP cells restored growth of MC38 ΔIdo1-G/RFP-6 cells in trans (Fig. 4h ). IHC staining detected dsRed-positive cells close to the expected percentage in mixed tumors (Fig. 4i, j ). Moreover, the prominent CD3 + T-cell infiltration in MC38 ΔIdo1-GFP tumors (Fig. 4d, e ) was abolished in mixed tumors (Fig. 4k ). IHC staining revealed strong infiltration of Granzyme B + immune cells in MC38 ΔIdo1-G/RFP-6 tumors, which was also abolished in mixed tumors (Fig. 4l ). These data demonstrate that Ido1 + MC38 CRC cells are able to promote immune escape of transplanted tumors. Fig. 4: Ablation of Ido1 in MC38 cells interferes with tumor formation in immunocompetent host mice. a qPCR for Ido1 mRNA expression in MC38 wt-GFP and MC38 ΔIdo1-GFP-2 cells 0, 1, and 24 h after IFNγ stimulation ( n = 3). b , c Tumor size ( b ) and final tumor weight ( c ) after subcutaneous injection of MC38 wt-GFP , MC38 ΔIdo1-GFP-2 , and MC38 ΔIdo1-GFP-6 cells into C57BL/6 hosts (MC38 wt-GFP : 7 tumors, 7 host mice; MC38 ΔIdo1-GFP-2 : 7 tumors, 4 host mice; MC38 ΔIdo1-GFP-6 : 7 tumors, 7 host mice). d , e IHC staining ( d ) and quantification ( e ) of CD3 + infiltrating cells in tumors of MC38 wt-GFP (7 tumors) and MC38 ΔIdo1-GFP-2 (6 tumors) cells. f , g Tumor size ( f ) and final tumor weight ( g ) of MC38 wt-GFP and MC38 ΔIdo1-GFP-2 tumors in NSG hosts (MC38 wt-GFP : 15 tumors, 15 host mice; MC38 ΔIdo1-GFP-2 : 8 tumors, 8 host mice). h Tumor weight of MC38 wt-GFP (15 tumors, 15 host mice), MC38 ΔIdo1-G/RFP-6 (8 tumors, 8 host mice), and 1 : 1 mixed tumors (7 tumors, 7 host mice) in C57BL/6 hosts. i , j IHC staining ( i ) and quantitation ( j ) of GFP + and dsRed(RFP) + tumor cells in mixed tumors ( n = 3). Expected percentages of positive cells are indicated by dashed lines. Scale bars indicate 100 µm. k , l Quantification of CD3 + ( k ) and Granzyme B + ( l ) immune cells in MC38 GFP-wt (three tumors each), MC38 ΔIdo1-G/RFP-6 (four tumors each), and mixed tumors (three tumors each). n.d.: not detectable. Bars represent mean ± SEM. Full size image Tumor cell-intrinsic Stat1-Ido1 favors progression of CRC We investigated correlations of STAT1 and IDO1 protein expression in IHC-stained biopsies of 149 human T3 and T4 CRC (Supplementary Table 2 ). A score of 0–4 was attributed to STAT1 and IDO1 levels in tumor and stroma compartments (Supplementary Fig. 6 ). STAT1 or IDO1 protein in tumor or stroma cells did not correlate with overall survival and metastasis-free survival of patients (Supplementary Table 2 ). However, a strong correlation was observed between protein expression of STAT1 and IDO1 in tumor cells (Fig. 5a ) and stroma cells (Fig. 5b ). Additional analysis of TCGA data was used to increase the sample size. A correlation plot, derived from TCGA data, confirmed strong co-expression (Spearman score of 0.797) at the RNA level (Fig. 5c ). Patient survival curves for Stat1 and Ido1 were similar (Fig. 5d, e ), which is consistent with the co-expression of both genes. Patients with Stat1 high or Ido1 high cancers showed a trend towards improved survival 50 months after diagnosis (Fig. 5d, e ), which might reflect beneficial effects of Stat1 in stromal immune cell activation. Consistently, immune metagene signature analysis 36 revealed strong infiltration of anti-tumor immune cells in Stat1 high tumors (Supplementary Fig. 7a ). However, they also showed increased numbers of Tregs (Supplementary Fig. 7b ), which could be due to tumor cell-intrinsic Stat1-Ido1 expression. Therefore, we looked for a surrogate marker to discriminate between tumor cell-intrinsic effects of Stat1-Ido1 expression and superimposing stromal effects. We have recently identified IFN-induced Protein with Tetratricopeptide 1 (Ifit1) as a surrogate marker for Stat1 expression in breast cancer cells, whereas stroma cells were Ifit1-negative 37 . We tested whether Ifit1 can also be employed as a specific surrogate marker for Stat1-Ido1 expression in the neoplastic epithelium of CRC. IHC staining of human biopsies demonstrated IFIT1 expression in CRC cancer cells but not in the tumor stroma (Supplementary Fig. 7c ). Importantly, IFIT1 expression correlated strongly with STAT1 and IDO1 expression in CRC cancer cells (Supplementary Fig. 7c ). Scoring of IDO1 and IFIT1 staining intensities (Supplementary Fig. 7d ) confirmed a strong correlation of protein expression (Supplementary Fig. 7e , Pearson coefficient = 0.541). We first evaluated the prognostic value of stromal Stat1 expression in Ifit1 low tumors. Ifit1 low Stat1 low and Ifit1 low Stat1 high tumors should display low Stat1-Ido1 expression in the neoplastic epithelium and low or high Stat1 expression in stroma cells, respectively. The Ifit1-based stratification significantly improved the prognostic value of Stat1 expression in CRC ( p = 0.03, Fig. 5f ). Patients with Ifit1 low CRC benefited from high stromal Stat1 expression immediately after diagnosis, indicating that low expression of tumor cell-intrinsic Stat1-Ido1 sensitizes tumors to immune attack. We next mimicked conditions of our mouse models and stratified TCGA data into Ifit1 high Stat1 low and Ifit1 low Stat1 high CRC. Ifit1 high Stat1 low CRC (tumor cell-intrinsic Stat1-Ido1↑, stromal Stat1↓, similar to Stat1 flox/flox Apc Min tumors) displayed a higher percentage of late-stage IV tumors than Ifit1 low Stat1 high CRC (tumor cell-intrinsic Stat1-Ido1↓, stromal Stat1↑, similar to Stat1 ∆IEC Apc Min tumors) (Fig. 5g, h ), suggesting that tumor cell-intrinsic Stat1-Ido1 promotes CRC progression. Moreover, we employed the DeMixT algorithm 38 to deconvolute tumor cell-intrinsic and stromal expression of Ido1 in TCGA data. We further stratified tumors into CMS1-4 consensus molecular subtypes 39 and reinvestigated immune cell marker expression. Similarly, immune cell markers were predominantly expressed in the stroma. Moreover, the strongest expression was detected in the stroma of CMS1, a subtype characterized by immune cell infiltration 39 , as exemplified for T-cell markers (Supplementary Fig. 8a, b ). Without tumor stroma deconvolution, patients with Ido1 high tumors showed a slight trend towards better prognosis (Fig. 5d , p = 0.55). After deconvolution, survival curves for stroma and tumor compartments were laterally reversed (Supplementary Fig. 8c ). Patients with strong Ido1 expression in the neoplastic epithelial cells showed a strong trend ( p = 0.074) towards bad prognosis. Taken together, these data suggest that tumor cell-intrinsic Stat1-Ido1 expression favors immune escape and progression of human CRC. Fig. 5: Use of Ifit1 as surrogate marker reveals a negative prognostic value of Stat1-Ido1 expression in the neoplastic epithelium of CRC. a , b One hundred and forty-nine IHC-stained sections of human CRC were used to assess the correlation between STAT1 and IDO1 protein expression in the tumor cell compartment ( a ) and the stroma ( b ). c Scatter plot of TCGA expression data showing a strong correlation between Stat1 and Ido1 mRNA expression in human CRC. d – f Patient survival curves for Stat1 high/low ( d ), Ido1 high/low ( e ), and Ifit low Stat1 high/low ( f ) CRC using TCGA data. The median expression was used for stratification. g , h Stratification of human CRC into Ifit1 low Stat1 high and Ifit1 high Stat1 low cancer subtypes, and correlation with the tumor stage using TCGA data. Full size image IDO1 + Paneth cells are present in normal crypts Similar to tumors, Ido1 mRNA expression was downregulated in small intestinal and colonic IEC preparations of Stat1 ∆IEC Apc Min mice (Fig. 6a ). Therefore, we wondered whether IDO1 + Paneth cells are present in normal crpyts. IHC and ISH analyses revealed IDO1 + vesicle-bearing Paneth cells in a subset of crypts of Stat1 flox/flox Apc Min mice (Fig. 6b ), which were abolished in Stat1 ∆IEC Apc Min crypts (Fig. 6b, c ). We detected up to three IDO1 + cells in crypts of the small intestine and up to seven cells in colonic crypts (Fig. 6e ). IDO1 + Paneth cells were also identified in Stat1 flox/flox mice (Fig. 6c ), demonstrating that their formation does not depend on the Apc Min allele. They were more abundant in the distal small intestine, which has a higher bacterial load, than in the proximal small intestine (Fig. 6c ). Moreover, their abundance was decreased in mice that were housed in an extra clean special pathogen-free (SPF) facility and treated with antibiotics (Fig. 6c ). Treated mice also displayed reduced numbers of Lysozyme + Paneth cells in the proximal small intestine (Fig. 6d ). These data suggest that the formation of IDO1 + Paneth cells is induced by the bacterial microbiome. Fig. 6: IDO1 + Paneth cells are present in intestinal crypts and reduced in Ifngr1 −/− mice. a qPCR for Ido1 mRNA expression in isolated intestinal epithelial cells of Stat1 flox/flox Apc Min (si: n = 4; colon: n = 5) and Stat1 ∆IEC Apc Min (si: n = 6; colon: n = 5) mice. b IHC staining for IDO1 (upper images) and double ISH for Ido1 and Lgr5 (lower images) in the small intestine of Stat1 flox/flox Apc Min and Stat1 ∆IEC Apc Min mice. IDO1 + Paneth cells are indicated by arrowheads. Scale bars indicate 20 µm. c – e Quantification of IDO1 + crypts ( c ), LYZ1 + Paneth cells/crypt ( d ), and IDO1 + cells/crypt ( e ) in different intestinal compartments of co-housed Stat1 flox/flox Apc Min ( n = 4), Stat1 ∆IEC Apc Min ( n = 5), Stat1 flox/flox ( n = 4), Stat1 ∆IEC ( n = 4), as well as C57BL/6 mice kept at an extra clean SPF facility with and without ABx treatment ( n = 3 each). f qPCR for Ido1 mRNA expression after IFNγ stimulation in intestinal organoids of Stat 1 flox/flox (three technical replicates, two mice) and Stat1 ∆IEC (three technical replicates, three mice) mice. g IF showing the induction of IDO1 upon IFNγ stimulation in tumor organoids of Stat1 flox/flox Apc Min mice. Scale bars indicate 50 µm. h – j Quantification of IDO1 + crypts ( h ), LYZ1 + Paneth cells/crypt ( i ), and IDO1 + cells/crypt ( j ) in different intestinal compartments of co-housed C57BL/6 ( n = 3) and Ifngr1 −/− mice ( n = 3). n.d.: not detectable; si: small intestine; SPF: special pathogen free; ABx: antibiotics. Bars represent mean ± SEM. Full size image It was previously shown that the TLR9 agonist ISS DNA can induce Ido1 in intestinal epithelial cells, which protects from colitis 40 . Therefore, we isolated intestinal organoids and stimulated them with immunostimulatory (ISS) DNA (ODN 1668). Lysozyme staining showed that Paneth cells were present in organoids but ISS DNA failed to induce Ido1 (Supplementary Fig. 8d, e ). However, bacteria also promote IFNγ production by CD4 + T cells in the intestinal lamina propria 41 . IFNγ readily induced Ido1 in all epithelial cells of organoids at the RNA and protein level (Fig. 6f, g and Supplementary Fig. 8e ). To further investigate the role IFNγ in vivo, we performed IHC for IDO1 in intestines of Ifngr1 −/− mice 42 , which lack functional IFNγ signaling. The number of IDO1 + crypt cells was reduced in different intestinal parts of Ifngr1 −/− mice (Fig. 6h, j ). Interestingly, Ifngr1 −/− mice displayed reduced numbers of Lysozyme + Paneth cells in the proximal small intestine (Fig. 6i ) similar to mice with antibiotic treatment (Fig. 6d ). These data demonstrate a contribution of IFNγ in induction of IDO1 + Paneth cells. Our data suggest that a bacteria/IFNγ axis is responsible for Ido1 induction in Paneth cells. Therefore, we analyzed single-cell RNA-seq (scRNA-seq) data of Haber et al. 43 to investigate the impact of bacterial infection on Ido1 induction in Paneth cells. Paneth cell clusters were identified in t-distributed stochastic neighbor embedding (t-SNE) maps using marker genes (Supplementary Fig. 9a, b ). Only two Ido1 + cells were found in t-SNE maps of healthy mice. Interestingly, they were both allocated to the Paneth-1 cell cluster, which is located in the distal small intestine and expresses the marker GM21002 43 . Haber et al. 43 derived also scRNA-seq data from bacteria- and helminth-infected mice. Analysis of these data revealed a prominent induction of Ido1 in Paneth cells of bacteria- but not helminth-infected mice (Fig. 7a, b, d, f ). Stat1 was induced in both infection models but more prominently by bacteria (Fig. 7a–c, e ). These data demonstrate that the formation of Ido1 + Paneth cells is induced by the bacterial microbiome. Fig. 7: Expression of Ido1 and Stat1 is induced in Paneth cells following bacterial infection. a – d t-SNE maps obtained from single-cell RNA-seq (full-length sequencing) of 389 epithelial cells from the small intestine of 7–10-week-old C57BL/6J mice without infection (control) and with infection ( Heligmosomoides polygyrus or Salmonella enterica ). Three paneth cell-enriched areas, which show increased expression of Ido1 following infection with Salmonella enterica are highlighted (red circles). e , f Violin plots for the expression of Stat1 and Ido1 (transcripts per million) in Paneth cells with and without infection. Single-cell RNA-seq data were generated by Haber et al. 43 (GEO database: GSE92332). g Model how IDO1 + Paneth cells promote immune escape of CRC (for details, see discussion). CTL: cytotoxic T lymphocyte; Treg: regulatory T cell. Full size image Discussion We identified an immune escape mechanism of CRC that is based on Stat1-dependent expression of Ido1 in Paneth cells. Paneth cell markers have previously been linked with intestinal tumorigenesis but the significance of the observations remained unclear. The markers Pla2g2a and Mmp7 were identified as modifiers of Min 44 and loss of Mmp7, which is essential for Paneth cell function 45 , interfered with Apc Min -induced tumor formation 46 . Moreover, expression of Paneth markers correlated with increased risk for dietary-induced sporadic intestinal cancer in mice 47 and a Paneth cell-associated gene expression pattern was identified in human intestinal tumors 48 . The presence of IDO1 + Paneth cells in intestinal cancers might provide an explanation for these observations. IDO1 increases local kynurenine levels and depletes tryptophan. Effector T cells respond to tryptophan depletion with cell cycle arrest 49 and kynurenine promotes Treg differentiation via the aryl hydrocarbon receptor AhR 29 . This promotes an immune-tolerant microenvironment with reduced CD8 + T-cell activities and expansion of Tregs 50 , 51 , 52 . Ido1 overexpression is commonly observed in human CRC and associated with reduced serum tryptophan levels, whereas kynurenine metabolites are increased 53 , 54 , 55 . Localization studies have shown that Ido1 is expressed by infiltrating myeloid cells and neoplastic epithelial cells 56 , 57 , 58 , and both cellular compartments could contribute to kynurenine production. Our results suggest that the neoplastic epithelium is an important source for kynurenine, because loss of IDO1 + Paneth cells in Stat1 ∆IEC Apc Min tumors resulted in significantly reduced kynurenine levels that were not compensated by stromal kynurenine production. We speculate that neoplastic cells are major producers of kynurenine in tumors, whereas stromal myeloid cells use different metabolic routes. A contribution of Ido1-expressing neoplastic epithelial cells to immune escape has also been found in pancreatic ductal adenocarcinomas 59 and high Ido1 expression in neoplastic epithelial cells at the invasive front is an independent adverse prognostic factor for overall survival and metastasis in CRC 58 , 60 , 61 . Stat1 is considered as a tumor suppressor in solid cancers 9 and we expected tumors of increased size in Stat1 ∆IEC Apc Min mice. However, tumors were smaller and contained reduced numbers of Tregs and increased numbers of CD8 + T cells. Similar neoplastic and immunologic aberrations were observed in Ido1 −/− Apc Min tumors 62 . This indicates that loss of IDO1 + Paneth cells and corresponding immunological consequences surpassed tumor-promoting effects of Stat1 deletion in Stat1 ∆IEC Apc Min tumors. Interestingly, tumor formation was not affected in Stat1 −/− Apc Min mice 63 but this study neglected compensating effects of stromal Stat1 deletion, which interferes with immunosurveillance and alleviates the need for immunosuppression. Most patients develop sporadic CRC, whereas colitis-associated CRC (CAC) affects only 1–2% of human cases. Recent studies demonstrated that specific deletion of Ido1 in intestinal epithelial cells interfered with AOM-DSS-induced CAC formation in mice 30 . The oncogenic function of Ido1 in CAC was attributed to tumor cell-intrinsic phosphatidylinositol-3-kinase–Akt-mediated nuclear translocation of β-catenin rather than immunosuppression 30 . Our results showed that Stat1 ablation and corresponding loss of IDO1 + tumor cells did not affect nuclear β-catenin levels in sporadic Apc Min tumors. This suggests that Ido1 promotes the formation of sporadic CRC and CAC through different mechanisms. Moreover, the neoplastic and immunological consequences of epithelial Stat1 deletion in CAC are different from sporadic CRC. We found increased AOM-DSS-induced CAC formation in male Stat1 ∆IEC mice 64 . Stat1 ∆IEC tumors contained reduced numbers of CD8 + T cells 64 , although IDO1 + Paneth cells were absent (unpublished data). Therefore, Stat1-dependent IDO1 + Paneth cells might be particularly important for the development of sporadic tumors but dispensable for CAC. It is challenging to deduce prognostic information of tumor cell-intrinsic Stat1 and Ido1 expression from CRC TCGA data, because both genes are expressed in neoplastic cells and immune cells. Correspondingly, good prognosis of CRC patients with Stat1 high tumors might be primarily caused by enhanced anti-tumor activity of Stat1 high immune cells 7 , whereas prognostic information of Stat1 expression in neoplastic cells is masked. Using IHC staining of human samples, we identified IFIT1 as a surrogate marker for STAT1-IDO1 expression in cancer cells of human CRC. IFIT1 was not detectable in stromal cells and regulates the replication of viruses, a function that should not impact on human CRC prognosis. Ifit1 surrogate expression enabled us to discriminate between tumor cell-intrinsic and stromal Stat1-Ido1 functions in bulk gene expression TCGA data. These analyses suggested that tumor cell-intrinsic Stat1-Ido1 expression promotes progression of human CRC, correlates positively with Treg numbers and desensitizes tumors to immune attack. Moreover, Stat1-Ido1 expression correlated with Lysozyme expression in human TCGA data and IDO1 + Paneth cells were present in adenomas of FAP patients, indicating that human CRC also contain neoplastic IDO1 + Paneth cells. Stat1-dependent IDO1 + Paneth cells were also found in normal murine crypts. They did not depend on Apc Min but the presence of the Min mutation affected their spatial distribution in the distal small intestine and colon. Extra clean SPF conditions and treatment of mice with antibiotics substantially reduced the number of IDO1 + crypts in all parts of the intestine. Moreover, IDO1 + Paneth cells in normal crypts were enriched in the distal small intestine, which has a high bacterial load, and we could identify Ido1 induction in Paneth cells of bacteria-infected mice using scRNA-seq data. This suggests that IDO1 + Paneth cells are induced by the local microbiome. TLR9 and IFN signaling are candidate pathways that could promote Ido1 expression in Paneth cells. The tumor studies with Ifnar1 ΔIEC mice and ISS DNA-treated organoids suggest that Ido1 is not induced by type I IFN or TLR9 signaling. In contrast, IFNγ readily induced Ido1 in epithelial cells of organoids and IDO1 + Paneth cells were reduced in the intestine of Ifngr1 −/− mice. However, in contrast to Stat1 ΔIEC mice, IDO1 + Paneth cells were not completely abolished in Ifngr1 −/− mice, indicating that additional factors are implicated in Stat1-Ido1 induction. Ido1 is an IFNγ-inducible gene in human and murine tumor cell lines 33 . The microbiome induces IFNγ production by mucosal T cells in mice 41 and in humans 65 , and depletion of bacteria reduces IFNγ levels 41 . This indicates that interaction of lamina propria cells with the microbiome leads to the production of type II IFN that induces Stat1-dependent Ido1 expression in Paneth cells of distinct intestinal crypts. In summary, we identified Stat1-dependent IDO1 + Paneth cells in intestinal tumors and normal intestinal crypts. They might represent bone fide Paneth cells but need Stat1 for Ido1 expression. IDO1 + Paneth cells could act as local immunosuppressors to prevent aberrant immune cell activation in response to bacteria. Hence, they could also provide immune-privileged niches for tumor formation (Fig. 7g ). Early adenomas might use these niches to shield anti-tumor immune attack during elimination and equilibrium phases of immunoediting. Consistent with this idea, neoplastic IDO1 + Paneth cells were particularly abundant in early adenomas of Apc Min mice. Of note, tumor formation was impaired in Apc Min mice kept under germ-free conditions 66 . Targeting IDO1 + Paneth cells might improve efficacy of immunotherapy in microsatellite-stable CRC patients. Besides representing a conceptual advance, our findings will improve precision oncology of CRC. Methods Mice Mice with floxed alleles of Stat1 24 or Ifnar1 34 were crossed to Villin-cre mice 23 . Villin-cre Stat1 flox/flox and Villin-cre Ifnar1 flox/flox animals were crossed with Apc Min mice 22 (Jackson Laboratory) to generate Villin-cre Stat1 flox/flox Apc Min/+ ( Stat1 ∆IEC Apc Min ) and Villin-cre Ifnar1 flox/flox Apc Min/+ ( Ifnar1 ∆IEC Apc Min ) mice. Mice were kept on a C57BL/6 genetic background and housed under standard conditions at the Dezentrale Biomedizinische Einrichtung of the Medical University Vienna ( Stat1 ∆IEC Apc Min ) and the Zentrale Versuchstieranlage of the Medical University Innsbruck ( Ifnar1 ∆IEC Apc Min ). Experiments were performed with adult (6–8 weeks old) or tumor-bearing (4 months old) male or female mice. To deplete commensal gut microbiota, adult wild-type C57BL/6J mice were given ampicillin (1 g/l), vancomycin (0.5 g/l), neomycin sulfate (1 g/l), and metronidazole (1 g/l) in drinking water for 28 days. Water was changed every third day to ensure antibiotic stability. All mouse experiments were performed in accordance with Austrian and European laws (license numbers BMWFW-66.009/0191-WF/V/3b/2015 and BMWFW-66.009/0189-WF/V/3b/2015) and with the general regulations specified by the Good Science Practices guidelines of the Medical Universities Vienna and Innsbruck. Human material Patient material from the Austrian Breast and Colorectal Cancer Study 91 (ABCSG trial 91, NCT00309543) was used, which is a prospective, multicenter, randomized trial comparing the efficacy of adjuvant chemotherapy in stage II colon cancer 67 . All patients provided written consent and the study was approved by the ethics committees at the participating institutions. Isolation of intestinal epithelial cells Intestinal epithelial cells were isolated from 10–12-week-old mice as described previously 68 . Histology, immunohistochemistry, and immunofluorescence Intestines were flushed with phosphate-buffered saline (PBS) and 4% paraformaldehyde, fixed and embedded in paraffin as swiss rolls. Swiss rolls were cut into 2 µm sections and IHC/IF-stained with standard procedures using antibodies against β-catenin (Becton Dickinson, 610153, 1 : 80), BrdU (BrdU In-Situ Detection Kit, Becton Dickinson, 550803), cleaved Caspase 3 (Cell Signaling, 9661, 1 : 200), Endomucin (eBioscience, 14-5851-82, 1 : 500), GR1 (Serotec, MCA771GA, 1 : 200), Granzyme B (Abcam, ab4059, 1 : 200), IDO1 (Biolegend, 122402, 1 : 80), iNOS (Biolegend, 610431, 1 : 200), Ki67 (Novocastra, NCL-KI67-P, 1 : 1000), Lysozyme (Dako, A009902, 1 : 100), p-STAT1 (Cell Signaling, 9167S, 1 : 100), p-STAT3 (Cell Signaling, 9145, 1 : 100), STAT1 (Santa Cruz, sc-592, 1 : 500), STAT3 (Santa Cruz, sc-7179, 1 : 80), Synaptophysin (GeneTex, GTX100865, 1 : 1000), GFP (Roche, 11814460001, 1 : 1000), red fluorescent protein (RFP) (Rockland antibodies and assays, 600-401-379S, 1 : 500), CD3 (Neomarker RM9107, 1 : 100), MMP7 (Cell Signaling, 3801, 1 : 100). IHC staining on human samples was performed using antibodies against IFIT1 (Sigma Aldrich, HPA055380, 1 : 500), IDO1 (Biolegend, 122402, 1 : 100), p-STAT1 (Cell Signaling, 9167S, 1 : 100), and STAT1 (Cell Signaling, 14994, 1 : 1000). CRISPR/Cas9 of MC38 cells and transplantation MC38 ∆Ido1-GFP cells were generated using CRISPR-Cas9 as described previously 69 . Ido1 exon 6 was targeted using the following oligos: 5′-CACCTCCTGGTGGGGACTGCGACA-3′ (forward) and 5′-AAACTGTCGCAGTCCCCACCAGGA-3′ (reverse). Frequency of insertions/deletions in the transfected cell pool was estimated using the TIDE analysis software. The following primers were used for target site amplification: 5′-AACTCAGGGCTTTGAGAATGT-3′ and 5′-TTCATCCACTAAGCCACCCC-3′. Single cells derived from the initially targeted cell pool were expanded independently, sequenced for DNA modifications using the above mentioned primers, and used for transplantation experiments. To label MC38 ∆Ido1-GFP cells with dsRed, 10 µg of DsRed-pLenti plasmid (gift from Venugopal Bhaskara), 8 µg of packaging vector (psPAX2, Addgene plasmid # 12260), 3 µg of envelope vector (pVSV-G, Addgene plasmid # 14888), and 61 µl of 2 M CaCl 2 were diluted to 500 µl in ddH 2 O. The solution was then mixed with 500 µl of 2× HBS (50 mM HEPES, 10 mM KCl, 12 mM Dextrose, 280 mM NaCl, 1.5 mM Na 2 HPO 4 pH 7.05) and incubated for 10 min at room temperature, before being added to HEK293T cells for virus production. Target cells were incubated with virus containing supernatant for 5 days. Cells 10 6 were injected subcutaneously into the flanks of 8–9-week-old male C57BL/6J mice or NSG mice. To account for host effects, cells of different genotypes were implanted in the left and right flanks of the same mouse. Mixed MC38 wt-GFP /MC38 ∆Ido1-G/RFP tumors were evaluated for the presence of GFP- and RFP-positive cells via IHC. Individual MC38 wt-GFP and MC38 ∆Ido1-G/RFP tumors with 100% GFP + or GFP + /dsRed + cells were used for normalization of data. MC38 cell stimulation MC38 wt-GFP and MC38 ∆Ido1-GFP cells were cultured in Dulbecco’s modified Eagle’s medium containing 10% fetal calf serum (FCS), 1% Penicillin/Streptomycin (10,000 U/ml) and 1% l -glutamine (200 mM). At 70–80% confluency, cells were stimulated with 100 ng/µl IFNγ (Immunotools, 12343536) for 1 h and 24 h, in triplicates. Quantification and grading of Apc Min tumors Swiss rolls were stained with hematoxylin and eosin, scanned with a Pannoramic Midi Slide Scanner (3D Histec), and histomorphometrically analyzed with Definiens TM Developer software (Definiens). Grading was performed by a board certified pathologist (L.K.). Flow cytometry Intestinal tumors from single mice were pooled, minced and digested in 2 ml PBS containing 0.25% (v/v) FCS and 0.25% (w/v) collagenase IV (Life technologies, 17104-019) for 45’ at 37 °C under shaking. After straining through a 70 µm mesh and washing twice with 30 ml PBS, cells were incubated with TruStain fcX (Biolegend, 101320) and Zombie Aqua Fixable Viability Kit (Biolegend, 423102). Extracellular staining was performed using antibodies against CD8a (Biolegend, 100728), CD45 (Biolegend, 103128), CD4 (Biolegend, 100408), CD3e (eBioscience, 35-0031-82), and CD25 (eBioscience, 25-0251-81). Cells were fixed (Fixation/Permeabilization Buffer, eBioscience, 00-5123-43), permeabilized (Permeabilization Buffer, eBioscience, 00-8333-56), and intracellular staining of FOXP3 (Biolegend, 320011) and Granzyme B (Biolegend, 515405) was performed. Data were collected using a FACS Fortessa (BD) and analyzed with FlowJo software. Enzyme-linked immunosorbent assay Single tumors of the small intestine and colon were homogenized in 60 µl PBS and centrifuged. Supernatants were used for kynurenine ELISA (EMELCA Bioscience, MBS043489) according to the manufacturer’s instructions. The results were normalized for the amount of protein. In-situ hybridization Duplex ISH was performed on formalin-fixed paraffin-embedded tissue samples using RNAscope 2.5 HD assay (ACD, 322436) with probes against Lgr5 (ACD, 312171) and Ido1 (ACD, 315971) according to the manufacturer’s instructions. RNA sequencing Total RNA from tumors was extracted using TRIzol Reagent (Thermo Fisher Scientific, 15596018) and processed for sequencing using the TruSeq RNA Sample Preparation Kit (Illumina, Inc.) according to the manufacturer’s protocol. mRNA was purified using poly(T)-oligo-attached magnetic beads, fragmented, and applied to first-strand complementary DNA (cDNA) synthesis. Second-strand cDNA synthesis was performed using DNA polymerase I and RNase H. cDNA was end-repaired, A-tailed, ligated to adapters, and amplified to create the final cDNA library for sequencing (HiSeq2000, Illumina, Inc). TopHat2 algorithm was used to align raw RNA-seq data to mm10. Aligned bam files were deposited in ArrayExpress database (E-MTAB-5083). Differentially expressed genes were identified using DeSeq2 algorithm. An adjusted p < 0.005 and a fold change >2 or < −2 were defined as cut-off for differentially expressed genes. GO enrichment analyses were performed using GOrilla software. scRNA-seq analysis Pre-processed droplet-based scRNA-seq datasets from Haber et al. 43 (GEO; GSE92332) were re-analyzed using the R package Seurat. For comparison of Ido1 expression in Paneth cell clusters, the infection model datasets “SH_Salmonella” and “SH_Hpoly” were used, as well as the according control sets. Different infection durations (3 days and 10 days) within the “SH_Hpoly” dataset were pooled. Furthermore, sequencing data of intestinal cells, specifically sorted with focus on large cells to improve Paneth cell yield, were analyzed. Dimensionality reduction was performed using gene expression data for a subset of variable genes. The variable genes were selected based on dispersion of binned variance to mean expression ratios using FindVariableGenes function of Seurat 70 followed by filtering of ribosomal protein and mitochondrial genes. Next, principal component analysis (PCA) was performed and the data were reduced to the top 15 PCA (infection model)/10 PCA (large cells) components (number of components was chosen based on SDs of the principal components—in a plateau region of an elbow plot). Graph-based clustering of the PCA reduced data with the Louvain Method was used after computing a shared nearest-neighbor graph 70 . The clusters were visualized on a two-dimensional map produced with t-SNE. The VlnPlot function was applied to show expression probability distributions across the clusters and the FeaturePlot function to visualize feature expression within the clusters on a t-SNE plot. These methods were performed for marker genes of our cells of interest to identify Paneth and goblet cell clusters. Violin plot expression levels are depicted on a log transcripts per million (TPM) scale per cluster. Feature plot depicts a color scale for average gene expression. To identify further clusters containing Ido1 + cells (TA/Stem, Tuft, enterocytes), the top 50 specific marker genes for each cluster identified and described in Haber et al. 43 were selected using the CaseMatch function and aggregated and matched to gene expression profiles of the clusters identified within this analysis using the MetaFeature function (calculation of relative contribution of each feature to each cell for given set of features). Polymerase chain reaction Wild-type, floxed and deleted Stat1 alleles were amplified using 5′-TAGGCTCCCTCTTTCCCTTC-3′, 5′-ACACCATTGGCTTCACCTTC-3′, and 5′-CCCCTGTCATCTGGAGTGAT-3′ primers. The Cre transgene was detected with 5′-CGGTCGATGCAACGAGTGATGAGG-3′ and 5′-CCAGAGACGGAAATCCATCGCTCG-3′ primers. Apc Min genotyping was performed using 5′-TCTCGTTCTGAGAAAGACAGAAGCT-3′ and 5′-TGATACTTCTTCCAAAGCTTTGGCTAT-3′ primers, and digestion of amplicons with HindIII. Quantitative PCR RNA was isolated with TRIzol (Life Technologies, 15596-018) and reverse transcribed with QuantiTect Reverse Transcription Kit (Qiagen, 205313). qPCR was performed using Fast SYBR Green Mastermix (Thermo Fisher Scientific, 4385616) and Applied Biosystems 7500 Fast Real Time PCR System with primers 5′-TGGTGAAATTGCAAGAGCTG-3′ and 5′-TGTGTGCGTACCCAAGATGT-3′ for Stat1, 5′-ATGTGGGCTTTGCTCTACCA-3′ and 5′-AAGCTGCCCGTTCTCAATCA-3′ for Ido1, and 5′-TGTTTGTGATGGGTGTG-3′ and 5′-TACTTGGCAGGTTTCTC-3′ for Gapdh. Statistics and reproducibility Sample sizes and numbers of replicates are described in detail in the figure legends. Biological replicates were defined as parallel measurements of biologically distinct samples (mice in most cases). Each experiment was repeated at least three times. All values are given as means ± SEM. Normal distribution of data was tested and appropriate tests were performed: comparisons of two groups were calculated with unpaired Student’s t -test or Mann–Whitney U -test. For more than two groups, one-way analysis of variance and Bonferroni’s post-hoc test or Kruskal–Wallis test, and Dunn’s post-hoc test were used. For analysis of the tumor and TMA gradings, χ 2 -test was used. Survival analyses using clinical data from CRC TCGA patients were performed using log-rank testing and GraphPad Prism 6 software. Correlation analyses of TCGA data were calculated using cor function of R3.2.1 software and visualization was performed using corrplot and ggplot2 packages. No sample size estimation was performed. Samples were excluded as outliers according to Grubbs’ test ( α = 0.05). Experiments were performed and analyzed in a blinded, randomized manner. Significant differences between experimental groups are stated as: * p < 0.05, ** p < 0.01, *** p < 0.001, or **** p < 0.0001. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability RNA sequencing data were deposited in ArrayExpress database, accession number E-MTAB-5083. Previously generated single-cell RNA sequencing data analyzed here can be found in GEO (GSE92332). Processed (MapSplice aligned, RSEM quantified and upper-quartile normalization standardized; Level 3 RnaSeqV2) RNA sequencing data of the COADREAD dataset were obtained from The Cancer Genome Atlas (TCGA) database. All other data that support the findings of this study are available from the corresponding author upon reasonable request.

A MedUni Vienna study group has identified a previously unknown mechanism involved in the development of intestinal cancer: The bacterial microbiome activates the so-called immune checkpoint Ido1 in Paneth cells, a special cell that is only found in the gastrointestinal tract, thereby preventing local intestinal inflammation. However, this also gives rise to immunosuppressed areas, in which intestinal tumors can develop. The Paneth cells are therefore a new cellular target for immune-based therapies against intestinal tumors. The study was recently published in the Nature journal Communications Biology. Cancer cells have developed mechanisms that allow them to go undetected by the body's immune system and so avoid their destruction. These mechanisms restrict humoral and cell-mediated immune activity in the surrounding connective tissue (stroma) and make cancers resistant to treatment. Sensitizing cancer cells for an immune attack is therefore an important strategy in ensuring the success of immunotherapy techniques. However, little is yet known about the underlying mechanisms by which cancers escape the immune system. Paneth cells are gland cells in the gastrointestinal area and support the division of intestinal stem cells. MedUni Vienna researchers, led by Robert Eferl from the Institute of Cancer Research (member of the Comprehensive Cancer Center CCC of MedUni Vienna and Vienna University Hospital) discovered that Paneth cells, which produce the enzyme Ido1 (indoleamine-2,3-dioxygenase-1), inhibit the action of the immune system against intestinal cancer in the stem cell area of intestinal crypts and tumors. The researchers knocked-out the transcription factor Stat1 in intestinal epithelial cells of so-called ApcMin mice, which develop intestinal tumors due to a mutation in the Apc gene. The result was that they developed smaller and less aggressive tumors. Moreover, these tumors were infiltrated by immune cells that play a role in tumor defense. These included cytotoxic T lymphocytes. The study group then studied these tumors in more detail using RNA sequencing and found that the Stat1-deficient tumors exhibit reduced expression of the Ido1 gene. Ido1 encodes an enzyme that produces the metabolite kynurenine. Kynurenine suppresses the immune response in the tumor, thereby promoting tumor growth. Further studies have shown that ApcMin tumors with an intact Stat1 gene contain special cells that express Ido1. Similar cells were also discovered in human intestinal polyps. These cells were not present in the Stat1-deficient ApcMin tumors. Further characterisation revealed that these special cells are Paneth-like cells. Paneth cells are also found in the normal intestine, in the stem cell areas that are responsible for regeneration of the intestinal mucosa. Ido1 expression in the Paneth cells is stimulated by interaction with the bacterial microbiome. Eferl explains: "According to our hypothesis, the bacterial microbiome induces the Ido1 immune checkpoint in the Paneth cells, thereby preventing local intestinal inflammation. However, this also gives rise to immunosuppressed areas, in which intestinal tumors can develop. Ido1-expressing Paneth cells are therefore a cellular target for immune-based therapies."

10.1038/s42003-020-0989-y

Biology

Diet affects the evolution of birds

Gustavo Burin et al. Omnivory in birds is a macroevolutionary sink, Nature Communications (2016). DOI: 10.1038/ncomms11250 Journal information: Nature Communications

http://dx.doi.org/10.1038/ncomms11250

https://phys.org/news/2016-04-diet-affects-evolution-birds.html

Abstract Diet is commonly assumed to affect the evolution of species, but few studies have directly tested its effect at macroevolutionary scales. Here we use Bayesian models of trait-dependent diversification and a comprehensive dietary database of all birds worldwide to assess speciation and extinction dynamics of avian dietary guilds (carnivores, frugivores, granivores, herbivores, insectivores, nectarivores, omnivores and piscivores). Our results suggest that omnivory is associated with higher extinction rates and lower speciation rates than other guilds, and that overall net diversification is negative. Trait-dependent models, dietary similarity and network analyses show that transitions into omnivory occur at higher rates than into any other guild. We suggest that omnivory acts as macroevolutionary sink, where its ephemeral nature is retrieved through transitions from other guilds rather than from omnivore speciation. We propose that these dynamics result from competition within and among dietary guilds, influenced by the deep-time availability and predictability of food resources. Introduction Variation of biodiversity across space and time is a trademark of the history of life on Earth and ultimately determined by speciation and extinction rates 1 , 2 . To better understand the dynamics of biodiversity we need to understand the roles of biotic and abiotic factors in determining speciation and extinction dynamics 3 . While examples of abiotic factors affecting diversification dynamics are numerous (for example, ref. 4 and references therein), few studies have explored biotic influences on macroevolutionary rates across large spatio-temporal scales 5 , 6 , 7 . Hence, the relevance of biotic interactions for diversification dynamics across deep timescales is still an open question. Understanding the role of biotic interactions is a daunting task, given the myriad of interactions (for example, antagonistic, mutualistic and competitive) that individuals of a given species can have with individuals of other species. However, characterizing and understanding the trophic habits of species is tractable and may also be of great importance to understand potential adaptive responses to food availability, (for example, ref. 8 ), as well as the effects of biotic interactions on macroevolutionary dynamics 9 . As such, the diet of a given species can be used as a first-order proxy to biotic interactions. It summarizes distinct morphological, physiological and behavioural traits of an organism, which jointly determine the way it interacts with the biotic and abiotic environment 10 , 11 , 12 . For example, birds that attend army ant raids have to deal with the unpredictability of those raids, and have developed cognitive and behavioural adaptations to surpass these challenges 11 . Similarly, many nectar-feeding species have evolved beaks that suit the morphology of the flowers on which they feed, (for example, ref. 12 ). Since flowering phenology is strongly constrained by seasonality, the variability in climate (for example, in temperature) strongly determines geographic distributions of guilds such as nectarivores 13 . More generally, the long-term availability of particular climates 14 , as well as the spatio-temporal predictability of food resources, (for example, refs 15 , 16 ), might influence evolutionary radiations and diversity dynamics 14 , 15 , with environmental instability setting a potential limit to the degree of specialization 17 . Dietary strategies have been crucial for understanding species formation because interspecific competition for similar food resources can explain character displacement and the evolutionary divergence of species 18 , 19 . Nevertheless, to date only few studies at macroevolutionary scales have tested how diets might affect diversification dynamics across whole clades, (for example, refs 9 , 20 , 21 ). The paucity of whole-clade investigations relating diet to macroevolutionary dynamics is partly due to the lack of data, but also due to methodological limitations. However, recently developed methods now allow us to explicitly address 22 , 23 , 24 or indirectly assess 25 , 26 the relationship between trait evolution and diversification rates, and various authors have therefore analysed this relationship. Collectively, those studies revealed the effects of numerous traits on diversification dynamics, including self-incompatibility in Solanaceae 27 , tank formation and photosynthesis type in bromeliads 28 , migratory behaviour in birds 29 and diet in mammals 9 , 20 . Hence, ecological and life history traits play a critical role for understanding macroevolutionary dynamics and broad-scale patterns of species coexistence 24 , 30 . One of the few papers that explicitly addressed the effect of diet on macroevolutionary dynamics has shown that coarse trophic levels (that is, herbivores, carnivores and omnivores) are characterized by different diversification rates in mammals 9 . These results suggest that omnivorous mammals have lower net diversification rates than carnivores and herbivores, and that transitions into omnivory are more frequent than into other trophic levels. Using a finer diet classification within ruminants (that is, giraffes, deer, buffaloes, antelope and relatives), it has further been shown that different feeding styles underwent differential diversification rates 21 . However, this analysis suggested that grazing and mixed feeding (a combination of browsing and grazing) have both higher diversification rates and more transitions into and from these diets than browsing. Overall, these studies highlight the potential association between dietary guilds and diversification dynamics, but they also suggest that a more generalist diet (for example, omnivory or mixed feeding) might not have the same straightforward macroevolutionary outcomes at different lineages or hierarchical levels. Birds represent a good model system for investigating the role of diet on speciation and extinction 18 , 19 , and more broadly, to understand the interplay between ecology and diversification. The clade Aves has an enormous taxonomic diversity (c. 10,300 species) with a large variability in ecological and life history traits 13 , 30 , 31 . The recently published whole-clade bird phylogeny 32 and the abundance of ecological information for Aves have allowed biologists to assess the evolutionary dynamics (either trait-dependent or trait-independent) of many bird lineages at different taxonomic levels, (for example, refs 29 , 33 , 34 ). Moreover, different types of diet have evolved multiple times within the clade 13 . Dietary adaptations range from specialized feeders such as some insectivores (for example, swifts and swallows), frugivores (for example, oilbirds), seed predators (for example, macaws), vertebrates (for example, peregrine falcons) and carrion-feeders (for example, vultures) that feed preferentially on one particular food type to omnivores such as the medium-sized, common raven Corvus corax (family Corvidae), which have a generalized diet by feeding on multiple food items such as insects, fruits, seeds, vertebrates and carrion 13 . Such variation in the degree of diet specialization is probably related to different physiological and anatomical adaptations required to deal with different food items 11 , 12 , 35 . For example, some nectarivorous and frugivorous species show specific preferences for different sugar contents related to enzyme activity and absorption rates 36 , which might eventually affect their food preference and hence their degree of specificity. Here we combine the most complete bird phylogeny 32 and a comprehensive global data set of the diets of the world’s bird species (ref. 31 , updated with ref. 37 ) to investigate the potential effect of different diets on the speciation and extinction rates of birds, and the evolutionary transition rates between all dietary guilds. Given that shifts to new diets result in different ways of interacting with the environment 8 , 38 , and that such shifts might also affect the degree of specialization within a given lineage 9 , 21 , 39 , we hypothesize that the evolution of different diets in birds will result in distinct speciation, extinction and transition dynamics. Even though a simple classification of diet has been shown to affect diversification rates of mammals, (for example, ref. 9 ), we know virtually nothing about the macroevolutionary effects of diet on such diverse groups of vertebrates such as birds where we have a more refined dietary categorization. Hence, investigating the role of diet on bird diversification will not only allow us to understand its effect on this extremely diverse lineage but also help us to begin evaluating how general the observed effects of diets are for macroevolutionary dynamics across tetrapods. For our analyses, we assigned each species to a different dietary guild based on its main diet (at least 50% of one particular food type; see also Supplementary Fig. 10 and Supplementary Note 1 ) for a sensitivity analysis regarding this dietary classification. When no item comprised more than 50% of the whole diet or if a given species consumed two food types equally, then it was considered an omnivore. By following this categorization we grouped species into carnivores, frugivores, granivores, herbivores, insectivores, nectarivores, omnivores, piscivores and scavengers. We then fitted Multiple State Speciation and Extinction (MuSSE) models in a Bayesian framework for 200 randomly sampled phylogenetic trees to incorporate phylogenetic uncertainty, and used the posterior distributions of diversification and transition rates to infer the relationship between diet and diversification (see also Supplementary Material for model testing, adequacy tests and sub-clade analysis). In addition, we used network analysis to further quantify the evolutionary diet transitions among guilds and a principal component analysis (PCA) of diet scores to assess the multidimensional similarity of diets. Our results indicate that dietary habits have influenced the diversification dynamics of birds, with omnivores experiencing higher extinction, lower speciation, and with transition rates being substantially higher into omnivores than into any other guild. Results Dietary guilds Bird species are not equally distributed among dietary guilds. Both the total number of species and the phylogenetic signal strength differs among guilds ( Table 1 ). This suggests that different dietary guilds might in fact have different diversification dynamics. The three most common dietary guilds are insectivores (55%), omnivores (12%) and frugivores (12%), and the least common is the scavenger guild (0.3%). Below we exclude scavengers from the results and discussion because their diversification rates were poorly estimated due to small sample size (33 species grouped in a few lineages such as New World and Old World vultures, some crows and a few phylogenetically isolated species). In general, all dietary habits seem to have multiple origins in Aves. However, there are at least two distinct evolutionary conservatism patterns in diets across the bird tree of life. Whereas omnivores are largely spread randomly across the bird phylogeny, all other dietary guilds are phylogenetically clustered to some extent ( Table 1 and Supplementary Figs 1–9 ). Table 1 Number and percentage of total species per dietary guild and mean phylogenetic signal of each dietary guild. Full size table Diversification rates Our results reveal that the net diversification rate of omnivores is lower than that of any other dietary guild ( Fig. 1a ). Underlying these dynamics is a lower speciation rate and a higher extinction rate of omnivores compared with other guilds ( Fig. 1b-i ). In addition, the net diversification rates for all dietary guilds are positive except for omnivores, where the median value of the net diversification rate is negative ( Fig. 1a ). Even though the distribution of net diversification rates for omnivores includes zero (specifically when looking at the posterior distribution peak, Fig. 1a ), this guild is the only one that has a large portion of negative values in its diversification rate posterior distribution. This reinforces the idea that omnivores have different dynamics, with net diversification rates being significantly lower than in other guilds. Figure 1: Diversification rates associated with different bird dietary guilds. ( a ) Posterior distributions of net diversification rates for each dietary guild and ( b – i ) corresponding posterior distributions of speciation and extinction rates. Bars on a represent the 95% CI of each distribution and the dots the median of the posterior distribution. In b – i colour-filled curves represent speciation rates and white-filled curves represent extinction rates. The filled and empty dots represent median values for speciation and extinction rates, respectively. Full size image The posterior distributions of all rates for almost all guilds are mono-modal, suggesting that parameter values well represent the estimated value for each rate. The main exception is the speciation rate for herbivores ( Fig. 1e ). Other distributions that are not mono-modal are the extinction rates for both insectivores and omnivores ( Fig. 1f-1h respectively). In the case of herbivores, the distribution has a large uncertainty that results from combining mono-modal posterior distributions for individual phylogenetic trees that converged into different values. For insectivores and omnivores, the bimodality of the extinction posterior distributions also arises from combining several mono-modal distributions from all sampled trees. However, this bimodality represents the effect of phylogenetic uncertainty and not the non-convergence of estimates, reinforcing the importance of our implemented modelling framework, which explicitly includes sources of phylogenetic uncertainty. Figure 2 shows the credibility intervals (CIs) at different significance levels (95, 90 and 80%) for the posterior distributions of differences between the rates of all guilds as compared with those of omnivores. This reveals that net diversification, speciation and extinction of omnivores differ from other guilds in most cases ( Fig. 2 ). Omnivores show a statistically significant lower diversification rate than all other guilds except insectivores where this difference is marginal ( Fig. 2a ). A similar pattern is found in speciation rates, where omnivore rates are lower than those of granivores, herbivores, nectarivores and frugivores (at 95% CI for the first three guilds and 90% CI only for the latter; Fig. 2b ). Even though extinction rate differences are not as striking for some guilds as those for speciation rates (compare Fig. 2c and Fig. 2b ), omnivores show higher extinction rates than carnivores, frugivores, granivores, nectarivores and piscivores (at a 90% CI). Omnivores also have extinction rates that are marginally higher than those of herbivores and insectivores. Figure 2: Distinct diversification dynamics associated with omnivory in birds. Differences between net diversification ( a ), speciation ( b ) and extinction ( c ) of all dietary guilds relative to omnivores. The differences in rates are calculated by incorporating phylogenetic uncertainty and therefore represent a posterior distribution of differences between the rate estimates of each guild compared with the rate estimate of omnivores. The thick lines, filled and hollow triangles indicate different credibility intervals, and the dashed line indicates 0 (no difference). Positive values mean that the considered rate is higher for each guild than the same rate for omnivores. Omnivores generally show lower speciation and higher extinction rates, although differences are significant at different degrees of credibility depending on the guild. Full size image Quantifying the transitions into different dietary guilds reveals a prevalence of transition rates into omnivores rather than into any other dietary guild ( Fig. 3 and Supplementary Figs 11 and 12 ). Herbivores and granivores show the highest transition rates into omnivory, insectivores almost no transitions into omnivory or any other diet, and other dietary guilds intermediate transition levels into omnivory. Overall, these results suggest that all dietary guilds preferentially shift into omnivores, except insectivores. This is also supported by a network analysis that shows that eigenvector centrality (a measure of whether network nodes—here dietary guilds—behave as preferential end points within a network) of omnivores is equal to 1, which is significantly higher than expected by chance (permutation test with 10,000 permutations, P <0.0001; Supplementary Fig. 10 ). Estimates for all other guilds show centrality values that are not significantly different from the null model ( Supplementary Fig. 10 ). It is interesting to note that the estimates of transition rates into omnivory suggest that the overall rate of transition into omnivory (summing up the transitions from all guilds) is at the same order of magnitude as the speciation rates for other guilds (compare the panels b–i of Fig. 1 with Fig. 3 ). Figure 3: Transition rates among different bird dietary guilds. Network depicting the estimated transition rates (links) between dietary guilds (nodes). The intensity of each directed link is proportional to the median of the posterior distributions of transition rates. All transition rates smaller than 0.001 were omitted in the figure for better visualization. Numbers above the links correspond to the median value of the posterior distribution of the corresponding rate. Transitions towards omnivores are more common than any other direction of transition, and omnivory is the only guild that is significantly more connected than expected by chance (null-model analysis, P <0.0001). Full size image Dietary niche overlap Each species has its dietary preferences described by a vector of diet items (that is, vertebrates, fruits, seeds, invertebrates and so on) whose scores sum up to 10, and each of these scores represent the proportion that a given food item is consumed in the diet of a given species. To explore the multidimensional dietary similarity among guilds we used a PCA on the complete vector of diets for each species. This analysis shows that within the first three PCA axes the omnivores occupy intermediate positions relative to all other dietary guilds, having a considerable overlap with them. In contrast, other guilds show little overlap with each other at least in one of the three PCAs ( Supplementary Fig. 14 ). Higher overlap of omnivores with other guilds is also reflected in the mean Euclidean distance between each species in the orthogonal space formed by the first three PCA axes ( Table 2 ). Omnivorous species show greater mean distance within their own guild than do non-omnivorous species within their own guilds. In addition, average distances between omnivores and species within each other guild are usually similar while the average distances between species of specialized guilds and of other guilds (including omnivores) can be highly variable and for many comparisons higher ( Table 2 ). These results mean that omnivore guild is more centrally positioned in a coarse dietary space ( Supplementary Fig. 14 ). Finally, we also explored the patterns of overlap between omnivores and species of other guilds. All omnivores include at least some insects in their diet ( Supplementary Fig. 15 ). Fruits and grains also show considerable prevalence in their diet, but carrion is rarely consumed by omnivores ( Supplementary Fig. 15 ). Overall, these results support the idea that diet overlap of omnivores with other guilds is high. Table 2 Average Euclidian distances between species calculated using the first three axes of a PCA on diet item scores per species within guilds (bold) and between guilds. Full size table Model performance and adequacy Our four auxiliary analyses showed that, in our particular case, it is very unlikely that the statistical methods (MuSSE) and the diet classification scheme produced spurious associations between diet and diversification dynamics. First, we show that simulations using empirical transition rates and no association between speciation/extinction rates and trait states do not recover the speciation and extinction dynamics seen in the empirical analyses ( Supplementary Figs 16 and 17 ). Second, a model adequacy test suggested that the simulations using all estimated parameters produced a range of diet proportions that encompass the proportions of diets as observed in the empirical data set ( Supplementary Fig. 18 ). Third, the diversification dynamics observed in sub-clades of the whole phylogenetic tree showed partial concordance with our main results, especially that extinction rates of omnivores tend to be higher than those of any other dietary guild ( Supplementary Fig. 19 ). The results for speciation and transition rates within species-rich sub-clades (Passeriformes, Piciformes, Psittaciformes and Charadriiformes) were inconclusive and difficult to interpret ( Supplementary Results and Supplementary Fig. 19 , and Supplementary Note 1 for a brief discussion). Higher transition rates into omnivory were sometimes also recovered in these sub-clades, but for the sub-clade analysis, as opposed to the whole-tree analysis, speciation rate became relatively more important on generating omnivore species than the transition rates. This change in relative importance (speciation being the main process of formation of new omnivore species) suggests that the speciation and transition dynamics are interrelated, making a comparison with the full phylogenetic tree not straightforward (see Supplementary Results for further discussion). Finally, we performed a fourth test to evaluate the sensitivity of our diet classification scheme. Using a more inclusive categorization of omnivory did not change our main results, that is, that omnivory can be seen as a macroevolutionary sink ( Supplementary Figs 20–24 ). Hence, for the analysis presented here we suggest that the MuSSE model provides reliable rate estimates and that the qualitative results and conclusions derived from the whole-tree analysis are robust. We therefore focus the discussion only on the main results. Discussion Diet has a clear association with the diversification dynamics of birds. Most prominently, omnivores show lower (and even negative) net diversification rates compared with the positive rates of all other guilds ( Fig. 1a ). Our results suggest that this distinct evolutionary dynamic exhibited by omnivores arises from the interplay between significantly lower speciation rates and significantly higher extinction rates when compared with other guilds ( Fig. 2 ). Estimating speciation and extinction rates from molecular phylogenies has limitations (ref. 40 , but see ref. 41 ), but we highlight that our main conclusions are based on qualitative differences between omnivores and other dietary guilds rather than on the precise rate estimates. Interestingly, we further observed that transitions into omnivory occur at much higher rates than into any other guild ( Fig. 3 ) and that those transition rates occur in the same order of magnitude as the estimated speciation rates of other guilds. This result suggests that omnivory acts as a macroevolutionary sink where generalized diets are only transient. This sink behaviour might be a more widespread pattern in tetrapods because similar dynamics have also been suggested for mammals 9 . Lower speciation rates and higher extinction rates of omnivores in mammals 9 were obtained by defining omnivory as eating similar proportions of plant and meat compared with two other trophic levels (that is, carnivores and herbivores). However, at lower taxonomic levels within the mammalian tree of life such results differ among lineages. For instance, diversification rates have also been found to be lower for more generalized bat lineages that complement their frugivorous diet with other food items (that is, nectar and pollen) relative to more specialized frugivore lineages 20 . In contrast, in ruminants the grazing and mixed-feeding strategies have both higher diversification rates than browsing 21 . In general, lower diversification rates of omnivores could be explained by the ecological tenet that generalist species might be at a disadvantage when competing with specialists 42 . Such a ‘jack of all trades is a master of none’ mechanism (species that can utilize several resources while performing poorly at utilizing specific resources 43 ) could leave a signature at the macroevolutionary scale. According to ref. 44 , two main characteristics determine the coexistence probability of two or more species in the same place: niche overlap and competition asymmetry. We suggest that omnivorous species are at competitive disadvantage relative to species of more specialized guilds due to both factors. For niche overlap, our diet similarity analysis shows that omnivorous birds have a considerable degree of diet overlap with species from at least two other dietary guilds. In fact, omnivorous species have, on average, equivalent distances to other omnivorous species or to species belonging to other guilds ( Table 2 ), indicated by similar average pairwise distances. In addition, omnivory has a more central position than other guilds on all three PCA axes of the diet analysis, and always some degree of overlap with other dietary guilds ( Supplementary Fig. 14 ). When considering competitive asymmetry, species within specialized dietary guilds should also show different levels of specialization. For example, within insectivores there are some highly specialized lineages. True antbirds (Thamnophilidae) are specialized on eating mostly terrestrial invertebrates escaping from army ant raids in tropical forests 11 , and flycatchers (for example, family Tyrannidae) are highly adapted to catching their insect prey in flight. Dietary specialization therefore plays an important role for competitive dynamics and thereby might also influence evolutionary dynamics. From an ecological point of view, several authors have proposed that the fitness of specialists (usually assessed via population size) is higher when compared with generalists 45 , 46 . This can be explained by trade-offs between performing well at acquiring a narrow range of resources (for example, hosts, food items and so on) or having a wide range of resources at the cost of being worse at acquiring them. When a specialist and a generalist species compete for the specialist’s preferred resource, the specialist species should ecologically outperform the other 46 . This explanation might be particularly true if resources are constantly available, for example, in relatively stable or aseasonal environments. In contrast, specialists might be at disadvantage in places or at times where the preferred resource is scarce or unpredictable 18 , 45 . If we expand this competitive scenario to a situation where omnivores share their resources with multiple different specialists, we hypothesize that over longer timescales omnivores would be systematically at a competitive disadvantage due to both high niche overlap and competition asymmetry. This would ultimately lead to very low abundances of generalist species 46 and possibly to local extinctions 47 . The simultaneous competition with multiple species might therefore translate into higher extinction rates at a macroevolutionary scale, resulting in a high macroevolutionary cost to omnivores. Assuming this scenario of multi-species competition, an omnivore would be a ‘jack of all trades’ (a species that can utilize several resources 43 ) trapped in an arena of ecological competition with multiple competitors belonging to different guilds. Such a ‘master of none’ mechanism (that is, species perform poorly at utilizing specific resources 43 ) would lead to macroevolutionary consequences at the species level, where the ‘jacks of all trades’ should show low speciation and/or high extinction rates. As outlined above, we suggest that higher extinction rates of omnivorous birds are the result of competition with species from multiple guilds. However, the generality of such a mechanism remains to be tested more widely given that a potential association between lower diversification rate and a more generalized dietary guild has so far only been examined in mammals, (for example, ref. 9 ) and birds. Assuming that body size is a proxy for ecological niche 48 , mammals might be responding to a similar mechanism as proposed here. Mammalian omnivorous species show both lower diversification rates 9 , as well as intermediate and overlapping body masses (that is, intermediate and overlapping ecological niches) when compared with herbivores and carnivores 49 . Hence, inter-guild competition might be an overlooked mechanism that is potentially important to explain lower diversification rates of omnivorous species. Although species-level mechanisms or outcomes were at first widely rejected as being drivers of macroevolutionary dynamics, they are now considered important mechanisms 6 , 50 , 51 and the ever-growing empirical studies that show a pattern of trait-based diversification 21 , 27 , 29 suggest that it might indeed be a common phenomenon in determining the evolutionary success of lineages with different traits 2 , 51 . Along with increased extinction rates, we also detected lower speciation rates of omnivores relative to other guilds. The mechanism behind the association between low speciation rates and omnivory is more elusive, but given that speciation and extinction rates are usually linked by the same mechanisms 52 , 53 , it is possible that inter-guild competition might also play a role here. If each guild is an adaptive zone ( sensu (ref. 54 )), where the speciation process results in the crowding of this adaptive zone, then higher rates of speciation from multiple specialized guilds might result in a compound ‘crowding’ effect that reduces speciation rates of omnivores at the macroevolutionary scale. Alternatively, the lower speciation rates of omnivores could also be explained by higher extinction rates at the population level, whereby populations experiencing high competition with multiple species are likely to go extinct. In this scenario, some populations that are going through a speciation process might not have enough time to be fully separated into two different species, resulting in lower speciation rates at the macroevolutonary scale, a process that might be referred to as ‘ephemeral speciation’ 55 . Given the macroevolutionary ‘costs’ associated with omnivory (that is, low speciation rates and high extinction rates), it might seem surprising that this dietary guild still constitutes such a considerable portion of extant bird diversity (1,158 species, circa 12%; Table 1 ). From a deep-time perspective, a lineage with low diversification rates—especially those with negative rates—should eventually disappear or at best reduce its diversity due to species sorting 6 , 56 . We hypothesize the reason why omnivory has not disappeared lies in the high transition rates into omnivory. Our results show that transition rates into omnivory are significantly higher than into any other dietary guild ( Fig. 3 ) and that they occur at the same order of magnitude as the speciation rates for other guilds. Moreover, network analysis reveals that omnivory is the most central guild and that diet shifts occur from all other dietary guilds into omnivory more than one would expect by random ( Supplementary Fig. 10 ). This suggests that omnivore lineages preferentially originate at the macroevolutionary scale via transitions, and not through speciation. The reason why such transition rates are so high could depend on selection driven by resource competition at the individual level. Omnivory could be favoured at times or places with low abundance of a preferred resource or when resource availability is highly unpredictable 17 , 45 . For instance, climate variability (for example, seasonality) clearly influences resource availability, and specialists might only survive if their resources are continuously available and highly predictable 15 , 17 . For example, specialized nectarivores and frugivores only survive in places where seasonality is low and hence resource availability relatively constant 13 . Granivores benefit in dry climates where seeds are constantly available, whereas insectivores perform well in the tropics where insects are available all year round 13 . Hence, a low spatio-temporal predictability of resources, as well as high environmental instability is likely to benefit omnivores by setting a limit to the degree of specialization 17 . At macroevolutionary scales, this will influence diversification dynamics and increase transition rates into omnivory. A mixed-feeding diet has been shown to be beneficial for individuals belonging to different herbivore species across different animal groups 57 , 58 , 59 . If individual-level selection is indeed an important factor for avian transitions into omnivory, we can expect ancestral lineages to feed on resources that were temporally limited, unpredictable, difficult to digest or with poor nutrition. In birds, most transitions into omnivory come from granivores and herbivores, and herbivores are represented with only few species ( Table 1 ). Given such low frequency and the fact that feeding exclusively on leaves might represent a poor diet 60 , selection pressure to add new, perhaps more nutritious, food items could indeed drive the evolution of omnivory from herbivorous ancestral lineages. In the case of granivores, it is more likely that resource availability plays an important role, but analogously to the hypothesis of transitions from herbivores this hypothesis remains to be properly tested. Interestingly, the transitions into omnivores ( Fig. 3 ) and the detailed information on their diets ( Supplementary Fig. 15 ) suggest that transitions into omnivory systematically include the addition of insects. Insects might represent a predictable and protein-rich resource, but insectivory might also pose evolutionary challenges such as the digestion of lipids 61 and the potential competition with more specialized insectivore species 11 . We propose that the diversification dynamics of different dietary guilds are driven by resource competition caused by deep-time temporal and spatial changes in resource availability and predictability. These fluctuations in resource availability and predictability might create evolutionary pressures at two levels of organization. At the individual/population level, these fluctuations might promote transitions into omnivory in times of food resource scarcity by selecting individuals/populations that do not rely on single food items 62 . At the species level, the same climate and resource fluctuations might result in more favourable conditions that would eventually bring back omnivore species in contact with species belonging to multiple dietary guilds. In times or places with relatively small changes in resource availability and predictability, the more specialized guilds can rapidly (re)colonize areas where omnivores emerged, possibly preventing the transitions of omnivores back into other more specialized guilds due to the velocity of migration in relation to selection. This would explain the higher extinction rates of omnivores. Such a selection mosaic ( sensu (ref. 63 )) of resource distribution and competition would therefore mediate the macroevolutionary fate of omnivores and specialized dietary guilds 9 . Even though it is challenging to directly test mechanistic hypotheses at a macroevolutionary scale, we suspect that such a competitive mechanism acting at both the species and individual level should not only result in specific macroevolutionary patterns (for example, higher extinction rates of omnivores) but also in macroecological predictions. At broad spatial scales, we therefore predict that the spatial distribution of omnivorous species peaks in places where co-occurrence of specialized dietary guilds is low. For instance, the relatively stable, long-term (Cenozoic) availability of rainforest climates in South America 14 coincides with a low diversity of omnivores and high diversity of species belonging to specialized dietary guilds such as granivores, frugivores, nectarivores, insectivores and carnivores 13 . Expanding these ideas into the Anthropocene where human-driven global change is homogenizing biological communities and eliminating the resources of many specialist species, we expect that a shift in the competitive dynamics between generalists and specialist species will occur. Globally, generalist bird species are at a much lower risk of extinction than specialists, and in birds there is a positive relationship between increased specialization and increased risk of human-driven extinction 64 . Hence, ongoing human-driven changes are likely to distort future macroevolutionary dynamics by changing diversification rates and favouring generalist species at the expense of specialists. Irrespective of the mechanism, our results support the notion that omnivory is a macroevolutionary sink, that is, a transient state in bird evolutionary history. This dynamic seems to be affected by two different hierarchical processes. On the one hand, species sorting through higher extinction rates and lower speciation rates will lower species richness of omnivores through time. On the other hand, selection — presumably driven by changes in resource abundance and predictability — brings species diversity of omnivores back and results in higher transition rates into omnivory at the macroevolutionary scale. The ecological mechanisms behind these macroevolutionary dynamics are difficult to test, but the available data suggest that the interplay between intra- and inter-guild competition might lie at the heart of this macroevolutionary game of the ‘jack of all trades is a master of none’. Methods Data set We used the bird phylogeny from Jetz et al . 32 , which encompasses almost all bird species (9,993 species, available online at The tree was built using molecular data from 6,670 species, and the remaining taxa with no molecular information were added to the phylogeny based on taxonomic information and simulated branching times from a pure birth (Yule) model of diversification 32 . The distribution of these inserted species spans the entire tree and virtually all clades ( Supplementary Fig. 13 ). The addition of those species should therefore not bias diversification estimates and at best only homogenize any real differences between different traits, making our tests more conservative with respect to finding true differences in diversification dynamics among guilds. A distribution of 10,000 trees with different topologies was obtained from the original paper 32 . To account for phylogenetic uncertainty, we randomly sampled 100 trees from this posterior distribution of trees for each of the two backbone trees, totalling 200 trees. Using these 200 trees diminishes any possible biases that the insertion process of species with no molecular data could bring into the phylogeny. We note that the two backbones from ref. 32 showed a similar amount of differences in topology as when both were compared with two other recently published high-order bird phylogenies ( Supplementary Fig. 25 and Supplementary Notes for methods, see Supplementary Methods ). A comprehensive bird diet database (ref. 31 ; updated with ref. 37 ) was used with numerical scores for different food types consumed by birds (including invertebrates, fruits, nectar, seeds, terrestrial vertebrates, fishes, carrion, plants (non-reproductive) and miscellaneous). The data came from over 250 ornithological books, as well as peer-reviewed articles compiled in a global ornithological database by C.H.S. (ref. 31 ; updated with ref. 37 ). The literature used includes synthetic works (for example, Handbook of the Birds of the World , The Birds of Africa , The Birds of South America , Australia/New Zealand Handbook of Birds , The Birds of Western Palearctic and all the books on bird families), which provide bird species accounts based on a summary of all literature on a particular bird species. Therefore, our diet classification was based on a comprehensive diet database that summarizes dietary preferences across a species range and across seasons. The scores of all diet items add up to 10 and represent the approximate proportion of each food type in the diet of a given species. A species was classified into a specific dietary guild if it had one food item with a score >5 (for sensitivity analysis see Supplementary Figs 19–23 ). Species with only two equally consumed food items in their diet or with no food item with a score >5 were classified as omnivores. Thus, all species were classified into nine dietary guilds: carnivores (feeding predominantly on vertebrates); frugivores (feeding predominantly on fleshy fruits); granivores (feeding predominantly on seeds); herbivores (feeding predominantly on non-reproductive plant material such as leaves, roots and shoots); insectivores (feeding predominantly on insects or other invertebrates); nectarivores (feeding predominantly on nectar); piscivores (feeding predominantly on fish); scavengers (feeding predominantly on carrion); and omnivores (the species that do not have a predominant diet). After matching the taxonomy of species with dietary data with the phylogeny, we finally used a total of 9,876 species in all analyses. Model fitting and parameter estimates MuSSE models were fitted across all sampled trees 23 . This class of models estimates the parameter values (speciation, extinction and transition rates) associated with each trait state in a phylogeny. The models were implemented in a Bayesian Monte Carlo Markov Chain (MCMC) framework to account for both phylogenetic and rate value uncertainties. Phylogenetic methods might underestimate extinction rates 40 , 65 , and to avoid rates to be equal to 0 (especially transition rates that are prone to be very small in a multi-state model) we used three Cauchy distributions as hyperpriors. These hyperpriors have a location parameter fixed to 0 and the scale parameter is estimated from MCMC analysis. This allowed rates to be very small, but not zero. All parameters were independently estimated, that is, with no constrains. A total of 1,500,000 steps (sampling every 1,000th step) were necessary to achieve an acceptable convergence of the majority of the parameters. The Bayesian analysis was run for separate trees in parallel on four computer servers. All analyses were conducted within the statistical environment R 66 using the diversitree package 23 and a new script designed to implement the MCMC analysis (available at ). There has been recently a debate over the performance of trait-based models 67 , 68 . The main critiques are related to the low presence of true replicas. Strong phylogenetic signal and few events of state change in a given character could lead to pseudoreplication 67 , and a high percentage of false positives in a class of trait-dependent speciation and extinction models due to rate heterogeneity throughout the tree could additionally bias rate estimates 68 . Although the latter limitation has only been proven to be true for binary-state characters, some authors suggest that it is a common limitation among all xxSSE models 67 , 68 . As an alternative to the trait-dependent methods, a recent study by Huang and Rabosky 33 estimated speciation and extinction rates using BAMM (Bayesian Analysis of Macroevolutionary Mixtures 26 ), a trait-independent method that estimates these rates using reversible-jump MCMC to identify shifts in diversification rates. With the BAMM results a significant relationship between the degree of sexual dichromatism in birds and diversification rates was found using comparative methods. However, using BAMM as an alternative solution to xxSSE models does not seem to be fully adequate for our analyses and the phylogenetic structure of the diet traits. BAMM does not estimate transitions between states of the analysed character when estimating speciation and extinction rates, and these rates seem to have an important role for the macroevolutionary dynamics in our analyses, and more broadly in evolutionary dynamics. In addition, given the phylogenetic overdispersion of omnivory in our phylogenetic trees (omnivore species usually appear as a isolated tip within a clade with species that belong to a more specialized dietary guild) and how BAMM operates (it finds a node where a shift in rate is justifiable) we suspect that it is virtually impossible to detect rate shifts associated with omnivory using BAMM. The reason is that within each group of species that contains omnivores the statistical power to detect any shifts in speciation and/or extinction rates for omnivorous species would be insufficient. We therefore suspect that in such a phylogenetic trait configuration the diversification rates obtained with BAMM for omnivores would potentially be biased in different directions depending on the diet of closely related species. This would turn any posterior analysis unprofitable. Last, a semi-parametric test to detect trait-dependent diversification was proposed by Rabosky and Huang 69 that relies on the rate estimates derived by BAMM to later estimate the relationship between a binary (or continuous) trait and the diversification rates. This test uses rate regime permutations to build null distributions of correlation coefficients. Even though this seems as an interesting alternative, this test was not used here since it is currently not available for multi-state discrete characters (ref. 69 , page 12). To assess the reliability of our MuSSE results in relation to the issues raised by ref. 68 we performed four additional analyses. In the first additional analysis, we tested if rate heterogeneity captured by our empirical phylogenetic trees might have led MuSSE to detect spurious relationships between trait states and diversification. To do this, we simulated the evolution of a discrete character with the same number of states as in our empirical data set on 10 randomly selected empirical bird trees, using the empirical transition rate estimates. We then tested for a statistical association between those neutral characters and the estimated rates ( Supplementary Methods section A1 ) to see whether the model detects similar associations between trait states and speciation and extinction rates. In the second set of additional analyses (model adequacy), we simulated 1,000 trees using the rates estimated in our main empirical analysis, to check whether the estimated empirical rate values would generate a proportion of trait (diet) states comparable to the empirical proportions ( Supplementary Methods section A2 and Supplementary table 1 ). In the third set of additional analysis (sub-clade analysis), we ran separate MuSSE analyses for the four major bird orders (Passeriformes, Piciformes, Psittaciformes and Charadriiformes) on 10 trees to investigate if the macroevolutionary patterns associated with different diets as obtained from the whole-tree analysis were also recovered at these sub-clades ( Supplementary Methods section A3 ). In the fourth additional analysis, we investigated the extent to which our results were affected by our dietary classification scheme. We used a different classification scheme to categorize species into discrete dietary guilds and then estimated all diversification rates using the same procedure as in our main analysis using the 10 sampled trees ( Supplementary Methods section B ). The complete description and results of these tests can be found in the Supplementary Methods . Posterior distributions of rates The posterior distributions of parameters from all 200 trees were combined into one single posterior distribution for every parameter (for example, speciation, extinction and transition rates and hyperprior parameters, adding up to 93 parameters in total). For net diversification rates r (speciation−extinction), the posterior distribution was built by calculating r for each sample of the MCMC, resulting in the same 1,500 values for each state of the trait. For all posterior distributions of speciation, extinction and net diversification rates the 95, 90 and 80% CIs (highest posterior density) were calculated. All results and discussion do not encompass rates from scavenger species because estimates were poor due to small sample size (33 species). Comparison of rates of dietary guilds To test whether or not speciation, extinction and net diversification rates of omnivores were significantly different from rates of all other guilds, we calculated the difference between each omnivore rate to the rate estimated for each other dietary guild. These differences in speciation, extinction and net diversification rates were calculated at each sampled MCMC step, building posterior distributions of differences. These distributions were then compared and analysed separately and the omnivore rate was considered different when the value 0 was not included in the CI for each rate difference comparison ( Fig. 2 ). The MuSSE analysis also allowed us to generate estimates for pairs of transition rates but not to explicitly test for any general asymmetry while considering all the transitions at the same time. Depending on how transition rates are organized among distinct dietary guilds, some guilds might constitute preferential routes of transition. In contrast, if there is no consistent pattern in the distribution of transition rates among guilds, no guild will show a higher transition rate into or from it. To evaluate if the empirical transition matrix significantly deviates from a null model where all transitions are expected to be balanced among nodes, we used a network theory approach. We depicted the transition rates as weighted links and dietary guilds as nodes of the transition network. If species from other guilds consistently shift to the same dietary guild, this latter dietary guild would show high levels of centrality in the transition network. In the transition network, eigenvector centrality describes how the transition rates lead, directly or indirectly, to a given dietary guild. We computed the eigenvector centrality of each dietary guild 70 , which varies from 0 (peripheral dietary guild) to 1 (central dietary guild). Thus, a highly central dietary guild can be viewed as an absorbing state to which species from other dietary guilds may evolve by changing resource use. To verify the significance of these centrality values, we built a null distribution of centrality values by randomly assigning to each link a value sampled from the estimated transition rates without replacement for each of the 10,000 replicas. We then compared the empirical centrality values to this null distribution and verified to which quantile the real value corresponded. Diet similarity analysis With the original diet scores for all species, we quantified the score frequencies of each food item within the diet of all omnivore species ( Supplementary Fig. 15 ). This was done to better characterize the diet of omnivorous species and to trace diet similarities between omnivores and all other guilds. We additionally performed a PCA using the full vector of diet scores (with each food item as a variable) to characterize omnivorous species and their multidimensional dietary similarity with other guilds ( Supplementary Fig. 14 ). This allowed us to assess the distribution of dietary guilds in niche space with reduced multidimensionality. We also calculated the Euclidean distance between each possible pair of species in the orthogonal space created by the first three principal components ( Table 2 ). These distances were then averaged within and between each and all guilds for further comparison. Additional information How to cite this article: Burin, G. et al . Omnivory in birds is a macroevolutionary sink. Nat. Commun. 7:11250 doi: 10.1038/ncomms11250 (2016).

How diet has affected the evolution of the 10,000 bird species in the world is still a mystery to evolutionary biology. A study by Daniel Kissling of the Institute for Biodiversity and Ecosystem Dynamics (UvA) and colleagues from the University of São Paulo and the University of Utah shows how diet preferences have influenced bird diversification over millions of years. The findings were published in Nature Communications. Since the seminal work by Charles Darwin, it is know that dietary habits of birds can affect the evolution of species, such as the beak sizes of Galapagos finches. However, birds show an astonishing diversity of species and dietary adaptations, ranging from very small nectar-feeding hummingbirds to large carnivorous eagles. How such diverse dietary preferences ultimately lead to differences in diversification dynamics (i.e. the balance between speciation and extinction) of different birds has not yet been examined. Diet dataset The researchers compiled an impressive diet dataset of almost all bird species in the world together with a large phylogenetic tree that represent the relatedness of all bird species. Using models of trait-dependent diversification, they then showed that omnivorous bird lineages (with species that feed on many different food items) have lower rates of speciation (i.e. generating less new species) and higher rates of extinction (i.e. losing more existing species) than species which prefer specific food items such as fruits, nectar, or insects. Furthermore, the researchers also found that over deep evolutionary time birds which are specialized on a particular food item often add other food items to their diets, resulting in evolving transitions into omnivory. Surprised 'I was really surprised to find that omnivores preferentially originate via transitions, and not through speciation', says lead author and PhD student Gustavo Burin from the University of São Paulo. Together with the low speciation rates and high extinction rates, these high transition rates indicate that omnivores originate from more specialized birds that expand their diets, rather than directly through speciation of omnivorous bird clades. 'We suggest that this is caused by resource competition, climate instability, and deep-time availability of food resources', says Burin. High transition rates towards omnivory may arise in times when food is harder to find or when it is temporally unavailable. Human activities Expanding these findings to the current human-driven changes on our planet the researchers expect that shifts in competitive dynamics between generalists and specialists will occur. 'Human activities such as habitat destruction and other global change drivers eliminate the resources of many specialist species', says Daniel Kissling from University of Amsterdam. This means that specialists are currently at higher risk of extinction than generalists. 'This will dramatically change the ecology and evolution of life on Earth because generalists are now favoured at the expense of specialists', explains Kissling. Ultimately, this will affect the functioning of ecosystems and the services that nature provides to humanity.

10.1038/ncomms11250

Space

New calibration tool will help astronomers look for habitable exoplanets

X. Yi et al. Demonstration of a near-IR line-referenced electro-optical laser frequency comb for precision radial velocity measurements in astronomy, Nature Communications (2016). DOI: 10.1038/ncomms10436 Journal information: Nature Communications

http://dx.doi.org/10.1038/ncomms10436

https://phys.org/news/2016-01-calibration-tool-astronomers-habitable-exoplanets.html

Abstract An important technique for discovering and characterizing planets beyond our solar system relies upon measurement of weak Doppler shifts in the spectra of host stars induced by the influence of orbiting planets. A recent advance has been the introduction of optical frequency combs as frequency references. Frequency combs produce a series of equally spaced reference frequencies and they offer extreme accuracy and spectral grasp that can potentially revolutionize exoplanet detection. Here we demonstrate a laser frequency comb using an alternate comb generation method based on electro-optical modulation, with the comb centre wavelength stabilized to a molecular or atomic reference. In contrast to mode-locked combs, the line spacing is readily resolvable using typical astronomical grating spectrographs. Built using commercial off-the-shelf components, the instrument is relatively simple and reliable. Proof of concept experiments operated at near-infrared wavelengths were carried out at the NASA Infrared Telescope Facility and the Keck-II telescope. Introduction The earliest technique for the discovery and characterization of planets orbiting other stars (exoplanets) is the Doppler or radial velocity (RV) method whereby small periodic changes in the motion of a star orbited by a planet are detected via careful spectroscopic measurements 1 . The RV technique has identified hundreds of planets ranging in mass from a few times the mass of Jupiter to less than an Earth mass, and in orbital periods from less than a day to over 10 years (ref. 2 ). However, the detection of Earth-analogues at orbital separations suitable for the presence of liquid water at the planet’s surface, that is, in the ‘habitable zone’ 3 , remains challenging for stars like the Sun with RV signatures <0.1 m s −1 (Δ V / c <3 × 10 −10 ) and periods of a year ( ∼ 10 8 sec to measure three complete periods). For cooler, lower luminosity stars (spectral class M), however, the habitable zone moves closer to the star which, by application of Kepler’s laws, implies that a planet’s RV signature increases, ∼ 0.5 m s −1 (Δ V / c <1.5 × 10 −9 ), and its orbital period decreases, ∼ 30 days ( ∼ 10 7 s to measure three periods). Both of these effects make the detection easier. But for M stars, the bulk of the radiation shifts from the visible wavelengths, where most RV measurements have been made to date, into the near-infrared. Thus, there is considerable interest among astronomers in developing precise RV capabilities at longer wavelengths. Critical to precision RV measurements is a highly stable wavelength reference 4 . Recently a number of groups have undertaken to provide a broadband calibration standard that consists of a ‘comb’ of evenly spaced laser lines accurately anchored to a stable frequency standard and injected directly into the spectrometer along with the stellar spectrum 5 , 6 , 7 , 8 , 9 . While this effort has mostly been focused on visible wavelengths, there have been successful efforts at near-IR wavelengths as well 10 , 11 , 12 . In all of these earlier studies, the comb has been based on a femtosecond mode-locked laser that is self-referenced 13 , 14 , 15 , such that the spectral line spacing and common offset frequency of all lines are both locked to a radio frequency standard. Thus, laser combs potentially represent an ideal tool for spectroscopic and RV measurements. However, in the case of mode-locked laser combs, the line spacing is typically in the range of 0.1–1 GHz, which is too small to be resolved by most astronomical spectrographs. As a result, the output spectrum of the comb must be spectrally filtered to create a calibration grid spaced by >10 GHz, which is more commensurate with the resolving power of a high-resolution astronomical spectrograph 8 . While this approach has led to spectrograph characterization at the cm s −1 level 16 , it nonetheless increases the complexity and cost of the system. In light of this, there is interest in developing photonic tools that possess many of the benefits of mode-locked laser combs, but that might be simpler, less expensive and more amenable to ‘hands-off’ operation at remote telescope sites. Indeed, in many RV measurements, other system-induced errors and uncertainties can limit the achievable precision, such that a frequency comb of lesser precision could still be equally valuable. For example, one alternative technique recently reported is to use a series of spectroscopic peaks induced in a broad continuum spectrum using a compact Fabry–Perot interferometer 17 , 18 , 19 . While the technique must account for temperature-induced tuning of the interferometer, it has the advantage of simplicity and low cost. Another interesting alternative is the so-called Kerr comb or microcomb, which has the distinct advantage of directly providing a comb with spacing in the range of 10–100 GHz, without the need for filtering 20 . While this new type of laser comb is still under development, there have been promising demonstrations of full microcomb frequency control 21 , 22 and in the future it could be possible to fully integrate such a microcomb on only a few square centimetres of silicon, making a very robust and inexpensive calibrator. Another approach that has been proposed is to create a comb through electro-optical modulation of a frequency-stabilized laser 23 , 24 . In the following, we describe a successful effort to implement this approach. We produce a line-referenced, electro-optical modulation frequency comb (LR-EOFC) ∼ 1559.9 nm in the astronomical H band (1,500–1,800 nm). We discuss the experimental set-up, laboratory results and proof of concept demonstrations at the NASA Infrared Telescope Facility (IRTF) and the W. M. Keck observatory (Keck) 10 m telescope. Results Comb generation A LR-EOFC is a spectrum of lines generated by electro-optical modulation of a continuous-wave laser source 25 , 26 , 27 , 28 , 29 which has been stabilized to a molecular or atomic reference (for example, f 0 = f atom ). The position of the comb teeth ( f N = f 0 ± Nf m , N is an integer) has uncertainty determined by the stabilization of f 0 and the microwave source that provides the modulation frequency f m . However, the typical uncertainty of a microwave source can be sub-Hertz when synchronized with a compact Rb clock and moreover can be global positioning system (GPS)-disciplined to provide long-term stability 12 . Thus, the dominant uncertainty in comb tooth frequency in the LR-EOFC is that of f 0 . The schematic layout for LR-EOFC generation is illustrated in Fig. 1 and a detailed layout is shown in Fig. 2 . All components are commercially available off-the-shelf telecommunications components. Pictures of the key components are shown in the left column of Fig. 1 . The frequency-stabilized laser is first pre-amplified to 200 mW with an Erbium-Doped Fibre Amplifier (EDFA, model: Amonics, AEDFA-PM-23-B-FA) and coupled into two tandem lithium niobate (LiNbO 3 ) phase modulators ( V π =3.9 V at 12 GHz, RF input limit: 33 dBm). The phase modulators are driven by an amplified 12 GHz frequency signal at 32.5 and 30.7 dBm, and synchronized by using microwave phase shifters. This initial phase modulation process produces a comb having ∼ 40 comb lines (≈2 π × V drive / V π ), or equivalently 4 nm bandwidth. This comb is then coupled into a LiNbO 3 amplitude modulator with 18–20 dB distinction ratio, driven at the same microwave frequency by the microwave power recycled from the phase modulator external termination port. The modulation index of π /2 is set by an attenuator and the phase offset of the two amplitude modulator arms is set and locked to π /2. Microwave phase shifters are used to align the drive phase so that the amplitude modulator gates-out only those portions of the phase modulation that are approximately linearly chirped with one sign (that is, parabolic phase variation in time). A nearly transform-limited pulse is then formed when this parabolic phase variation is nullified by a dispersion compensation unit using a chirped fibre Bragg grating with 8 ps nm −1 dispersion. A 2 ps full-width at half-maximum pulse is measured after the fibre grating using an autocorrelator. Owing to this pulse formation, the duty cycle of the pulse train reaches below 2.5%, boosting the peak intensity of the pulses. These pulses are then amplified in a second EDFA (IPG Photonics, EAR-5 K-C-LP). For an average power of 1 W, peak power (pulse energy) is 40 W (83 pJ). The amplified pulses are then coupled into a 20 m length of highly nonlinear fibre with 0.25±0.15 ps nm −1 km −1 dispersion and dispersion slope of 0.006±0.004 ps nm −2 km −1 . Propagation in the highly nonlinear fibre causes self-phase modulation and strong spectral broadening of the comb 30 . Comb spectra span and envelope can be controlled by the pump power launched into the highly nonlinear fibre. A typical comb spectrum with >600 mW pump power from the 1,559.9 nm laser is shown in Fig. 3a , with >100 nm spectral span. Moreover, by using various nonlinear fibre and spectral flattening methods, broad combs with level power are possible 31 . Figure 1: Conceptual schematics of the line-referenced electro-optical frequency comb for astronomy. Vertically, the first column contains images of key instruments. ( a – e ) The images are reference laser, Rb clock (left) and phase modulator (right), amplitude modulator, highly nonlinear fibre and telescope. A simplified schematic set-up is in the second column. Third and fourth columns present the comb state in the frequency and temporal domains. The frequency of N -th comb tooth is expressed as f N = f 0 + N × f m , where f 0 and f m are the reference laser frequency and modulation frequency, respectively. N is the number of comb lines relative to the reference laser (taken as comb line N =0), RV is radial velocity and δf N , δf 0 and δf m are the variance of f N , f 0 and f m . ( a ) The reference laser is locked to a molecular transition, acquiring stability of 0.2 MHz, corresponding to 30 cm s −1 RV . ( b ) Cascaded phase modulation (CPM) comb: the phase of the reference laser is modulated by two phase modulators (PM), creating several tens of sidebands with spacing equal to the modulation frequency. The RF frequency generator is referenced to a Rb clock, providing stability at the sub-Hz level ( δf m <0.03 Hz at 100 s). ( c ) Pulse forming is then performed by an amplitude modulator (AM) and dispersion compensation unit (DCU), which could be a long single mode fibre (SMF) or chirped fibre Bragg grating (FBG). ( d ) After amplification by an erbium-doped fibre amplifier (EDFA), the pulse undergoes optical continuum broadening in a highly nonlinear fibre (HNLF), extending its bandwidth >100 nm. ( e ) Finally the comb light is combined with stellar light using a fibre acquisition unit (FAU) and is sent into the telescope spectrograph. The overall comb stability is primarily determined by the pump laser. Full size image Figure 2: Detailed set-up of line-referenced electro-optical frequency comb. ( a ) The entire LR-EOFC system sits in a 19 inch instrument rack. Optics and microwave components in the rack are denoted in orange and black, respectively. Small components were assembled onto a breadboard. These included the phase modulators (PM), amplitude modulator (AM), fibre Bragg grating (FBG), photodetector (PD), variable attenuator (VATT), attenuator (ATT), highly nonlinear fibre (HNLF), microwave source, microwave amplifier (Amp), phase shifter (PS) and band-pass filter (BPS). The reference laser, erbium-doped fiber amplifier (EDFA), rubidium (Rb) clock, counter, optical spectrum analyser (OSA) and servo lock box are separately located in the instrument rack. ( b ) A simplified schematic of the fibre acquisition unit (FAU) is also shown. Stellar light is focused and coupled into a multimode fibre (MMF). The comb light from a single mode fibre (SMF), together with the stellar light in the MMF, are focused on the spectrograph slit and sent into the spectrograph. Full size image Figure 3: Comb spectra and stability of the C 2 H 2 and HCN reference lasers. ( a ) A typical comb spectrum from the 1,559.9 nm laser with >100 nm span generated with 600 mW pump power. The insets show the resolved line spacing of 12 GHz or ∼ 0.1 nm. ( b ) Experimental set-up: BP, optical band-pass filter; PD, photodiode. All beam paths and beam combiners are in single mode fibre. ( c ) Time series of measured beat frequencies for the two frequency-stabilized lasers with 10 s averaging per measurement. The x axes are the dates in November of 2013 and May/June of 2014, respectively. ( d ) Allan deviation, which is a measure of the fractional frequency stability, computed from the time series data of c . Right-side scale gives the radial velocity precision. Full size image The LR-EOFC system is mounted on an aluminum breadboard (18" × 32", or equivalently 45.7 × 81.3 cm) in a standard 19-inch instrument rack (see Fig. 2 ) for transport and implementation with the spectrograph at the NASA IRTF and at Keck II on Mauna Kea in Hawaii. The system is designed to provide operational robustness matching the requirements of astronomical observation. All optical components before the highly nonlinear fibre are polarization maintaining fibre-based, so as to eliminate the effect of polarization drift on spectral broadening in the highly nonlinear fibre. Moreover, no temperature control is required at the two telescope facilities. As a result, the comb is able to maintain its frequency, bandwidth and intensity without the need to adjust any parameters. During a 5 day run at IRTF, the comb had zero failures and the intensity of individual comb teeth was measured to deviate less than 2 dB, including multiple power-off and on cycling of the optical continuum generation system (see Fig. 4b ). Figure 4: Experimental results at IRTF. ( a ) Comb spectrum produced using 1,559.9 nm reference laser. The insets on top left and right show the resolved comb lines on the optical spectrum analyser. Comb spectra taken by the CSHELL spectrograph at 1,375, 1,400, 1,670 and 1,700 nm are presented as insets in the lower half of the figure. The blue circles mark the estimated comb line power and centre wavelength for these spectra. Comb lines are detectable on CSHELL at fW power levels. ( b ) Comb spectral line power versus time is shown at five different wavelengths. During the 5 day test at IRTF, no parameter adjustment was made, and comb intensity was very stable even with multiple power-on and -off cycling of the optical continuum generation system. ( c ) An image of the echelle spectrum from CSHELL on IRTF showing a 4 nm portion of spectrum ∼ 1,670 nm. The top row of dots are the laser comb lines, while the broad spectrum at the bottom is from the bright M2 II–III giant star β Peg seen through dense cloud cover. ( d ) Spectra extracted from c . The solid red curve denotes the average of 11 individual spectra of β Peg (without the gas cell) obtained with CSHELL on the IRTF. The regular sine-wave like blue lines show the spectrum from the laser comb obtained simultaneously with the stellar spectrum. The vertical axis is normalized flux units. Full size image Comb stability As noted above, the frequency stability of the LR-EOFC is dominated by the stability of the reference laser frequency f 0 . We explored the use of two different commercially available lasers (Wavelength

Promising new calibration tools, called laser frequency combs, could allow astronomers to take a major step in discovering and characterizing earthlike planets around other stars. These devices generate evenly spaced lines of light, much like the teeth on a comb for styling hair or the tick marks on a ruler—hence their nickname of "optical rulers." The tick marks serve as stable reference points when making precision measurements such as those of the small shifts in starlight caused by planets pulling gravitationally on their parent stars. Yet today's commercially available combs have a significant drawback. Because their tick marks are so finely spaced, the light output of these combs must be filtered to produce useful reference lines. This extra step adds complexity to the system and requires costly additional equipment. To resolve these kinds of issues, Caltech researchers looked to a kind of comb not previously deployed for astronomy. The novel comb produces easily resolvable lines, without any need for filtering. Furthermore, the Caltech comb is built from off-the-shelf components developed by the telecommunications industry. "We have demonstrated an alternative approach that is simple, reliable, and relatively inexpensive," says paper coauthor Kerry Vahala, the Ted and Ginger Jenkins Professor of Information Science and Technology and Applied Physics as well as the executive officer for Applied Physics and Materials Science in Caltech's Division of Engineering and Applied Science. The kind of frequency comb used by the researchers previously has been studied in the Vahala group in a different application, the generation of high-stability microwaves. "We believe members of the astronomical community could greatly benefit in their exoplanet hunting and characterization studies with this new laser frequency comb instrument," says Xu Yi, a graduate student in Vahala's lab and the lead author of a paper describing the work published in the January 27, 2016, issue of the journal Nature Communications. Scientists first began widely using laser frequency combs as precision rulers in the late 1990s in fields like metrology and spectroscopy; for their work, the technology's developers (John L. Hall of JILA and the National Institute of Standards and Technology (NIST) and Theodor Hänsch of the Max Planck Institute of Quantum Optics and Ludwig Maximilians University Munich) were awarded half of the Nobel Prize in Physics in 2005. In astronomy, the combs are starting to be utilized in the radial velocity, or "wobble" method, the earliest and among the most successful methods for identifying exoplanets. The "wobble" refers to the periodic changes in a star's motion, accompanied by starlight shifts owing to the Doppler effect, that are induced by the gravitational pull of an exoplanet orbiting around the star. The magnitude of the shift in the starlight's wavelength—on the order of quadrillionths of a meter—together with the period of the wobble can be used to determine an exoplanet's mass and orbital distance from its star. These details are critical for assessing habitability parameters such as surface temperature and the eccentricity of the exoplanet's orbit. With exoplanets that pass directly in front of (or "transit") their host star, allowing their radius to be determined directly, it is even possible to determine the bulk composition—for example, if the planet is built up primarily of gas, ice, or rock. In recent years, so-called mode-locked laser combs have proven useful in this task. These lasers generate a periodic stream of ultrashort light pulses to create the comb. With such combs, however, approximately 49 out of every 50 tick marks must be blocked out. This requires temperature- and vibration-insensitive filtering equipment. The new electro-optical comb that the Caltech team studied relies on microwave modulation of a continuous laser source, rather than a pulsed laser. It produces comb lines spaced by tens of gigahertz. These lines have from 10 to 100 times wider spacing than the tick marks of pulsed laser combs. To see how well a prototype would work in the field, the researchers took their comb to Mauna Kea in Hawaii. In September 2014, the instrument was tested at the NASA Infrared Telescope Facility (IRTF); in March 2015, it was tested with the Near Infrared Spectrometer on the W. M. Keck Observatory's Keck II telescope with the assistance of UCLA astronomer Mike Fitzgerald (BS '00) and UCLA graduate student Emily Martin, coauthors on the paper. The researchers found that their simplified comb (the entire electro-optical comb apparatus requires only half of the space available on a standard 19-inch instrumentation rack) provided steady calibration at room temperature for more than five days at IRTF. The comb also operated flawlessly during the second test—despite having been disassembled, stored for six months, and reassembled. "From a technological maturity point of view, the frequency comb we have developed is already basically ready to go and could be installed at many telescopes," says paper coauthor Scott Diddams of NIST. The Caltech comb produces spectral lines in the infrared, making it ideal for studying red dwarf stars, the most common stars in the Milky Way. Red dwarf stars are brightest in infrared wavelengths. Because red dwarfs are small, cool, and dim, planets orbiting these types of stars are easier to detect and analyze than those orbiting hotter sun-like stars. NASA's Kepler space observatory has shown that almost all red dwarf stars host planets in the range of one to four times the size of Earth, with up to 25 percent of these planets located in the temperate, or "habitable," zone around their host stars. Thus, many astronomers predict that red dwarfs provide the best chance for the first discovery of a world capable of supporting life. "Our goal is to make these laser frequency combs simple and sturdy enough that you can slap them onto every telescope, and you don't have to think about them anymore," says paper coauthor Charles Beichman, senior faculty associate in astronomy and the executive director of the NASA ExoPlanet Science Institute at Caltech. "Having these combs routinely available as a modest add-on to current and future instrumentation really will expand our ability to find potentially habitable planets, particularly around very cool red dwarf stars," he says. The research team is planning to double the frequency of the prototype comb's light output—now centered around 1,550 nanometers, in the infrared—to reach into the visible light range. Doing so would allow the comb also to calibrate spectra from sun-like stars, whose light output is at shorter, visible wavelengths, and thus seek out planets that are Earth's "twins." Other authors of the paper are Jiang Li, a visitor in applied physics and materials science, graduate students Peter Gao and Michael Bottom, and scientific research assistant Elise Furlan, all from Caltech; Stephanie Leifer, Jagmit Sandhu, Gautam Vasisht, and Pin Chen of JPL; Peter Plavchan (BS '01), formerly at Caltech and now a professor at Missouri State University; G. Ycas of NIST; Jonathan Gagne of the University of Montréal; and Greg Doppmann of the Keck Observatory.

10.1038/ncomms10436

Medicine

Should STEMI patients recover in the ICU?

BMJ (2019). DOI: 10.1136/bmj.l1927 Journal information: British Medical Journal (BMJ)

http://dx.doi.org/10.1136/bmj.l1927

https://medicalxpress.com/news/2019-06-stemi-patients-recover-icu.html

Abstract Objective To evaluate the effect of intensive care unit (ICU) admission on mortality among patients with ST elevation myocardial infarction (STEMI). Design Retrospective cohort study. Setting 1727 acute care hospitals in the United States. Participants Medicare beneficiaries (aged 65 years or older) admitted with STEMI to either an ICU or a non-ICU unit (general/telemetry ward or intermediate care) between January 2014 and October 2015. Main outcome measure 30 day mortality. An instrumental variable analysis was done to account for confounding, using as an instrument the additional distance that a patient with STEMI would need to travel beyond the closest hospital to arrive at a hospital in the top quarter of ICU admission rates for STEMI. Results The analysis included 109 375 patients admitted to hospital with STEMI. Hospitals in the top quarter of ICU admission rates admitted 85% or more of STEMI patients to an ICU. Among patients who received ICU care dependent on their proximity to a hospital in the top quarter of ICU admission rates, ICU admission was associated with lower 30 day mortality than non-ICU admission (absolute decrease 6.1 (95% confidence interval −11.9 to −0.3) percentage points). In a separate analysis among patients with non-STEMI, a group for whom evidence suggests that routine ICU care does not improve outcomes, ICU admission was not associated with differences in mortality (absolute increase 1.3 (−0.9 to 3.4) percentage points). Conclusions ICU care for STEMI is associated with improved mortality among patients who could be treated in an ICU or non-ICU unit. An urgent need exists to identify which patients with STEMI benefit from ICU admission and what about ICU care is beneficial. Introduction Survival after ST elevation myocardial infarction (STEMI) has increased by nearly 20% over the past two decades. 1 Complications from STEMI, such as cardiogenic shock or life threatening arrhythmias, have also fallen markedly. 2 This improvement is usually attributed to the accessibility and implementation of early reperfusion therapy, which typically occurs before patients are admitted to an intensive or coronary care unit (ICU). Nevertheless, 75% of patients with STEMI in the US are admitted to an ICU. 3 The costs of this practice are enormous. ICU admissions are on average 2.5 times more costly than non-ICU admissions, and critical care services now account for almost 1% of the US gross domestic product. 4 Whether ICU care for patients with STEMI provides a benefit over lower levels of care, such as general, telemetry, or intermediate care, is poorly understood. 5 Guidelines for STEMI care also provide conflicting advice. Recent European guidelines recommended admitting all patients with STEMI to an ICU. 6 Previous American guidelines suggested that STEMI patients at low risk may not need ICU level care, 7 but recent guideline updates did not specifically discuss the role of ICU care for STEMI. 1 This uncertainty is reflected in practice. Wide variation exists among hospitals in the use of ICUs for STEMI, both in the US and worldwide. 3 8 In this context, we sought to evaluate the effect of ICU admission on mortality for patients with STEMI in the US. As the decision to use an ICU is inherently linked to a patient’s severity of illness, previous observational studies were limited by confounding by indication. 9 To overcome this, we used an instrumental variable analysis to examine the effect of ICU admission on patients with borderline or discretionary ICU needs (that is, those patients who could reasonably receive care in an ICU or non-ICU unit). As mortality after STEMI is declining owing to broader use of reperfusion therapy, 10 we hypothesized that ICU admission would not be associated with a mortality benefit. Methods Study cohort We did a retrospective cohort study of all fee-for-service Medicare beneficiaries aged 65 years and older who were admitted to a hospital in the US for STEMI between January 2014 and October 2015. We identified patients with STEMI by using international classification of diseases, ninth revision, clinical modification (ICD-9-CM) primary diagnosis codes for STEMI (supplementary table A). 10 We excluded patients with STEMI admitted to hospitals without ICU capabilities or admitted directly (that is, transferred in) from other acute care hospitals. We also excluded patients for whom data necessary for analyses were missing (such as ZIP codes (n=1415), hospital characteristics (n=805), or hospital identifiers (n=2)) (supplementary figure A). Data source We linked inpatient claims from the Medicare Provider Analysis and Review File to mortality data in the Medicare Beneficiary Summary File. 11 Characteristics of hospitals came from the American Hospital Association’s Annual Surveys and the Healthcare Cost Reporting Information Systems. 12 13 We obtained population and geographic information by linking the patient’s ZIP code of residence to 2010 US census data. Patient and public involvement Patients were not involved in the development of the research question, study design, or outcome measures. Patients were not involved in the recruitment, conduct, or interpretation of the study. There are no specific plans to disseminate the results of the research to study participants or to relevant patients beyond the usual channels of publication. Exposure variable, outcome variable, and covariate definitions We used room and board charges for each Medicare beneficiary to establish the level of care for which the patient was billed. Thus, we relied on the US Centers for Medicare and Medicaid Services’ definition of an ICU, which is based on two criteria. 14 Firstly, an ICU must have lifesaving equipment available for immediate use and be geographically and identifiably separate from general routine care areas. Secondly, a single nurse must take care of no more than two patients, and nursing staff cannot be shared between an ICU and other units that provide lower levels of care. The exposure variable was admission to ICU (the presence of an ICU or coronary care unit (CCU) revenue center code in the administrative billing record). 15 We defined non-ICU admission as any admission to a general/telemetry (the lack of any ICU or CCU revenue center code) or intermediate care ward (the presence of an intermediate ICU or intermediate CCU revenue center code). 16 The primary study outcome was 30 day all cause mortality, measured from the time of hospital admission. To account for differences between patients admitted to ICU or non-ICU units, we used patients’ characteristics such as age, sex, race/ethnicity, comorbid illness, 17 severity of illness, cardiac procedures performed during the hospital admission, median household income, and urbanicity for adjustment. We captured severity of illness through secondary ICD-9-CM diagnosis and procedural codes for acute organ dysfunction or mechanical ventilation (supplementary table A). 18 19 We identified percutaneous coronary intervention, coronary artery bypass graft procedure, and thrombolytics by using ICD-9-CM procedural codes. 20 21 Median household income was based on the patient’s ZIP code of residence, using 2010 US census data. Urbanicity was defined by the National Center for Health Statistics Urban-Rural Classification Scheme. 22 We also used characteristics of hospitals, including geographic region, teaching hospital status by resident to hospital bed ratio, hospital size by number of beds, ICU size by proportion of total hospital beds, proportion of Medicaid patients among all admitted patients, nursing ratio (nursing full time equivalents per 1000 patient days averaged over the entire hospital), and annual hospital STEMI case volume, for adjustment. Instrumental variable analysis Because admission to an ICU is likely to be correlated with severity of illness (that is, sicker patients are admitted to an ICU and are also more likely to die) and observational data often lack all variables needed for adjustment, standard multivariable regression is likely to be biased through confounding by indication. 9 We found that a multivariable adjusted model yielded biased estimates relative to those from an instrumental variable model using the Wooldridge’s score test of endogeneity (F 1,1726 =5.1; P=0.02). 23 Therefore, we used an instrumental variable analysis to account for confounding. In an instrumental variable analysis, an instrument is used to adjust a patient’s likelihood of receiving the treatment. In this study, we used differential distance as the instrument. It has been used previously in myocardial infarction by McClellan and colleagues to examine the effect of cardiac catheterization on mortality. 24 Conceptually, distance to a hospital acts as an instrument because, in general, most people choose their residence for reasons unrelated to nearby hospitals. However, when a person has a STEMI or other acute illness, they are likely to be taken to the nearest hospital. Differential distance represents the additional distance that a patient would need to travel, beyond the closest hospital, to arrive at a hospital in the top quarter of ICU admission rates for STEMI. We calculated differential distance as the difference between the distance from a patient’s residence to the nearest high ICU use hospital (that is, a hospital in the top quarter of ICU admission rates for STEMI) and the distance from a patient’s residence to the nearest hospital of any type. Distances were measured by using the linear arc distance function, which calculates the distance between the geographic coordinates of the hospital and the centroid of the patient’s residential ZIP code. The median differential distance to a high ICU use hospital was 6.8 miles. To show its validity, an instrument must meet two conditions. 25 Firstly, the instrument must be associated with the treatment. Secondly, the instrument should have no relation with the outcome, except through the treatment. As an example, in a randomized trial, the randomization tool may act as the perfect instrument, increasing or decreasing the probability that a person receives a given treatment while also being otherwise unrelated to the study outcome. In this study, we showed the strength of the instrument in three ways. Firstly, differential distance was highly correlated with ICU admission (partial F 1,1726 =64; P<0.001). An F statistic greater than 10 generally indicates that an instrument is strongly associated with the treatment. 25 Secondly, for every 10 mile increase in differential distance, the probability of ICU admission decreased by 1.8%. Thirdly, more patients who lived close to a high ICU use hospital were admitted to an ICU than patients who lived far from a high ICU use hospital: 73.8% (40 354/54 691 patients) who lived less than the median (6.8 miles) were admitted to an ICU compared with 63.8% (34 873/54 684 patients) whose differential distance was more than 6.8 miles ( table 1 ). Table 1 Patients’ characteristics by median differential distance to high intensive care unit (ICU) hospital * . Values are numbers (percentages) unless stated otherwise View this table: View popup View inline No empiric method exists to show that the instrument is not associated with the outcome other than through the treatment. 25 The recommended way of evaluating this condition is to stratify observed characteristics by the instrument and then carefully examine balance. 25 27 Balance of observed characteristics across the instrument provides confidence that unobserved characteristics are similarly balanced. We examined whether patients’ characteristics were balanced across the distribution of the instrument by using standardized differences ( table 1 ). Generally, standardized differences less than 0.1 indicate balance, 26 and we showed reasonable balance between the two groups, except for race/ethnicity and urbanicity. Imbalances in race/ethnicity and urbanicity are recognized to be inherent to the use of distance instruments. 27 Interpretation of instrumental variable results The results of a standard regression represent the treatment effect for the average patient. The results of an instrumental variable analysis represent the treatment effect for the statistically marginal patient. 28 Marginal patients represent those whose likelihood of receiving the treatment depended on the instrument. A patient in a randomized trial might only receive an experimental treatment if randomized to it, and a marginal patient in this study received ICU care only because they lived close to a hospital with a high ICU admission rate for STEMI. Thus, statistically marginal patients might be considered clinically to have borderline or discretionary ICU needs—they might receive care in an ICU or outside of an ICU depending on the hospital to which they are admitted because different providers might reasonably disagree about the best location for the patient’s care. Marginal patients cannot be identified within an instrumental variable analysis. 28 However, we estimated the size of the marginal population of patients and their population level characteristics by using the method of Angrist and Pischke. 29 The statistical code for these estimates is given in the supplementary methods. Statistical analysis To account for characteristics of patients and hospitals, we used multivariable logistic regression models. In the instrumental variable analysis, we used two stage least squares regression with adjustment for the same patient and hospital characteristics. To overcome the imbalances in race/ethnicity and urbanicity when using a distance instrument, we specifically included these characteristics in all adjusted models, including the instrumental variable analysis. 27 We used predictive margins to estimate adjusted absolute differences in outcomes. All models estimated robust standard errors with clustering at the hospital level. Subgroup, sensitivity, and falsification analyses To test whether the results were consistent for specific target populations, to assess for effect measure modification, and to examine mechanisms for identified differences, we did several subgroup, sensitivity, and falsification analyses. Firstly, to evaluate whether severely ill patients could be driving the association between ICU admission and mortality, instrumental variable analyses were stratified by organ failure score and also repeated after exclusion of patients with ICD-9-CM diagnosis or procedural codes for respiratory failure or shock. Secondly, given the imbalance in race/ethnicity and urbanicity when using a distance instrument, we did separate subgroup analyses with these characteristics. Thirdly, to examine whether patients with life limiting treatment preferences could influence the results, we did separate instrumental variable analyses stratified by whether or not patients were aged 80 years or older, excluding patients who received a billing code for palliative care, 30 or excluding patients who did not receive percutaneous coronary intervention, coronary artery bypass graft, or thrombolytics. To ensure that differences in outcomes based on age were not missed by dichotomizing a continuous variable, we did a non-linear two stage residual inclusion instrumental variable analysis and then used predictive margins to estimate adjusted absolute differences in 30 day mortality at 5 year age intervals. Fourthly, to assess whether particular hospital capabilities could be influencing the results, we did separate subgroup analyses excluding hospitals without percutaneous coronary intervention capabilities or intermediate care. Fifthly, to assess the consistency of the results to the modeling method, we repeated the instrumental variable analysis using a non-linear two stage residual inclusion model. 31 Confidence intervals for the two stage residual inclusion model were based on 3000 non-parametric bootstrap samples with replacement. Finally, as a falsification test, we repeated the instrumental variable analysis for non-STEMI patients, a group for which increasing data suggest that routine ICU care does not improve outcomes. 32 We used SAS 9.3 and Stata 14.2 for data management and analyses. The analytic code is included in the supplementary methods. All tests were two sided with a P value less than 0.05 considered significant. Results Between January 2014 and October 2015, 109 375 patients with STEMI were admitted to 1727 hospitals (supplementary figure A). Among these patients, 75 227 (68.8%) were admitted to an ICU. Patients admitted to an ICU were more likely to be younger, to be male, and to be sicker by the number of organ failures than non-ICU patients, although both had a similar number of Elixhauser comorbidities ( table 2 ). ICU patients were also more likely to receive procedures such as percutaneous coronary intervention, coronary artery bypass graft, or thrombolytics, as well as other procedures such as renal replacement therapy, 33 mechanical cardiac support, 34 cardiac arrest, 35 or targeted temperature management ( table 2 ). 35 STEMI patients from the south were more likely to be admitted to an ICU than were other patients. Differences were appropriately balanced by the instrument, except for race/ethnicity and urbanicity ( table 1 ). We identified 431 (25%) hospitals as having high ICU use for STEMI, with ICU admission rates of 85% or greater ( fig 1 ). High ICU use hospitals had lower rates of intermediate care use and were more likely to have fewer than 100 hospital beds and a higher proportion of hospital beds that were ICU beds ( table 3 ). Table 2 Patients’ characteristics by intensive care unit (ICU) admission. Values are numbers (percentages) unless stated otherwise View this table: View popup View inline Fig 1 Distribution of intensive care unit (ICU) admission rates for ST elevation myocardial infarction (STEMI). Each circle represents an individual hospital, based on its ICU admission rate for STEMI and then ranked by its ICU admission rate for STEMI Download figure Open in new tab Download powerpoint Table 3 Hospitals’ characteristics by intensive care unit (ICU) use. Values are numbers (percentages) unless stated otherwise View this table: View popup View inline Patients with STEMI who were admitted to an ICU had higher unadjusted 30 day mortality than STEMI patients admitted to non-ICU units (18.2% for ICU admission versus 13.8% for non-ICU admission) ( table 4 ). In a multivariable regression adjusted for characteristics of patients and hospitals, ICU admission for STEMI was associated with an increase in 30 day mortality compared with non-ICU care (17.0% for ICU admission versus 16.5% for non-ICU admission; absolute increase 0.5 (95% confidence interval 0.3 to 1.0) percentage points). Table 4 Association of intensive care unit (ICU) admission with 30 day mortality View this table: View popup View inline We estimated that approximately one in 10 patients in this study had borderline ICU needs (that is, received ICU or non-ICU care dependent only on their proximity to a given hospital) (supplementary table B). The population of patients with borderline ICU needs was more likely to be over the age of 85, have no organ failures, live in the western US, or have a median household income by ZIP code of more than $100 000 (£77 000; €90 000) (supplementary table C). In the instrumental variable analysis, ICU admission was associated with lower 30 day mortality compared with non-ICU admission (14.9% for ICU admission versus 21.0% for non-ICU admission; P=0.04), with an absolute reduction in 30 day mortality of 6.1 (−11.9 to −0.3) percentage points) ( table 4 ). Subgroup analyses showed results consistent with the main effect. Point estimates were consistent with the main result for subgroups of race/ethnicity and urbanicity. Point estimates favored a benefit to ICU admission across organ failures and age strata; after exclusion of severely ill patients, patients who received a billing encounter for palliative care, patients who did not receive an intervention for STEMI, or patients admitted to hospitals without percutaneous coronary intervention or intermediate care capabilities; and when the association of ICU admission was estimated using a two stage residual inclusion model. A falsification test in which the instrumental variable analysis was repeated for non-STEMI patients showed no mortality benefit associated with ICU admission (absolute increase 1.3 (−0.9 to 3.4) percentage points) ( fig 2 ; supplementary table D, supplementary figure B). Fig 2 Subgroup, sensitivity, and falsification analyses. All models used an instrumental variable analysis and adjusted for characteristics of patients and hospitals. Error bars represent 95% confidence intervals for absolute 30 day mortality differences (intensive care unit (ICU) v non-ICU care). STEMI=ST elevation myocardial infarction Download figure Open in new tab Download powerpoint Discussion Admission to ICU was associated with an absolute survival benefit of 6.1 percentage points at 30 days for STEMI patients with borderline or discretionary ICU needs. This association persisted in several subgroup and falsification analyses. Contrary to the prespecified hypothesis, we found that ICU care may be underused for certain patients with STEMI. This finding has important implications given the rising costs and use of critical care services globally. Findings in context Whether patients with STEMI benefit from ICU care has been uncertain. No randomized trials have evaluated the use of ICU care for STEMI, and guidelines do not offer consistent recommendations about whether to admit STEMI patients to an ICU. 1 6 7 As a result, tremendous variation exists in whether ICUs are used for patients with myocardial infarction, both in the US and globally. 3 8 This variation is clinically relevant because, despite the decrease in mortality for myocardial infarction over time, 10 our ability as clinicians to identify patients at high risk of decompensation has not necessarily improved. 36 37 Thus, although inpatient care for STEMI patients may be safer overall, certain populations of patients may continue to be at risk, and efforts to reduce the number of STEMI patients who receive ICU care on average may place these patients at even higher risk in the future. The results of this study apply specifically to statistically marginal patients and not to those with obvious indications for or against ICU care. 28 Marginal patients are those who were admitted (or not admitted) to an ICU solely on the basis of their proximity to a given hospital. Clinically, these patients likely have borderline or discretionary ICU needs. In other words, they may be admitted to an ICU in some but not all hospitals because clinicians may disagree about the optimal location of care. Thus, the borderline patients in this study represent a group of patients for whom there is clinical equipoise and unwarranted variation exists in care. Instrumental variable analyses are unable to identify individual patients who are “marginal” or “borderline.” Despite this, we were able to identify population level characteristics associated with these patients. For example, we found that the borderline population was more likely to be older than 85 and have no organ failures. This suggests that the population of patients in this study with those characteristics was particularly vulnerable to local hospital practices and, as a result, was more likely to benefit from ICU admission. Future research can use this profile of borderline patients to design a prospective trial to test whether expanded ICU access improves outcomes for patients with STEMI. What about ICU care might be beneficial to STEMI patients remains unclear. 5 Historically, CCUs were developed to monitor for post-infarction complications, such as life threatening ventricular arrhythmias or mechanical sequelae. 38 The incidence of these complications has been greatly reduced by early reperfusion therapy, 39 and most STEMI patients receive definitive reperfusion therapy before admission to an ICU or CCU. Thus, neither the existing literature nor our study precisely identifies what about ICU care might be particularly beneficial to these patients. ICUs in the US are primarily defined by their level of nursing care and by their ability to provide aggressive, lifesaving treatment. STEMI patients with borderline ICU needs, compared with other types of myocardial infarction patients, may benefit from the enhanced nursing care available in ICUs, allowing for earlier detection of complications or decompensation. 40 The results may also reflect the growing complexity of STEMI patients, who might also present with or develop non-cardiac conditions (for example, pneumonia), for which ICU care may be beneficial. 41 42 Finally, ICUs may be capable of providing more timely access to particular treatments or may have more effective protocols to ensure the provision of important care (for example, discharge drugs) than non-ICU units. 43 Thus, the results of this study may have more to do with limitations of non-ICU care rather than the direct benefits of ICU care, suggesting that certain hospitals may be better prepared to care for STEMI patients across different units. 44 Previous observational studies evaluating the role of ICU care are important to note for two reasons. Firstly, on comparison, the mortality rate in this study may seem unusually high relative to results from recent clinical trials. However, our 30 day mortality rates were consistent with other studies of myocardial infarction among the Medicare population, which is older and at higher risk than the general population. 45 46 47 Secondly, most previous studies suggested increased mortality among patients admitted to an ICU, but these studies are at risk for confounding by indication and for estimating average rather than the more clinically relevant marginal treatment effects (that is, the effect on patients who could reasonably be treated in either ICU or non-ICU settings). 48 49 50 Our unadjusted and adjusted results also showed increased mortality associated with ICU admission. Other instrumental variable analyses have similarly shown a shift in the effect compared with conventional regression analyses. 42 51 Strengths and limitations of study The quality of an instrumental variable analysis depends on the validity of the instrument. 25 We showed that differential distance was strongly correlated with ICU admission, showed balance of observed patients’ characteristics as a result of the instrument, and did several analyses to evaluate for residual confounding. We used the instrument to look at a separate condition for which ICU care may not be beneficial—non-STEMI—and found no association between ICU admission and mortality. 32 This study should be considered in the context of several limitations. Firstly, we used administrative data. Thus, certain clinically relevant variables such as infarct size, blood pressure, left ventricular ejection fraction, Killip class, 52 culprit vessel, or time to reperfusion were unavailable. Residual confounding based on these factors cannot be ruled out. However, the excellent balance shown by the instrument suggests that unmeasured differences may be balanced as well. In addition, we did several subgroup and falsification analyses, which showed consistent results. Collectively, these results suggest that the effect of differential distance on the outcome operated through changes in the use of ICU care and not through other potential confounders. Secondly, the study’s cohort consisted of US Medicare beneficiaries and may not generalize to STEMI patients younger than 65 or to non-US STEMI patients. Thirdly, this study treated intensive care and coronary care units as the same. In many real world settings, these units are often combined to care for critically ill patients; differences between the two units at some hospitals may affect patient care and should be considered in future work. Similarly, key differences may exist between intermediate and general ward/telemetry care that were not assessed in this study. Finally, most hospitals in this study were non-teaching hospitals, and whether clinical outcomes for STEMI patients may differ between teaching and non-teaching hospitals remains controversial. 53 54 55 Implications and conclusions This study has important implications for clinicians and health system leaders. Conventional wisdom in the US suggests that ICU care is generally overused and that efforts must be made to reduce the number of patients receiving ICU care. 5 56 However, this study, in combination with others, 42 57 indicates instead that ICU care may often be misdirected, with some patients experiencing underuse while others experience overuse. ICU admission for STEMI patients with borderline or discretionary ICU needs was associated with improved survival at 30 days. Methods to identify STEMI patients who might benefit from ICU care are needed and should be followed by randomized trials to test the effect of expanded access to ICU and the mechanisms that result in benefit from ICU care. What is already known on this topic On average, 75% of patients with ST elevation myocardial infarction (STEMI) in the US are admitted to an intensive or coronary care unit (ICU) Most patients with STEMI receive definitive reperfusion treatment before ICU admission Whether ICU admission is beneficial for STEMI patients in contemporary practice is unknown What this study adds This study suggests that ICU admission improves mortality for STEMI patients who could be treated in an ICU or non-ICU unit STEMI patients who might benefit from ICU care should be identified, and the effect of expanded ICU access for these patients should be tested Footnotes Contributors: TSV and BKN were responsible for study concept and design. TSV obtained funding. TSV and CRC acquired the data, and all authors were involved in analysis and interpretation of data. TSV and BKN drafted the manuscript, and all authors revised it critically for important intellectual content. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. TSV is the guarantor. Funding: This work was supported by NIH K23 HL140165 (TSV), K12 HL138039 (TJI), and R01 HL137816 (CRC). The funding organizations had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript. Competing interests: All authors have completed the ICMJE uniform disclosure form at (available on request from the corresponding author) and declare: no support from any organization for the submitted work other than that described above; no financial relationships with any organizations that might have an interest in the submitted work in the previous three years; and no other relationships or activities that could appear to have influenced the submitted work. Ethical approval: The Institutional Review Board for the University of Michigan approved the study and provided a waiver of consent (HUM00053488). Data sharing: The statistical code for the analyses is included in the supplementary materials. Additional code can be obtained from the corresponding author on request. Medicare data are not publicly available but can be obtained through the Center for Medicare and Medicaid Services. Transparency: The lead author affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. Disclaimer: This manuscript does not necessarily represent the views of the US Government or the Department of Veterans Affairs. This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: .

A trip to an intensive care unit can be more than twice as costly as a stay in a non-ICU hospital room, but a new study finds intensive care is still the right option for some vulnerable patients after a severe heart attack. The difficulty lies in determining which people are best served in the ICU while they recover. The new Michigan Medicine (University of Michigan) research, published in The BMJ, found ICU admission was associated with improved 30-day mortality rates for patients who had a STEMI heart attack and weren't clearly indicated for an ICU or non-ICU unit. "For these patients who could reasonably be cared for in either place, ICU admission was beneficial," says lead author Thomas Valley, M.D., M.Sc., an assistant professor of internal medicine at Michigan Medicine, who cares for patients in the intensive care unit. But Valley cautions against simply continuing to send nearly everyone to the ICU. "ICU care is a treatment just like any medication," Valley says. "Providers need to know whether it's right for an individual person just like we try to do with a prescription drug." The researchers analyzed Medicare data from more than 100,000 patients hospitalized with STEMI, or ST-elevation myocardial infarction, a dangerous heart attack that requires quick opening of the blocked blood vessel to restore blood flow. Those patients were hospitalized at 1,727 acute care hospitals across the U.S. in a nearly two-year period from January 2014 to October 2015, and most were sent to the ICU after treatment. "A lot of the focus is on getting these people to the cardiac catheterization lab as soon as possible to open up the blood vessel, but less is known about what you do after that," Valley says. Current U.S. guidelines don't address whether to send patients to the ICU, while European guidelines recommend the ICU. Valley says providers could use more clear guidance on how to make these decisions. In this study, the mortality rate was 6.1% lower after 30 days for those admitted to their hospital's ICU. Valley says the surprising results—in the face of other studies that show ICU overuse—demonstrate that ICU care is misdirected. 'An important debate in cardiology' This study addresses an important issue in ICU care, says Michael Thomas, M.D., an assistant professor of internal medicine who runs the Cardiac ICU at Michigan Medicine's Cardiovascular Center. "At Michigan Medicine, all of our STEMI patients are admitted to the Cardiac ICU," says Thomas, who was not involved with the BMJ paper. "However, knowing where to send these patients after STEMI is an important debate in cardiology right now." "Some recent studies suggest many patients don't need ICU level of care and that it wastes resources. But before we pull back from this model, we need to understand this problem more fully," he says. Across the nation, 75% of STEMI heart attack patients are sent to the ICU, most of the time after reperfusion treatment in the cath lab to open up the blocked vessel. ICU vs. non-ICU care People recovering from a STEMI are some of the very sick patients ICUs were originally designed for, so providers may not even think about disrupting the longtime status quo, Valley says. "The historical thinking was, 'Why not send everyone to the ICU?' Now, we see that there are risks associated," Valley says. "For example, in the ICU, you're more likely to have a procedure, whether you need it more or not. "We must also consider the risk of infection, sending someone to a unit full of really sick patients who might have C. diff or other serious infections." The sleep quality as people are recovering from their heart attacks may also be lower in the ICU, because patients are given such close nursing care, Valley says. That's necessary for the sickest patients, but it might be disruptive to those people on the bubble who could be getting better rest on a regular floor, he says. Medicare has requirements for what constitutes ICU care, such as high nurse staffing levels and access to lifesaving care. "Because of Medicare requirements, ICUs tend to be more similar across hospitals than non-ICUs," Valley says. "Perhaps some hospitals can take care of patients anywhere, while others really need to use the ICU at high rates in order to provide safe care." A clear benefit for some, increased cost for others Valley says these data show a clear benefit of ICU care for vulnerable patients, as opposed to non-STEMI patients studied who did not have a significant difference in mortality rates with or without ICU admission. "Physicians might look at STEMI patients and wonder, 'Do they really need the ICU? Could it harm them? Is it a good use of resources?'" Valley says. Valley, a member of U-M's Institute for Healthcare Policy and Innovation, has previously found ICU overuse occurred for less critical patients hospitalized for a flare-up of chronic obstructive pulmonary disease (COPD) or heart failure. In that study, ICU admission dramatically increased cost of care without an increased survival benefit. The next step, according to Valley, is to determine what is beneficial about the ICU for those patients who benefit from it. He says that could lead to hospitals adopting some ICU care practices on non-ICU floors. Valley hopes making non-ICU floors more similar to the ICU in some ways could improve outcomes while reducing cost of care and infection risk.

10.1136/bmj.l1927

Earth

Study shows Southern Arizona once looked like Tibet

James B. Chapman et al, Geochemical evidence for an orogenic plateau in the southern U.S. and northern Mexican Cordillera during the Laramide orogeny, Geology (2019). DOI: 10.1130/G47117.1 Journal information: Geology

http://dx.doi.org/10.1130/G47117.1

https://phys.org/news/2019-12-southern-arizona-tibet.html

Abstract Previous studies of the central United States Cordillera have indicated that a high-elevation orogenic plateau, the Nevadaplano, was present in Late Cretaceous to early Paleogene time. The southern United States Cordillera and northern Mexican Cordillera share a similar geologic history and many of the same tectonic features (e.g., metamorphic core complexes) as the central United States Cordillera, raising the possibility that a similar plateau may have been present at lower latitudes. To test the hypothesis of an elevated plateau, we examined Laramide-age continental-arc geochemistry and employed an empirical relation between whole-rock La/Yb and Moho depth as a proxy for crustal thickness. Calculations of crustal thickness from individual data points range between 45 and 72 km, with an average of 57 ± 12 km (2σ) for the entire data set, which corresponds to 3 ± 1.8 km paleoelevation assuming simple Airy isostasy. These crustal thickness and paleoaltimetry estimates are similar to previous estimates for the Nevadaplano and are interpreted to suggest that an analogous high-elevation plateau may have been present in the southern United States Cordillera. This result raises questions about the mechanisms that thickened the crust, because shortening in the Sevier thrust belt is generally not thought to have extended into the southern United States Cordillera, south of ∼35°N latitude. INTRODUCTION High-elevation orogenic plateaus like the modern Tibetan, Anatolian, and Altiplano-Puna plateaus commonly develop in the interior, or hinterland, of convergent orogenic systems. Construction of such plateaus is arguably among the most significant tectonic phenomena of Cenozoic time. The plateaus influence Earth systems in a wide variety of ways, including disrupting atmospheric circulation patterns ( Molnar et al., 1993 ), driving past climate change ( Raymo and Ruddiman, 1992 ; Strecker et al., 2007 ), concentrating metallic and other natural resources ( Hou and Cook, 2009 ), and altering plate motions ( Patriat and Achache, 1984 ; Iaffaldano et al., 2006 ). A similar plateau—the Nevadaplano, analogous to the Altiplano in the central Andes—was present in the Great Basin region of the central United States Cordillera (∼35°N–45°N latitude; Fig. 1 ) during Late Cretaceous to early Paleogene time ( DeCelles, 2004 ; Best et al., 2009 ; Snell et al., 2014 ). The Nevadaplano was isostatically supported by thickened continental crust formed by retroarc shortening in the Sevier thrust belt and in precursor thrust belts such as the Luning-Fencemaker ( Jones et al., 1998 ; DeCelles, 2004 ). Previous estimates for crustal thickness in the Nevadaplano using igneous geochemical proxies, similar to the technique employed in this study, range from 55 to 65 km ( Chapman et al., 2015 ). Structural restoration of late Paleogene to Holocene extension, including zones of high-magnitude extension associated with the Cordilleran metamorphic core complexes ( Fig. 1 ), also suggests that the Nevadaplano was supported by thick (40–65 km) crust ( Coney and Harms, 1984 ). Some researchers have suggested that the Nevadaplano may have extended farther southeast ( Whitney et al., 2004 ; Copeland et al., 2017 ). This idea is contentious, however, because the Sevier thrust belt does not extend into the southern United States Cordillera and northern Mexican Cordillera ( Fig. 1 ; Yonkee and Weil, 2015 ; Fitz-Díaz et al., 2018 ). Instead, Late Cretaceous to Paleogene contractional strain in these areas is generally characterized by basement-involved uplifts accompanying high-angle reverse faulting, formed during the Laramide orogeny ( Krantz et al., 1989 ; Clinkscales and Lawton, 2017 ; Fitz-Díaz et al., 2018 ). Horizontal shortening recorded by these high-angle reverse faults is insufficient to have significantly thickened the crust ( Davis, 1979 ). Thus, if the Laramide orogeny in southern Arizona and northern Sonora resulted in thickened crust, mechanisms in addition to tectonic shortening are required. To evaluate the possibility of thick crust supporting an orogenic plateau in the southern United States and northern Mexican Cordillera, we used whole-rock La/Yb ratios in Laramide-age, intermediate-composition continental-arc rocks as a proxy for crustal thickness ( Fig. 2 ), following the method of Profeta et al. (2015) . As crustal thickness increases, whole-rock heavy rare earth element (HREE) concentrations decrease and light rare earth element (LREE) concentrations increase, due to the high-pressure stabilization of HREE-enriched phases such as amphibole and garnet at the expense of LREE-enriched phases such as plagioclase ( Hu et al., 2017 ; Müntener and Ulmer, 2018 ). Despite the low concentration of LREE in plagioclase, the mineral is important because of its abundance in continental-arc rocks and because it is unstable at higher pressure, in contrast to other LREE-bearing accessory phases such as monazite. GEOLOGIC BACKGROUND Some early studies of the southern United States Cordillera suggested that the Sevier thrust belt was continuous from southern Nevada to northern Chihuahua ( Drewes, 1978 ). However, subsequent research has demonstrated that many of the thrust faults used to support a thrust belt interpretation are low-angle normal faults or other types of geologic contacts such as unconformities ( Dickinson, 1984 ; Krantz et al., 1989 ; Clinkscales and Lawton, 2017 ). As a result, most researchers now believe that the Sevier thrust belt terminates in the Mojave region of southern California ( DeCelles, 2004 ; Yonkee and Weil, 2015 ) and that shortening in southern Arizona and northern Sonora was temporally restricted to the Laramide orogeny and predominantly occurred along high-angle reverse faults, in part reactivating Late Jurassic to Early Cretaceous rift-related structures ( Davis, 1979 ; Krantz et al., 1989 ; Lawton, 2000 ; Favorito and Seedorff, 2018 ; Fitz-Díaz et al., 2018 ). Continental-arc magmatism migrated eastward through the study area ( Fig. 2 ) during the Laramide orogeny as the subduction angle of the Farallon slab shallowed ( Coney and Reynolds, 1977 ). Laramide igneous rocks in southern Arizona and northern Sonora are mainly intermediate, metaluminous, and calc-alkaline rocks that have radiogenic isotopic compositions inherited from the lithospheric province into which they were emplaced ( Lang and Titley, 1998 ; González-León et al., 2011 ; Chapman et al., 2018 ). During mid-Eocene time, toward the end of the Laramide orogeny and after arc magmatism had migrated through the region, silicic (SiO 2 >70 wt%), peraluminous granitoids were emplaced as sills, dikes, and plutons interpreted as products of crustal melting ( Miller and Bradfish, 1980 ; Haxel et al., 1984 ; Miller and Barton, 1990 ; Fornash et al., 2013 ). METHODS AND RESULTS To estimate crustal thickness, we utilized an empirical relation between igneous whole-rock La/Yb and Moho depth for modern continental arcs ( Profeta et al., 2015 ). The method was limited here to intermediate-composition rocks in order to avoid generally more mafic rocks that originated directly from the mantle and generally more felsic rocks that originated by partial melting of middle to upper crust or from highly fractionated melts. We used data only from rocks within the compositional ranges of SiO 2 = 55–70 wt%, MgO = 1–4 wt%, and Rb/Sr = 0.05–0.25 ( Chapman et al., 2015 ; Profeta et al., 2015 ). These ranges exclude all analyses of Eocene peraluminous rocks. The filtered data set consisted of 105 whole-rock geochemical analyses, 16 new and 89 compiled from literature sources. Only minimally altered and unmineralized samples were included. Sample information, geochemical data for new and compiled analyses, and analytical methods are presented in the GSA Data Repository 1 . Samples analyzed in this study came from locations between 108.5°W and 114°W longitude and 30°N and 35°N latitude ( Fig. 2 ). Crystallization ages of samples that passed through geochemical filters ranged from 81 to 50 Ma (see the Data Repository). All available data were used in the calculation of crustal thickness except for analyses from intrusive rocks associated with the Ray copper porphyry deposit in central Arizona ( Fig. 2 ; Lang and Titley, 1998 ). These were excluded because they exhibit anomalous REE trends compared to the rest of southern Arizona and northern Sonora ( Fig. 3 ). The unusual REE patterns at the Ray deposit may be related to derivation from or assimilation of the 1.4 Ga Ruin Granite, a HREE-enriched A-type granite that is the predominant component of the upper crust around Ray ( Banks et al., 1972 ). Estimates of crustal thickness from individual rock analyses range from 45 to 72 km, with an average uncertainty (1σ) of 10 km ( Fig. 2 ; Table DR2 in the Data Repository). Average crustal thickness estimates and average crystallization ages for areas with ≥ 5 samples are labeled in Figure 2 to provide a representation of data variability. Crustal thickness estimates for specific areas range from 48 to 62 km ( Fig. 3 ), with uncertainty ranging from 10 to 12 km (2σ). No clear correlation of age, location, and calculated crustal thickness was distinguishable in the data set, although the resolution of the data does not preclude possible correlations (Fig. DR1). As a result, we suggest that the best estimate of crustal thickness during the Laramide orogeny is obtained by considering the data collectively. Figure 4 presents all of our crustal thickness estimate results from the entire data set as a histogram and a kernel density estimate (KDE), which uses an adaptive bandwidth. The KDE is characterized by a large population of thicknesses (∼40% of data) centered on 61 km and a broad distribution of thickness values between 47 and 57 km. Average crustal thickness for the entire data set is 57 ± 12 km (2σ), where the reported uncertainty is the average single measurement uncertainty and standard deviation added in quadrature. Application of the empirical relation between whole-rock Sr/Y and crustal thickness as presented in Chapman et al. (2015) to the filtered data set indicates a median crustal thickness and median absolute deviation of 58 ± 16 km (1σ). We focused on the evaluation based on La/Yb primarily because of the lower uncertainty. DISCUSSION AND CONCLUSIONS Our estimate, 57 ± 12 km, for the average crustal thickness in the southern United States Cordillera and northern Mexican Cordillera during the Laramide orogeny is consistent with the concept of a high-elevation orogenic plateau. This crustal thickness estimate is similar to previous estimates (55–65 km) for the Nevadaplano during the same time span, ca. 90–45 Ma ( Chapman et al., 2015 ). The crustal thickness reported in this study is slightly higher than crustal thickness estimates (40–55 km) based on structural restoration of Cenozoic extension in southern Arizona ( Coney and Harms, 1984 ). For comparison, the average crustal thickness beneath the highest elevations in the Altiplano in the Central Andes is 60–70 km ( Ryan et al., 2016 ). An outstanding question is whether the southern United States and northern Mexican Cordillera was a true plateau—a low topographic relief surface—during the Laramide orogeny. Evidence for crustal anatexis during Late Cretaceous to Eocene time ( Miller and Bradfish, 1980 ; Haxel et al., 1984 ) suggests hot and low-viscosity middle to lower crust, which would have favored development of a low-relief orogenic plateau ( Bird, 1991 ). Using average values for continental crust density (2800 kg/m 3 ), upper mantle density (3300 kg/m 3 ), and continental crust thickness (37 km; Rudnick and Gao, 2003 ) and assuming Airy isostatic compensation, 57-km-thick crust corresponds to a paleoelevation of ∼3 km during or immediately following the Laramide orogeny. This paleoelevation falls within the range of estimates (1.5–4.5 km) based on carbonate δ 18 O for the Eocene American Southwest ( Licht et al., 2017 ). Sedimentary provenance and paleocurrent data from gravels along the Mogollon Rim of central Arizona (“rim gravels”; Fig. 2 ) also indicate an uplifted sediment source region south of the Colorado Plateau during the Laramide orogeny ( Elston and Young, 1991 ). Paleoaltimetric studies for the Nevadaplano suggest minimum elevations of 2.2–3.1 km during Late Cretaceous time based on carbonate clumped-isotope thermometry ( Snell et al., 2014 ) and as much as 3.5 km elevation during Oligocene time based on δD values of ancient meteoric water preserved in ignimbrite glasses ( Cassel et al., 2014 ). The major implication of this study is that some mechanism is required to have thickened the crust in the southern United States and northern Mexican Cordillera during the Laramide orogeny. Unlike the central United States Cordillera (the location of the Nevadaplano), the southern United States Cordillera was marked by extension associated with the Bisbee-McCoy-Sabinas rift system during Late Jurassic to Early Cretaceous time ( Dickinson and Lawton, 2001 ). Aptian–Albian marine limestone in southern Arizona suggests that the region was at or below sea level during mid-Cretaceous time ( Dickinson and Lawton, 2001 ) and only transitioned to a contractional regime at the start of the Laramide orogeny ( Chapman et al., 2018 ). Structural studies in the area indicate that deformation primarily occurred by folding or slip associated with high-angle reverse faults, which only accumulated a few to several tens of kilometers of horizontal shortening ( Davis, 1979 ; Krantz et al., 1989 ; Clinkscales and Lawton, 2017 ; Favorito and Seedorff, 2018 ). Conversely, parts of the thin-skinned Sevier thrust belt in the central United States Cordillera accommodated ≥ 300 km of horizontal shortening ( DeCelles and Coogan, 2006 ). To significantly thicken the crust in southern Arizona and northern Sonora, either (1) low-angle thrust faults are more prevalent than currently recognized, or (2) processes other than crustal shortening may have prevailed. Late Cretaceous, north-south–directed shortening in the Maria fold-and-thrust belt ( Spencer and Reynolds, 1990 ; Tosdal, 1990 ; Fig. 2 ) may have locally thickened the crust in west-central Arizona, but cannot explain elevated thicknesses elsewhere in the region. Despite the view that Laramide shortening in southern Arizona and northern Sonora primarily occurred by high-angle reverse faulting, low-angle thrust faults have been locally documented, including in the Catalina-Rincon Mountains (e.g., Gehrels and Smith, 1991 ; Arca et al., 2010 ; Spencer et al., 2011 ) and in and around the Baboquivari Mountains ( Haxel et al., 1984 ; Goodwin and Haxel, 1990 ; Fig. 2 ). More work is needed to document shortening magnitudes in these areas and test whether they are representative of a larger structural province or only local complexities. Erdman et al. (2016) suggested that magmatic additions in central Arizona during the Laramide orogeny may have thickened the crust and formed a Nevadaplano-like feature. Other mechanisms to form orogenic plateaus include underthrusting/underplating ( Zhou and Murphy, 2005 ), crustal inflation by lateral channel flow ( Bird, 1991 ; Husson and Sempere, 2003 ), and intracontinental subduction ( Tapponnier et al., 2001 ). None of these processes has been explicitly evaluated in the southern United States and northern Mexican Cordillera, which highlights how the postulated plateau challenges existing paradigms of the tectonic and geodynamic history of the region and raises questions as to whether this plateau is the southern extension of the Nevadaplano or a separate, distinct feature—the Arizonaplano. ACKNOWLEDGMENTS Chapman acknowledges support from U.S. National Science Foundation grant EAR-1928312. Constructive reviews by Peter Copeland, Tim Lawton, Ryan Crow, Science Editor Jim Schmitt, and anonymous reviewers improved the manuscript. 1 GSA Data Repository item 2020047, new and compiled geochemical data, data references, analytical methods, and supplementary figures; and new and compiled data in an Excel file, is available online at , or on request from [email protected] . © 2019 The Authors. Gold Open Access: This paper is published under the terms of the CC-BY license. /foreach /foreach /.widget-items /.module-widget <h3>Affiliations</h3> Archive Early Publication About the Journal Geology Science Editors Instructions for Authors Permissions About the Society Events Join the Society Publisher Bookstore Publisher Homepage Contact the Society Open Access Policy Online ISSN 1943-2682 ISSN 0091-7613 Copyright © 2023 Geological Society of America About GSW Our Story Contact Us How to Subscribe Privacy Policy Resources For Librarians For Industry For Society Members For Authors Help FAQ Terms & Conditions Explore Journals Books GeoRef OpenGeoSci Connect Facebook LinkedIn Twitter YouTube 1750 Tysons Boulevard, Suite 1500 McLean, VA 22102 Telephone: 1-800-341-1851 Copyright © 2023 GeoScienceWorld .global-footer-link-list { float: left; margin-right: 10px; } .global-footer-link-list { margin-left: 6.5rem !important; } .global-footer-link-list:first-of-type { margin-left: 0 !important; } .global-footer-link-list .list-title { font-family: "Lato",sans-serif; } .global-footer-link-wrap { align-items: stretch; display: flex; flex-direction: row; flex-wrap: wrap; justify-content: space-between; } .global-footer-link-list li { margin-bottom: 1.25rem !important; } .global-footer-link-list li a { color: #343434; font-size: 0.9rem; } .global-footer-site-details li { color: #343434; margin-bottom: 10px; } .global-footer-block .global-footer-site-details { display:block; } .global-footer-block .global-footer-site-details li { float:none; } .global-footer-block .global-footer-site-details li:first-of-type { text-align:left; } .global-footer-site-info a { display: block; height: 40px; margin-bottom: 10px; width:255px; } .pg_article .fb-featured-image { height: 154px; width: 120px; } /* SS padding fix */ .pg_sspage .page-column--center { padding:2rem; } /* Umbrella footer fix */ body[data-sitename="geoscienceworld"] .widget-SitePageFooter { background: #00436d; min-height: 70px; } body[data-sitename="geoscienceworld"] .ad-banner-footer { display:none; } /* HP layout fixes */ .pg_index .widget-instance-Home_MainContentB1 .widget-SelfServeContent { margin: 38px 25px 0 0!important; } .widget-instance-Home_MainContentB2 .widget-SelfServeContent.widget-instance-rmg_Home_Row3_Left { width:65%!important; } .pg_index .society-second-row .widget-SelfServeContent { flex: unset!important; } body[data-sitename="geologicalsocietyofamerica"].pg_index .widget-instance-Home_MainContentB1 .widget-dynamic-inner-wrap>div:nth-child(1) { max-width:100%!important; } body[data-sitename="geologicalsocietyofamerica"].pg_index .widget-dynamic__header { display:block!important; } .pg_index .row-related-titles .related-title-text a { color:#00436d!important; } .widget-instance-Home_MainContentB0 { background-repeat: repeat; } body.pg_Index[data-sitename="geoscienceworldjournals"] .widget-instance-Home_MainContentB0 div[class*="_Home_Row1_Middle"] { left: unset; bottom: unset; margin:auto!important; } body.pg_Index[data-sitename="geoscienceworldbooks"] .widget-instance-Home_MainContentB0 div[class*="_Home_Row1_Middle"] { left: unset; bottom: unset; margin:auto!important; } @media (max-width: 1024px) { body.pg_Index .widget-instance-Home_MainContentB1 .widget-SelfServeContent .homepage-panel-wrap .homepage-panel-image { margin-bottom:1rem; } } body.pg_Index ul.homepage-list { margin: 0!important; } @media (max-width: 900px) { body.pg_Index[data-sitename="geoscienceworld"] ul.homepage-list, body.pg_Index[data-sitename="geoscienceworldbooks"] ul.homepage-list, body.pg_Index[data-sitename="geoscienceworldjournals"] ul.homepage-list { column-count:2; } } @media (max-width: 600px) { body.pg_Index[data-sitename="geoscienceworld"] ul.homepage-list, body.pg_Index[data-sitename="geoscienceworldbooks"] ul.homepage-list, body.pg_Index[data-sitename="geoscienceworldjournals"] ul.homepage-list { column-count:1; } } .pg_Index .widget-SelfServeContent .global-footer-link-wrap ul { list-style-type: none; } /* Mobile menu fix */ @media (max-width: 930px) { .site-theme-header-menu { top: 50%!important; } } /* Footer logo fix */ body[data-sitename="claysandclayminerals"] .journal-footer-colophon, body[data-sitename="southafricanjournalofgeology"] .journal-footer-colophon { width: 100%; margin-top: 1rem; } /* My Account logo fix */ body.pg_MyAccount .global-nav { width: auto; } /* GeoRef Record spacing fix */ .geoRefRecordSummary { line-height:1.5rem; } /* Book TOC By - DOI consistency */ .pg_Book .book-bottom-section .chapter-authors-prefix { font-size: 14px; } .pg_Book .book-bottom-section .chapter-doi .chapter-doi-label { font-weight: 400; } /* Advanced Search Spacing fix */ .pg_AdvancedSearch { line-height: 1.5rem; } .pg_AdvancedSearch label.inline { padding: .5625rem 0; } .pg_AdvancedSearch .advanced-search-text { background-color:#F6F6F6; } /* My Account Spacing fix */ .pg_MyAccount h2 { margin-bottom:1rem; } .pg_MyAccount table { margin-top:1rem; } .pg_MyAccount input { border:1px solid #ccc; border-radius:4px; } .pg_MyAccount .input-wrap { margin:2rem 0; display:flex; align-items:center; } .pg_MyAccount .change-email-address-wrap input, .pg_MyAccount .change-email-address-wrap select { width: 400px; height: auto; margin-top:0.25rem; margin-bottom:0; } .pg_MyAccount label, .label-placeholder { font-weight:bold; width:250px; margin-right:1rem; text-align:right; } .pg_MyAccount .offset-wrap { margin: 1rem 0; margin-left:266px; } .pg_MyAccount .success-message { display:none; } /* SS h tag fix */ .pg_SelfServePage .content-main_content .widget-SelfServeContent h1, .pg_SelfServePage .content-main_content .widget-SelfServeContent h2, .pg_SelfServePage .content-main_content .widget-SelfServeContent h3, .pg_SelfServePage .content-main_content .widget-SelfServeContent h4, .pg_SelfServePage .content-main_content .widget-SelfServeContent h5 { width: auto; float: none; } /* Ordered Lists fix */ ol.number { margin-left:2rem; } /* Superscript fix */ sup { font-size:12px; } /* Save Search Modal fix */ .solrSearchSaveModel.reveal-modal.small.open div { margin: 1rem auto; } /* Splitview - Standard view icon fixes*/ .toolbar-wrap a.standard-view::before { font-family: white-label; content: "\e941"; } .toolbar-wrap a.split-screen::before { font-family: white-label; content: "\e940"; } /* CrossRef fixes */ .pg_CrossRefCitingArticles .content-main_content { margin: 0 auto; line-height: 1.5; float: none; max-width: 1200px; padding: 0; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby { padding: 1rem; margin: -2rem 0 2rem 0; background: white; border: 1px solid #ddd; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby-citedArticleCitation { margin: 1rem 0 } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby-citedByNoteText { padding: .75rem; margin: 1rem 0; border: 1px solid #ddd; border-radius: .25rem; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry { padding: 2rem 0; border-bottom: 1px solid #ddd; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry:last-of-type { border-bottom: none; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry-title { font-weight: bold; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry-author li { display: inline; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry-author li:after { content: ", "; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby__entry-author li:last-of-type:after { content: " "; } .pg_CrossRefCitingArticles .content-main_content .crossref-citedby-noResultsMessage { margin: 3rem 0; } table th, table td { white-space: unset; } .table-wrap table th, .table-wrap table td { white-space: nowrap; } .license {text-align: left;} [data-sitename="segdiscovery"].pg_Index .widget-info-wrap .article-issue-info .volume, [data-sitename="segdiscovery"].pg_Index .widget-info-wrap .article-issue-info .issue { display: none; } .sticky-footer .icon-general-close {display: none;} .token-wrap .institution-token-redeem-container .token-redeem-item:last-of-type { border-bottom: none !important; } .token-account-link-wrap { border-bottom: 1px solid #dbdbdb; } .token-wrap { padding: 1rem 1rem 0 1rem; } Full Text Full Text data&figures Data & Figures contents Contents georef GeoRef supplements Supplements references

A University of Wyoming researcher and his colleagues have shown that much of the southwestern United States was once a vast high-elevation plateau, similar to Tibet today. This work has implications for the distribution of natural resources, such as copper, and provides insight into the formation of mountains during the subduction of tectonic plates. "We normally think of southern Arizona and the surrounding areas as hot, cactus-laden deserts with relatively low base elevations, below 3,000 feet," says Jay Chapman, an assistant professor in UW's Department of Geology and Geophysics. "However, our recent research suggests that, during the Late Cretaceous to Early Paleogene period (80-50 million years ago), the region may have had elevations in excess of 10,000 feet and looked more like the Tibetan plateau north of the Himalayan Mountains or the Altiplano in the Andes Mountains in South America." Chapman is lead author of a paper, titled "Geochemical evidence for an orogenic plateau in the southern U.S. and northern Mexican Cordillera during the Laramide Orogeny," which was published online Nov. 22 in the journal Geology. The print version will be published in February. Roy Greig, a Ph.D. student in the Department of Geosciences at the University of Arizona, and Gordon Haxel, a U.S. Geological Survey scientist based in Flagstaff, Ariz., are co-authors of the paper. Chapman and his colleagues analyzed the chemistry of igneous rocks to determine how thick Earth's crust was in the past and then related the thickness to elevation. "Earth's crust floats in the mantle just like an iceberg floats in the water, with a little bit sticking out above the surface," Chapman says. "When the crust is thicker, the height of mountains and the elevation of the land surface are higher, just like the height of an iceberg sticking out of the water is taller if the overall iceberg is larger." The study determined that the crust in southern Arizona was once almost 60 kilometers thick, which is twice as thick as it is today—and comparable to how thick the crust is in parts of the Himalayas. "While the ancient mountains were forming, magma intruded into the crust and formed rocks like granite," Chapman says. "When the crust was really thick, the magmas experienced extreme pressure from the weight of all the rocks above them, which caused distinctive changes in the types and the chemistry of the minerals that formed those rocks." One of the interesting questions the study raises is how the crust in southern Arizona became so thick in the past. "The most common way to make really thick crust is for tectonic plates to converge or collide, which produces large earthquakes and faults that stack rock masses overtop one another," he says. "The prevailing view of southern Arizona is that there was never enough faulting in the area to make the thickness of crust we observe. It is a bit of a conundrum as to how such thick crust was generated." Adam Trzinski, a first-year Ph.D. student at UW, is now tackling this problem and searching for ancient faults in southern Arizona that could help explain how the thick crust became so thick. In addition to helping understand plate tectonic processes, the study may help explain why copper is so abundant in southern Arizona. "Several previous studies have noted a correlation between large copper ore deposits and regions of thick crust," Chapman says. "For example, there are many copper mines in the Andes Mountains in Chile. The results of this study strengthen that correlation and may aid in exploration efforts."

10.1130/G47117.1

Computer

Autonomous steering system keeps human drivers engaged

Tomohiro Nakade et al, Haptics based multi-level collaborative steering control for automated driving, Communications Engineering (2023). DOI: 10.1038/s44172-022-00051-2

https://dx.doi.org/10.1038/s44172-022-00051-2

https://techxplore.com/news/2023-01-autonomous-human-drivers-engaged.html

Abstract Increasing the capability of automated driving vehicles is motivated by environmental, productivity, and traffic safety benefits. But over-reliance on the automation system is known to cause accidents. The role of the driver cannot be underestimated as it will ultimately be the most relevant aspect for trust building and social acceptance of this technology. Here we introduce a driver-oriented automation strategy to achieve collaborative steering. Our approach relies on three major functionalities: interaction, arbitration, and inclusion. The proposed control strategy is grounded in the concept of shared control enabling driver intervention over the automation without deactivation. Well-defined physical human-robot interaction types are made available with the arbitration strategy. The automated driving trajectory is adapted to include the driver intent into the tactical level of trajectory planning. This enables driver initiated rerouting and consistent coordination of all vehicle actuators. In this way, automated vehicles, which rely on sight only, are augmented with the incorporation of the driver intent. The driver is neither replaced by nor excluded from the automation, rather their role remains active to the benefit of trust building and driving safety. Introduction Advanced driver assistance systems (ADAS) used in partial automation are intended to reduce the driver workload without causing disengagement. Level-2 automation splits the responsibility of the real-time operational and tactical functions to operate a vehicle safely in on-road traffic 1 , where the driver is responsible for the ‘object and event detection and response (OEDR)’, while the vehicle operates the sustained lateral and longitudinal motion control 2 . A combination of two ADAS is used to comply with this definition. Active cruise control regulates the vehicle to a predefined speed and slows down to maintain a preset distance with any slower moving vehicles ahead. Lane centering assistance (LCA) operates the steering system to track the trajectory computed by the automation (or AD trajectory), which typically is the center position of the lane in which the vehicle is traveling. Moreover, assistance for lane change is available in some vehicles. The automated lane change (ALC) function provides guidance to support the driver when the traffic condition is safe. Level-0 ADAS functions, such as automatic emergency braking for the longitudinal displacement or lane keeping assistance (LKA) for the lateral deviation complete the active safety envelope. Providing an interactive environment with the steering system, where manual and automated inputs can coexist alleviates the risk of disengagement. Hence, lateral control of the vehicle is often shared so that manual steering over the guidance torque of the automation is possible without deactivation. Here, shared control is defined following 3 : ‘human(s) and robot(s) are interacting congruently in a perception-action cycle to perform a dynamic task that either the human or the robot could execute individually under ideal circumstances. This definition excludes full automation (where there is no human) or manual control (where there is no automation)’. The concept of haptic shared control (HSC) has received significant attention due to the anticipated benefits to safety for partial and conditional automation levels 4 , 5 , 6 , 7 , 8 . Haptic communication through the steering interface is suggested to be the most practical channel to bond driver and vehicle because of its bilateral and dynamic characteristics 4 , 6 , 9 . Most partially automated vehicles use a blended control scheme for HSC (Fig. 1 a). It finds its origin in robotic force control under the name of ‘parallel force/position control’ 10 , 11 and is based on the idea that the driver and the automation can apply a torque command independently to the same actuator. In terms of control, the automation is a feedback loop of the steering displacement, in which a manual torque input is seen as an external disturbance to be rejected. Conceptually, blended control consists in modulating the angle controller impedance to enable driver intervention. The ADAS functions are realized through conditional operation of the blended control scheme. Typically, the gain G t and eventually the angle controller gains are programmed to satisfy the operating condition of each ADAS function. For example, the assistance provided from the LCA function is obtained by operating the steering system in shared control mode with 0 < G t < 1. The reaction torque to the driver is proportional to the tracking error and its derivative. This angular error is caused by manual intervention or variation of the tracking reference. Therefore, the reaction torque of the LCA represents haptic guidance directed towards the AD trajectory, which is intended to reduce the driver workload. Fig. 1: Two configurations of steering HSC between human driver and automation and overview of an automated driving control framework including the proposed collaborative steering control. The dashed lines represent the human driver control. a Blended control: The driver torque T d and the automation torque T a track their target angles θ d and θ a with the feedback of the measured angle θ p . The respective tracking efforts are superposed to form an electric power steering (EPS) motor torque command. The gain G t is used for attenuating the automation effort and enabling driver intervention in shared mode. G t is set to zero when operating the EPS in manual mode in the event of an override. b Admittance control: A virtual plant is used to estimate the manual deviation θ m from the measured driver torque input T t b . The angle control attempts to enforce the superposition angle of θ a and θ m by applying the command torque T m o t to the EPS. The reaction torque perceived by the driver is designed with the virtual plant and its load T a . Similarly to blended control, the gain G t is set to zero for manual steering. c The black blocks with the vision loops illustrate the typical structure of an automated vehicle control system. The proposed control is represented with the turquoise blocks. Arbitration allocates the control authority of the automation based on the interaction type. The driver and the automation interact through the virtual EPS. The resulting manual deviation is input into the steering angle control so as to enforce the superposition of the driver intent to the AD trajectory. Additionally, this deviation is propagated to the inclusion block to assimilate the driver intent into the trajectory planning. Full size image If the surrounding traffic situation allows for a safe lane change, ALC is activated upon confirmation that the driver holds the steering wheel and activation of the turn indicator. The ALC consists in the application of a predefined trajectory change toward the adjacent lane center. When the lane change is completed, LCA is again activated to maintain the vehicle centered in the new lane. LKA functions are often realized through brake activation to prevent lane departure. Torque vectoring is used to generate a vehicle yaw motion toward the lane center by applying asymmetrical commands to the individual brakes. This is a conservative approach that corrects the vehicle heading while reducing speed. In vehicles that employ the steering system for LKA, correction of the vehicle heading is achieved by adding a torque overlay 12 . As reported in the review of shared control for automated vehicles 8 , there are more than 100 contributions focusing on shared steering control, thus revealing the wide range of applications and implementations of the HSC concept. Nevertheless, most contributions focus on particular issues, such as how to prioritize the driver versus the automation and how to manage conflict, therefore providing only limited answers toward a unified and holistic approach. Further, state-of-the-art blended control has major disadvantages due to the dual role of tracking and regulation of the angle controller, which is typically a PID 13 . Ideally, perfect tracking is expected in the absence of driver input, while low rejection performance is required to enable manual intervention. Modulation of the control gain as a function of the driver activity is technically challenging because no sensor is available nor sufficiently reliable for this application (Supplementary Note 1) . Current practice is to consider each ADAS independently. This results in a discontinuous operation, which makes the driving experience uncomfortable. Consequently, drivers tend to display a low acceptance rate of ADAS technology 14 : LCA uses proportional and derivative gains independent of the driver input, while only the integrator is switched on to ensure zero steady-state error in the absence of driver intervention and switched off to avoid windup on manual input. Furthermore, the proportional and derivative gains are set to relatively low values to enable manual input, which lowers the tracking performance. Consequently, most partially automated vehicles have limited capability in tracking the lane in the case of road curvature. However, this centering torque is bounded by the driver input. Shared control is available under a preset driver torque threshold. Input above this threshold results in an override that deactivates the ADAS by returning the steering mode to manual ( G t = 0). When the driver torque decreases below the threshold, the ADAS is reactivated automatically by switching the steering mode back to HSC. Therefore, shared control for LCA is only available over a limited driver torque range resulting in discontinuous operation of the ADAS function 15 . Although ALC provides comfort during regular operation, the steering operation is switched to manual in the case of driver intervention. The assistance interruption is uncomfortable and eventually requires driver reactivation 16 , 17 . A torque overlay or offset is applied for LKA, which, in the worst case, results in the vehicle bouncing between the left and right lane markings. Shared control is not used because the low tracking performance of blended control does not guarantee lane tracking (as explained for LCA above) and therefore is not reliable for lane-keeping support in all road conditions (e.g., curves). While LCA and LKA share the control objective of centering the vehicle in the lane, they are not combined, increasing the risk of driver confusion. Although technical limitations of mass produced cars justify some of these design choices, partially automated vehicles are characterized by limited functional integration of shared control and discontinuous operation of the ADAS 18 . Therefore, a generic control framework for collaborative steering, consistent across tactical and operational vehicle controls and across all levels of automation is required to address these issues. In order to go beyond the classical form of driver-automation interaction this paper proposes a collaborative steering control framework within the limitation of mass produced steering hardware, that is based on the following functions: Interaction consists in providing the capability of haptic shared control to the steering system. Admittance control (Fig. 1 b) is applied to enable the driver to deviate the vehicle from the AD trajectory without impairing the tracking performance of the angle control. Arbitration refers to the allocation of roles among the driver and the automation when attempting to share the lateral control of the vehicle. There are four types of interaction: cooperation, co-activity, collaboration and competition (see “Methods” section for their definitions). Based on a preselected type of interaction, an arbitration rule is used to set the reaction torque of the automation according to the motor control of the driver. The parameters of the driver motor control: goal (or target angle) and impedance, have to be estimated with the sensors available in mass produced vehicle. Inclusion consists in adapting the AD trajectory to the driver intervention. If the manual deviation is sufficiently large and persistent in time, the automation assimilates this correction in the trajectory planning. Similarly to human-human collaboration 19 , 20 , the above three functions are essential for the realization of collaborative steering. An interactive control environment is a prerequisite for haptic communication with the driver 21 . Arbitration provides the capability of interacting in different manners according to the road, traffic, and driver conditions. Then, inclusion assimilates the deviation resulting from the interaction into the AD trajectory. While the frequency bandwidth of the interaction has to be compatible with that of the driver torque, inclusion occurs at the lower bandwidth of the vehicle motion. There have been various attempts to provide human-robot collaboration, but none of them have combined the three functions of interaction, arbitration and inclusion. For example, the literature 22 , 23 , 24 , 25 , 26 addresses the interaction and arbitration problem following different approaches, but omit inclusion. Collaboration cannot be achieved because the robot does not assimilate the human intent in its trajectory. As soon as the human stops interacting, the robot returns to its predefined trajectory. Conversely, the literature 27 , 28 , 29 , 30 proposes adaption of the trajectory based on manual intervention (human force or torque) without arbitration. While driver triggered re-routing of the AD trajectory becomes available, it is performed at the relatively slow dynamics of the vehicle, which is inappropriate for haptic interaction. Indeed, fine-tuning of a vehicle steering behavior is a subjective process that defines the vehicle performance. Degraded steering feel caused by low frequency interaction control is unacceptable. The main contribution of this work is the integration of the interaction, arbitration, and inclusion functions into a generic multi-level control framework that is applicable to mass-produced vehicles within the limitation of the available hardware. The control framework features the following advantages: Compatible with all levels of automation 0–4, where the human can still take part in the driving. Integration of the ADAS functions and continuous operation in shared control mode (override-free ADAS). ADAS functions that satisfy multi-objective requirements related to vehicle motion and driver intent to effectively contribute to better traffic safety. Figure 1 c gives an overview of the proposed control. The black blocks and the bottom half of the figure (gray background) represent the plant (detailed in “System dynamics” section) and the basic controls for automated driving. These controls are based on state feedback (positions and their derivatives) and rely on vision sensors (camera, radar, lidar, etc.). The turquoise blocks and the top half of the figure (light turquoise background) show the proposed control framework. The closed loop made with the arbitration and interaction blocks corresponds to the torque feedback of the admittance control illustrated in Fig. 1 b and detailed in “Interactive steering control” section. The arbitration block allocates the control authority of the automation based on the estimation of the motor control of the driver (“Estimation of driver motor control” section) and on a preset type of interaction (“Arbitration” section). Propagation of the manual deviation resulting from the interaction to the trajectory planning is realized with the inclusion block (“Inclusion of driver intent into the trajectory adaptation” section), which closes the haptic loop. Following the explanation of experimental configurations, the paper continues with the performance and experimental validation of the proposed control in the “Results” section. The “Discussion” section offers the contributions and limitations of the proposed control and the paper is concluded. Results Experimental configurations Four experiments have been conducted on different setups to test and validate the proposed control framework: virtual driver, human driver, driving simulator, and test vehicle (Fig. 2 ). These are summarized as follows and the symbols and parameters used for these experiments are listed in Supplementary Table 1 and Table 2 . Virtual driver configuration. The first experimental configuration consists of a column-type EPS and an electric motor to replicate the driver input (Fig. 2 a). Instead of the driver, an impedance-controlled motor is used for the validation of the estimation performance of the driver motor control (“Estimation of the driver motor control” section). The reference target angle and impedance of the virtual driver can be compared to their estimated values. Human driver configuration. The second setup uses the same equipment, but the impedance-controlled motor is replaced by a human driver (Fig. 2 b). The human driver is required to execute a sine-shaped maneuver through the different preset types of interactions. The estimation of the driver impedance validated in the first test configuration is confirmed in the case of manual steering and used to verify the arbitration rules (Eq. ( 12 )). Driving simulator configuration. The trajectory adaptation algorithm is validated on a static driving simulator (Fig. 2 c). The control environment includes trajectory planning, tracking control, and the shared control framework. The vehicle motion is simulated and displayed on a screen for visual immersion. The driving scenario is a double lane change on a three-lane 1.5 km straight course. The nominal trajectory of the automation lies in the center of the middle lane and the vehicle is controlled to track this nominal trajectory at 60 km h −1 using the Stanley trajectory tracking model 31 . The driver is required to operate the steering wheel only and is free to change lanes. Test vehicle configuration. This configuration concerns the implementation of the previously validated admittance control, arbitration rule, and trajectory adaptation in an actual test vehicle (Fig. 2 d). The vehicle tracks a predefined trajectory (nominal AD trajectory) at 60 km h −1 using cruise control on the same driving scenario and algorithm as that of the driving simulator configuration and position feedback from a high precision global navigation satellite system (GNSS). The driver is free to intervene and to deviate the vehicle away from the nominal AD trajectory. For the quantitative study (“Driver quantitative study” section), the driving scenario is the double lane change with 100 m intervals, as illustrated in Fig. 2 d. Fig. 2: Test equipment. a Virtual driver configuration. An impedance-controlled motor is used instead of the driver for the validation of the estimation of the driver motor control (driver goal and impedance). b Human driver configuration for the validation of the actual driver motor control and of the arbitration rules. c Driving simulator configuration for the validation of the trajectory adaptation. d Test vehicle configuration used for the proof of concept and the quantitative evaluation, and driving scenario for the quantitative evaluation. The following abbreviations are used: EPS for electric power steering, GNSS for global navigation satellite system, and AD for automated driving. Full size image Performance of the driver motor control estimation The performance of the estimation of the driver target angle (Eq. ( 15 )) and impedance (Eq. ( 18 ) and Eq. ( 19 )) were measured individually on the experimental configuration shown in Fig. 2 a. For the impedance estimation, the goal of the virtual driver was set to a sine wave and that of the automation to zero. For the goal estimation, the impedance was set randomly, as shown in Fig. 3 a. The estimation results are plotted in the same figure. The accuracy of the approximated driver goal varies as the driver impedance changes. When using the target angle of the virtual driver as input, the estimation of the driver stiffness and damping converge toward an oscillatory behavior about the set value. These oscillations stem from the driver impedance (Eq. ( 5 )) that is undefined when either driver input or tracking error goes to zero 32 . Fig. 3: Independent and combined estimations of the driver motor control measured on the test bench (virtual driver configuration). a The estimation of the driver target angle \({\hat{\theta }}_{d}\) is computed from the actual driver impedance Z d ,1 and Z d ,2 , while the estimated driver impedance \({\hat{Z}}_{d,1}\) and \({\hat{Z}}_{d,2}\) are calculated with the actual driver target angle θ d . b The estimation of the driver target angle \({\hat{\theta }}_{d}\) is computed from Eq. ( 13 ) and the estimated driver impedance \({\hat{Z}}_{d,1}\) and \({\hat{Z}}_{d,2}\) are calculated with the estimated driver target angle. Full size image Figure 3 b shows the combined estimations under the same test conditions when the approximated driver goal is used as input for the estimation of the impedance. While the performance of the combined estimation is impaired, the impedance variations can still be extracted. Two major errors can be observed. First is the overestimation of the driver impedance in the steady-state conditions that is a consequence of the underestimation of the driver goal (Eq. ( 5 )). In practice, the control is tuned for safe operation. Indeed, an overestimated impedance would amplify the role allocation from the arbitration rule (Eq. ( 12 )). The second error is the amplification of the oscillatory behavior. This is caused because the model used in the extended Kalman filter (EKF) is different from that used for the approximation of the driver goal. This class of oscillatory problems caused by modeling error is well-known 33 . Verification of the arbitration rule Setting the automation effort with the arbitration rule according to a type of interaction (Eq. ( 12 )) is verified in this experiment. The human driver configuration shown in Fig. 2 b is used so that the human can take part in the experiment. The driver was asked to perform a steady slalom maneuver while the automation had the objective of driving in a straight line (Fig. 4 ). The type of interaction was changed every fifteen seconds in the following order: (i) co-activity, (ii) collaboration, and (iii) competition. Furthermore, the driver was asked to take his hands off the steering wheel during the last five seconds of each interaction type. The automation impedance is set to be constant ( κ = 0) for co-activity during the first 15 s. The measurements show that the pinion angle tracks the average angle between those of the driver and the automation in this particular case, where the driver accommodates the automation. Since the automation impedance is constant, the driver torque is simply proportional to the angular deviation from the automation target. During the next fifteen seconds, κ = 1, which corresponds to collaboration. The automation impedance is adapted based on the estimated driver impedance: the larger the driver impedance, the smaller the automation impedance. During the manual intervention, the driver perceives resistance from the higher authority of the automation at first. Then, as the automation detects driver engagement, the control authority is gradually transferred to the driver. Conversely, when the automation detects that the manual intervention fades, the automation impedance is recovered and the automation target angle is tracked. This demonstrates how automation backs up the human in the driving task with a continuous estimation of the driver motor control. From 30 s onward, κ = −1 which sets the competition type of interaction. The automation impedance increases according to that of the driver in order to oppose manual intervention. The driver has to apply higher torque to accomplish the same maneuver. Smaller values of κ enable stronger resistance and virtually full rejection of the driver intervention. Fig. 4: Interaction performance for different types of interaction measured on the human driver configuration. The estimation of the target angle \({\hat{\theta }}_{d}\) is computed from the road information and the measured driver torque T t b with Eq. ( 13 ), and the estimated driver impedance \({\hat{Z}}_{d,1}\) and \({\hat{Z}}_{d,2}\) is calculated with the estimated driver target angle \({\hat{\theta }}_{d}\) . Based on the estimated driver impedance \({\hat{Z}}_{d,1}\) and \({\hat{Z}}_{d,2}\) and a preselected type of interaction, an arbitration rule (Eq. ( 12 )) is used to modulate the automation impedance Z a ,1 and Z a ,2 , which finally generates the automation torque T a to track its target angle θ a with the feedback of the measured angle θ p . The type of interaction is set to co-activity ( κ = 0) from 0 to 15 s, to collaboration ( κ = 1) from 15 to 30 s, and to competition ( κ = −1) from 30 s onward. The sections with the gray background indicate time periods where the driver has his/her hands off the steering wheel. Full size image During the three time periods in which the driver is not holding the steering wheel (hands-off), the control authority is naturally returned to the automation (nominal impedance). This highlights that the automation works as a backup to the driver but also that sustained effort is required for any deviation away from the AD trajectory. Automation backup is suitable for automated driving level 3 or more but not at level 2, where the driver is required to be engaged in the driving task, as it may increase the risk of misuse. Performance of the trajectory adaptation This section summarizes the results obtained for the trajectory adaptation on the static driving simulator, shown in Fig. 2 c. Co-activity role allocation ( κ = 0) is chosen to focus on the trajectory adaptation without the estimation of the driver motor control. The measured torque, the AD trajectory, and the actual vehicle trajectory are compared when the adaptation is deactivated and activated for a double lane change maneuver (Fig. 5 ). When deactivated, the AD trajectory is fixed on the initial lane and the driver has to continuously apply torque to deviate the vehicle away from the AD trajectory (Fig. 5 a). The double lane change maneuver is performed at the expense of sustained effort as the automation continuously pulls the driver back to the AD trajectory. This interaction is of interest because it provides guidance to the driver. While large deviation may result in high interaction torque, it is fundamental as a haptic cue during local deviation. Fig. 5: Comparison of driver inputs over a double lane change maneuver with the trajectory adaptation inactive and active measured on the driving simulator (driving simulator configuration). The starting of the lane changes correspond to the points where the measured driver torque T t b arises. a The automation (AD) trajectory y r , o p t is not adapted, and the driver and the automation track the actual vehicle trajectory Δ y v . b The AD trajectory y r , o p t is adapted according to the driver intervention, and the driver and the automation track the actual vehicle trajectory Δ y v . Full size image As the trajectory shifts towards the next lane with the adaptation algorithm activated, the interaction torque relaxes and the vehicle is centered on that new lane (Fig. 5 b). The driver applies torque to initiate a local deviation, which triggers an adaptation of the trajectory if sufficiently large. The bounded interaction torque and the adaptive guidance constitute the relevant haptic cues for collaborative steering. Proof of concept on the test vehicle The previously validated arbitration and inclusion control algorithms were implemented on the test vehicle shown in Fig. 2 d. The driving scenario is the same double-lane change as that in the previous section and both arbitration and inclusion controls were active. Two responses are provided for the arbitration rule set to collaboration (Fig. 6 a) and to competition (Fig. 6 b). The practical verification of the proposed multi-level haptic control demonstrates a consistent response of the test vehicle. First, the driver interacts with the automation under the control allocation set with the arbitration rule. Because the driver and the automation impedances are complementary in collaboration mode, the torque peak that is observed is lower than that during co-activity (Fig. 5 b). Second, the lateral deviation induced from the interaction is propagated to the trajectory planning via the inclusion control. Consequently, the vehicle tracks the left and right lanes without sustained manual torque. Fig. 6: Performance of the complete multi-level haptic control measured on the test vehicle (test vehicle configuration). All symbols are compatible in Fig. 3 –Fig. 5 . a The type of interaction is set to collaboration ( κ = 1). b The type of interaction is set to competition ( κ = −2). Full size image In competition mode, the automation impedance varies together with that of the driver. Because of the relatively low value of κ , the driver cannot apply a torque high enough to deviate the vehicle from the AD trajectory without reaching the maximum capacity of the steering system. No assimilation of the manual intervention is verified, and the vehicle tracks the trajectory developed without haptic contribution. In summary, the haptic cue communicated to the driver is twofold. First, the driver feels the set role allocation and may be able to generate a lateral deviation of the vehicle. Second, if this deviation is large enough, inclusion occurs by means of haptic adaptation of the trajectory. As a consequence, the interaction torque remains bounded, indicating to the driver that his intervention has been assimilated. These haptic cues contribute to the realization of intuitive collaborative steering. Driver quantitative study This section presents an evaluation of several individuals to verify the usefulness of the proposed multi-level haptic control framework. Five participants with an average age of 35 (from 29 to 44) years old took part in the driving assessment on the test vehicle shown in Fig. 2 d. All participants were experienced drivers and reported an average annual travel distance of 5800 km. The participants were required to execute double lane change maneuvers at 60 km h −1 , as illustrated in Fig. 2 d, with the four different control modes (Table 1 ) set in random order. All participants drove twice under each control mode and the averages of these trials are used for this study. Table 1 Four control modes for the quantitative study Full size table Two criteria are proposed for the assessment of collaborative steering: driver effort (DrE) and steering entropy (StE). DrE corresponds to the driver torque steering effort throughout the test duration t s c 28 , 34 , while StE is a criterion representing the smoothness of the evolution of the steering angle that is commonly used to quantify maneuverability 35 , 36 . Their respective formulations are given in the “List of KPIs” section. Control modes with lower DrE and StE allow for a smoother operation with less effort for the driver. Statistical differences between the control modes were analyzed using one-way analysis of variance (ANOVA), and multiple comparisons between specific control modes were executed via paired samples t-tests. The results obtained with each control mode are summarized in Fig. 7 . Figure 7 a shows that the DrE significantly decreases in the order of modes 1–4 ( p < 0.001) according to the ANOVA results. In particular, there is a large gap in DrE under modes 3 and 4 compared to modes 1 and 2. This is interpreted as the result of the implementation of inclusion, where the AD trajectory was adapted to match the driver intention so that sustained manual torque was no longer required during the lane change maneuver. Fig. 7: Quantitative evaluation by driver effort (DrE) and steering entropy (StE) measured on the test vehicle (test vehicle configuration). DrE and StE are normalized with the min–max normalization method where the maximum value is the mean of mode 1 and the minimum value is zero, i.e. the lower bound of the physical range. The error bars represent the standard deviations between participants. a DrE evaluation between control modes. b StE evaluation between control modes. c DrE evaluation between control modes and participants. d StE evaluation between control modes and participants. Full size image Further, Fig. 7 b suggests that the application of arbitration reduces the average of StE both with and without inclusion, i.e. the maneuver was executed more smoothly. In particular, the lowest StE is obtained under mode 4, compared to mode 1 ( p < 0.04) and mode 3 ( p < 0.02), according to the t -test results. StE shown in Fig. 7 b suggests that the variability between participants was higher than that of DrE (Fig. 7 a), especially under modes 1 and 2. In addition, the participants can be classified into two groups: group 1 includes participants 1 and 2, and group 2 includes the others. The StE is smaller and the DrE larger in group 1 compared to group 2 (Fig. 7 c, d). From these trends, it can be inferred that participants of group 1 applied more effort to achieve a smooth maneuver, while participants of group 2 operated with a smaller torque input at the expense of smoothness. These variations in driver behavior suggest that control modes 1 and 2 may lead to a low rate of acceptance, whereas control modes 3 and 4, which yield a smaller variability, are likely to be accepted by a wide range of drivers. The comprehensive analysis of these two criteria (DrE and StE) suggests that the proposed control framework based on arbitration and inclusion has the potential to achieve smooth maneuvering with less effort for a wide variety of drivers. Discussion The proposed control framework enables collaborative steering through haptic control integration at the operational and tactical levels of automated driving vehicle. A broad spectrum of interactions between the driver and the automation is made available through the arbitration rules, and manually induced deviations are consistently assimilated and updated via trajectory planning. Compared to the literature 22 , 23 , 24 , 25 , 26 , which consider interaction and arbitration only, the proposed control framework prevents the vehicle from returning to the nominal AD trajectory after a driver intervention. Inclusion assimilates intervention as an additional factor alongside vision information for the rerouting of the AD trajectory (Fig. 5 b and Fig. 6 a). This enables maintaining continuous shared steering operation in the event of a manually induced lane or route change. For example, by tuning the inclusion parameters and selecting the appropriate interaction type, LCA, ALC, and LKA can be integrated consistently in partially automated vehicles. Compared to control algorithms that consider solely interaction and inclusion 29 , 30 , the proposed framework permits full rejection of the driver input by setting the interaction type to competition (Fig. 6 b). This means that it can accommodate any level of automation and the development of multi-objective ADAS functions. It is the arbitration that provides this capability (Fig. 7) . For partially automated vehicles, the LKA function can be enhanced with a high temporary reaction torque to prevent a collision. Also, this applies to highly automated vehicles where the automation can take the responsibility of the OEDR. The automation will have the authority to collaborate or compete with the driver depending on the road and traffic situation. Compared to control schemes that merely rely on inclusion 27 , 28 , the proposed control framework enables independent optimizations of the driver and vehicle responses. Driver reaction torque is tuned with the interaction and arbitration functions, while the vehicle motion is adapted with the inclusion function. This alleviates the tuning trade-off and results in higher overall performance. With the admittance control, the trade-off between the acceptance of the driver input and the tracking accuracy, which is found in blended control, is solved, and the tuning range is widened significantly. The implemented framework requires the interaction type between the driver and the automation to be set by a higher-level controller based on endogenous (driver state) and exogenous (road and traffic conditions) information. Since the selection of the interaction type is out of scope for this work, the appropriate interaction type setting according to the driving situation remains to be addressed in a future task. The adaptation of the automation impedance based on the preset type of interaction is simplified with Eq. ( 12 ) in comparison to the optimization method 37 (Fig. 4) . This approach allows the validation of the comprehensive concept of arbitration while satisfying the implementation requirement on mass production hardware. However, to faithfully realize the interaction types originally defined in the literature 37 , a control theory to minimize the cost function consisting of the angle tracking error and effort of the driver and the automation defined for each type of interaction is required. In a future task, this could be achieved by using non-linear model predictive control (MPC) 25 , 26 . Although the accuracy of the driver goal approximation used for arbitration is limited, it proves to be sufficiently rich to extract the dynamics of the driver impedance (Fig. 4) . Moreover, the implemented approximation method, which merely relies on an admittance model, requires relatively low computational power. Nevertheless, the driver goal and impedance are abstract concepts, and it has not been verified whether the values estimated by the current algorithm match the actual driver motor control. However, this is a conceptual proposal to roughly approximate how strongly the driver is steering the vehicle, in order to implement the arbitration strategy. A further limitation is that the EKF tuning for the driver impedance estimation is based on the assumption that the stiffness and damping of driver change simultaneously. This means that the EKF cannot capture situations where the driver operation causes an extreme change only in his stiffness or damping. This limitation could be improved by the comparison with measured driver operation information captured via electroencephalography (EEG) or electromyography (EMG) sensors. Using the manual angle as input to the vehicle model to estimate the yaw rate is robust to modeling errors compared to using the driver torque as suggested in the literature 27 , 28 . Furthermore, as the manual deviation is related to the type of interaction, propagation of this deviation to the trajectory planning enhances haptic consistency. Vehicle tests demonstrated the capability of interacting under the role allocated by the arbitration and the inclusion to manual intervention. However, the timing for that propagation from the initial manual intervention to the trajectory adaptation should be carefully adjusted to guarantee an acceptable steering feel. Hence, further fine-tuning and customization of the proposed control strategy is essential for intuitive haptic communication and driver acceptance. The analysis of DrE and StE suggests that the proposed control framework (Mode 4 in Table 1) can achieve smoother maneuvers with less effort for a wide variety of drivers compared to controls that use arbitration only (Mode 2 in Table 1) 22 , 23 , 24 , 25 , 26 or which consider solely inclusion (Mode 3 in Table 1) 27 , 28 , 29 , 30 . However, since the test samples are relatively small, it would be worthwhile to validate the proposed control with a larger number of participants to obtain a quantitative evaluation of greater statistical relevance. Conclusion A driver-centered automation control has been proposed to address the concept of collaborative steering in automated driving without alteration of the hardware available in mass-produced vehicles. According to a preset type of interaction, the driver steering intention is reflected in the automation impedance and trajectory planning. Because the implication of manual intervention affects both operational and tactical levels of automated driving control, intuitive haptic communication is made available to the driver and consistent integration across all vehicle actuators is supported. The originality of the proposed implementation is summarized as follows: The proposed multi-level control framework enables consistent integration of the ADAS functions while continuously operating in shared control mode. Furthermore, the high-performance angle control, combined with the large spectrum of interaction, makes this framework compatible with all automation levels where the driver can still be part of the driving. The admittance control has been applied to a steering system to enable interaction between driver and automation. The interactive nature of admittance control alleviates the trade-off found in blended control while ensuring superior tracking performance of both AD trajectory and driver intervention. Furthermore, the interactions taking place in the virtual plant are isolated from hardware limitations resulting in robust performance. A large spectrum of interaction between independent agents has been made available with the proposed rules of arbitration. Consideration of the context of collaborative steering enables the assumptions of independent interacting agents. The observability issue of the combined estimation of driver goal and impedance is avoided by considering the agent goals as boundary conditions and consequently, impedance modulation can be achieved. Practical reconsideration of classical two-level steering control within the context of collaborative steering resulted in the development of a simple approximation of the driver goal. The manual deviation from the AD trajectory resulting from the interaction is consistently propagated to the trajectory planning by using the manual angle as input. Through quantitative evaluation with five participants, the proposed multi-level haptic control has been validated in a vehicle. The assessment suggests a significant potential to provide smooth collaborative steering with less effort for a wide range of drivers. While the proof of concept on the test vehicle demonstrates the capability of this multi-level collaborative steering control, fine-tuning and customization is required to render the steering feel comfortable and consistent for a safer and more reliable shared driving experience. Finally, the application of the proposed control framework for the development of ADAS functions can be considered with the objective of encouraging driver engagement at partial automation or providing continuous automation back up to the driver at higher automation levels. Methods System dynamics The system enabling collaborative steering is composed of the driver, the automation, and an electric power steering (EPS) system, which represents the mechatronic interface. The EPS is composed of a steering wheel, a motor, gears, and angle and torque sensors, as shown in Fig. 8 . The dynamics of the EPS system can be described as: $${T}_{d}+{i}_{s}{T}_{mot}+\epsilon ={{\Psi }}$$ (1) where T m o t is the motor torque command, T d is the driver input torque, i s is the ratio of the reduction gear, ϵ is white noise in the driver and the automation torque, and Ψ is the dynamics of the EPS. Assuming that the components from the lower side of the torque sensor to the front wheel are stiff, Ψ can be simplified to a two-inertia system 38 : $${J}_{sw}{\ddot{\theta }}_{sw}={T}_{d}-{T}_{tb}$$ (2) $${T}_{tb}={K}_{tb}({\theta }_{sw}-{\theta }_{p})$$ (3) $${J}_{p}{\ddot{\theta }}_{p}={T}_{tb}+{i}_{s}{T}_{mot}+{T}_{ld}$$ (4) where J s w and J p are the steering wheel and the lower part of the torque sensor inertia, respectively, θ s w and θ p are the steering wheel and the measured pinion shaft angles, T t b and K t b are the torque sensor output and its stiffness, and T l d is a disturbance consisting of internal nonlinearities (friction, backlash, etc.) and the road load. Fig. 8: Structure of a dual pinion type electric power steering. In manual operation, the motor is controlled so that less effort is required for the driver when turning the wheels. For collaborative steering, the automation, which input is computed in the motor control unit (MCU), is controlled so as to support appropriately the driver. Full size image Both agents, the driver, and the automation are assumed to track their own trajectory based on individual impedance control loops. The motor control of the driver holding the steering wheel is formulated as follows: $${T}_{d}=-{Z}_{d}^{{\prime} }{\xi }_{d},\,{Z}_{d}=\left[\begin{array}{c}{Z}_{d,1}\\ {Z}_{d,2}\end{array}\right],\,{\xi }_{d}=\left[\begin{array}{c}{\theta }_{sw}-{\theta }_{d}\\ {\dot{\theta }}_{sw}-{\dot{\theta }}_{d}\end{array}\right]$$ (5) where Z d is the driver impedance ( \({Z}_{d}\in {{\mathbb{R}}}_{\ge 0}^{2}\) ), ξ d is the tracking error of the driver, and θ d is the target angle or goal of the driver. \({}^{{\prime} }\) represents a transpose matrix. The effort T a is the torque input of the automation in Fig. 1 b. It constitutes one of the components of the EPS motor torque T m o t (see the next section): $${T}_{a}=-{Z}_{a}^{{\prime} }{\xi }_{a},\,{Z}_{a}=\left[\begin{array}{c}{Z}_{a,1}\\ {Z}_{a,2}\end{array}\right],\,{\xi }_{a}=\left[\begin{array}{c}{\theta }_{p}-{\theta }_{a}\\ {\dot{\theta }}_{p}-{\dot{\theta }}_{a}\end{array}\right]$$ (6) where Z a is the automation impedance ( \({Z}_{a}\in {{\mathbb{R}}}_{\ge 0}^{2}\) ), ξ a is the tracking error of the automation, and θ a is the target angle or goal of the automation. To represent how the driver and the automation impedances may evolve over time, the following dynamic models are introduced 39 . $${\dot{Z}}_{d}(t)=-{T}_{z,d}^{-1}{Z}_{d}(t)+{T}_{z,d}^{-1}{Z}_{d}(t-1)$$ (7) $${\dot{Z}}_{a}(t)=-{T}_{z,a}^{-1}{Z}_{a}(t)+{T}_{z,a}^{-1}{Z}_{a}(t-1)$$ (8) where T z , d and T z , a are time-constant parameters for modulating the driver and the automation impedances. Interactive steering control The interpretation of “physical human-robot interaction” (pHRI) has significantly evolved over the past decades. While safety was originally the main concern in the case of physical contact with a robot, pHRI has been considered as an implicit means to communicate the human intention to a robot with the objective of jointly completing a task 40 . The literature 41 groups control strategies for pHRI into two categories: “indirect force control” and “direct force control”. The former controls the force through motion feedback, with typical applications of impedance and admittance controls. The latter has the objective of controlling the interaction force to the desired value based on the feedback from the actual force measurement. The objectives of the interactive control of the steering actuator are twofold: High-angle tracking performance Enabling manual deviation from the AD trajectory without impairing the angle-tracking performance An admittance control framework (Fig. 9 a) is proposed for the interactive steering control to overcome the limitations of blended control. Although admittance control is not commonly used for haptic interaction 42 , it is appropriate for the application of automated steering because of the high performance of position tracking and the availability of the measurement of the driver torque. Assuming that a lower-level controller linearizes and decouples the plant dynamics, a linear two-degree of freedom controller (feedback and feedforward) with a single set of gains is sufficient to guarantee constant position tracking performance under any operating condition. One of the advantages of admittance control is that the inner angle control loop is purposefully made stiff so as to ensure high tracking performance. Consequently, the AD trajectory is tracked accurately in the absence of interaction. Conversely, the outer torque loop is naturally closed in the presence of interaction 43 . The reference position of the automation θ a is corrected with an estimated manual deviation θ m computed from the dynamics of the virtual plant. $${J}_{vp}{\ddot{\theta }}_{m}={T}_{tb}+{T}_{a}$$ (9) The angle reference of the inner loop θ c m d is defined as the superposition of commands from the automation and the driver: $${\theta }_{cmd}={\theta }_{a}+{\theta }_{m}$$ (10) Hence, the automation angle control attempts to enforce the angle superposition of θ a and θ m by applying the motor torque command T m o t . Fig. 9: Detailed representation of the admittance control structure for haptic shared control and equivalent interaction dynamics of an admittance-controlled electric power steering (EPS). a The dashed lines represent the driver control. The inner loop is an angular position control, which is purposefully made stiff. The outer loop is activated only when the driver inputs torque. The virtual EPS computes an estimation of the manual deviation, which reflects the driver intent under the preset type of interaction. The manual deviation is superposed to the angular command of the automation (AD trajectory) to form the command of the inner loop. b The equivalent interaction dynamics is a two-inertia system coupled with the torque sensor, which stiffness is K t b . The steering wheel (inertia J s w ) represents the interface with the driver motor control (goal θ d and impedance Z d ), while the reaction from the automation (goal θ a and impedance Z a ) is applied to the virtual inertia J v p . Full size image The closed-loop system dynamics are obtained by substituting Eq. ( 2 ), Eq. ( 5 ), and Eq. ( 6 ) into Eq. ( 9 ) and assuming perfect tracking ( θ c m d ≈ θ p ). $$\begin{array}{l}-{J}_{sw}{\ddot{\theta }}_{sw}-{Z}_{d}^{{\prime} }{\xi }_{d}={J}_{vp}{\ddot{\theta }}_{m}+{Z}_{a}^{{\prime} }{\xi }_{a}\end{array}$$ (11) This equivalent two-inertia system is illustrated in Fig. 9 b. It shows that the torque felt by the driver ( \({T}_{d}={Z}_{d}^{{\prime} }{\xi }_{d}\) ) when interacting with the automation can be controlled by tuning the virtual plant and the automation effort ( \({T}_{a}={Z}_{a}^{{\prime} }{\xi }_{a}\) ). For stability reasons, the bandwidth of the outer torque control loop should be set lower than that of the inner angle control loop 42 , 43 , 44 . In practice, the inertia of the virtual plant is set to a value higher than that of the actual plant ( J v p > J p ). Hence, it is the automation effort that is modulated to render the interaction. As shown in the next section, an arbitration rule is used to allocate the automation control authority according to a preset type of interaction. In consequence, the admittance control framework enables manual deviation θ m of the vehicle from the AD trajectory θ a . When the manual intervention ends ( θ m = 0), the steering returns back to the AD trajectory ( θ c m d = θ a ). Arbitration Arbitration in pHRI is required to regulate the control authority of the robot when attempting to accomplish a common task according to a preset type of interaction. The literature 37 proposes a taxonomy of the types of interactions based on neuroscience and game theory: Assistance is an extreme case of cooperation where, typically, the robot (slave) is used to amplify the physical capability of the human (master). Cooperation takes place when the two agents work towards a common end and need each other to reach the goal. Part of cooperation is the education role arbitration, which is critical for gaining new capability by ensuring a certain degree of engagement. Co-Activity occurs when the two interacting agents, without knowledge of each other’s actions, incidentally succeed in a common task. Collaboration features no fixed roles distribution but rather adapts the distribution to accommodate the other while still considering its own perspective. Competition, similarly to collaboration, is a symmetric arbitration where the role distribution opposes the other while considering its own perspective. Review 40 cites numerous contributions for each interaction type. The type of interaction is likely to vary dynamically over the completion of a joint task. Endogenous (driver state) and exogenous (road and traffic conditions) information is used in a higher-level controller to set the interaction type, which is, however, outside the scope of this work. The objective of the arbitration is to define how the automation has to react to the driver intent based on the preset type of interaction. From Eq. ( 11 ) and assuming constant virtual plant inertia, two variables, the automation angle θ a , and its impedance Z a , are available for adjusting the reaction torque as a function of the driver goal θ d and impedance Z d . Two approaches have been proposed in the literature. The literature 24 focuses solely on the interaction and avoids consideration of the boundary conditions by opting for constant human and automation impedances. In this way, an arbitration rule was established based on the human goal and resulted in a large spectrum of interaction but with limited dynamic performance. Conversely, the literature 23 addresses the application of robotic rehabilitation, which relies on cooperation, with the human goal assumed to be equivalent to that of the robot. Under this assumption, impedance modulation was developed for this specific type of interaction. Here, the proposed arbitration considers that driver and automation are two independent agents. Therefore, their respective goals are boundary conditions that need to be identified separately. Then, the driver impedance can be estimated with the EKF when knowing the driver goal (detailed in the next section). Here, the following arbitration rule is proposed for the adaptation of the automation impedance: $${Z}_{a}={Z}_{a,0}-\kappa {\hat{Z}}_{d}$$ (12) where Z a ,0 is the nominal automation impedance, \({\hat{Z}}_{d}\) is the estimated driver impedance, and \(\kappa \in {\mathbb{R}}\) is a parameter used to set the type of interaction. For κ = 0, the automation impedance is constant, which corresponds to co-activity. This is the natural type of interaction obtained from the admittance control. For κ > 0, the automation adapts and supports the driver. This is the collaboration type of interaction. The opposite behavior or competition is obtained for κ < 0. Here the automation impedance increases with that of the driver, resulting in a rejection of the manual intervention. With this approach, the range of interactions from competition to co-activity and collaboration is made available. However, cooperation (including assistance) is not applicable because of the assumption made regarding the independent goals of driver and automation. However, note that cooperation-type interaction is already being used in the EPS control for manual operation: the EPS (automation) amplifies the manual torque so as to assist the driver in reaching their goal. Estimation of the driver motor control Realization of the arbitration relies on the availability of the driver goal and impedance. The observability issue of a combined estimation of these two variables with the interaction dynamics 32 (Eq. ( 11 )) is avoided with independent estimations. Indeed, it is assumed that the contextual nature of the joint task of driving defines the boundary conditions (driver and automation goals) of the interaction dynamics. Then, the driver and the automation vary their impedances as they interact under the constraints of their respective goals. The driver goal and impedance are abstract representations of how the driver interacts with the automation. Although numerous driver models have been proposed, their objectives are to represent the driver under particular conditions. These objectives range from vehicle tracking of a given trajectory with a virtual driver model to more elaborated driver models, which include trajectory planning with optimization preference (time, acceleration, braking, rpm, etc.) 45 . However, there is no practical and generic approach available that could predict where a driver intends to go in any situation. Similarly, various attempts to describe and identify the driver impedance have been proposed but they rely either on additional sensors (e.g. EMG, grip force, driver torque) or are laboratory-based setups with limited practical relevance 46 , 47 . The literature 48 proposes the identification of the driver impedance while driving under the assumption of a constant driver goal. Unfortunately, these approaches are not suitable for the estimation of the driver impedance while interacting with the automation. Considering the economic constraints of mass-produced vehicles with a limited number of sensors available, the driver goal and impedance can, at best, only get approximated roughly. Here, an approximation of the driver goal is computed at first. The sensors available in mass-produced vehicles are limited, so the two-level model of steering 49 is applied in the context of collaborative steering 7 . The driver anticipatory visual open-loop control is assumed to track the center of the lane as an inherent environmental constraint. Any deviation from it is considered as originating from a driver intent in the compensatory closed-loop control. Hence, the estimate of the driver goal \({\hat{\theta }}_{d}\) is composed of an environmental constraint θ env and of the driver intent θ int . $${\hat{\theta }}_{d}={\theta }_{{{{{\rm{env}}}}}}+{\theta }_{{{{{\rm{int}}}}}}$$ (13) A steady-state model of the vehicle motion with longitudinal speed v x and road curvature ρ as inputs is used for the computation of the environmental constraint: $${\theta }_{env}=\left(1-\frac{{M}_{v}{v}_{x}^{2}}{2{({l}_{f}+{l}_{r})}^{2}}\frac{{l}_{f}{C}_{f}-{l}_{r}{C}_{r}}{{C}_{f}{C}_{r}}\right)({l}_{f}+{l}_{r}){i}_{o}\rho$$ (14) where M v is the mass of the vehicle, l f and l r are the distance from the gravity center to the front and rear axles, respectively, C f and C r are the fronts and rear cornering stiffness, and i o is the overall gear ratio from the steering angle to the tire angle. A driver intent estimator is introduced to generate an approximation of the manual deviation away from the environmental constraint. It is assumed that a simple admittance model used to convert the driver torque to a desired future angle propagated by some time interval will provide a rough approximation of the driver intent 50 : $${\theta }_{int}={\iint} _{t}^{t+{t}_{i}}\frac{{T}_{tb}(t)}{{J}_{sw}+{J}_{d}}dt$$ (15) where J d is the driver inertia and t i is the propagation time. Both parameters can be tuned. With the available measurements of the torque T t b and pinion angle θ p as well as the estimate of the driver goal \({\hat{\theta }}_{d}\) , the EKF 51 is developed for estimating the driver impedance 52 . The measurement of the pinion angle allows the decoupling of the steering wheel inertia from the dynamics of the pinion 53 . Consequently, Eq. ( 2 ), Eq. ( 3 ), Eq. ( 5 ), and Eq. ( 7 ) are discretized at time interval Δ t to form the plant model for the estimation. $${x}_{t+1}={f}_{t}({x}_{t})+{w}_{t}$$ (16) $${y}_{t}={h}_{t}({x}_{t})+{v}_{t}$$ (17) where, $${f}_{t} =\left[\begin{array}{c}{\theta }_{sw,t}+{\dot{\theta }}_{sw,t}\Delta t\\ {\dot{\theta }}_{sw,t}+{J}_{sw}^{-1}({T}_{d,t}-{T}_{tb,t})\Delta t\\ {\hat{\theta }}_{d,t}+{\dot{\hat{\theta }}}_{d,t}\Delta t\\ {\dot{\hat{\theta }}}_{d,t}\\ {Z}_{d,1,t}+{T}_{z,d}^{-1}(-{Z}_{d,1,t}+{Z}_{d,1,t-1})\Delta t\\ {Z}_{d,2,t}+{T}_{z,d}^{-1}(-{Z}_{d,2,t}+{Z}_{d,2,t-1})\Delta t\end{array}\right]\\ {h}_{t} =\left[\begin{array}{c}{T}_{tb,t}\\ {\hat{\theta }}_{d,t}\\ {\dot{\hat{\theta }}}_{d,t}\end{array}\right]\\ {x}_{t} ={\left[\begin{array}{cccccc}{\theta }_{sw,t}&{\dot{\theta }}_{sw,t}&{\hat{\theta }}_{d,t}&{\dot{\hat{\theta }}}_{d,t}&{Z}_{d,1,t}&{Z}_{d,2,t}\end{array}\right]}^{{\prime} }$$ The following observer is formulated for the estimation of the driver impedance. $${\hat{x}}_{t+1/t}={f}_{t}({\hat{x}}_{t/t})$$ (18) $${\hat{x}}_{t/t}={\hat{x}}_{t/t-1}+{K}_{t}({ \, y}_{t}-{h}_{t}({\hat{x}}_{t/t-1}))$$ (19) The EKF gain is calculated as: $${K}_{t}={P}_{t/t-1}{\hat{H}}_{t}^{{\prime} }({\hat{H}}_{t}{P}_{t/t-1}{\hat{H}}_{t}^{{\prime} }+R)$$ where P can be obtained by solving the Riccati equations: $${P}_{t+1/t} ={\hat{F}}_{t}{P}_{t/t}{\hat{F}}_{t}^{{\prime} }+Q\\ {P}_{t/t} ={P}_{t/t-1}-{P}_{t/t-1}{H}_{t}^{{\prime} }{({\hat{H}}_{t}{P}_{t/t-1}{\hat{H}}_{t}^{{\prime} }+R)}^{-1}{\hat{H}}_{t}{P}_{t/t-1}$$ where \(\hat{x}\) is the state estimated by the EKF and \(\hat{F}\) and \(\hat{H}\) are Jacobian matrices, defined as follows. $${\hat{F}}_{t}={\left(\frac{\partial {f}_{t}({x}_{t})}{\partial {x}_{t}}\right)}_{{x}_{t} = \hat{{x}_{t}}},\quad{\hat{H}}_{t}={\left(\frac{\partial {h}_{t}({x}_{t})}{\partial {x}_{t}}\right)}_{{x}_{t} = \hat{{x}_{t}}}$$ where Q and R are the covariance matrices of the process noise w and observation noise v respectively, which have to be tuned based on the modeling error and the noise level of the target system. Through computation of the prediction and correction 33 with Eq. ( 18 ) and Eq. ( 19 ), the last two components of \(\hat{x}\) are estimated as the driver impedance Z d . Inclusion of driver intent into the trajectory planning The arbitration rule allocates the control authority of the automation according to the preselected type of interaction. Manual intervention causes a deviation from the AD trajectory. Sustained input from the driver results in a steady interaction torque, and when released, the vehicle returns to the AD trajectory. This section presents the inclusion of driver intervention into trajectory planning to realize collaborative steering. For example, during a manually triggered lane change maneuver, it is necessary to reflect the driver intent in the trajectory planning. Hence, the reaction torque remains bounded along the maneuver and the driver does not have to apply a sustained torque to keep the vehicle in the new lane. These effects on the reaction torque represent haptic cues that communicate to the driver how the automated steering collaborated during the maneuver. The proposed approach is inspired by the literature 27 , 28 for the integration of the driver intent into trajectory planning. However, rather than using the driver torque for the trajectory planning because of the absence of interactive steering control, the proposed approach uses the angular deviation resulting from the interaction. Consequently, collaborative steering is available only when the type of interaction enables a manual deviation from the AD trajectory, such as co-activity and collaboration. In the following, only the differences from the literature 27 , 28 are presented. Inclusion consists in adding a term that represents the manual intent into the trajectory planning. At first, the yaw rate of the vehicle γ m caused by the manual intervention is computed from a single track vehicle model 54 with the driver angle θ m as input (Fig. 10 a): $${\dot{x}}_{v} ={A}_{v}{x}_{v}+{B}_{v}{u}_{v}\\ {x}_{v} =\left[\begin{array}{l}\beta \\ {\gamma }_{m}\\ \end{array}\right]\,{u}_{v}={\delta }_{m}=\frac{{\theta }_{m}}{{i}_{o}}\\ {A}_{v} =\left[\begin{array}{cc}{a}_{11}&{a}_{12}\\ {a}_{21}&{a}_{22}\\ \end{array}\right],\,{B}_{v}=\left[\begin{array}{c}{b}_{11}\\ {b}_{21}\\ \end{array}\right]\\ {a}_{11} =\frac{-2({C}_{r}+{C}_{f})}{{M}_{v}{v}_{x}},\,{a}_{12}=\frac{2({l}_{r}{C}_{r}-{l}_{f}{C}_{f})}{{M}_{v}{v}_{x}^{2}}-1,\\ {a}_{21} =\frac{2({l}_{r}{C}_{r}-{l}_{f}{C}_{f})}{{I}_{z}},\,{a}_{22}=\frac{-2({l}_{r}^{2}{C}_{r}+{l}_{f}^{2}{C}_{f})}{{I}_{z}{v}_{x}},\\ {b}_{11} =\frac{2{C}_{f}}{{M}_{v}{v}_{x}},\,{b}_{21}=\frac{2{l}_{f}{C}_{f}}{{I}_{z}}$$ (20) where β is the side slip angle, v x is the longitudinal velocity, and I z is the yaw moment of inertia of the vehicle. Second, a constant turn ratio and velocity (CTRV) model is used for converting the calculated yaw rate into a driver desired lateral deviation. The CTRV model enables the computation of the lateral deviation Δ y d when the vehicle moves forward during a time horizon t s in stationary condition with constant vehicle longitudinal velocity v x and yaw rate γ m . The kinematics are given as follows 55 : $$\Delta {y}_{d}=\Delta {y}_{v}+\frac{{v}_{x}}{{\gamma }_{m}}(1-\cos ({t}_{s}{\gamma }_{m}))$$ (21) where Δ y d and Δ y v represent the lateral error between the driver desired lateral position and the AD trajectory and that between the current vehicle position and the AD trajectory as illustrated in Fig. 10 b. The inclusion of the driver intent uses this estimate of the lateral deviation as a new corrective term into the trajectory planning. The cost function used to select the optimal lateral trajectory y r , o p t from a predefined set of candidates y r ( i , k ) is augmented with the new corrective term. $${C}_{y}(i,k)= \,{k}_{j}{J}_{y}(i,k)+{k}_{t}{t}_{f}(k)\\ +{k}_{a}{({ \, y}_{rf}(i))}^{2}+{k}_{m}{({ \, y}_{rf}(i)-\Delta {y}_{d})}^{2}$$ (22) where \({k}_{j},{k}_{t},{k}_{a},{k}_{m}\in {\mathbb{R}}\) are the weights of the cost function components. k j J y ( i , k ) is the jerk-related term to account for driving comfort. k t t f ( k ) is the time-related term. \({k}_{a}{({y}_{rf}(i))}^{2}\) and \({k}_{m}{({y}_{rf}(i)-\Delta {y}_{d})}^{2}\) account for the deviation errors from both agents. The final lateral position y r f ( i ) is used with the completion time t f ( k ) for the computation of the trajectory candidates y r ( i , k ). Fig. 10: Vehicle and constant turn ratio and velocity (CTRV) model. a Representation of the single track vehicle model used for the calculation of the yaw rate γ m from the manual deviation θ m . β is the side slip angle, v x is the longitudinal velocity, and i o is the overall gear ratio from the steering angle to the tire angle δ m . l f and l r are the distance from the gravity center to the front and rear axles, respectively. b CTRV model for the calculation of the driver desired lateral deviation Δ y d when the vehicle moves with the constant yaw rate γ m and longitudinal velocity v x during a time horizon t s . Representation of the lateral deviation caused by the manual intervention is made in the Frenet coordinate. The AD trajectory is represented on the s-axis and any deviation from it corresponds to a relative displacement along the d -axis as Δ y v . Full size image Notice that the selected optimal lateral trajectory is tracked during no driver intervention only. In the case of manual intervention, the optimal lateral trajectory is continuously computed at a frequency higher than the completion time t f . Inclusion of the driver intent into the trajectory planning is realized with the term of the lateral position error from the driver in the cost function (Eq. ( 22 )). Consequently, the deviation caused by the driver intervention is propagated to the trajectory planning, thus preventing the occurrence of excessive and sustained interaction torque. Moreover, this assimilation transfers the manual correction of the AD trajectory consistently to the other actuators of the vehicle, such as the brakes and the accelerator (Fig. 1 c). List of KPIs The KPIs used for the driver quantitative study are listed as follows: Driver effort (DrE) Driver torque steering effort during the time of manoeuver: $$DrE=\int\nolimits_{0}^{{t}_{sc}}{T}_{tb}^{2}dt$$ (23) Steering entropy (StE) Algorithm to calculate the entropy: 1. Obtain the time-series steering angle data for each sampling time d t ( d t was set to 150 ms in this study with reference to the literature 56 ). 2. The future steering angle is predicted by quadratic Taylor expansion from the past three data points of the steering angle, and the prediction error between the predicted future steering angle and the actual steering angle is obtained. 3. Determine the 90 percentile value α centered at 0 degrees ( α was set to 0.25 from the average prediction error distribution of all participants when driving in conventional manual mode). 4. Divide the frequency distribution of the prediction error into nine bins based on the range of α (−5 α , −2.5 α , α , −0.5 α , 0.5 α , α , 2.5 α , 5 α ). 5. Calculate StE from the proportion P i of each bin using the following formula: $$StE=\mathop{\sum }\limits_{i=1}^{9}{P}_{i}lo{g}_{9}{P}_{i}$$ (24) Data availability The data that support the findings of this study are available from the corresponding authors upon reasonable request. Code availability The code that supports the findings of this study is available from the corresponding authors upon reasonable request.

Researchers from EPFL and JTEKT Corporation have developed an automated driving system based on the concept of "collaborative steering," which aims to increase transportation safety, efficiency, and comfort by encouraging active interaction between autonomous vehicles and their human drivers. Autonomous driving technologies have already been integrated into many mass-produced vehicles, providing human drivers with steering assistance in tasks like centering a vehicle in its lane. But the little data available on automated driving safety shows that placing too much control of a vehicle in the hands of automation can do more harm than good, as disengagement by human drivers can increase the risk of accidents. "Current vehicles on the market are either manual or automated, and there is no clear way of making their control a truly shared experience. This is dangerous, because it tends to lead to driver over-reliance on automation," explains Jürg Schiffmann, head of the Laboratory for Applied Mechanical Design in the School of Engineering. Now, researchers from the lab have collaborated with Japanese steering system supplier JTEKT Corporation to develop and successfully road-test a haptics-based automated driving system that integrates different modes of human-robot interaction. The researchers hope that their approach will increase not only the safety of automated driving, but also social acceptance of it. "This research was based on the idea that automation systems should adapt to human drivers, and not vice versa," says EPFL Ph.D. student and JTEKT researcher Tomohiro Nakade, who is also the first author on a recent paper describing the system published in the journal Communications Engineering. Credit: Ecole Polytechnique Federale de Lausanne Nakade adds that a good metaphor for the new system can be drawn from a transportation mode that predates automation: "A vehicle must be open to negotiation with a human driver, just as a horseback rider conveys his or her intention to the horse through the reins." Interaction, arbitration, and inclusion Unlike current automated driving systems, which use only cameras for sensory input, the researchers' more holistic approach integrates information from a car's steering column. It also encourages continuous engagement between driver and automation, as opposed to current automated systems, which are typically either switched on or off. "In automation in general, when humans are just monitoring a system but not actively involved, they lose the ability to react," says Robert Fuchs, a former EPFL Ph.D. student who is now an R&D general manager at JTEKT Corporation. "That's why we wanted to actively improve driver engagement through automation." The researchers' system achieves this thanks to three functionalities: interaction, arbitration, and inclusion. First, the system distinguishes between four different types of human-robot interaction: cooperation (the automation supports the human in achieving a goal); coactivity (the human and automation have different goals, but their actions impact one another); collaboration (human and automation assist one another in achieving different goals); and competition (human and automation activities are in opposition). Next, as the driver operates the vehicle, the system arbitrates, or moves between different interaction modes depending on the evolving situation on the road. For example, the car might switch from collaboration to competition mode to avoid a sudden collision threat. Finally, and still within the same control framework, the system integrates an 'inclusion' function: it recomputes the vehicle's trajectory whenever the driver intervenes—by turning the steering wheel, for example—rather than perceiving it as an override and switching off. A practical solution To test their system, the researchers developed experiments involving a simulated virtual driver and a human driver using a detached power steering system, a full driving simulator, and even field tests with a modified test vehicle. The field tests were carried out with the participation of five drivers on a JTEKT test course in Japan's Mie prefecture, by connecting the researchers' system to a standard sedan via an external controller. The researchers specifically tested drivers' experiences of steering smoothness and lane-changing ease, and their results confirmed the system's significant potential for increasing comfort and reducing effort for drivers through collaborative steering. "This is a very practical concept—it's not just research for research's sake," says Schiffmann, adding that the software-based system can be integrated into standard mass-produced cars without any special equipment. "It's also a beautiful example of a fruitful partnership between our lab and JTEKT, with whom EPFL has collaborated since 1998."

10.1038/s44172-022-00051-2

Medicine

Study of complex genetic region finds hidden role of NCF1 in multiple autoimmune diseases

Jian Zhao et al, A missense variant in NCF1 is associated with susceptibility to multiple autoimmune diseases, Nature Genetics (2017). DOI: 10.1038/ng.3782 Journal information: Nature Genetics

http://dx.doi.org/10.1038/ng.3782

https://medicalxpress.com/news/2017-02-complex-genetic-region-hidden-role.html

Abstract Systemic lupus erythematosus (SLE) is a heterogeneous autoimmune disease with a strong genetic component characterized by autoantibody production and a type I interferon signature 1 . Here we report a missense variant (g.74779296G>A; p.Arg90His) in NCF1 , encoding the p47 phox subunit of the phagocyte NADPH oxidase (NOX2), as the putative underlying causal variant that drives a strong SLE-associated signal detected by the Immunochip in the GTF2IRD1 – GTF2I region at 7q11.23 with a complex genomic structure. We show that the p.Arg90His substitution, which is reported to cause reduced reactive oxygen species (ROS) production 2 , predisposes to SLE (odds ratio (OR) = 3.47 in Asians ( P meta = 3.1 × 10 −104 ), OR = 2.61 in European Americans, OR = 2.02 in African Americans) and other autoimmune diseases, including primary Sjögren's syndrome (OR = 2.45 in Chinese, OR = 2.35 in European Americans) and rheumatoid arthritis (OR = 1.65 in Koreans). Additionally, decreased and increased copy numbers of NCF1 predispose to and protect against SLE, respectively. Our data highlight the pathogenic role of reduced NOX2-derived ROS levels in autoimmune diseases. Main Dozens of SLE-associated loci have been identified by genome-wide association studies (GWAS) and included on the Immunochip for fine-mapping 3 , 4 . Using the Immunochip, we genotyped DNA samples from SLE cases and healthy controls from Chinese, European-American and African-American ancestry groups. In Chinese, the strongest association signal was detected at rs73366469 (minor allele frequency of 28.7% in cases versus 12.6% in controls, P = 3.8 × 10 −29 , OR = 2.88) within the GTF2IRD1 – GTF2I intergenic region at 7q11.23 rather than SLE-associated GWAS loci ( Fig. 1a ), consistent with another Asian Immunochip study 5 . This association was confirmed in European Americans at a modest significance level ( P = 7.5 × 10 −3 , OR = 1.32), but not in African Americans ( Supplementary Table 1 ). Figure 1: The GTF2IRD1 – GTF2I – NCF1 region at 7q11.23. ( a ) Association plot of Immunochip variants. The locations of 1000 Genomes Project variants are indicated on the top. The allelic P value (–log 10 P ) for each Immunochip variant assessed for association with SLE in Chinese is plotted as a circle according to the location of the variant. A map of the SLE-associated SNPs described in this study is shown in the box outlined by a blue dashed line. ( b ) Large duplications at 7q11.23 containing NCF1 , NCF1B and NCF1C . Duplications are highlighted as red boxes in which the location of NCF1 , NCF1B or NCF1C is indicated by a triangle. The region shown in a is highlighted by a blue dashed box. ( c ) LD (shown as r 2 ) analyses in African-American (AfrAm), Chinese and European-American (EurAm) subjects with a 4:2 ratio of ΔGT/GTGT ( n = 100 for each ancestry group). Full size image Because rs73366469 has no strong functional implication (RegulomeDB score = 5) and is not in linkage disequilibrium (LD; r 2 < 0.04) with any Immunochip SNP, we hypothesized that the underlying causal variant(s) were not on the Immunochip and were in strong, modest and weak LD with rs73366469 in Chinese, European Americans and African Americans, respectively. In the 1000 Genomes Project data set, we found two non-Immunochip SNPs (rs117026326 and rs12667901) that were in stronger LD with rs73366469 in Asians than in Europeans and Africans ( Fig. 1 and Supplementary Table 2 ). In a subset of Chinese and European-American subjects genotyped using the Immunochip (Chinese-1 and EurAm-1) and independent replication data sets (Chinese-2 and EurAm-2), rs117026326, located in intron 9 of GTF2I , exhibited stronger association with SLE ( P meta Chinese = 4.4 × 10 −40 , OR = 2.94; P meta EurAm = 1.2 × 10 −4 , OR = 2.83; Supplementary Tables 3 and 4 ) than rs73366469 and rs12667901, and conditioning on rs117026326 rather than rs73366469 or rs12667901 eliminated association signals for the other two SNPs, suggesting that association of rs73366469 with SLE might be attributed to rs117026326. However, rs117026326 was nearly non-polymorphic in African Americans and was not associated with SLE ( Supplementary Table 5 ), which explains the lack of association between rs73366469 and SLE in African Americans. Located 62 kb from rs117026326 is NCF1 (neutrophil cytosolic factor 1), encoding the regulatory p47 phox subunit necessary for activation of the phagocytic NOX2 complex, which is likely an SLE-related gene because nonfunctional NOX2 exacerbates lupus features in lupus-prone MRL. Fas lpr mice and induces a lupus-like type I interferon signature, autoantibody production and immune complex deposition in the kidneys of BALB/c. Ncf1 m1J mice 6 , 7 ; NCF2 , encoding another NOX2 regulatory subunit, p67 phox , harbors the missense variant p.His389Gln that is associated with SLE risk in European Americans 8 . However, rs117026326 genotype was not associated with the transcript levels of NCF1 or two other neighboring genes, GTF2I and GTF2IRD1 , in the peripheral blood mononuclear cells (PBMCs) of patients with SLE and healthy controls ( Supplementary Fig. 1 ). Notably, the NCF1 region is barely covered by variants in the 1000 Genomes Project Phase 1 data set ( Fig. 1a ), probably because NCF1 has 98% sequence identity with the nonfunctional pseudogenes NCF1B and NCF1C at 7q11.23 that are located within large DNA duplications 9 ( Figs. 1b and 2a ). The strong association of rs117026326 with SLE and functional implication of NCF1 led us to further hypothesize that rs117026326 might tag causal variant(s) of NCF1 not present in the 1000 Genomes Project. Figure 2: Highly homologous sequence among NCF1 , NCF1B and NCF1C . ( a ) Alignment of the DNA sequences of NCF1B and NCF1C with that of NCF1 . Sequences of NCF1B and NCF1C that are not identical to that of NCF1 are shown in red. This figure was generated using the BLAT tool in the UCSC Genome Browser. ( b ) NCF1 variants GTGT, p.Arg90His and p.Ser99Gly and the corresponding sequences at NCF1B and NCF1C (highlighted in red). Sequences are based on human reference genome build GRCh38/hg38. Full size image The GTGT sequence within exon 2 of NCF1 is a well-characterized variant that distinguishes this gene from NCF1B and NCF1C , which contain a GT deletion (ΔGT) 10 ( Fig. 2b ). However, reciprocal crossover results in the presence of GTGT-containing NCF1B and NCF1C and ΔGT-containing NCF1 ( Fig. 3a ) 11 , 12 . To obtain the correct genotypes for NCF1 variants, we specifically amplified the NCF1 sequence using PCR targeting GTGT ( Supplementary Fig. 2 ) and measured the ΔGT/GTGT ratio by real-time PCR to exclude subjects carrying GTGT-containing NCF1B and NCF1C (3:3 or 2:4 ratio) or ΔGT-containing NCF1 (5:1 ratio) ( Fig. 3b,c ). Using this approach followed by Sanger sequencing, we resequenced the entire 15.5-kb NCF1 region in 45 Chinese subjects and identified 67 SNPs ( Supplementary Table 6 ). Of these SNPs, only two encoding p.Arg90His (rs201802880) and p.Ser99Gly (rs17295741) in exon 4 and two intronic SNPs (intronic-1 (without a dbSNP ID) and intronic-2 (rs199789198)) of NCF1 showed r 2 > 0.1 with rs117026326. We hypothesized that at least one of these four NCF1 SNPs was the causal variant tagged by rs117026326 in Chinese and shared by European and African Americans and that the causal SNP(s) might be located by leveraging the different LD patterns in these ancestry groups. Figure 3: Determination of the ΔGT/GTGT ratio. ( a ) ΔGT/GTGT ratios. ( b ) Distribution of the ΔGT/GTGT ratio in all studied subjects. ( c ) Plot of the ΔGT/GTGT ratio in the different ancestry groups. Full size image LD analysis in African Americans suggested that either p.Arg90His or p.Ser99Gly might be the causal variant because intronic-1 and intronic-2 were in complete LD with non-polymorphic rs117026326 ( Fig. 1c ). To confirm this, we assessed these four variants, by performing nested PCR and TaqMan assays and measuring the ΔGT/GTGT ratio, for association with SLE in African Americans. Of the variants, only p.Arg90His was associated with SLE (15.7% versus 8.3%, P = 2.9 × 10 −5 , OR = 2.02; Table 1 and Supplementary Table 5 ). As expected, intronic-1 and intronic-2 were nearly non-polymorphic and were not associated with SLE in African Americans, suggesting that they might not be the causal variant shared by different ancestry groups, and they were excluded in subsequent analyses. Table 1 Association of SNPs in the GTF2IRD1 – GTF2I – NCF1 region with autoimmune diseases in different ancestry groups Full size table Next, we assessed p.Arg90His and p.Ser99Gly for association with SLE in Asians and European Americans. In two Chinese data sets and one Korean data set, p.Arg90His exhibited stronger association with SLE (Chinese-1: 38.1% versus 15.6%, P = 2.6 × 10 −23 , OR = 3.35; Chinese-2: 41.6% versus 16.7%, P = 1.5 × 10 −41 , OR = 3.27; Korean: 46.6% versus 18.1%, P = 2.6 × 10 −43 , OR = 3.82; P meta Asian = 3.1 × 10 −104 , OR = 3.47; Table 1 and Supplementary Table 3 ) than p.Ser99Gly, and conditioning on p.Arg90His eliminated association signals detected at p.Ser99Gly and the GTF2IRD1 – GTF2I region. In two European-American data sets, p.Arg90His consistently showed stronger association with SLE (EurAm-1: 5.5% versus 2.1%, P = 9.5 × 10 −5 , OR = 2.83; EurAm-2: 6.0% versus 2.4%, P = 5.0 × 10 −4 , OR = 2.42; P meta EurAm = 1.9 × 10 −7 , OR = 2.61; Table 1 and Supplementary Table 4 ) than p.Ser99Gly and explained association signals at p.Ser99Gly and the GTF2IRD1 – GTF2I region in a conditional test. Furthermore, the allele encoding p.Arg90His was dose dependently associated with early age of onset in Korean and European-American patients with SLE ( Supplementary Fig. 3 ). These data support p.Arg90His as a likely causal variant for SLE susceptibility shared across Asian, European-American and African-American populations within the GTF2IRD1 – GTF2I – NCF1 region. In addition to SLE, p.Arg90His was associated with other autoimmune diseases, including primary Sjögren's syndrome (MIM 270150 ) in Chinese (37.8% versus 18.3%, P = 7.2 × 10 −17 , OR = 2.45) and European Americans (4.8% versus 2.2%, P = 9.7 × 10 −4 , OR = 2.35; Table 1 and Supplementary Table 7 ), which explains the reported association of rs117026326 with Sjögren's syndrome in Chinese 13 , and seropositive rheumatoid arthritis (MIM 180300 ) in Koreans (26.6% versus 18.1%, P = 2.5 × 10 −8 , OR = 1.65; Table 1 and Supplementary Table 8 ), but with a modest effect size. Arg90 of p47 phox , located in a phosphoinositide-binding pocket of the PX domain, has a crucial role in the membrane translocation of cytosolic p47 phox and resultant activation of NOX2 for ROS production 14 . The substitution of evolutionarily conserved Arg90 with a histidine residue encoded by the SLE risk allele was predicted to be deleterious ( Supplementary Fig. 4 ), a prediction that is supported by in vitro mutation studies indicating that changing Arg90 to histidine 2 , or to lysine, leucine or alanine 14 , 15 , 16 , reduces ROS production. These data suggest that p.Arg90His might confer risk for SLE by reducing NOX2-derived ROS levels. Consistent with this idea, the SLE risk allele encoding p.His389Gln in p67 phox causes reduced ROS production in transfection assays 8 . However, p.Arg90His was not associated with intracellular ROS levels in neutrophils from controls ( Supplementary Fig. 5 ), probably because of the impact of mitochondrial ROS 17 . The GTGT to ΔGT mutation in NCF1 (rs273585651) leads to a frameshift and a premature stop codon at residue 51 ( Fig. 2b ) (ref. 11 ). Because of an absence of functional p47 phox and failure of ROS production, homozygous carriers of ΔGT in NCF1 develop a rare disease, chronic granulomatous disease (CGD), and have increased risk of developing SLE ( Fig. 3a ) (ref. 7 ). NCF1B and NCF1C are transcribed but do not produce functional protein because they contain ΔGT 10 . However, GTGT-containing NCF1B and NCF1C are believed to produce intact p47 phox similar to functional NCF1 , although this has not yet been experimentally validated 12 . Analyzing the ΔGT/GTGT ratio allowed us to assess copy number variation (CNV) in NCF1 . One copy of NCF1 (5:1 ratio) was associated with increased risk of SLE in Koreans ( P = 0.032), Chinese ( P = 0.011) and European Americans ( P = 5.9 × 10 −4 , OR = 3.91; Supplementary Table 9 ). In contrast, having ≥3 copies of NCF1 (3:3 and 2:4 ratios) was protective against developing SLE in Koreans ( P = 3.7 × 10 −5 ), Chinese ( P = 2.8 × 10 −3 , OR = 0.28), European Americans ( P = 0.038, OR = 0.85) and African Americans ( P = 0.018, OR = 0.73). These data support the notion that reduced ROS production is a risk factor for SLE. Association between CNV of NCF1 and Sjögren's syndrome was not detected in European Americans, probably owing to the limited sample size. ROS can be a double-edged sword in autoimmunity. High levels of ROS, predominantly produced by NOX2 in phagocytes for host defense, may lead to inflammatory tissue damage, but ROS are also signaling molecules regulating T cell differentiation, B cell proliferation and antigen processing in dendritic cells 18 . Our findings suggest that reduced NOX2-derived ROS production increases the risk of developing autoimmune diseases, consistent with previous reports 2 , 6 , 7 , 8 , but the underlying mechanism remains elusive. Of interest, NOX2-derived ROS are required for LC3-associated phagocytosis 19 and nonfunctional NOX2 causes defective clearance of dying cells and lupus-like phenotypes 20 , providing a possible explanation for the pathogenic role of reduced ROS levels in SLE. Of note, NCF1 variants have not been correctly called in studies using short sequence reads, such as the 1000 Genomes Project ( Supplementary Fig. 6 ) and the Exome Aggregation Consortium, owing to the presence of NCF1B and NCF1C . Consequently, we would like to emphasize that NCF1 variants need to be assessed with great caution to exclude the impact of NCF1B and NCF1C . In this study, we identified p.Arg90His by NCF1 -specific PCR and Sanger sequencing. However, because of the complexity of this region and the difficulty of long-range PCR, we only resequenced NCF1 in Chinese subjects, and potential causal variants within the NCF1 -neighboring region or specific for European and African Americans might not have been discovered. Given that most lupus-associated variants show OR <2 in GWAS 21 , the chance of finding a causal variant within the noncoding region of NCF1 that can explain the association of p.Arg90His with SLE (OR > 3 in Asians and OR > 2 in European and African Americans) is small. Although we cannot exclude the possibility of additional causal NCF1 variants, their association with SLE should be independent from that of p.Arg90His. In summary, we identify a p.Arg90His substitution encoded in NCF1 as a novel risk variant for SLE, Sjögren's syndrome and rheumatoid arthritis, and we show that decreased and increased copy numbers of NCF1 predispose to and protect against SLE, respectively. Our data highlight the pathogenic role of reduced ROS production in autoimmune diseases and indicate the presence of missing heritability within complex genomic regions. Methods Subjects. Discovery stage. Patients with SLE and healthy controls in the discovery stage were recruited from the University of California Los Angeles (UCLA), the Oklahoma Medical Research Foundation (OMRF) and the Medical University of South Carolina (MUSC). All patients with SLE met at least 4 of the 11 American College of Rheumatology (ACR) criteria for the classification of SLE 22 . The final data set after quality control comprised subjects from three different ancestry groups, including Chinese (1,010 cases and 848 controls recruited from UCLA), African Americans (532 cases and 367 controls recruited from OMRF and MUSC) and European Americans (930 cases and 1,107 controls recruited from UCLA and OMRF). All these samples were genotyped using the Immunochip, and we estimated that there was power of >90% to identify an SLE-associated variant with MAF >10% and OR >2.0 at the GWAS significance level of P < 5 × 10 −8 in Chinese and European Americans. Of the subjects, all African Americans and a subset of the Chinese (441 cases and 589 controls; data set Chinese-1) and European Americans (716 cases and 578 controls; data set EurAm-1) were genotyped using TaqMan assays for SNPs not included on the Immunochip. Replication stage. To replicate the result in data set Chinese-1, an independent cohort comprising 746 Chinese SLE cases and 1,034 healthy Chinese controls (data set Chinese-2) was recruited from Shanghai Renji Hospital. In addition, another Asian replication cohort comprising 614 Korean SLE cases and 692 healthy Korean controls was recruited from the Hanyang University Hospital for Rheumatic Diseases (HUHRD). To replicate the result in data set EurAm-1, an independent European-American cohort comprising 875 SLE cases and 540 healthy controls (data set EurAm-2) was recruited from UCLA, MUSC and OMRF. To assess p.Arg90His and p.Ser99Gly for association with other autoimmune diseases, 863 Korean patients with rheumatoid arthritis were recruited from HUHRD. All patients with rheumatoid arthritis were positive for antibodies to citrullinated peptide and fulfilled the ACR 1987 revised criteria for the classification of rheumatoid arthritis 23 . In addition, 382 European-American patients with primary Sjögren's syndrome, described in a previous GWAS 24 , were recruited from OMRF and 449 Chinese patients with Sjögren's syndrome and 469 healthy Chinese controls were recruited from Peking University People's Hospital. All patients with Sjögren's syndrome fulfilled the American-European Consensus Group (AECG) criteria for primary Sjögren's syndrome 25 . To measure the ΔGT/GTGT ratio, additional Chinese (198 patients with SLE and 471 controls recruited from the First Affiliated Hospital of Nanjing Medical University), African Americans (184 patients with SLE and 39 controls) and Koreans (215 patients with SLE and 90 controls) were used, but these subjects were not genotyped for SNPs. Each participating institution had institutional review board (IRB) approval to recruit subjects. All subjects provided written informed consent. Immunochip genotyping and quality control. Samples in the discovery stage were genotyped using the Immunochip according to Illumina's protocols at the University of Texas Southwestern Medical Center, HudsonAlpha or OMRF, and all samples were reclustered for genotype calling as a single project at OMRF. We excluded SNPs with a call rate <95% in cases or controls and removed samples with a SNP call rate <90%. SNPs were also excluded if they showed deviation from Hardy–Weinberg equilibrium ( P HWE < 0.001 in controls, P HWE < 0.00001 in cases) or they had significantly different call rates in cases and controls (call rate < 98% and P < 0.05). On the basis of the remaining SNPs, we identified related samples (shared identity by descent (PI_HAT) > 0.25; estimated using PLINK v1.07) and samples showing mismatch between the reported and estimated sex and excluded them from subsequent analyses. To identify ancestry outliers, the remaining samples were assessed by principal-component analysis (implemented in EIGENSOFT 4.2) based on 7,500 randomly selected autosomal SNPs with MAF >1%, low LD ( r 2 <0.1 with each other) and no evidence of association with SLE ( P > 0.01), and 1000 Genomes Project samples were used as reference populations (including 286 Asians (97 CHB, 100 CHS, 89 JPT), 379 Europeans (85 CEU, 98 TSI, 89 GBR, 93 FIN, 14 IBS) and 246 Africans (88 YRI, 61 ASW, 97 LWK)) ( Supplementary Fig. 7 ). Outliers of each ancestry (>6 s.d. from the mean principal component) were excluded. Principal components showing significant differences between cases and controls were included as covariates in the association test. The Korean samples were analyzed in an independent Immunochip study and a GWAS 5 , 26 . Principal components for the Korean samples were obtained from these two studies. NCF1 -specific PCR for DNA sequencing. To identify NCF1 variant(s) tagged by rs117026326, a total of 45 Chinese subjects, including 23 homozygous for the risk allele and 22 homozygous for the non-risk allele at rs117026326, were selected for sequencing. Among these subjects, there was probability of 90% and 64% of discovering NCF1 variants with MAF of 5% and 1%, respectively. To distinguish NCF1 from NCF1B and NCF1C , we confirmed that all 45 subjects showed a 4:2 ratio of ΔGT/GTGT and amplified NCF1 -specific DNA sequence by three PCR reactions ( Supplementary Fig. 2 ) using the LongRange PCR kit (206402, Qiagen). The PCR reactions contained 200 nM of each primer, 600 μM dNTP, 2.75 mM Mg 2+ , 1× buffer and 1 U Taq polymerase. PCR reactions were run on the SimpliAmp thermal cycler (Thermo Fisher Scientific) with the following conditions: 3 min at 93 °C followed by 50 cycles of 15 s at 93 °C, 1 min at 62 °C and 7 min at 68 °C. PCR products were sequenced on the 3730xl DNA Analyzer (Thermo Fisher Scientific). Primer sequences are shown in Supplementary Table 10 . NCF1 -specific PCR and TaqMan genotyping. rs117026326, rs73366469 and rs12667901 were directly genotyped using TaqMan assays (Thermo Fisher Scientific). Because of the presence of NCF1B and NCF1C , genotypes for p.Arg90His, p.Ser99Gly, intronic-1 and intronic-2 in NCF1 were obtained by nested PCR and TaqMan assay. We first PCR amplified an NCF1 -specific fragment by targeting the GTGT sequence in exon 2 of NCF1 (P2 or P2*, as shown in Supplementary Fig. 2 ), and each PCR fragment was subjected to agarose gel electrophoresis to assess the quality of specific amplification. Next, 1 μl of the PCR fragment diluted 1:10,000 was used as the template in a TaqMan assay for SNP genotyping. Samples that failed in the first PCR reaction were removed from TaqMan analysis. TaqMan assays were run on either ABI 7900HT or the QuantStudio 6 Flex RT-PCR System (Thermo Fisher Scientific). DNA sequences for TaqMan assays are shown in Supplementary Table 11 . Raw TaqMan data for p.Arg90His are shown as Supplementary Data . Determination of the ΔGT/GTGT ratio. To exclude subjects carrying GTGT-encoding NCF1B and NCF1C or ΔGT-encoding NCF1 and determine the CNV of NCF1 , NCF1B and NCF1C , we measured the ΔGT/GTGT ratio using TaqMan CNV assays described in previous studies 2 , 12 ( Fig. 3 and Supplementary Table 11 ). In a duplex real-time PCR reaction, the TaqMan assay targeting either ΔGT or GTGT was run simultaneously with a copy number reference assay (RNase P) targeting the ribonuclease P RNA component H1 gene ( RPPH1 ) known to exist in two copies in the genome (4403326, Thermo Fisher Scientific). Each reaction was run in quadruplicate on the QuantStudio 6 Flex RT-PCR System (Thermo Fisher Scientific). The copy numbers of ΔGT and GTGT were compared with that of RNase P and calculated by the comparative C t method using CopyCaller Software v2.0 (Thermo Fisher Scientific). We combined all subjects ( n = 6,914) for multiplate analysis and selected the “without calibrator sample” option in CopyCaller by assuming four copies of ΔGT and two copies of GTGT as the most frequent copy numbers. Ratios between the copy numbers of ΔGT and GTGT were normally distributed around 5, 2, 1 and 0.5 ( Fig. 3b ). According to the distribution, subjects showing a ΔGT/GTGT ratio of >4, 1.3–2.6, 0.7–1.2 and <0.7 were assigned a theoretical ratio of 5:1, 4:2, 3:3 and 2:4, respectively ( Fig. 3b,c ). Association analyses. All African-American subjects ( Supplementary Table 5 ), all European-American subjects ( Supplementary Tables 4 and 7 ), all Korean patients with SLE and controls ( Supplementary Table 3 ) and a subset of subjects in the Chinese-1 data set ( Supplementary Table 3 ) were assessed for the ΔGT/GTGT ratio. Association analyses were carried out either in subjects with the normal ΔGT/GTGT ratio of 4:2 (carrying 2 copies each of NCF1 , NCF1B and NCF1C ) or in all subjects. Given that the 5:1, 3:3 and 2:4 ratios were found in less than 2% of Asians ( Fig. 3c and Supplementary Table 9 ) and that their impact on association analyses of p.Arg90His in the Asian data sets was negligible, Korean patients with rheumatoid arthritis ( Supplementary Table 8 ) and some of the Chinese subjects ( Supplementary Tables 3 and 7 ) were not assessed for the ΔGT/GTGT ratio. In each ancestry group, SNPs were assessed for association with disease using an additive model in logistic regression in which principal components showing significant differences between cases and controls (available for Chinese, European Americans and African Americans in the discovery stage and Koreans in the replication stage) and sex were included as covariates. Haplotype-based conditional association tests were performed to detect independent association signals, and meta-analysis was conducted to combine multiple data sets. All analyses described above were performed using PLINK v1.07. In addition, we calculated the Bayes factor and posterior probability for each SNP using SNPTEST v2.5.2. Pairwise LD values were calculated using Haploview 4.2. CNV of NCF1 was assessed for association with SLE and Sjögren's syndrome using Fisher's exact test, and the 4:2 ratio of ΔGT/GTGT was used as the reference genotype (OR = 1). Quantitative real-time PCR. Total mRNA was extracted from the PBMCs of patients with SLE and controls using the All-Prep DNA/RNA mini kit (Qiagen) and then reverse transcribed into cDNA (Thermo Fisher Scientific). Transcript levels of NCF1 , GTF2I , GTF2IRD1 and GAPDH were measured by quantitative real-time PCR using TaqMan assays (Hs00165362_m1, Hs01073660_m1, Hs00249456_m1 and Hs03929097_g1, Thermo Fisher Scientific, respectively). The relative expression levels of NCF1 , GTF2I and GTF2IRD1 , normalized to those of the housekeeping gene GAPDH , were calculated by the comparative C t method. Measurement of ROS levels in neutrophils. We measured ex vivo ROS levels in neutrophils for association with p.Arg90His genotypes. Fresh blood samples collected in vacutainer tubes containing EDTA were obtained from healthy Chinese subjects ( n = 101) recruited at Shanghai Renji Hospital. Neutrophils were isolated from blood by density gradient centrifugation using PolymorphPrep (Axis-Shield) and cultured at 37 °C in RPMI-1640 medium (11875-093, Gibco) supplemented with 10% FBS. ROS levels were determined using DCFH-DA dye (S0033, Beyotime), which can be oxidized to fluorescent DCF by intracellular peroxides. Two million neutrophils were incubated with 5 μM DCFH-DA for 20 min and then stimulated with 30 ng/ml phorbol myristate acetate (PMA) (P1585, Sigma) for 1 h. Cells were washed twice with PBS and assayed for mean fluorescence intensity using flow cytometry (BD Biosciences). Data were processed using FlowJo software. DNA samples were extracted using the TIANamp Blood DNA kit (DP348-03, TIANGEN) and assessed for p.Arg90His genotype. Data availability. Sequencing data and summary-level association data are shown in Supplementary Tables 1 and 3–9 . Raw TaqMan data for p.Arg90His are shown as Supplementary Data . URLs. 1000 Genomes Project, ; RegulomeDB, ; PLINK, ; SNPTEST, ; Haploview, ; EIGENSOFT, ; UCSC Genome Browser, .

Investigators at the Medical University of South Carolina (MUSC) report pre-clinical research showing that a genetic variant encoded in neutrophil cystolic factor 1 (NCF1) is associated with increased risk for autoimmune diseases, including systemic lupus erythematosus (SLE), rheumatoid arthritis, and Sjögren's syndrome, in the January 2017 issue of Nature Genetics. Data indicate that increased NCF1 protects against SLE while decreased NCF1 raises SLE risk and highlight the pathogenic role of reduced reactive oxygen species in autoimmune disease development. Single-nucleotide polymorphisms (SNPs - pronounced 'snips') are the most common type of human genetic variation; each one represents a small difference in a nucleotide - the building blocks of our DNA. The Immunochip for fine-mapping is an important tool for conducting genome-wide association studies of the genetic components of disease. Researchers use the Immunochip to investigate DNA samples from people with a particular disease for linkage disequilibrium (LD) signals that illuminate associations between specific SNPs and the disease. Autoimmune diseases such as SLE are known to have a strong genetic component and, to date, dozens of SNPs associated with SLE have been identified and included on the Immunochip. The Achilles heel is, of course, that the Immunochip cannot identify associations with SNPs that it does not include. When MUSC researchers genotyped DNA samples from Chinese, European-American, and African-American SLE patients, they found a strong signal in the Chinese sample at the rs73366469 locus in the GTF2IRD1-GTF2I intergenic region at 7q11.23. This was puzzling because that locus was not consistent with SLE loci identified by other genome-wide association studies. Furthermore, the very strong signal in the Chinese sample appeared as a modest signal in the European-American sample and did not appear at all in the African-American sample. "A true risk gene should be the same in all populations," explained Betty Pei-tie Tsao, Ph.D., Richard M. Silver Endowed Chair for Inflammation Research at MUSC and senior author on the article. "And for such a strong signal, we wondered, 'why hasn't anyone else seen it?' We wanted to find out if what we were seeing was true and explain it." The team confirmed their finding using a different genotyping platform in an independent Asian sample provided by Nan Shen, M.D., Ph.D., professor of medicine and director of the Shanghai Institute of Rheumatology at Shanghai Jiao Tong University's School of Medicine. But, because rs73366469 did not show LD with any SNPs in the Immunochip, the researchers hypothesized that the SNP containing the true underlying risk factor was not included in it. "We came into the study from our Asian samples and then started looking for this signal in other populations," said Tsao. "Every ethnic group has a different ancestral background and different LD patterns. We used the LD signal strength as a guide to find our way to the true risk gene - the particular variant that actually caused the increased risk for lupus." Because the SNP they were looking for was most likely not included in the Immunochip, the team turned to the 1000 Genomes Project dataset, where they found two SNPs that were not only not on the Immunochip, but also produced stronger LD signals with rs73366469 in Asian patients than European or African patients. One of these two, rs117026326 located on intron 9 of GTF2I, showed a stronger association with SLE than either the original or the other locus from the 1000 Genomes Project. As the researchers focused in on rs117026326, they saw that the NCF1 gene was nearby. This was important because NCF1, which encodes a subunit of NOX2, is thought to be related to SLE due to its role in activating the phagocytic complex NOX2. Preclinical studies have shown that non-functional NOX2 exacerbates lupus in mice. Furthermore, NCF2, which encodes another subunit of NOX2, is associated with SLE risk in European Americans. Betty Pei-tie Tsao, Ph.D. (front, center), Richard M. Silver Endowed Chair for Inflammation Research at MUSC and senior author on the Nature Genetics article, with first author Jian Zhao, Ph.D. (to Dr. Tsao's left) and second author Yun Deng, M.D. (to Dr. Tsao's right). Credit: Medical University of South Carolina The strong association of rs117026326 with SLE and the functional implications of nearby NCF1 took the team to their next hypothesis: that the rs117026326 SNP might tag causal variants of NCF1 that were not present in the 1000 Genomes Project database. But unraveling this mystery was not going to be easy. "This is a very complex genomic region," explained Tsao. "The NCF1 gene has two nearly identical twins - NCF1B and NCF1C - that are 98% the same. But they are non-functional pseudo-genes. This makes working in this region of the human genome very difficult. That's why the next-generation sequencing method that the 1000 Genomes Project has been doing doesn't pertain to this region." The researchers believed that mapping techniques commonly used by the larger projects, while efficient, limited their ability to find unique sequences among all the copies and duplications in this region. So, they decided to set up their own, novel PCR assay. "You can't easily sequence this region using the next-generation techniques," said Tsao. "So, we had to do it the old-fashioned way, which was very time consuming and labor intensive. To genotype the region correctly, we used PCR to selectively amplify the NCF1 copies and conduct copy number variation tests. Then we only used samples with no copy number variation to examine the NCF1 variant. This method ensured that what we identified as an NCF1 variant was truly a variant." Using this strategy, the team identified 67 SNPs, four of which had a strong association with rs117026326. After conducting a long series of multiple tests in samples from various ethnic populations, they gradually eliminated three of the four SNPs and determined that the one called p.Arg90His was the likely genetic variant causing SLE susceptibility across all populations. In addition, p.Arg90His was associated with increased risk for other autoimmune diseases, including rheumatoid arthritis and Sjögren's syndrome. The team also found that having only one copy of NCF1 was associated with a higher SLE risk, but having three or more NCF1 copies was associated with reduced SLE risk. Finally, while the underlying mechanism is unclear, the team found that having reduced NOX2-derived reactive oxygen species also raised the risk for these autoimmune diseases. Tsao notes that perseverance was a critical component of this work. This work was started years ago when the team was at the University of California Los Angeles and was completed after moving to MUSC. "We just stuck with it as a labor of love. Our lead author, Jian Zhao, devoted several years of his life to this project," explained Tsao." At the time we started, we didn't know it was going to be so complex. We just wanted to explain what we were seeing. It turned out to be quite a chase and very interesting and rewarding to finally bring this project to this point." This work also points out an important unmet need in the field of genetic mapping. "We need a more efficient platform to screen complex genome regions for variants. For a lot of diseases we've identified some, but not all, of the variants. There may be more variants hiding in these complex regions," said Tsao. "You have to sort it out like a puzzle. Autoimmune diseases share certain risk factors but also have unique genetic variants that drive the molecular pathogenesis of the disease. Each time you find a variant, you get more puzzle pieces and you can start to understand more about that disease and other autoimmune diseases as well."

10.1038/ng.3782

Medicine

Expectations and dopamine can affect outcome of SSRI treatment

Olof R. Hjorth et al, Expectancy effects on serotonin and dopamine transporters during SSRI treatment of social anxiety disorder: a randomized clinical trial, Translational Psychiatry (2021). DOI: 10.1038/s41398-021-01682-3 Journal information: Translational Psychiatry

http://dx.doi.org/10.1038/s41398-021-01682-3

https://medicalxpress.com/news/2021-11-dopamine-affect-outcome-ssri-treatment.html

Abstract It has been extensively debated whether selective serotonin reuptake inhibitors (SSRIs) are more efficacious than placebo in affective disorders, and it is not fully understood how SSRIs exert their beneficial effects. Along with serotonin transporter blockade, altered dopamine signaling and psychological factors may contribute. In this randomized clinical trial of participants with social anxiety disorder (SAD) we investigated how manipulation of verbally-induced expectancies, vital for placebo response, affect brain monoamine transporters and symptom improvement during SSRI treatment. Twenty-seven participants with SAD (17 men, 10 women), were randomized, to 9 weeks of overt or covert treatment with escitalopram 20 mg. The overt group received correct treatment information whereas the covert group was treated deceptively with escitalopram, described as an active placebo in a cover story. Before and after treatment, patients underwent positron emission tomography (PET) assessments with the [ 11 C]DASB and [ 11 C]PE2I radiotracers, probing brain serotonin (SERT) and dopamine (DAT) transporters. SAD symptoms were measured by the Liebowitz Social Anxiety Scale. Overt was superior to covert SSRI treatment, resulting in almost a fourfold higher rate of responders. PET results showed that SERT occupancy after treatment was unrelated to anxiety reduction and equally high in both groups. In contrast, DAT binding decreased in the right putamen, pallidum, and the left thalamus with overt SSRI treatment, and increased with covert treatment, resulting in significant group differences. DAT binding potential changes in these regions correlated negatively with symptom improvement. Findings support that the anxiolytic effects of SSRIs involve psychological factors contingent on dopaminergic neurotransmission while serotonin transporter blockade alone is insufficient for clinical response. Introduction Selective serotonin reuptake inhibitors (SSRIs) are commonly prescribed for depression and anxiety but it has been widely debated to what extent SSRI efficacy can be attributed to expectancies of improvement—a key mechanism of placebo effects [ 1 , 2 , 3 , 4 , 5 , 6 ]. This question has been discussed extensively in the field of depression, but it is relevant also for anxiety conditions [ 7 , 8 ] including social anxiety disorder (SAD) [ 9 ]. Meta-analyses support that SSRIs are efficacious for these disorders [ 10 , 11 ] but the clinical effect of SSRIs in double-blind RCTs may, at least partly, reflect an enhanced placebo response because of perceived side effects by participants in the active drug arm, compromising the integrity of the blind and increasing response expectancies [ 12 ]. While this notion has been questioned [ 6 , 13 ], it is supported by trials using active placebo, mimicking the side effects of the active substance, and by experimental research demonstrating that expectancies affect therapeutic outcomes [ 14 , 15 , 16 ]. Further research is needed to clarify the magnitude of the SSRI clinicial effect, to what extent it can be attributed to the drug itself, and the neural mechanisms underlying symptom remission with SSRIs. Research designs involving deception have been used to separate drug from expectancy effects in clinical as well as neuroimaging trials [ 9 , 15 , 16 , 17 , 18 ]. We previously demonstrated enhanced anti-anxiety effects of overt as compared to covert SSRI treatment with escitalopram in patients with SAD [ 9 ]. Patients were treated with equivalent clinical doses of escitalopram for 9 weeks, but only one group was correctly informed about the treatment received and its effectiveness. Using a credible cover story, the other group was led to believe that they were treated with an “active placebo” (a neurokinin-1 receptor antagonist) expected to induce similar side effects as the SSRI while lacking anxiolytic properties. In the present study, we further investigated the therapeutic mechanisms underlying SSRI efficacy by analyzing how serotonin and dopamine transporters (DATs) are affected by response expectancies. The SSRIs are held to exert their therapeutic effects by blocking serotonin uptake via the serotonin transporter (SERT) [ 19 ] and clinical doses typically result in 76–85% SERT occupancy in the striatum [ 20 ]. However, the downstream therapeutic mechanisms of action are still not elucidated, and in this respect dopaminergic mechanisms may also be important as the serotonin and dopamine systems have reciprocal functional influences on each other [ 21 , 22 ]. SAD patients show increased expression and co-expression of SERTs and DATs in comparison to healthy controls [ 23 ]. Molecular neuroimaging studies suggest that SSRIs exert effects also on the DAT [ 24 , 25 , 26 , 27 , 28 ]. It is, however, unclear to what extent SERTs and DATs are affected by the SSRI itself or by psychological processes like expectancies. Here, in a subsample of our previous SSRI deception study of SAD [ 9 ], we examined if giving correct or incorrect information about the drug affects serotonergic and dopaminergic neurotransmission as assessed with positron emission tomography (PET) and the highly selective radioligands [ 11 C]DASB and [ 11 C]PE2I, probing SERTs and DATs respectively. Effects on monoamine transporter proteins and clinical responsiveness were evaluated when escitalopram was administered with and without clear expectations of improvement, i.e., overt vs. covert SSRI treatment. Methods Participants We studied a PET subsample of a previously reported SAD treatment cohort, and for methodological details, we refer to that paper [ 9 ]. Also, PET baseline comparisons of SAD patients vs. healthy controls have been reported elsewhere [ 23 ]. Here, 27 right-handed patients with SAD (17 men, 10 women; mean ± SD age, 31.1 ± 10.3 years) underwent [ 11 C]DASB and [ 11 C]PE2I PET imaging before and after 9 weeks of escitalopram treatment—see Fig. S1 and Table S1 in the Supplementary. Of these, one female participant could not be included in posttreatment [ 11 C]DASB analyses due to scanner failure. In addition to the included sample, two patients were assessed by PET at baseline but were excluded from analyses due to magnetic resonance imaging (MRI) contraindications, and withdrawal from the study before completed MRI, respectively. Between March 17th 2014 and May 22nd 2015, participants were recruited through advertisements in newspapers, public billboards and the internet. Exclusion criteria were age <18 or >65 years, earlier PET-scan, contraindications for MRI, pregnancy, menopause, substance abuse or dependency, any ongoing severe somatic disease or serious psychiatric disorder, and ongoing or recently terminated (<3 months) psychiatric treatment. Participants were screened using an extensive online form and those not meeting the initial exclusion criteria were administered an excerpt from the Structured Clinical Diagnostic Interview for the DSM-IV [ 29 ] and the full Mini-International Neuropsychiatric Interview [ 30 ] via telephone to verify a DSM-IV primary diagnosis of SAD. Social anxiety symptom severity was measured with the self-report version of the Liebowitz Social Anxiety Scale [ 31 ], LSAS-SR (pre-treatment mean ± SD: 84.96 ± 20.37). Treatment design The study was an investigator-initiated clinical trial with SAD patients, matched for age and sex, randomized to either overt ( n = 14) or covert ( n = 13) SSRI-treatment. The experimental manipulation was verbal instructions of whether the drug was expected to be effective or not. After baseline scans (Fig. 1A ), one group was instructed that they would receive escitalopram, demonstrated to be effective for SAD, and the other group that they would receive a non-effective neurokinin-1-receptor antagonist, in the cover story described as an active placebo with similar side effects as escitalopram but out of which no symptom-improvement could be expected (Fig. 1B ) [ 9 ]. However, both groups were treated with 20 mg escitalopram per day, starting with 10 mg the first week. All accepted their allocated group. All participants and observers were blinded to manipulation except the study clinician who supervised medication and debriefed participants when the cover story was revealed [ 9 ]. Fig. 1: Study design and main results. A Shows the whole-sample distribution of serotonin (SERT) and dopamine (DAT) transporters, expressed as non-displacable binding potentials (BP ND ) at the baseline PET assessment. B Illustrates the experimental manipulation; high or low response expectancies were induced by different verbal instructions. All patients were then treated under randomized conditions with escitalopram 20 mg for 9 weeks, correctly described as an effective SSRI for the overt group and incorrectly described as an active placebo in the covert group. C Shows the treatment effects on PET measures. Overt as compared to covert SSRI treatment resulted in lowered DAT availability, the significant cluster in the right putamen/pallidum is shown together with percent BP ND change from pre- to posttreatment. In contrast, the average escitalopram SERT occupancy levels were similar in both groups after treatment. D Shows the results of the clinical evaluation. Overt as compared to covert treatment resulted in a significantly higher percentage responders and lowered (pre-post) social anxiety as assessed with the Liebowitz Social Anxiety Scale, self-report (LSAS-SR) administered online. Error bars reflect 95% confidence intervals. Full size image Participants revisited the clinic after 1 week and were then handed their supply of the medication for the remainder of the study period. Blood serum analyses were performed to examine escitalopram and metabolite concentrations at posttreatment and compliance was further assessed by counting of remaining capsules at the posttreatment visit—see [ 14 ]. Treatment randomization and preparation of escitalopram was prepared by APL, Stockholm, Sweden. The study was approved by the Regional Ethical Review Board in Uppsala, the Radiation Safety Committee at Uppsala University Hospital and the Medical Products Agency in Sweden. All participants were informed both verbally and in writing regarding study objectives, comparing escitalopram and “active placebo”, as well as side-effects of drugs and risks of neuroimaging methods. The full written and verbal information is provided in the supplementary appendix to our previous paper [ 9 ]. All participants were offered additional treatment with internet-delivered cognitive-behavior therapy after the study period. Written consent was required for inclusion. Imaging procedure Positron emission tomography A Siemens ECAT EXACT HR + (Siemens/CTI) was used to acquire the PET images with 63 contiguous planes of data and slice thickness of 2.46 mm resulting in a total axial field of view of 155 mm. Participants fasted for at least 3 h and refrained from alcohol, nicotine and caffeine for at least 12 h before the scan. At posttreatment, participants were instructed to take the escitalopram dose 24 h before the PET scan. Participants were positioned supine in the scanner with their head gently fixated and a venous catheter for tracer injections was inserted. A 10 min transmission scan for attenuation correction was performed using three retractable germanium ( 68 Ge) rotating line sources. Participants were injected with on average 327 ± 27 MBq of [ 11 C]PE2I (N-(3-iodopro-2E-enyl)-2b-carbomethoxy-3b-(4-methyl-phenyl)nortropane) through an intravenous bolus and 22 frames of data were acquired over 80 min of data (4 × 60 s, 2 × 120 s, 4 × 180 s, 12 × 300 s). Following a 45–60 min waiting period to allow for sufficient decay of the radioactivity (i.e., >6 radioactive half-lifes), acquisition commenced for [ 11 C]DASB (3-amino-4-(2-dimethylaminomethylphenylsulfanyl)-benzonitrile), using an identical injection procedure and an average activity of 333 ± 20 MBq. In total, 22 frames of data were acquired over 60 min (1 × 60 s, 4 × 30 s, 3 × 60 s, 4 × 120 s, 2 × 180 s, 8 × 300 s). Magnetic resonance imaging Participants underwent an anatomical T1-weighted MR scan used for anatomical referencing of PET data (echo time (TE) = 50 ms; repetition time (TR) = 500 ms; Field of view = 240 × 240 mm 2 ; voxel size = 0.8 × 1.0 × 2.0 mm 3 ; 170 contiguous slices) on a Philips Achieva 3.0 T whole body MR-scanner (Philips Medical Systems, Best, The Netherlands) with an 8-channel head-coil. Five participants were scanned with a 32-channel head-coil due to a scanner upgrade. PET data preprocessing Ordered subset expectation maximization with six iterations and eight subsets and a 4 mm Hanning post-filter with appropriate corrections was used to reconstruct dynamic images. Voxel-wise parametric images of non-displaceable binding potentials (BP ND ) were calculated for both radioligands with the cerebellum as reference region using reference Logan [ 32 ] for [ 11 C]DASB (time interval 30–60 min) and receptor parametric mapping [ 33 ] for [ 11 C]PE2I. Cerebellar gray matter was selected as reference region for both radioligands because of the negligible levels of SERTs and DATs. It was automatically outlined on each participant’s anatomical T1-weighted image using the PVElab software [ 34 ]. The pre and post treatment [ 11 C]DASB BP ND and [ 11 C]PE2I BP images were co-registered to the anatomical T1-weighted MR image using Statistical Parametric Mapping 8 (SPM8; (Wellcome Department of Cognitive Neurology, University College London, ) implemented in Matlab (Mathworks Inc., Nantucket, MA, USA). The T1-image was then segmented and normalized to the Montreal Neurological Institute (MNI) standard space and the transformation parameters applied to the [ 11 C]DASB and [ 11 C]PE2I BP ND images, resulting in parametric images with 2 mm isotropic voxels. Images were then smoothed using a 12 mm Gaussian kernel. Statistical analysis Behavioral treatment outcome was assessed with mixed repeated measures ANOVA of LSAS-SR, and Fisher’s exact test of number of responders fulfilling the criteria for clinically significant improvement [ 35 ]. Participants were deemed to be responders if they were within two standard deviations of the normal population after treatment (LSAS score < 39), and exhibited a Reliable Change Index larger than 1.96 [ 35 ]. As in our recent PET-study [ 23 ], the a priori regions of interest (ROIs) for both radiotracers were the amygdala, hippocampus, caudate nucleus, putamen, nucleus accumbens (NAcc), pallidum and thalamus, and for [ 11 C]DASB also the anterior cingulate cortex (ACC), insula cortex and raphe nuclei. Anatomical regions were defined using the Automated Anatomical Labeling library from the Wake Forest University Pickatlas [ 36 ] except for the NAcc and raphe nuclei which were defined by the Hammersmith atlas [ 37 ] and PVElab software [ 34 ] respectively. For voxel-wise analyses, SERT occupancy [(pre-post)/pre] images and percentage change in DAT binding potential [(post-pre)/pre] images were calculated. To examine group differences before treatment and changes with treatment, two-sample t -tests were performed on BP ND data for both tracers separately in SPM8 with age and sex as covariates. Correlations between LSAS-SR and brain measures were performed using Pearson’s product-moment correlations for [ 11 C]DASB and [ 11 C]PE2I separately. The statistical threshold was set at P < 0.05 and analyses were corrected for familywise error (FWE) within the ROIs. We used Fisher-transformed partial Pearson’s product-moment correlations to examine voxel-wise relations between SERT occupancy and percent change in DAT BP ND with the statistical threshold set to P < 0.05 [ 23 ]. Analyses were performed in MatlabR2018a. Results Serotonin transporter binding Distribution of [ 11 C]DASB binding, probing SERT availability at baseline, is shown in Fig. 1A . Groups did not differ in initial SERT BP ND . After expectancy manipulation and 9 weeks of treatment (Fig. 1B ), no between-group (overt vs. covert) differences in SERT occupancy were detected and escitalopram SERT occupancy was significant in all evaluated ROIs with an average of 78% when accounting for total volume (Fig. 1C , Table S2 ). There were no correlations ( P > 0.10) between SERT occupancy and symptom improvement as assessed with LSAS-SR. Dopamine transporter binding Distribution of [ 11 C]PE2I binding, probing DAT availability at baseline, is shown in Fig. 1A . Groups did not differ in initial DAT BP ND in any region except for the right thalamus (MNI x,y,z: 4,−10,10, P FWE = 0.001, Z = 4.13, 1584 mm 3 ). Following expectancy manipulation and treatment (Fig. 1B ), between-group analyses showed a differential response with relative decreases in DAT BP ND in the overt group and increases in the covert group, in the right putamen, extending into pallidum, and also in the left thalamus (Fig. 1C , Table 1 ). Follow-up analysis showed that reduced DAT binding in these clusters correlated with social anxiety symptom improvement (Fig. 2 ). In addition, within-group analyses revealed significantly decreased (pre-post) DAT BP ND in the right amygdala in the overt SSRI group and increased DAT BP ND in the bilateral pallidum, left thalamus and left hippocampus in the covert group (Table 1 ). Between-group differences in the right amygdala (MNI x,y,z: 24,2,−12, P = 0.017, Z = 2.11, 280 mm 3 ), and left hippocampus (MNI x,y,z: 18,−30,−4, P = 0.009, Z = 2.34, 32 mm 3 ) were significant at an uncorrected statistical threshold, with relatively higher increases in binding in the covert group. Table 1 Brain regions showing differences in dopamine transporter binding potential change after overt and covert SSRI treatment. Full size table Fig. 2: Brain-behavior correlations. Significant correlations are shown between decreased dopamine transporter (DAT) availability, expressed as percent pre-post change in binding potential (BP), and symptom improvement expressed as higher scores on the Liebowitz Social Anxiety Scale self-report (LSAS-SR) before as compared to after treatment. Significant correlations were noted in the right putamen/pallidum cluster (left panel) and in the left thalamus (right panel). Full size image Concomitant changes in serotonin-dopamine transporter binding with treatment Correlations between SERT occupancy and percentage change in DAT BP ND within each treatment group, as well as significant group differences in these correlations, are listed in Table 2 . Significant group differences were noted in the bilateral pallidum, left putamen and right thalamus (Fig. 3 ). Level of SERT occupancy correlated with decreased DAT BP ND in the overt group and increased DAT BP ND in the covert SSRI group. To statistically evaluate involvement of SERT-DAT interactions in response expectancies, follow-up logistic regression analyses were conducted with transporter changes as independent variables and group (overt/covert) as dependent variable. These confirmed that inclusion of SERT×DAT interaction terms to the models with main effects of SERT and DAT, drastically increased the McFadden R 2 explained variance in all regions (interaction/main effects: putamen = 0.28/0.02; left pallidum = 0.19/0.03, right thalamus = 0.41/0.11), except the right pallidum (0.27/0.27). Table 2 Brain regions showing significant correlations between serotonin transporter occupancy and percent dopamine transporter binding potential change after overt and covert SSRI treatment. Full size table Fig. 3: Concomitant serotonin-dopamine transporter changes with overt and covert SSRI treatment. Significant group differences in correlations between percent sertonin transporter (SERT) occupancy and change in dopamine transporter binding (DAT BP) from pre-to post-treatment were noted in the left putamen, bilateral pallidum, and right thalamus (top panel). Scatterplots are shown in the lower panel. Full size image Blood serum analyses Groups did not differ significantly in blood serum concentrations of escitalopram ( t = −0.78, P = 0.44, 95% CI: −60.3–27.2) or S-desmethylcitalopram ( t = 0.55, P = 0.59, 95% CI: −10.1–17.5) at posttreatment—see Supplementary Appendix . Clinical evaluation After treatment, there were significantly more responders in the overt (8/14; 57%) than the covert (2/13; 15%) group (Fisher’s exact test: OR = 0.15, P = 0.046), according to conservative response criteria [ 35 ]—see Fig. 1D . On the main outcome measure (LSAS-SR), groups did not differ in pre-treatment scores ( t (24.71) = 0.44, P = 0.67, 95% CI: −12.96–19.94) and ANOVA revealed a significant Group × Time interaction ( F (1,25) = 13.20, 95% CI of group difference = 11.15–40.07, P = 0.001) with larger improvement in the overt (M diff ± SD = 47.07 ± 19.23, Cohen’s d = 2.33) as compared to the covert (M diff = 21.46 ± 17.25, Cohen’s d = 0.93) group over 9 weeks of treatment (Fig. 1D ). Thus, as in the larger cohort [ 9 ], superiority of overt (>covert) SSRI administration was noted. Discussion Verbally-induced response expectancies had a significant influence on SSRI-efficacy and dopamine, but not serotonin, transporter availability. Overt was clinically superior to covert SSRI-treatment, with almost a fourfold higher response rate, resulting in relatively lowered DAT binding in striatal and thalamic brain regions that correlated significantly with symptom improvement. In contrast, groups did not differ in levels of SERT occupancy after treatment, and escitalopram/S-desmethylcitalopram serum concentrations were also similar. The present findings support that dopamine neurotransmission is crucially involved in the therapeutic mechanisms of SSRIs and that the anxiolytic properties can be attributed largely to psychological factors. DAT binding in the putamen, pallidum, and thalamus increased with covert SSRI treatment while it decreased in the overt group, with reductions linearly coupled to symptom improvement, suggesting slower clearance and/or increased release of dopamine when expectancies are higher, resulting in better improvement. Previous SPECT studies, including a study of SAD, have generally noted increased striatal DAT binding after acute or stable SSRI treatment [ 24 , 25 , 26 , 27 , 28 ]. It should be noted, however, that the radioligands used in these SPECT studies are affected by SSRIs, and are not as specific and sensitive as the current [ 11 C]PE2I PET ligand [ 38 ]. Moreover, expectancies were not assessed in previous studies. Other lines of evidence also support that SSRIs have measurable dopaminergic effects, although the direction has varied. For example, in a study of dopaminergic challenges in SAD, an acute dose of pramipexole but not sulpiride, attenuated anxiety levels during a behavioral test in SSRI-treated patients, suggesting desensitization of dopamine D3 receptors [ 39 ]. Some side effects of SSRIs have previously been indentifed as dopamine-dependent [ 40 ]. Further, animal studies show that serotonin agonists and SSRIs increase extracellular dopamine levels in the striatum, hypothalamus and prefrontal cortex [ 41 , 42 ] and that SSRI antidepressant effects are abolished by dopamine depletion [ 43 ]. Previous research also indicates promiscuity between monoamine transporters [ 22 ] and that serotonin can be transported by DATs when SERTs are blocked by SSRIs [ 44 ]. This may be counterbalanced by decreased DAT availability when response expectancies are high, or reinforced when expectancies are low. Striatal regions are important for reward processing, receiving input from the thalamus while also relaying information to the thalamus through pallidum [ 45 ]. Higher expectancies with overt treatment may come with more optimistic cognitions, remoralization, enhanced approach motivation and willingness to engage in self-exposure, enhancing reinforcement learning and dopamine-dependent reward function. Indeed, reward expectancy and approach motivation activate the striatal dopamine system [ 46 , 47 ] as do placebo effects [ 47 ]. Conversely, animal studies report reduced striatal dopamine release during passive coping with stressful situations [ 48 ]. The overt group also exhibited significantly decreased DAT BP ND in the right amygdala, a central hub in threat processing. The association between decreased amygdala-striatal DAT availability and better anxiolytic effects is congruent with our recent PET study in which baseline DAT BP ND correlated positively with anxiety severity, indicative of dopamine hypoactivity in SAD [ 23 ]. Dopaminergic hypofunction has also been suggested to underlie at least some subgroups of treatment resistant depression for which dopamine agonists could be effective [ 49 ]. The present findings suggest that pharmacologic SERT-blockade is, by itself, not sufficient for adequate clinical improvement. Because the SSRIs are effective in SAD [ 10 ] and block the SERT in a dose-dependent manner [ 20 ], and because PET studies show increased SERT availability in SAD [ 23 , 50 ], it could be expected that the anti-anxiety effects of SSRIs are SERT-mediated. However, despite the large difference in clinical efficacy, SERT occupancy was equally high in the overt and covert SSRI groups in all evaluated brain regions, and did not correlate with reduced social anxiety. This was not explained by attrition or poor SSRI compliance as both groups had comparable and expected blood serum concentrations of escitalopram and S-desmethylcitalopram. Consistently, several molecular imaging studies have failed to demonstrate a relationship between SERT occupancy and clinical response to SSRIs [ 20 , 51 , 52 ]. Similary, pharmacologic SERT blockade occurs within hours after acute SSRI intake while the clinical response is delayed several weeks [ 53 ]. Nonetheless, as some improvement occurred also in the covert group [ 9 ], ample SERT occupancy could still be a prerequisite for SSRI-induced anxiety relief but other mechanisms are also likely to be involved. A previous study of SSRI-treated patients with SAD reported that lowering of serotonin by tryptophan depletion increased anxiety induced by an autobiographical script, but not by a stressful speaking task [ 54 ]. In contrast, PET data from our group suggested that serotonin synthesis was reduced and tied to symptom improvement following SSRI and other pharmacological treatments [ 55 ]. Here, we found that superior improvement with overt SSRI administration was tied to decreased DAT availability occurring in parallel to increased SERT occupancy e.g., in the striatum. This suggests that serotonin-DAT interactions are involved, not only in the pathogenesis of SAD [ 23 ], but also in response expectancies. The full clinical SSRI response may thus result from expectancy effects on dopamine and serotonin-dopamine interactions, in addition to pharmacological SERT blockade. The drug-expectancy relation could be additive or synergistic [ 52 ]. It should be noted that PET-data on transporter proteins are limited to brain regions with adequate tracer uptake and do not provide detailed information about neural signaling, also preventing conclusion about dynamics within and across specific serotonin and dopamine paths as well as tonic-phasic interplay. Further research on pre- and postsynaptic processes is needed to clarify how the monoamines contribute to anxiety and symptom improvement with treatment. The complexity of this issue calls for studies that use a variety of methodologies like multimodal neuroimaging, genetic approaches and pharmacological challenges. The sample size was relatively small in the present study, due to high costs involved with PET, thereby restricting statistical power. This could increase the risk for type 1 and 2 errors, i.e., either that the between-group null findings on SERT occupancy were false negatives or that significant DAT results were false positives. Since levels of SERT occupancy were highly similar, it is unlikely that significant overt-covert group differences would have emerged with an increased number of subjects. With regard to DAT changes, we observed significant between-group differences as well as significant correlations with symptom improvement at the behavioral level. Moreover, the treatment-related SERT-DAT correlations in striatal and thalamic regions were in opposite direction in the two groups. This coherent pattern of results supports that overt vs. covert SSRI-treatment had dissimilar effects on dopaminergic signaling, arguing against false positives although replication in a larger sample is warranted. Among the study limitations it should also be mentioned that we, for ethical and practical reasons, could only use two of the four arms in the balanced placebo design [ 16 ]. Thus the “told SSRI/given placebo”, and “told placebo/given placebo” conditions were lacking. Also, we did not measure expectancies during the course of treatment because we were wary that this would reveal the study design [ 9 ]. Thus, we could not evaluate the relationship between subjective expectancies and imaging or clinical outcomes. Assessment of clinical efficacy was based essentially on a subjective self-report measure (LSAS-SR) and additional objective measures, like cortisol levels or heart rate variability, could have been added. Finally, the generalizability of the present results to other disorders, pharmaceuticals, or treatment modalities is not known and we cannot determine if the SSRI has a long-term or normalizing effect on transporter densities. This would require additional measurements after drug discontinuation. In conclusion, the anti-anxiety properties of SSRIs appear to be largely dependent on expectancy effects on dopamine signaling while SERT blockade is not sufficient for symptom remission. This provides new insights on the key therapeutic mechanisms of SSRIs, incorporating psychological effects on dopamine neurotransmission.

Levels of dopamine and the placebo effect can determine whether patients with social anxieties improve when treated with SSRIs. A new study shows the effect was four times higher for patients with high expectations of the medication compared with patients with low expectations. This was true even though the groups received the same medical treatment. Although SSRIs influence levels of serotonin in the brain, the effects on dopamine had the greatest impact for improvement. Selective serotonin reuptake inhibitors (SSRIs) are an established and effective medication for treating depression and anxiety. The placebo effect, where the positive effects of a treatment can increase when a patient expects to be helped, is a well-known phenomenon. The effect can be significant, and it is unclear how much of the improvement results from expectations of the SSRI treatment. It is also unclear whether the expectations use the same mechanism in the brain as SSRI medications (the inhibiting of the transporter protein for serotonin) or whether other neurotransmitters are involved. The new study points to the transporter protein for dopamine being the key. Researchers at Uppsala University confirmed in a study on social anxiety published in Translational Psychiatry that the placebo effect had a major impact on the anti-anxiety effect of the SSRI drug escitalopram. The surprising finding in the study was that the improvement after SSRI treatment can largely be linked to effects on dopamine rather than to the serotonin transporters. In the study, all the participants were treated with the same clinical dose of escitalopram for nine weeks, but they had different expectations. Half received accurate information about the drug and its effectiveness, while a cover story was used for the other half. Participants in the second group were told that the drug was an 'active placebo' that causes similar side effects as SSRIs but was not expected to alleviate their social anxiety. "The results showed that almost four times as many patients responded to the treatment when correct information about the drug was given. This is consistent with previous research showing that expectations affect treatment outcome," says researcher Olof Hjorth. Positron emission tomography (PET) brain scanning showed that the SSRI drug had the same effect on serotonin and blocked about 80 per cent of serotonin transporters in both groups. This was true even for the group that had low expectations and did not improve. "This indicates that the pharmacological effect was identical in both groups and that this cannot explain why correct information gave better treatment results. So, inhibiting serotonin transporters is insufficient for achieving good clinical relief from social anxiety using SSRI drugs." When assessing the transporter protein for dopamine after treatment, however, a clear difference between the groups was observed. Participants who began the treatment knowing that it was an effective drug showed a reduced availability of dopamine transporters in the striatum, a part of the cerebrum, while the opposite was true in the group that was given the cover story. One explanation may be that expectations affected the release of dopamine in the brain's reward pathways. This may have led to differences in the two groups in the proportion of dopamine transporters available after treatment. "The results indicate that positive expectations arising in the relationship between doctor and patient affect dopamine and enhance the effect of SSRI treatment," says Professor Tomas Furmark, who led the study.

10.1038/s41398-021-01682-3

Medicine

Exploring how the immune system does battle in the intestines to keep bacteria in check

Daniel Sorobetea et al, Inflammatory monocytes promote granuloma control of Yersinia infection, Nature Microbiology (2023). DOI: 10.1038/s41564-023-01338-6 Journal information: Nature Microbiology

https://dx.doi.org/10.1038/s41564-023-01338-6

https://medicalxpress.com/news/2023-03-exploring-immune-intestines-bacteria.html

Abstract Granulomas are organized immune cell aggregates formed in response to chronic infection or antigen persistence. The bacterial pathogen Yersinia pseudotuberculosis ( Yp ) blocks innate inflammatory signalling and immune defence, inducing neutrophil-rich pyogranulomas (PGs) within lymphoid tissues. Here we uncover that Yp also triggers PG formation within the murine intestinal mucosa. Mice lacking circulating monocytes fail to form defined PGs, have defects in neutrophil activation and succumb to Yp infection. Yersinia lacking virulence factors that target actin polymerization to block phagocytosis and reactive oxygen burst do not induce PGs, indicating that intestinal PGs form in response to Yp disruption of cytoskeletal dynamics. Notably, mutation of the virulence factor YopH restores PG formation and control of Yp in mice lacking circulating monocytes, demonstrating that monocytes override YopH-dependent blockade of innate immune defence. This work reveals an unappreciated site of Yersinia intestinal invasion and defines host and pathogen drivers of intestinal granuloma formation. Main Microbial pathogens utilize diverse mechanisms to subvert host immunity to replicate and spread to new hosts. While acute infections can be cleared rapidly by the immune system, some pathogens evade immune defences to cause chronic disease. Chronic infections often result in formation of structures termed granulomas that limit pathogen dissemination and tissue damage 1 . Granulomas are characterized by the presence of activated phagocytes, notably monocytes and macrophages, and form in response to a wide variety of infections 2 . Monocytes are rapidly recruited to infected tissues, where they produce inflammatory cytokines and antimicrobial effector molecules, contributing to defence against multiple pathogens 3 , 4 , 5 , 6 . Some pathogens, however, exploit monocytes as a means of dissemination, including Salmonella enterica , Yersinia pestis and Mycobacterium species 7 , 8 , 9 . The pathogen-specific signals that induce granuloma formation remain poorly defined. Enteropathogenic Yersinia , including Y. pseudotuberculosis ( Yp ) and Y. enterocolitica ( Ye ), cause self-limiting gastroenteritis and mesenteric lymphadenopathy following enteric infection 10 , 11 . In immune-compromised patients, however, bacteria disseminate and cause a systemic plague-like disease, indicating that the intestinal immune system is critical for control of acute infection. Indeed, the intestine constitutes a bottleneck against Yersinia dissemination, as bacteria in systemic organs are thought to originate predominantly from the intestinal lumen rather than gut-associated lymphoid tissues 12 . A hallmark of Yersinia infections is the presence of chronic pyogranulomas (PGs) in lymphoid tissue, characterized by nodular infiltrates of activated monocytes and macrophages surrounding a core of activated neutrophils 13 . Notably, the contribution of monocytes to PG formation and their role in Yersinia restriction are unclear 14 , 15 . In this Article, we report that PGs form acutely in the murine intestinal mucosa during enteric Yersinia infection. PGs are enriched in neutrophils and inflammatory monocytes, and contain live bacteria at levels comparable to Peyer’s patches (PPs). Notably, CCR2-deficient mice, which lack circulating inflammatory monocytes 16 , 17 , form disorganized necrosuppurative lesions rather than defined PGs, are unable to contain bacteria within the lesions and succumb rapidly to infection. Moreover, mice lacking circulating monocytes exhibit reduced levels of interleukin (IL)-1 cytokines and surface expression of the neutrophil activation marker CD11b within intestinal PGs. Yp lacking either the virulence plasmid (pYV) encoding the type III secreted Yersinia Outer Proteins (Yops), or lacking Yops that block phagocytosis and the reactive oxygen burst, do not induce detectable PGs, indicating that PGs are induced in response to Yersinia blockade of innate immune defence. Notably, CCR2-deficient mice infected with bacteria lacking the virulence factor YopH, which blocks actin cytoskeleton dynamics, were able to restrict bacterial burdens and form defined granulomatous lesions, accompanied by restored neutrophil CD11b surface expression. Neutrophil depletion in CCR2-deficient animals abrogates control of YopH-mutant Yersinia , demonstrating that inflammatory monocytes overcome YopH-mediated disruption of neutrophil function. Altogether, our study identifies an unappreciated site of Yersinia colonization within the murine intestinal mucosa, and reveals an essential function for inflammatory monocytes in maintainance of PG architecture during Yp infection. Results Intestinal PGs form upon oral Yersinia infection Yp colonizes gut-associated lymphoid tissues, resulting in acute PG formation following oral infection 18 . Interactions between Yersinia and immune cells within systemic tissues have been extensively documented 19 , 20 , 21 . In our efforts to dissect intestinal immune responses to Yp , we observed numerous macroscopically visible nodular lesions in the gastrointestinal tract 5 days post infection, which appeared as punctate areas of increased opacity (Fig. 1a ). Lesions were most prevalent in the jejunum and ileum, ranged in number from two to over 40 in individual mice and also formed in response to Ye infection (Fig. 1b,c ). Histology of Yp- infected intestines revealed focal inflammation characterized by crypt hyperplasia, oedema and submucosal to transmural cellular infiltration, whereas non-lesional areas of infected intestines appeared largely unaffected (Fig. 1d ). The lesions contained infiltrates of macrophages and neutrophils surrounding colonies of coccobacilli, similar to structures that we and others observed in lymphoid tissues and have termed PGs (Fig. 1e ) (refs. 15 , 18 , 22 ). Consistently, flow cytometric analysis of intestinal punch biopsies containing PGs (PG+), adjacent non-granulomatous tissue (PG−), and uninfected control tissue revealed that neutrophils were the most enriched cell type within PG+ tissue, followed by macrophages and inflammatory monocytes (Fig. 1f,g ). We also observed increases in eosinophils, dendritic cells and CD4 + T cells in PG+ tissue, although the relative frequencies of these populations were decreased, due to even larger increases in neutrophils, monocytes and macrophages (Extended Data Fig. 1a,b ). Fig. 1: Intestinal PGs form upon oral Yersinia infection. a , Small intestinal segments from uninfected (uninf) and Yp -infected mice with arrows depicting lesions, and a magnified lesion with dotted circle depicting size of punch biopsies. Scale bars, 3 mm (left) and 0.5 mm (right). Representative of more than three independent experiments. b , Frequency distribution of lesions along the intestine, with graphical key of anatomical segments. Each coloured bar represents the mean frequency of lesions in a given segment ( n = 20 mice). Only mice with >9 total lesions (>80% of mice) were included. Pooled from three independent experiments. c , Quantification of total intestinal lesions at day 5 post infection with Yp or Ye . Each circle represents one mouse ( n = 16–17). Lines represent median. Pooled from three independent experiments. d , H&E-stained paraffin-embedded longitudinal small-intestinal sections from uninf and Yp -infected mice. Scale bars, 200 μm. a is lesion, b is crypt hyperplasia, c is submucosal inflammation and d is oedema. Representative of three independent experiments. e , H&E-stained paraffin-embedded small-intestinal PG depicting an encircled bacterial colony. Scale bar, 10 μm. Representative of three independent experiments. f , Flow cytometry plots identifying CD11b + Ly-6G + neutrophils, CD11b + CD64 + Ly-6C + monocytes and CD11b + CD64 + Ly-6C - MHC-II + macrophages in small intestinal tissue. Representative of four independent experiments. g , Frequency and total number of neutrophils, monocytes and macrophages in small intestinal tissue. Each circle represents one mouse ( n = 15–19). Lines represent median. Pooled from four independent experiments. h , Fluorescently labelled small intestinal PG from a Ccr2 gfp/+ mouse. White ( Yp- mCherry), magenta (Ly-6G-AF647) and green (CCR2-GFP). Scale bars, 100 μm. Representative of two independent experiments. i , Bacterial burdens in small intestinal tissue. Each circle represents one mouse ( n = 24–25). Lines represent geometric mean. Dotted line represents detection limit. Pooled from four independent experiments. j , Cumulative bacterial burdens in PG+ tissue and PP. Each circle represents one mouse ( n = 29–30). Lines represent geometric mean. Pooled from five independent experiments. Wilcoxon test (two-tailed) was performed for paired analyses (PG− versus PG+). Mann–Whitney U test (two-tailed) was performed for remaining statistical analyses. * P < 0.05, ** P < 0.01, **** P < 0.0001; NS, not significant; ND, not detected. Full size image In lymphoid tissues, Yp PGs consist of a central bacterial colony, surrounded by neutrophils, which are bordered in turn by monocytes and macrophages 15 , 22 . Confocal microscopy demonstrated that intestinal PG also contained a central Yp colony surrounded by a dense population of Ly-6G + neutrophils (Fig. 1h ). Interestingly, in contrast to lymphoid tissues, CCR2 + monocytes and macrophages formed a mesh-like network of cells that both overlapped with and bordered the neutrophils (Fig. 1h ). Consistent with the presence of bacterial colonies detected by histology and fluorescence microscopy, PG+ tissue harboured high numbers of viable bacteria, comparable to that found in the PPs (Fig. 1i,j ). Since PPs are a major entry point and replicative niche for enteropathogenic Yersinia following oral innoculation 23 , 24 , altogether, these findings reveal intestinal PGs as a previously unappreciated location of Yersinia invasion within the intestinal mucosa and a potential site of bacterial restriction or dissemination. Intestinal inflammation is spatially restricted to PGs To test whether PGs exhibit location-specific inflammatory responses, we next performed RNA sequencing of PGs, adjacent non-PG tissue and uninfected tissue. Principal component analysis showed distinct clustering by sample type (Fig. 2a ), and comparison of PG+ and PG− samples revealed 355 upregulated and 363 downregulated genes (Fig. 2b ). Top upregulated genes included granulocyte- and monocyte-recruiting chemokines ( Cxcl1, Cxcl2, Cxcl3 and Cxcl5 ), pro-inflammatory cytokines ( Il1b and Il22 ), metal-sequestration proteins ( S100a8 and S100a9 ) and matrix metalloproteases ( Mmp3 ) (Fig. 2c ). Gene Ontology analysis indicated that chemotaxis of myeloid cells and defence against bacterial pathogens predominated the top upregulated responses within PG+ biopsies (Fig. 2d and Extended Data Table 1 ). These responses were strikingly similar to previously reported Yp -infected PPs 25 . Consistently, gene set-enrichment analysis indicated that a set of 50 upregulated genes previously reported in Yp -infected PPs were also enriched in PG+ samples (Fig. 2e ). Furthermore, protein levels of the pro-inflammatory cytokines IL-1α, IL-1β, IL-6, TNF and and CCL2 were significantly elevated within PG+ biopsies (Fig. 2f ). Consistent with histology and microscopy, the inflammatory transcriptional response was localized to PG+ tissue, as PG− samples did not exhibit enrichment of genes or ontology terms related to myeloid cell migration or innate immune-cell activation (Extended Data Fig. 2 and Extended Data Table 2 ). Likewise, production of pro-inflammatory cytokines was not detected in PG− tissue (Fig. 2e ). Altogether, these data indicate that the pro-inflammatory response to Yp infection in the gut mucosa is spatially restricted to PG. Fig. 2: Intestinal inflammation is spatially restricted to PGs. a , Principal component analysis of PG+ (pink), PG− (green) and uninfected (uninf; grey) samples at day 5 post infection. Five biopsies were pooled per mouse. b , Heat map of all differentially expressed genes in PG+ compared with PG− samples. FDR <0.05 using Benjamini–Hochberg procedure. Pink and orange bars denote two clusters grouped based on Pearson correlation. c , Heat map of top 30 significantly upregulated genes in PG+ compared with PG− samples in descending order by fold change. FDR <0.05 using Benjamini–Hochberg procedure. d , Gene Ontology analysis of top 30 upregulated genes by fold change only in PG+ compared with PG− samples. p , P value. e , Gene set enrichment analysis of top 50 upregulated genes in Yp -infected PP. NES, normalized enrichment score. FDR, false discovery rate. f , Cytokine levels in homogenates of tissue punch biopsies at day 5 post infection. Lines represent group mean. Each circle represents one mouse ( n = 6–9). Statistical analysis by Mann–Whitney U test (two-tailed). * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001; NS, not significant. Data from one or two pooled independent experiments. Full size image Inflammatory monocytes maintain PGs to restrict infection Inflammatory monocytes promote host defence by differentiating into phagocytes and antigen presenting cells 4 , 6 , 26 , 27 , producing pro-inflammatory mediators 3 and modulating other immune cell functions 5 , 28 . Monocyte-derived cells can also promote pathogen replication or dissemination to new sites 7 , 29 . Mice lacking the chemokine receptor CCR2 have a tenfold reduction in circulating monocytes due to defective bone marrow egress 16 , 17 . Monocytes are rapidly recruited to intestinal PGs (Fig. 1 ) and sites of systemic Yp infection 15 , 18 , 22 , raising the question of their role in Yp infection. CCR2 deficiency has been associated with more rapid bacterial clearance from the mesenteric lymph node (MLN) following enteric Yersinia infection 15 , but increased susceptibility to intravenous Yp infection 14 . Interestingly, we found that while Ccr2 gfp/gfp mice, which lack circulating mononcytes due to lack of CCR2 expression 30 , 31 , had similar overall numbers of macroscopic intestinal lesions as wild-type (WT) mice (Fig. 3a and Extended Data Fig. 3a ), their lesions had a disorganized appearance, and exhibited central caseation with tissue necrosis (Fig. 3b ). In contrast to WT PG that exhibited robust inflammatory infiltrates and a defined cellular organization encapsulating central bacterial colonies, Ccr2 gfp/gfp intestinal lesions contained expanded coalescing bacterial colonies with limited immune cell recruitment (Fig. 3b,c ). PPs of Ccr2 gfp/gfp mice had similarly disorganized lesions with central tissue necrosis, suggesting that monocytes are required to establish or maintain organized PGs during Yp infection (Extended Data Fig. 3b ). Importantly, two independent CCR2-deficient mouse lines showed significantly higher bacterial burdens in PG+ and PG− tissues compared with WT mice (Fig. 3d and Extended Data Fig. 3c ). Moreover, acute depletion of monocytes in WT mice with anti-CCR2 specific antibodies resulted in increased PG− and PG+ bacterial burden (Extended Data Fig. 3d,e ), demonstrating that the requirement for monocytes in control of Yp is not due to developmental defects in CCR2-deficient mice. Interestingly, at day 3 post infection, Ccr2 gfp/gfp mice had elevated bacterial burdens in PG− tissue but not PG+ tissue, and did not exhibit overt signs of tissue necrosis in PGs (Extended Data Fig. 3f,g ), indicating that the defect in bacterial control and PG architecture develops between days 3 and 5. Together, these data suggest that monocytes enable maintenance of PG architecture and restrict Yp within intestinal PG. Fig. 3: Inflammatory monocytes maintain PGs to restrict infection. a , Quantification of total intestinal lesions at day 5 post infection. Each circle represents one mouse ( n = 20). Lines represent median. Pooled data from three independent experiments. b , H&E-stained small-intestinal sections from Yp -infected mice at day 5 post infection: 1 denotes the Yersinia microcolony, 2 denotes necrotic tissue. Scale bars, 250 μm (top) and 50 μm (bottom). Representative images of two independent experiments. c , Histopathological scores of small intestinal tissue at day 5 post infection. Each mouse was given a score between 0 and 4 (healthy to severe) for presence of bacterial colonies free from immune cell infiltrate. Each circle represents one mouse ( n = 7–9 for uninfected (uninf) and 9–16 for Yp ). Lines represent median. Pooled data from two independent experiments. d , Bacterial burdens in PG− and PG+ tissue at day 5 post infection. Each circle represents the mean Yp -CFU of 3–5 pooled punch biopsies from one mouse ( n = 31–37). Lines represent geometric mean. Pooled data from six independent experiments. e , Total numbers and frequencies of indicated cells in small intestinal uninf, PG− and PG+ tissue at day 5 post infection. Each circle represents the mean of three to ten pooled punch biopsies from one mouse ( n = 24). Lines represent median. Pooled data from five independent experiments. f , Fluorescently labelled PG+ tissue from Yp -infected Ccr2 gfp/+ (top) and Ccr2 gfp/gfp (bottom) mice at day 5 post infection. Scale bars, 100 μm. Representative images of two independent experiments. g , PG+ neutrophil surface CD11b expression (mean fluorescent intensity, MFI) at day 5 post infection. Each circle represents the mean of three to ten pooled punch biopsies from one mouse ( n = 6). Lines represent median. Representative of four independent experiments. h , Cytokine levels in homogenates of tissue punch biopsies were measured by cytometric bead array at day 5 post infection. Each circle represents the mean of three to ten pooled punch biopsies from one mouse ( n = 18–21). Lines represent median. Pooled data from three independent experiments. Mann–Whitney U test (two-tailed) was performed for all statistical analyses. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001; NS, not significant. Full size image Consistent with their reduced overall cellularity, Ccr2 gfp/gfp lesions exhibited decreased frequency and numbers of viable CD45 + hematopoietic cells compared with WT PG (Fig. 3e ). Consistent with the important role of CCR2 in promoting monocyte egress from bone marrow 16 , 17 , Ccr2 gfp/gfp intestinal lesions contained significantly lower numbers of monocytes, macrophages and dendritic cells compared with PG from WT mice (Fig. 3e and Extended Data Fig. 4a ). Notably, CCR2 deficiency did not impact T- or B-cell numbers in intestinal PGs, indicating that the defect in enteric control of Yp in CCR2-deficient mice was independent of adaptive immune cells (Extended Data Fig. 4a ). Ccr2 gfp/gfp intestinal lesions also showed a trend toward reduced neutrophil numbers (Fig. 3e ), suggesting that monocytes promote recruitment, retention or survival of neutrophils within intestinal PG during Yp infection. This defect was specific to the intestinal lesions, as we observed similar frequencies of neutrophils in the MLN and spleen (Extended Data Fig. 4b ). Immunofluorescence microscopy of the lesions indicated that neutrophils were unable to effectively contain Yp in the absence of monocytes, as the bacterial colony expanded outside the range of the neutrophil marker Ly-6G. This contrasted with WT PGs, in which Yp was fully encapsulated by neutrophils and CCR2 + cells (Fig. 3f ). Intriguingly, surface expression of the integrin CD11b, a well-established marker of neutrophil activation 32 , 33 , 34 , was significantly reduced in both PGs and MLN of Ccr2 gfp/gfp mice compared with the WT counterparts (Fig. 3g and Extended Data Fig. 4c ), suggesting a defect in neutrophil activation in the absence of monocytes. CD11b is present on the neutrophil cell surface and membranes of intracellular granules, and increased CD11b surface expression occurs in inflammatory settings 35 , 36 , 37 . Notably, PG neutrophils in CCR2-deficient mice exhibited increased intracellular CD11b levels, suggesting that translocation of CD11b from intracellular granules to the cell surface is defective in the absence of monocytes (Extended Data Fig. 4d ). Interestingly, total IL-1α and IL-1β levels were significantly reduced in Ccr2 gfp/gfp PGs, whereas other pro-inflammatory cytokines were unaffected (Fig. 3h and Extended Data Fig. 4e ), indicating that monocytes or monocyte-derived cells specifically produce IL-1, or promote IL-1 production by other cells within intestinal PGs. Notably, intracellular levels of IL-1 cytokines, TNF and lipocalin were unaffected in PG neutrophils in Ccr2 gfp/gfp mice (Extended Data Fig. 4f ), suggesting that inflammatory monocytes do not regulate neutrophil-intrinsic inflammatory cytokine and antimicrobial protein production. Altogether, these results demonstrate that inflammatory monocytes or monocyte-derived cells promote maintenance of functional granulomas that limit intestinal bacterial replication and dissemination. Inflammatory monocytes control systemic Yersinia Following systemic dissemination, Yp colonizes and induces PGs in lymphoid tissues 15 , 18 , 22 . Critically, both CCR2-deficient and anti-CCR2 depleted mice had significantly higher Yp burdens in the MLN and systemic organs (Fig. 4a and Extended Data Fig. 5 a,b ). Similar to intestinal tissue, systemic bacterial burdens in Ccr2 gfp/gfp mice were unaffected at day 3 post infection, indicating that this defect in control develops between days 3 and 5 (Extended Data Fig. 5c ). Consistent with our findings that monocytes were required for maintenance of intestinal PG architecture, infected Ccr2 gfp/gfp spleens exhibited widespread tissue necrosis, free bacterial colonies and sparse immune cell recruitment, in contrast to WT spleens where neutrophils and monocytes effectively encapsulated Yp microcolonies within organized PGs (Fig. 4b,c ). Notably, mice lacking CCR2 succumbed rapidly to acute infection (Fig. 4d and Extended Data Fig. 5d ). Importantly, co-housed littermate Ccr2 +/+ and Ccr2 gfp/+ mice were equally resistant to Yp infection while Ccr2 gfp/gfp littermates succumbed (Fig. 4e ), indicating that increased susceptibility of Ccr2 gfp/gfp to Yp infection is not due to differences in composition of a vertically transmitted intestinal microbiota. Collectively, our findings demonstrate that inflammatory monocytes maintain organized PGs in host tissues, thereby limiting tissue necrosis and systemic bacterial dissemination, ultimately enabling bacterial control and host survival following oral Yp infection. Fig. 4: Inflammatory monocytes control systemic Yersinia . a , Bacterial burdens in indicated organs at day 5 post infection. Each circle represents one mouse ( n = 20–25). Lines represent geometric mean. Pooled data from four independent experiments. b , H&E-stained paraffin-embedded longitudinal spleen sections from WT and Ccr2 gfp/gfp mice at day 5 post infection. Dashed circle denotes area with bacterial microcolonies and neutrophils. Scale bars, 500 μm (top) and 50 μm (bottom). Representative images of two independent experiments. c , Histopathological scores of spleens from uninfected and Yp -infected mice at day 5 post infection. Each mouse was given a score between 0 and 4 (healthy–severe) for presence of bacterial colonies free from immune cell infiltrate. Each circle represents one mouse ( n = 9 for uninfected and 16 for Yp ). Lines represent median. Pooled data from two independent experiments. d , Survival of infected WT ( n = 16) and Ccr2 gfp/gfp ( n = 13) mice. Pooled data from two independent experiments. e , Survival of infected littermate WT Ccr2 +/+ ( n = 20), heterozygous Ccr2 gfp/+ ( n = 19) and homozygous Ccr2 gfp/gfp ( n = 15) mice. Pooled data from three independent experiments. Statistical analyses by Mann–Whitney U test (two-tailed) ( a and c ) and Mantel–Cox test ( d and e ). * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001; NS, not significant. Full size image Yersinia virulence factors induce intestinal PGs Yersinia inject Yops, which are encoded on a virulence plasmid (pYV), into target cells 38 to block phagocytosis, production of reactive oxygen species (ROS) and degranulation 20 . Interestingly, Yp lacking the pYV (pYV−) still acutely colonize the intestine without causing disease 39 . Since granulomas form in response to pathogens that thwart immune defences, we hypothesized that intestinal PGs may be triggered by the activity of Yp effector proteins. Indeed, even at a tenfold higher infectious dose, we did not observe intestinal lesions at day 5 post infection with pYV− bacteria (Fig. 5a ), despite detectable (although reduced) intestinal colonization (Extended Data Fig. 6a ). Fig. 5: Yersinia virulence factors induce intestinal PGs. a , Quantification of total intestinal lesions at day 5 post Yp infection. Each symbol represents one mouse ( n = 10–43). Lines represent median. Pooled from two to six independent experiments. b , H&E-stained paraffin-embedded transverse small-intestinal sections from WT and Ccr2 gfp/gfp mice infected with WT or YopH R409A Yp at day 5 post infection depicting encircled bacterial colonies. Scale bars, 100 μm. Representative of two independent experiments. c , Frequency and total number of neutrophils in small-intestinal PG+ tissue at day 5 post WT or YopH R409A Yp infection. Each circle represents one mouse ( n = 12–14). Lines represent median. Pooled from three independent experiments. d , Mean fluorescent intensity (MFI) of surface CD11b expression on neutrophils in PG+ tissue at day 5 post WT or YopH R409A Yp infection. Each symbol represents one mouse ( n = 7–10). Lines represent median. Pooled from two independent experiments. e , Bacterial burdens in small intestinal PG− and PG+ tissue at day 5 post WT or YopH R409A Yp infection. Each symbol represents one mouse ( n = 25–26). Lines represent geometric mean. Pooled from four independent experiments. f , Bacterial burdens in indicated organs at day 5 post WT or YopH R409A Yp infection. Each symbol represents one mouse ( n = 15–22). Lines represent geometric mean. Pooled from four independent experiments. g , Survival of WT mice infected with WT ( n = 20) or YopH R409A ( n = 20), Yp and Ccr2 −/− mice infected with WT ( n = 10) or YopH R409A ( n = 17) Yp . Pooled from two independent experiments. Statistical analyses by Kruskal–Wallis test with Dunn’s post-test ( a and c – f ), and Mantel–Cox test ( g ). * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001; NS, not significant. Full size image pYV− bacteria still induced monocyte recruitment to both the intestinal mucosa and MLN, indicating that lack of detectable PG was not due to an absence of immune infiltration (Extended Data Fig. 6b ). However, neutrophil accumulation was absent upon pYV− infection, demonstrating that neutrophil recruitment occurs in response to Yp virulence (Extended Data Fig. 6c ). Although comparable in the intestinal mucosa, pYV− bacterial burdens were reduced in other gut-associated and systemic lymphoid tissues (Extended Data Fig. 6d ), illustrating that intestinal PG formation is dispensable for control of pYV− Yersinia . Intestinal lesions formed at WT levels in mice infected with Yp individually deficient in either YopM or YopJ enzymatic activity, which block pyrin inflammasome assembly or nuclear factor kappa B and mitogen‑activated protein kinase signalling 40 , 41 , 42 , respectively, suggesting that neither YopM nor YopJ is singly required for PG formation (Extended Data Fig. 6e ). Several Yops function together to disrupt the actin cytoskeleton, thereby blocking phagocytosis and the reactive oxygen burst 20 . These Yops (E, H and T), have partially overlapping functions and can compensate for one another in certain settings 20 . Notably, Yp with combined point mutations in each of the catalytic residues of these Yops (YopE R144A , YopT C139A and YopH R409A ) did not induce intestinal lesions and had burdens similar to pYV− infection (Fig. 5a and Extended Data Fig. 6a,d ). In contrast, bacteria lacking YopE and YopT, or YopE alone, induced WT numbers of intestinal lesions (Fig. 5a ) and were only attenuated in systemic organs (Extended Data Fig. 6d ), indicating that YopH is sufficient, in the absence of YopE and YopT, to induce PG formation. Interestingly, YopH R409A mutants that lack YopH tyrosine phosphatase activity induced fewer PGs than WT bacteria (Fig. 5a ), suggesting that YopH is sufficient and partially responsible for PG formation. In addition, MLN colonization was abrogated in the absence of YopH activity, whereas YopE was dispensable for MLN colonization (Extended Data Fig. 6d ). Intriguingly, bacteria lacking both YopE and YopH ( yopEH ) induced no detectable lesions (Fig. 5a ) and showed similar levels of colonization to yopETH bacteria (Extended Data Fig. 6a,d ), suggesting that the host response to YopE and YopH drives PG formation. YopH blocks innate cell phagocytosis and ROS production 20 , 43 , 44 , 45 , 46 . YopH-deficient Yp are therefore more susceptible to neutrophil killing in vitro and are attenuated in vivo 47 , 48 , 49 . Consequently, neutrophil depletion increases susceptibility to Yp and restores virulence to YopH-mutant bacteria 43 , 45 , indicating that neutrophil defences are an important target of YopH in vivo. However, whether monocytes combat YopH-dependent blockade of neutrophil function is unknown. As in WT mice, we found that YopH-mutant bacteria induced significantly fewer intestinal lesions in Ccr2 gfp/gfp mice than WT bacteria (Extended Data Fig. 6f ). Strikingly, despite the lack of monocytes, YopH R409A infection of CCR2-deficient mice restored neutrophil recruitment and induced neutrophil-rich lesions without necrosis (Fig. 5b,c and Extended Data Fig. 6g ). Intriguingly, CD11b surface expression was restored in PG+ tissue of CCR2-deficient animals infected with YopH R409A bacteria, indicating that YopH limits neutrophil activation in the absence of monocytes (Fig. 5d ). Furthermore, CCR2-deficient mice effectively controlled bacterial burdens in intestinal and systemic tissues following infection with YopH R409A bacteria (Fig. 5e,f and Extended Data Fig. 6h ). Consistently, while CCR2-deficient mice rapidly succumbed to WT Yp , they survived infection with YopH R409A bacteria at levels similar to WT mice infected with WT Yp , suggesting that monocytes control Yersinia infection by overcoming the virulence of YopH (Fig. 5g ). Altogether, these data indicate that intestinal PGs form in response to Yersinia disruption of innate immune defence, and that monocytes counteract YopH activity in vivo. Neutrophils control YopH-deficient Yersinia in absence of monocytes While monocytes are dispensable for the control of YopH-deficient Yp , the cell type that mediates this control in the absence of monocytes is unknown. Neutrophil recruitment to PGs was restored during infection of Ccr2 gfp/gfp mice with YopH-deficient Yp , raising the possibility that neutrophils mediate control of Yersinia in the absence of monocytes. Since Ccr2 gfp/gfp mice largely lack circulating monocytes, anti-Gr-1, which depletes both neutrophils and monocytes and is highly efficient at depleting neutrophils in infectious or inflammatory settings 50 , 51 , allowed us to interrogate the role of neutrophils in control of YopH-mutant Yp (Fig. 6a and Extended Data Fig. 7a ). Strikingly, Ccr2 gfp/gfp mice injected with anti-Gr-1 had elevated burdens of YopH-mutant Yp in PG- tissue (Fig. 6b ). Anti-Gr-1 did not further reduce blood monocyte frequencies in Ccr2 gfp/gfp mice (Extended Data Fig. 7b ), indicating that increased susceptibility of these mice was not attributable to depletion of remaining Ly-6C + monocytes. CCR2-deficient mice infected with YopH R409A bacteria had very few detectable PG (Fig. 6c ), making it difficult to robustly analyse bacterial burdens in this tissue. Intriguingly, Ccr2 gfp/gfp mice injected with anti-Gr-1 had elevated numbers of macroscopic intestinal lesions during infection with YopH R409A bacteria (Fig. 6c ). Moreover, in peripheral organs, Ccr2 gfp/gfp mice injected with anti-Gr-1 had elevated bacterial burdens following YopH-mutant infection (Fig. 6d ), demonstrating that, in the absence of monocytes, neutrophils play a key role in control of YopH-deficient bacteria. Notably, mice infected with WT bacteria did not exhibit increased systemic bacterial burdens upon neutrophil depletion (Fig. 6d ), consistent with effective blockade of neutrophil functions by YopH. Collectively, these data indicate that neutrophils control YopH-deficient Yp in the absence of monocytes. Fig. 6: Neutrophils control YopH-deficient Yersinia in absence of monocytes. a , Frequency of neutrophils in blood at day 5 post infection was determined by flow cytometry. Each symbol represents one mouse ( n = 10–11). Lines represent median. Pooled data from three independent experiments. b , Bacterial burdens in small intestinal PG− tissue at day 5 post WT or YopH R409A Yp infection. Each symbol represents one mouse ( n = 10–12). Lines represent geometric mean. Pooled data from three independent experiments. c , Quantification of total number of intestinal lesions at day 5 post infection. Each symbol represents one mouse ( n = 10–12). Lines represent median. Pooled data from three independent experiments. d , Bacterial burdens in indicated organs at day 5 post WT or YopH R409A Yp infection. Each symbol represents one mouse ( n = 10–12). Lines represent geometric mean. Pooled data from three independent experiments. All statistical analyses by Kruskal–Wallis test with Dunn’s post-test. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001; NS, not significant. Full size image Discussion Granulomas are a conserved response to persistent or long-term infectious and non-infectious stimuli 1 , 2 . The natural rodent and human pathogen Yp induces formation of neutrophil-rich PGs in infected lymphoid tissues 11 , 13 . Here we report that PG form throughout the intestine during murine enteropathogenic Yersinia infection. PPs are considered the primary site of Yersinia intestinal infection. However, intestinal PGs harboured a similar total bacterial burden as the PPs, suggesting that PGs are a previously unrecognized niche for enteropathogenic Yersinia intestinal colonization. Notably, WT mice almost entirely restricted intestinal Yp and inflammation to these granulomatous foci. Interestingly, the transcriptional profile of PGs was similar to Yp- infected PPs 25 , indicating a shared response to intestinal Yp infection driven by recruitment and activation of innate immune cells, which may serve to limit bacterial spread and tissue damage within the gut mucosa. While initial formation of PGs was similar in CCR2-deficient and WT mice, monocytes were critical for maintaining architectural integrity of PGs and bacterial control in the intestine and deeper tissues during Yersinia infection, as monocyte deficiency was associated with widespread tissue necrosis and elevated bacterial burdens in intestinal and systemic tissues that occurred between day 3 and day 5. Interestingly, Y. pestis , which causes more severe disease than Yp , is commonly associated with necrotic and necrosuppurative lesions 52 , suggesting that Y. pestis has evolved to overcome monocyte-mediated host defence and PG formation to enhance systemic dissemination and transmission. Monocytes were previously reported to be dispensable for acute control of Yp following oral inoculation 15 . These studies employed a Ccr2 −/− mouse line 17 , 27 originally generated on the 129 mouse background and backcrossed to C57BL/6 (ref. 16 ), raising the possibility that distinct polymorphisms in this mouse line might account for this difference. Ccr2 gfp/gfp mice were generated directly on the C57BL/6 J background 30 , making it unlikely that our findings are due to immunologically impactful polymorphisms that co-segregate with the Ccr2 locus. Infection of co-housed littermates also demonstrated that susceptibility of CCR2-deficient mice is not due to differences in maternally transmitted or environmentally acquired microbiota. Additionally, acute depletion of monocytes with anti-CCR2 antibodies phenocopied CCR2-deficient mice in abrogating control of intestinal and systemic bacterial burdens, demonstrating that monocytes are critical for acute control of enteric Yersinia infection. Lack of both YopE and YopH abrogated PG formation, whereas either effector alone was sufficient to induce PG formation, implying a key role for actin cytoskeleton disruption in PG formation. Notably, ablation of YopH activity alone significantly reduced numbers of intestinal PGs, indicating that YopH was predominantly responsible for PG formation. Precisely how YopH and YopE lead to PG formation remains to be determined. Monocytes and neutrophils comprise a large proportion of Yop-injected cells in vivo 53 . YopH blocks neutrophil degranulation, ROS production, phagocytosis and release of neutrophil extracellular traps 20 , 43 , 45 , 49 , 54 . The virulence of YopH-deficient bacteria is restored upon neutrophil depletion or ROS deficiency 43 , 45 , suggesting that YopH potently targets neutrophil function in tissues during infection. In line with these observations, PGs of CCR2-deficient mice exhibited decreased numbers of neutrophils and reduced surface expression of the activation marker CD11b, which were restored in the setting of YopH deficiency. CD11b is expressed basally on the plasma membrane and in neutrophil tertiary granules, which are recruited to the surface upon neutrophil activation 35 , 36 , 37 , 55 . Our findings imply that monocytes enhance neutrophil degranulation to overcome YopH-dependent blockade of their antimicrobial functions. Consistently, while monocytes were required for host survival in response to WT bacteria, they were dispensable for controlling YopH-mutant Yp . Furthermore, depletion of neutrophils in YopH R409A -infected CCR2-deficient mice restored bacterial virulence. Overall, our findings imply that monocytes play an important role in host defence by promoting maintenance of PG architecture and neutrophil function in the face of YopH-dependent blockade. Our study reveals a previously unappreciated site of Yersinia colonization within the intestine and provides insight into granuloma formation and function during Yersinia infection. Methods Animals C57BL/6 WT and Ccr2 gfp/gfp mice 30 (in which insertion of enhanced green fluorescent protein into the translation initiation site of Ccr2 disrupts its expression) were acquired from the Jackson Laboratory and bred at the University of Pennsylvania. Ccr2 −/− mice 31 were provided by Dr Sunny Shin (University of Pennsylvania). Unless specifically noted, all animals were bred by homozygous mating and housed separately by genotype. Mice of either sex between 8 and 12 weeks of age were used for all experiments. All animal studies were performed in accordance with University of Pennsylvania Institutional Animal Care and Use Committee-approved protocols (protocol no. 804523). Bacteria WT Yp (clinical isolate strain 32777, serogroup O1) (ref. 56 ) and isogenic mutants were provided by Dr James Bliska (Dartmouth College). Ye (strain 8081, serogroup O8) 57 was provided by Dr Stanley Falkow (Stanford University). Additional mutants lacking YopE ( ΔyopE ), enzymatic activity of YopH (YopH R409A ), both (denoted yopEH ) or YopE/YopT/YopH (YopE R144A T C139A H R409A , denoted yopETH ) were generated by two-step allelic recombination as previously described 58 with plasmids provided by Dr James Bliska. Fluorescent Yp (mCherry + ) was generated from plasmids provided by Dr Kimberly Davis (Johns Hopkins University). Infections Yp and Ye were cultured to stationary phase at 28 °C and 250 r.p.m. shaking for 16 h in 2× YT broth supplemented with 2 μg ml −1 triclosan (Millipore Sigma). Mice were fasted for 16 h and subsequently inoculated by oral gavage with 100–200 μl phosphate-buffered saline (PBS). All bacterial strains were administered at 2 × 10 8 colony-forming units (CFU) per mouse with the exception of pYV−, yopETH and yopEH , which were administered at 2 × 10 9 CFU per mouse. Antibody-mediated depletions Mice were given daily rat IgG2 isotype control or depletion antibodies in 100 μl PBS by intraperitoneal injections from day −1 to day 4 post infection. To deplete monocytes, mice were given 20 μg rat anti-mouse CCR2 (ref. 59 ) (clone MC-21 AK). To deplete neutrophils in CCR2-deficient animals, mice were given 200 μg rat anti-mouse Gr-1 (RB6-8C5; Bio X Cell). Protein quantifications Cytokines were measured in supernatants from homogenized tissue using Cytometric Bead Array (BD Biosciences) according to the manufacturer’s instructions with the following modification: the amount of capture beads, detection reagents and sample volumes was scaled down tenfold. Data were collected on an LSRFortessa flow cytometer (BD Biosciences) with FACSDiva v9.0 (BD Biosciences) and analysed with FlowJo v10 (BD Biosciences). Tissue preparation and cell isolation Blood was collected by cardiac puncture upon euthanasia and collected in 250 U ml −1 heparin solution (Millipore Sigma) before erythrocyte lysis with Red Blood Cell Lysing Buffer (Millipore Sigma). Lymph nodes and spleens were homogenized through a 70 μm cell strainer (Fisher Scientific), then flushed with R10 buffer consisting of Roswell Park Memorial Institute Medium 1640 (Millipore Sigma) supplemented with 10 mM HEPES (Millipore Sigma), 10% foetal bovine serum (Omega Scientific), 1 mM sodium pyruvate (Thermo Fisher Scientific) and 100 U ml −1 penicillin + 100 μg ml −1 streptomycin (Thermo Fisher Scientific). Intestines were excised, flushed luminally with sterile PBS to remove the faeces, opened longitudinally along the mesenteric side and placed luminal side down. Small intestinal tissue containing macroscopically visible PGs (PG+), adjacent non-granulomatous areas (PG−) and uninfected control tissue (uninf) were excised using a 2 mm-ø dermal punch biopsy tool (Keyes). Biopsies within each mouse were pooled, suspended in epithelial-dissociation buffer consisting of calcium- and magnesium-free Hanks’ Balanced Salt Solution (HBSS) (Thermo Fisher Scientific) supplemented with 15 mM HEPES, 10 mg ml −1 bovine serum albumin (BSA) (Millipore Sigma), 5 mM ethylenediaminetetraacetic acid (EDTA) (Millipore Sigma) and 100 U ml −1 penicillin + 100 μg ml −1 streptomycin, then incubated for 30 min at 37 °C under continuous agitation. To isolate immune cells from the lamina propria, the tissue was enzymatically digested in R10 buffer, along with 0.5 Wünsch units ml −1 liberase Thermolysin Medium (Roche), 30 μg ml −1 DNase I (Roche) and 5 mM CaCl 2 for 20 min at 37 °C under continuous agitation. The resulting cell suspensions were filtered through 100 μm cell strainers (Fisher Scientific) and subjected to density-gradient centrifugation using Percoll (GE Healthcare). Briefly, cells were suspended in 40% Percoll and centrifuged over a 70% Percoll layer for 20 min at 600 g , with lowest brake at room temperature. Cells collected between the layers were washed with R10 for downstream analysis. Flow cytometry Unspecific constant fragment (Fc) binding was blocked for 10 min on ice with anti-CD16/CD32 (93; Thermo Fisher Scientific). Cells were subsequently stained for 30 minutes on ice with the following antibodies and reagents: PE-conjugated rat anti-mouse Siglec-F (E50-2440; BD Biosciences), PE-TxR or PE-Cy5-conjugated rat anti-mouse CD11b (M1/70.15; Thermo Fisher Scientific), PE-Cy5-conjugated mouse anti-mouse NK1.1 (PK136; BioLegend), PE-Cy5.5 or PE-Cy7-conjugated rat anti-mouse CD4 (RM4-5; Thermo Fisher Scientific), PE-Cy7-conjugated rat anti-mouse CD3 (17A2; BioLegend), BV510-conjugated rat anti-mouse CD3e (145-2C11; BioLegend), FITC-conjugated Armenian hamster anti-mouse CD11c (N418; BioLegend), PerCP-Cy5.5-conjugated rat anti-mouse Ly-6C (HK1.4; Thermo Fisher Scientific), PB-conjugated rat anti-mouse CD90.2 (53-2.1; BioLegend), BV510-conjugated rat anti-mouse CD19 (1D3; BD Biosciences), BV605-conjugated Armenian hamster anti-mouse TCRβ (H57-597; BD Biosciences), BV650-conjugated rat anti-mouse I-A/I-E (M5/114.15.2; BD Biosciences), BV711-conjugated rat anti-mouse CD8α (53-6.7; BD Biosciences), BV785-conjugated rat anti-mouse Ly-6G (1A8; Thermo Fisher Scientific), AF647-conjugated mouse anti-mouse CD64 (X54-5/7.1; BD Biosciences), AF700-conjugated mouse anti-mouse CD45.2 (104; BioLegend) and PE-CF594-conjugated rat anti-mouse CD45R/B220 (RA3-6B2; BD Biosciences) along with eF780 viability dye (BioLegend) diluted in PBS. Antibodies were used at 1:200 dilution and viability dye at 1:1,500 dilution. For intracellular staining, cells were incubated for 3 h at 37 °C with 5% CO 2 in R10 buffer supplemented with 0.33 μl ml −1 GolgiStop (BD Biosciences) and 15 μg ml −1 DNase I. Surface proteins were stained as above, then cells were fixed for 20 min on ice with Cytofix/Cytoperm Fixation/Permeabilization solution (BD Biosciences). Lipocalin-2 was stained using biotin-conjugated rat anti-mouse lipocalin-2 (NGAL; BioLegend) on ice for 1 h followed by BV711-conjugated streptavidin (BD Biosciences) at 4 °C overnight. Intracellular cytokines were stained at 4 °C overnight with PerCP-e710-conjugated rat anti-mouse IL-1β (NJTEN3; Thermo Fisher Scientific), eF450-conjugated rat anti-mouse TNF (MP6-XT22; Thermo Fisher Scientific) and PE-conjugated Armenian hamster anti-mouse IL-1α (ALF-161; BioLegend). All intracellular antibodies were diluted 1:200 in Perm/Wash Buffer (BD Biosciences). Streptavidin was diluted 1:400 in Perm/Wash Buffer. Cells were acquired on an LSRFortessa flow cytometer with FACSDiva v9.0, and data were analysed with FlowJo v10. Dead and clustered cells were removed from all analyses. Histology Tissues were fixed in 10% neutral-buffered formalin (Fisher Scientific) and stored at 4 °C until further processing. Tissue pieces were embedded in paraffin, sectioned by standard histological techniques and stained with hematoxylin and eosin (H&E) for subsequent histopathological disease scoring by blinded board-certified pathologists. Tissue sections were given scores between 0 and 4 (healthy to severe) for multiple parameters, including degree of inflammatory cell infiltration, necrosis and free bacterial colonies along with tissue-specific parameters such as villus blunting and crypt hyperplasia. Healthy mice were characterized by and subsequently scored as having none or low levels of the parameters described, whereas severely afflicted mice presented with high amounts of the respective parameters. Fluorescence microscopy Small intestinal tissue was dissected and flushed with PBS. Intestines were opened longitudinally and ~0.5 cm tissue pieces containing macroscopically visible lesions were excised. Tissues were fixed in 1% paraformaldehyde overnight, then blocked for 2 h at room temperature in blocking solution containing 10% bovine serum albumin, 1 μg ml −1 anti-CD16/32 and 0.5% normal rat IgG in PBS. AF647-conjugated anti-Ly-6G antibody (1A8; BioLegend) was added at 0.01 mg ml −1 in 100 μl PBS per sample, then whole tissue was stained for 24 h at 4 °C. Samples were washed three times with PBS, mounted whole onto slides in Prolong Glass Antifade Mountant (Thermo Fisher Scientific) and cured for 2 days at room temperature. Images were acquired on a DMI 6000 laser-scanning confocal microscope (Leica) with a 20× NA 0.75 oil-immersion objective. The centre of the sample was determined in the Z direction, then imaged. Images were analysed using ImageJ v2.1. Adjustments for brightness and contrast were applied to the entire image. No threshold manipulation was performed. Green and magenta channels were pseudo-coloured. RNA sequencing Intestinal punch biopsies were collected as described above. Five biopsies per sample type were pooled for each mouse. RNA was extracted using the RNeasy Plus Mini Kit (Qiagen). Sequence-ready libraries were prepared using the Illumina TruSeq Stranded Total RNA kit with Ribo-Zero Gold rRNA depletion (Illumina). Quality assessment and quantification of RNA preparations and libraries were carried out using an Agilent 4200 TapeStation and Qubit 3, respectively. Samples were sequenced on an Illumina NextSeq 500 to produce 150 bp single-end reads with a mean sequencing depth of 9 million reads per sample. Raw reads from this study were mapped to the mouse reference transcriptome (Ensembl; Mus musculus GRCm38) using Kallisto v0.46.2 (ref. 60 ). Raw sequence data are available on the Gene Expression Omnibus (GEO; accession no. GSE194334 ). All subsequent analyses were carried out using the statistical computing environment R v4.0.3 in RStudio v1.2.5042 and Bioconductor 61 . Briefly, transcript quantification data were summarized to genes using the tximport package 62 and normalized using the trimmed mean of M values method in edgeR 63 . Genes with <1 counts per million (CPM) in three samples were filtered out. Normalized filtered data were variance stabilized using the voom function in limma 64 , and differentially expressed genes were identified with linear modelling using limma (false discovery rate (FDR) ≤0.05; absolute log 2 fold change ≥1) after correcting for multiple testing using Benjamini–Hochberg. Statistics Statistical analyses were performed using Prism v9.0 (GraphPad Software). Independent groups were compared by Mann–Whitney U test or Kruskal–Wallis test with Dunn’s multiple comparisons test. Survival curves were compared by Mantel–Cox test. Statistical significance is denoted as * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001 or NS (not significant, P > 0.05). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability Raw RNA sequencing data are available on the Gene Expression Omnibus (accession no. GSE194334 ). All other raw data are available upon request to the corresponding author. Code availability Code for RNA sequencing analysis is available in the Supplementary Code File.

Yersinia bacteria cause a variety of human and animal diseases, the most notorious being the plague, caused by Yersinia pestis. A relative, Yersinia pseudotuberculosis, causes gastrointestinal illness and is less deadly, but naturally infects both mice and humans, making it a useful model for studying its interactions with the immune system. These two pathogens, as well as a third close cousin, Y. enterocolitica, which affects swine and can cause food-borne illness if people consume infected meat, have many traits in common, particularly their knack for interfering with the immune system's ability to respond to infection. The plague pathogen is blood-borne and transmitted by infected fleas. Infection with the other two depends on ingestion. Yet the focus of much of the work in the field had been on interactions of Yersinia with lymphoid tissues, rather than the intestine. A new study of Y. pseudotuberculosis, led by a team from Penn's School of Veterinary Medicine and published in Nature Microbiology demonstrates that in response to infection, the host immune system forms small, walled-off lesions in the intestines called granulomas. It's the first time these organized collections of immune cells have been found in the intestines in response to Yersinia infections. The team went on to show that monocytes, a type of immune cell, sustain these granulomas. Without them, the granulomas deteriorated, allowing the mice to be overtaken by Yersinia. "Our data reveal a previously unappreciated site where Yersinia can colonize and the immune system is engaged," says Igor Brodsky, senior author on the work and a professor and chair of pathobiology at Penn Vet. "These granulomas form in order to control the bacterial infection in the intestines. And we show that if they don't form or fail to be maintained, the bacteria are able to overcome the control of the immune system and cause greater systemic infection." The findings have implications for developing new therapies that leverage the host immune system, Brodsky says. A drug that harnessed the power of immune cells to not only keep Yersinia in check but to overcome its defenses, the researchers say, could potentially eliminate the pathogen altogether. A novel battlefield Y. pestis, Y. pseudotuberculosis, and Y. enterocolitica share a keen ability to evade immune detection. "In all three Yersinia infections, a hallmark is that they colonize lymphoid tissues and are able to escape immune control and replicate, cause disease, and spread," Brodsky says. Earlier studies had shown that Yersinia prompted the formation of granulomas in the lymph nodes and spleen but had never observed them in the intestines until Daniel Sorobetea, a research fellow in Brodsky's group, took a closer look at the intestines of mice infected with Y. pseudotuberculosis. "Because it's an orally acquired pathogen, we were interested in how the bacteria behaved in the intestines," Brodsky says. "Daniel made this initial observation that following Yersinia pseudotuberculosis infection, there were macroscopically visible lesions all along the length of the gut that had never been described before." The research team, including Sorobetea and later Rina Matsuda, a doctoral student in the lab, saw that these same lesions were present when mice were infected with Y. enterocolitica, forming within five days after an infection. A biopsy of the intestinal tissues confirmed that the lesions were a type of granuloma, known as a pyogranuloma, composed of a variety of immune cells, including monocytes and neutrophils, another type of white blood cell that is part of the body's front line in fighting bacteria and viruses. Granulomas form in other diseases that involve chronic infection, including tuberculosis, for which Y. pseudotuberculosis is named. Somewhat paradoxically, these granulomas—while key in controlling infection by walling off the infectious agent—also sustain a population of the pathogen within those walls. The team wanted to understand how these granulomas were both formed and maintained, working with mice lacking monocytes as well as animals treated with an antibody that depletes monocytes. In the animals lacking monocytes "these granulomas, with their distinct architecture, wouldn't form," Brodsky says. Instead, a more disorganized and necrotic abscess developed, neutrophils failed to be activated, and the mice were less able to control the invading bacteria. These animals experienced higher levels of bacteria in their intestines and succumbed to their infections. Groundwork for the future The researchers believe the monocytes are responsible for recruiting neutrophils to the site of infection and thus launching the formation of the granuloma, helping to control the bacteria. This leading role for monocytes may exist beyond the intestines, the researchers believe. "We hypothesize that it's a general role for the monocytes in other tissues as well," Brodsky says. But the discoveries also point to the intestines as a key site of engagement between the immune system and Yersinia. "Previous to this study, we knew of Peyer's patches to be the primary site where the body interacts with the outside environment through the mucosal tissue of the intestines," says Brodsky. Peyer's patches are small areas of lymphoid tissue present in the intestines that serve to regulate the microbiome and fend off infection. In future work, Brodsky and colleagues hope to continue to piece together the mechanism by which monocytes and neutrophils contain the bacteria, an effort they're pursing in collaboration with Sunny Shin's lab in the Perelman School of Medicine's microbiology department. A deeper understanding of the molecular pathways that regulate this immune response could one day offer inroads into host-directed immune therapies, by which a drug could tip the scales in favor of the host immune system, unleashing its might to fully eradicate the bacteria rather than simply corralling them in granulomas. "These therapies have caused an explosion of excitement in the cancer field," Brodsky says, "the idea of reinvigorating the immune system. Conceptually we can also think about how to coax the immune system to be reinvigorated to attack pathogens in these settings of chronic infection as well."

10.1038/s41564-023-01338-6

Medicine

Stress-tolerant cells shown to drive tumor initiation in pancreatic cancer

David Cheresh, Pancreatic cancer cells upregulate LPAR4 in response to isolation stress to promote an ECM-enriched niche and support tumour initiation, Nature Cell Biology (2023). DOI: 10.1038/s41556-022-01055-y. www.nature.com/articles/s41556-022-01055-y Journal information: Nature Cell Biology

https://dx.doi.org/10.1038/s41556-022-01055-y

https://medicalxpress.com/news/2023-01-stress-tolerant-cells-shown-tumor-pancreatic.html

Abstract Defining drivers of tumour initiation can provide opportunities to control cancer progression. Here we report that lysophosphatidic acid receptor 4 (LPAR4) becomes transiently upregulated on pancreatic cancer cells exposed to environmental stress or chemotherapy where it promotes stress tolerance, drug resistance, self-renewal and tumour initiation. Pancreatic cancer cells gain LPAR4 expression in response to stress by downregulating a tumour suppressor, miR-139-5p. Even in the absence of exogenous lysophosphatidic acid, LPAR4-expressing tumour cells display an enrichment of extracellular matrix genes that are established drivers of cancer stemness. Mechanistically, upregulation of fibronectin via an LPAR4/AKT/CREB axis is indispensable for LPAR4-induced tumour initiation and stress tolerance. Moreover, ligation of this fibronectin-containing matrix via integrins α5β1 or αVβ3 can transfer stress tolerance to LPAR4-negative cells. Therefore, stress- or drug-induced LPAR4 enhances cell-autonomous production of a fibronectin-rich extracellular matrix, allowing cells to survive ‘isolation stress’ and compensate for the absence of stromal-derived factors by creating their own tumour-initiating niche. Main Tumour-initiating cells (TICs) are critical contributors to tumour growth, recurrence and metastatic spread by virtue of their capacity to overcome various forms of stress 1 , 2 . Such features are not only derived from intrinsic stem-like traits but can be influenced by tumour stroma-derived immune cell factors within a cancer stem cell (CSC) niche 3 . One such factor is lysophosphatidic acid (LPA) 4 , a bioactive lipid that regulates cell fate transitions for a variety of stem cell types, including CSCs 5 , 6 . We are particularly interested in how TIC properties may fluctuate in response to external cues 7 , such as changes in vascularization that impact oxygenation and access to nutrients, or application of therapeutic pressures designed to eliminate vulnerable populations of tumour cells. At the earliest stage of tumour formation or during seeding of metastatic cells in distant organs, individual tumour cells must overcome the challenging consequences of isolation-induced stresses such as hypoxia, nutrient deprivation and oxidative stress, while compensating for the loss of matrix adhesion, cell–cell contact or immune cell and stromal-produced survival factors 8 , 9 . While TICs may inherently possess resistance to such stresses, we considered whether non-TICs may undergo adaptive reprogramming to overcome isolation stress and thereby gain tumour-initiating properties. Pancreatic cancer is a lethal cancer that is notoriously resilient to therapy due in part to poor vascularization and dense stroma in the tumour microenvironment 10 . LPA and the enzyme that produces LPA, autotaxin, are particularly enriched in pancreatic cancer 11 , 12 , which can exploit LPA produced by neighbouring pancreatic stellate cells to drive pacreatic ductal adenocarcinoma (PDAC) growth in vivo 12 . Functions of the six G-protein-coupled LPA receptors (LPAR1–6) do not appear to overlap in pancreatic cancer cells. LPAR1 promotes cell migration and metastasis 13 , 14 , while LPAR2 suppresses migration 15 . Interestingly, LPAR4 suppresses mobility in PANC1 cells 16 but is also necessary for cardiac dedifferentiation and heart tissue repair 17 , suggesting a context-dependent role. In this Article, we discovered that LPAR4 becomes upregulated by the stress-induced loss of a tumour-suppressive microRNA (miRNA) and is capable of conferring TIC properties, stress tolerance and drug resistance. Mechanistically, LPAR4 expression endows solitary tumour cells with the ability to create their own fibronectin-containing niche to support tumour initiation. The molecular basis for this adaptive reprogramming mechanism may provide therapeutic opportunities to slow pancreatic cancer progression and sensitize tumours to therapeutic intervention. Results Pancreatic cancer cells upregulate LPAR4 in response to stress Tumour cells must overcome various stressors during tumour initiation and progression, including hypoxia, nutrient stress, loss of adhesion, oxidative stress and therapeutic stress. Given the role of autotaxin/LPA in pancreatic cancer progression 12 , we asked whether one or more LPA receptor(s) are linked to tumour initiation. To create tumour growth conditions that would selectively favour the survival of TICs in vivo, nu/nu mice received a subcutaneous injection of 300 pancreatic cancer cells, a cell number we established produces palpable tumours in approximately 50% of the injection sites after 14 days (Fig. 1a ). Compared with an injection of 1 million cells resulting in tumour growth at all injection sites, we propose that tumours formed from a limiting number of cells create a state of ‘isolation stress’ that highlights the contribution of tumour-initiating and/or self-renewing cells. Among the six LPA receptors, only LPAR4 was enriched in isolation-stressed tumours formed by 300 cells (Fig. 1a,b ). Interestingly, LPAR4 expression was significantly higher in early-stage (day 20) lesions versus established (day 35) tumours (Fig. 1c ), indicating that LPAR4 is transiently expressed when tumour cells are experiencing ‘isolation stress’ but downregulated as tumours become established. Compared with cells grown in 2D adherent culture conditions, LPAR4 messenger RNA (mRNA) was highly enriched in the self-renewing cells that form secondary tumourspheres (Fig. 1d ). Thus, LPAR4 is upregulated under in vitro and in vivo conditions that enrich for TICs. Fig. 1: Pancreatic cancer cells selectively upregulate LPAR4 in response to isolation stress. a , log 2 fold change (log 2 FC) for all LPAR s mRNA expression in ‘self-renewing’ tumours formed by subcutaneous injection of 300 cells of Colo-357 (pink dots, n = 6 biologically independent samples), 79E (black dots, n = 4 biologically independent samples) or 34E (blue dots, n = 6 biologically independent samples) with respect to ‘bulk’ tumours formed by injection of 1 million (1 M) cells of Colo-357 ( n = 6 biologically independent samples), 79E ( n = 6 biologically independent samples) or 34E ( n = 5 biologically independent samples). b , Representative images of immunohistochemistry staining of LPAR4 in 79E xenograft tumours derived from 1 M cells ( n = 5 biologically independent samples) or 300 cells ( n = 4 biologically independent samples). Scale bar is 50 µm or 10 µm as indicated. c , Relative LPAR4 mRNA level (normalized to housekeeping genes) in xenograft tumours collected at early timepoint (day 20, early lesions) versus late timepoint (day 35, established tumours). Data are mean ± standard deviation (s.d.) ( n = 3, 4 and 3 biologically independent samples from early lesions for Colo-357, 79E and 34E cells, respectively; n = 4 biologically independent samples from established tumours for all three cell lines). d , All LPARs mRNA expression in cells grown as secondary tumourspheres with respect to cells grown in 10% serum on 2D ( n = 3 independent experiments for each line). e , f , LPAR4 mRNA level in cells treated with various doses of gemcitabine ( e ) or paclitaxel ( f ), relative to cells treated with vehicle control. Data are mean ± s.d. ( n = 3 independent experiments). g , LPAR4 protein expression in cells treated with different stress as indicated. Data are representative of three independent experiments. h , Experimental flow showing the procedure of PDX implantation to the pancreas of NSG mice following by treatment with vehicle or gemcitabine in vivo up to 6 weeks. Bar graph showing relative LPAR4 mRNA expression in PDXs treated with vehicle or gemcitabine. Data are mean ± s.d. ( n = 3 biologically independent samples for vehicle control and n = 5 for gemcitabine-treated group). Bars represent median value per cell line ( a and d ). Statistical analyses were performed using two-tailed unpaired one-sample t -test ( a , c – f and h ). Source numerical data and unprocessed blots are available in source data. Source data Full size image We next considered whether LPAR4 upregulation represents a general mechanism that pancreatic cancer cells use to overcome the effects of cellular stresses encountered during tumour initiation. Whereas the expression of each LPAR in the absence of stress shows heterogeneity among individual pancreatic cancer cell lines (Extended Data Fig. 1a ), subjecting cells to nutrient stress in combination with non-adherent culture conditions selectively upregulated LPAR4 mRNA expression across all cell lines tested (Extended Data Fig. 1b ). LPAR4 mRNA and protein expression was selectively induced in cells exposed to various stresses (Extended Data Fig. 1c ) or sublethal doses of chemotherapy (Fig. 1e–g and Extended Data Fig. 1d,e ). To investigate whether therapeutic stress induces LPAR4 expression in vivo, we orthotopically implanted 1–2 mm 3 patient-derived xenograft (PDX) tumour fragments into the pancreas of NOD.Cg- Prkdc scid Il2rg tm1Wjl /SzJ (NSG) mice. Once each tumour reached 50–100 mm 3 as evaluated by ultrasound, mice received twice-weekly injections of either vehicle or 100 mg kg −1 gemcitabine. After 6 weeks of treatment, LPAR4 expression was significantly higher in tumours from gemcitabine-treated mice compared with the control group (Fig. 1h ). Together, these findings indicate that LPAR4 is a stress-responsive gene in pancreatic cancer cells. LPAR4 is necessary and sufficient to drive tumour initiation For the TNMplot public dataset 18 , LPAR4 gene expression was significantly higher in pancreatic adenocarcinoma (PAAD) than in normal pancreas (Fig. 2a ). While normal pancreas cores were negative, 6 out of 15 PDAC cores showed a relatively high level LPAR4 expression (Extended Data Fig. 2a ). Clinically, analysis of the TCGA PAAD dataset revealed that high LPAR4 mRNA expression significantly correlated with a shorter relapse-free survival (RFS, P = 0.017), with a similar trend for overall survival ( P = 0.18) (Fig. 2b ), indicating that high LPAR4 expression is associated with worse outcomes in patients with PAAD. Fig. 2: LPAR4 expression in patients with pancreatic cancer and its link to tumour initiation. a , TNMplot showing LPAR4 gene expression in PAAD and normal pancreas. Bars represent median value for each group. b , RFS and overall survival for patients with low versus high LPAR4 expression in the TCGA PAAD dataset. c , In a subcutaneous tumour model, TIC frequency (95% interval) between control cells and LPAR4 -expression-manipulated cells was calculated using ELDA software. P values were obtained by Pearson’s chi-squared two-tailed test. d , In an orthotopic tumour model, TIC frequency (95% interval) between Colo-357-sh-CTRL + luciferase cells and Colo-357-sh-R4.1 + luciferase cells was calculated using ELDA software. P values were obtained by Pearson’s chi-squared two-tailed test. e , Effect of ectopic LPAR4 expression on tumoursphere formation for cells grown in methylcellulose medium, plotted as fold change relative to cells expressing the empty vector control. Data are mean ± s.d. ( n = 3, 4 and 5 independent experiments for MiaPaCa2, Colo-357 and 79E cells, respectively). f , Fold change for the number of colonies for each group relative to control (no stress). Data are mean ± s.d. ( n = 5 independent experiments for 79E treated with H 2 O 2 and n = 4 for 79E cells treated with serum deprivation, n = 3 independent experiments for both Colo-357 and 34E cells). g , Median fluorescence intensity (MFI) of MitoSOX signalling in +EV or +R4 cells grown in serum-free medium for 48 h. Data are mean ± s.d. ( n = 3 independent experiments). h , Relative MFI of MitoSOX signalling in 79E cells stably transfected with sh-CTRL, sh-R4.1 or sh-R4.2 cultured in serum-free medium for 48 h. Data are mean ± s.d. ( n = 3 independent experiments). Statistical analyses were performed using two-tailed unpaired one-sample t -test. Source numerical data are available in source data. Source data Full size image To evaluate the biological significance of LPAR4 , we stably knocked down LPAR4 using short hairpin RNA (shRNA) (sh-R4) or ectopically expressed LPAR4 (+R4) in multiple pancreatic cancer cell lines and PDX models (Extended Data Fig. 2b ). Using a limiting dilution assay to gauge the contribution of LPAR4 to subcutaneous tumour initiation in vivo, TIC frequency was reduced by LPAR4 knockdown and enhanced by ectopic LPAR4 expression (Fig. 2c and Supplementary Table 1 ). Using an orthotopic limiting dilution tumour initiation assay with non-invasive bioluminescence imaging, we similarly found that TIC frequency was 90% lower for cells with LPAR4 knockdown (Fig. 2d , Supplementary Table 2 and Extended Data Fig. 2c ). While manipulation of LPAR4 expression had no effect on in vitro growth in the absence of stress (Extended Data Fig. 2d,e ) or tumour growth in vivo (Extended Data Fig. 2f ), ectopic expression of LPAR4 could help cells to overcome the challenge of isolation stress imposed during tumoursphere formation in 3D culture (Fig. 2e ). Similarly, forms of stress that induce LPAR4 expression (hydrogen peroxide and serum deprivation) also enhanced tumoursphere formation in an LPAR4 -dependent manner (Fig. 2f ). As an additional readout for stress tolerance, LPAR4 was both necessary and sufficient to mitigate the accumulation of mitochondrial superoxide (MitoSOX) in cells challenged with serum deprivation stress (Fig. 2g,h ). In contrast, knockdown of LPAR1 , a constitutively expressed LPAR on Colo-357 pancreatic cancer cells (Extended Data Figs. 1a and 2g ), equally impaired cell growth in both 2D and 3D (Extended Data Fig. 2h ), suggesting that LPAR1 plays a more generalized role in cell survival or proliferation. Together, these findings demonstrate that pancreatic cancer cells can overcome the growth limiting conditions during tumour initiation and isolation stress by upregulating LPAR4 . Stress suppresses miR-139-5p to release a brake on LPAR4 miRNAs regulate gene expression allowing reprogramming in response to stress or injury 19 and contribute to the function of stem-like cells in pancreatic 20 and other cancers 21 . To determine whether a miRNA may account for the upregulation of LPAR4 in response to stress, we used multiple miRNA-target prediction tools to generate a list of putative LPAR4-targeting miRNAs and then evaluated how these correspond to survival for the TCGA PAAD dataset. The top hit from this approach was miR-139-5p, for which the median survival of the high cohort (50.1 months) greatly exceeded that for the low cohort (19.9 months) (Supplementary Table 3 and Fig. 3a ). Considering that miR-139-5p had been identified as a tumour suppressor in pancreatic cancer 22 , and since hsa-miR-139-5p was predicted to recognize an 8mer binding sequence on the 3′ untranslated region (UTR) of the LPAR4 promoter (Fig. 3b ), we considered how stress would impact miR-139-5p. Remarkably, multiple stressors upregulated LPAR4 and downregulated miR-139-5p, while another miRNA predicted to bind LPAR4 (miR-138-5p) was not stress responsive (Fig. 3c ). Additionally, self-renewing cells (that is, secondary spheres) that were enriched for LPAR4 mRNA expression showed a significantly lower level of miR-139-5p, but not miR-138-5p, compared with cells grown under 2D conditions (Fig. 3d ). Manipulation of LPAR4 did not alter miR-139-5p levels (Extended Data Fig. 3a ), indicating that miR-139-5p is not a downstream target of LPAR4 . Fig. 3: Stress suppresses miR-139-5p to release the brake on LPAR4 expression. a , Overall survival probability for high versus low miR-139-5p for the TCGA PAAD dataset. b , Eight oligonucleotides are paired between LPAR4 3′ UTR and hsa-miR-139-5p. c , Relative mRNA levels of LPAR4 , miR-139-5p and miR-138-5p in Colo-357 cells treated with serum deprivation stress ( n = 3 independent experiments), hypoxia ( n = 3 independent experiments except n = 2 for miR-138-5p) or oxidative stress ( n = 4 independent experiments), normalized to cells treated with no stress. d , Relative mRNA levels of miR-139-5p and miR-138-5p in secondary methylcellulose spheres grown from Colo-357 or 34E cells as compared with their expressions in cells grown on 2D. e , Relative mRNA level of LPAR4 in Colo-357 ( n = 7 independent experiments) or 79E cell lines ( n = 4 independent experiments) treated with scrambled control miRNA or miR-139-5p mimic in normoxia versus hypoxia. f , LPAR4 protein level in three cell lines treated with scrambled control miRNA or miR-139-5p mimic in normoxia versus hypoxia. Data are representative of three independent experiments. g , The mRNA level of LPAR s and FOXO1 in 34E cells treated with miR-139-5p inhibitor relative to cells treated with scrambled control ( n = 3 independent experiments). h , The relative luciferase values in 34E ( n = 3 independent experiments) or Colo-357 ( n = 4 independent experiments) treated with miR-139-5p mimic or anti-miR-139-5p, along with their scrambled controls. i , Impact of miR-139-5p inhibitor on the mRNA level of LPAR4 and the number of tumourspheres in Colo-357 cells with or without LPAR4 knockdown ( n = 3 independent experiments for tumoursphere assay and n = 4 for qRT–PCR assay). j , Impact of miR-139-5p mimic on the mRNA level of LPAR4 and on the number of tumourspheres in Colo-357 cells with or without LPAR4 ectopic expression ( n = 3 independent experiments for tumoursphere assay and n = 5 for qRT–PCR assay). Data are presented as mean ± s.d. in c – e and g – i , statistical analyses were performed using two-tailed unpaired one-sample t -test ( c – e and g – j ). Source numerical data and unprocessed blots are available in source data. Source data Full size image Cells were transfected with a miR-139-5p mimic to counteract the downregulation of miR-139-5p during stress. The miR-139-5p mimic prevented hypoxia induced expression of LPAR4 (Fig. 3e,f ) while the anti-miR-139-5p inhibitor was sufficient to increase LPAR4 expression in the absence of stress (Fig. 3g and Extended Data Fig. 3b ), as well as a known miR-139-5p target, FOXO1 (refs. 23 , 24 ). Using an LPAR4-3′ UTR luciferase reporter assay, we observed miR-139-5p direct binding to the LPAR4 3′ UTR (Extended Data Fig. 3c ), which was decreased by the miR-139-5p mimic and increased by anti-miR-139-5p (Fig. 3h ). In unstressed cells, LPAR4 expression and tumoursphere formation that was increased by anti-miR-139-5p could be reversed by shRNA-mediated knockdown of LPAR4 (Fig. 3i and Extended Data Fig. 3d ), while ectopic LPAR4 could rescue the inhibitory effect of the miR-139-5p mimic on tumoursphere formation (Fig. 3j ). Thus, the effect of miR-139-5p on tumoursphere formation is primarily due to its regulation of LPAR4 . In support of this, among several gene targets of miR-139-5p in tumour cells, LPAR4 was the most consistently decreased by the miR-139-5p mimic (Extended Data Fig. 3e ). Together, these findings indicate that LPAR4 expression depends on the stress-inducible loss of miR-139-5p expression, revealing the existence of a stress-response pathway that pancreatic cancer cells can use to overcome isolation stress. LPAR4 gene signature includes ECM-related genes To further investigate how pancreatic cancer cells utilize LPAR4 to gain stem-like properties, patient-derived 79E pancreatic cancer cells with stable ectopic expression of LPAR4 (+R4) or control cells were treated with or without the canonical LPAR4 ligand, LPA, and then collected for RNA sequencing (RNA-seq) analysis. As expected, control cells responded to LPA by upregulating a cluster of differentially expressed genes (DEGs) (gene cluster 1) that promote Gene Ontology (GO) terms such as cell proliferation, growth factor activity, cell–cell signalling and signal transduction (Fig. 4a ). To our surprise, a set of genes (gene cluster 3) were differentially expressed in +R4 cells compared with +EV cells and independent of LPA, notably including multiple GO terms related to extracellular matrix (ECM) proteins and organization (Fig. 4a ). The genes in this cluster were validated in four paired cell lines using individual quantitative polymerase chain reaction (qPCR) assays (Fig. 4b and Extended Data Fig. 4a ). Thus, cells that upregulate LPAR4 gain a subset of pro-tumour functions regardless of their exposure to LPA, supporting the notion that LPAR4 allows cells to overcome isolation stress during tumour initiation when there is limited access to LPA derived from vascular, stromal or immune cells. Fig. 4: DEGs common to LPAR4 -expressing cells and patient tumours include ECM-related genes. a , Vector control (+EV) cells and LPAR4 (+R4) cells were treated with or without 1 μM LPA for 2 h. Samples were analysed by RNA-seq and DEseq2 differential expression analysis was performed. Unsupervised clustering analysis reveals unique sets of DEGs associated with LPA treatment and LPAR4 expression. Top GO terms illustrate LPA-induced or LPAR4 -induced effect on gene expression. b , qRT–PCR confirmation of LPAR4-regulated genes associated with ECM in two pairs of +EV versus +R4 cells, grown in charcoal-stripped FBS-containing media. All mRNA level was normalized to +EV cells. c , Overlap analysis was performed to identify DEGs that are commonly upregulated or downregulated for 79E +R4 cells and LPAR4 -high tumours from the TCGA PAAD dataset. Full list of DEGs that overlap between 79E +R4 cells and TCGA R4-high tumours is presented in Supplementary Table 4 . d , Graphs showing half-maximal inhibitory concentration (IC 50 ) values for H 2 O 2 or gemcitabine between 79E +EV and 79E +R4 cells, presented as mean ± s.d. ( n = 3 biologically independent experiments), calculated by GraphPad Prime 9. e , Effect of ectopic LPAR4 expression on tumoursphere formation for cells grown in methylcellulose medium topped with charcoal-stripped FBS-containing medium, plotted as fold change relative to cells expressing the empty vector control. Data are presented as mean ± s.d. ( n = 4 independent experiments for 79E cells and n = 5 for both Colo-357 and MiaPaCa2). Statistical analyses were performed using two-tailed unpaired one-sample t -test ( b , d , e ). Source numerical data are available in source data. Source data Full size image We next performed an overlap analysis to identify DEGs in common between 79E +R4 cells and LPAR4-high patient tumours from the TCGA PAAD dataset (Fig. 4c and Supplementary Table 4 ). Upregulated DEGs included several ECM genes with documented roles as contributors to pancreatic cancer progression, such as fibronectin ( FN1 ) (ref. 25 ). DEGs that were lower in LPAR4 + cells and tumours included the pancreatic lineage marker FOXA3 (ref. 26 ), suggesting that LPAR4 expression was associated with an immature or dedifferentiated phenotype. Notably, the upregulated DEGs were associated with positive hazard ratios (indicating poor RFS) while the downregulated DEGs were associated with negative hazard ratios (Supplementary Table 4 ). All hazard ratios achieved statistical significance ( P < 0.05) except as indicated, indicating the genes identified as part of the LPAR4 gene signature play a role in human pancreatic cancer progression. In line with the observation that the pro-tumour gene signature regulated by LPAR4 does not require exogenous LPA treatment, ectopic expression of LPAR4 allowed cells grown in medium containing either charcoal-stripped serum (to deplete bioactive lipids including LPA) or no serum to tolerate higher levels of stress induced by H 2 O 2 or gemcitabine (Fig. 4d and Extended Data Fig. 4b ). Similarly, exogenous LPA was not required for the ability of ectopic LPAR4 to promote 3D tumoursphere formation in methylcellulose or suspension culture (Fig. 4e and Extended Data Fig. 4c ). Accordingly, expression of the antioxidant gene SOD3 was significantly higher in +R4 cells in the absence of LPA (Extended Data Fig. 4d ). Interestingly, 34E +R4 cells had significantly enriched expression of the pancreatic CSC markers PROM1 , EPCAM and CD44 , while 79E +R4 cells showed higher expression of ALDH1A1 (Extended Data Fig. 4d ), suggesting the phenotype of +R4 cells overlaps to some extent with CSC populations, and this appears to be LPA independent. Together, these findings suggest pancreatic cancer cells that gain LPAR4 can autonomously produce pericellular matrices and acquire properties of TIC. Notably, this appears to occur even when there is limited access to LPA within the surrounding microenvironment, such as one lacking blood vessels, immune cells or stromal cells that serve as sources of LPA in pancreatic cancer 12 , 27 . LPAR4 promotes the cell-autonomous production of FN1 ECM proteins have a profound impact on cancer development and progression 28 . Among the ECM-related genes upregulated by LPAR4 , we were especially interested in FN1 because of its links to cancer stemness and tumour initiation 29 , 30 , 31 , biological features that are also induced by LPAR4 . In fact, there was a positive correlation between LPAR4 and FN1 mRNA expression for the TCGA PAAD dataset ( P < 0.0001) (Fig. 5a ) and orthotopic PDX tumours ( P = 0.0180), while gemcitabine-treated PDX tumours expressed high levels of both LPAR4 and FN1 (Fig. 5b ). As observed for LPAR4 , FN1 expression was also significantly higher in PAAD tumours than in normal pancreas (Extended Data Fig. 5a ). Importantly, analysis of the TCGA PAAD dataset revealed that high FN1 expression significantly correlated with lower overall survival and RFS (both P < 0.05) (Fig. 5c ). Fig. 5: LPAR4 expression promotes the cell-autonomous production of FN1. a , Expression correlation between LPAR4 and FN1 from the TCGA PAAD dataset ( n = 177 samples). Pearson correlation coefficient r = 0.63 and two-tailed P < 0.0001. b , Expression correlation between LPAR4 and FN1 from the PDX tumours from the orthotopic mouse models treated with vehicle control ( n = 3 biologically independent samples) or gemcitabine ( n = 5 biologically independent samples). Pearson correlation coefficient r = 0.80 and two-tailed P = 0.018. c , Overall survival and RFS for patients with low versus high FN1 expression in the TCGA PAAD dataset. d , Representative images for human FN1 immunohistochemistry staining in 79E ( n = 10 biologically independent samples), Colo-357 ( n = 4 biologically independent samples) or 34E ( n = 5 biologically independent samples) xenograft tumours formed from 1 million (1 M) cells injected or 300 cells injected. Scale bar is 50 µm. e , f , Western blot and immunostaining showing FN1 protein expression for pancreatic cancer cells with stable ectopic expression of LPAR4 (+R4) versus empty vector control (+EV) grown in charcoal-stripped FBS-containing medium for 72 h. Data are representative of three biological experiments. g , Representative images for human FN1 immunohistochemistry staining in Colo-357 +EV versus Colo-357 +R4 tumours collected at day 15 post implantation in the pancreas of nu/nu mice. Images in the bottom panel highlight FN1 staining in matrix areas. Data are representative of three biological samples. Scale bar is 100 µm or 20 µm as indicated. h , Effects of hypoxia and LPAR4 knockdown on FN1 protein expression. Data are representative of three biological experiments. Source numerical data and unprocessed blots are available in source data. Source data Full size image FN1 has up to 20 distinct isoforms due to alternative splicing, and isoforms that contain the extra domain A (EDA) or extra domain B (EDB) regions appear to be more tumourigenic 32 . Thus, we asked whether LPAR4-induced FN1 mRNA contains EDA, or EDB, or both domains. As shown in Extended Data Fig. 5b , +R4 cells expressed comparable mRNA levels of FN1 , FN1-EDA and FN1-EDB , suggesting LPAR4-induced FN1 probably contains both the pro-tumour EDA and EDB domains. As shown for LPAR4 (Fig. 1a ), FN1 expression was higher in isolation-stressed tumours formed from the injection of 300 cells compared with those formed by 1 million cells (Fig. 5d ). Consistent with the LPAR4 gene signature, +R4 cells showed higher levels of FN1 protein in the absence of exogeneous LPA (Fig. 5e,f ). In an orthotopic tumour model, there was more abundant FN1 staining in the matrix surrounding Colo-357 +R4 tumour cells than for Colo-357 +EV tumour cells (Fig. 5g ), providing evidence of LPAR4-induced tumour-intrinsic FN1 production and deposition in vivo. Furthermore, cells exposed to hypoxia showed increased FN1, and this was prevented by LPAR4 knockdown (Fig. 5h and Extended Data Fig. 5c ). These findings indicate that LPAR4 expression changes the phenotype of pancreatic cancer cells, providing them with a cell-autonomous source of pro-tumour FN1, creating a TIC niche to support survival during the earliest stages of tumour formation. FN1 is indispensable for LPAR4-induced TIC properties Next, we asked what signalling event(s) activated by LPAR4 drive FN1 expression. cAMP response element-binding protein (CREB) is a known stress-responsive transcription factor that can bind to the FN1 gene exon 1 (refs. 33 , 34 ). Indeed, LPAR4 knockdown prevented hypoxia-induced FN1 protein expression and CREB phosphorylation at serine 133, an indicator of CREB activity 31 (Fig. 6a ). In addition, all +R4 cells showed higher CREB activity in the absence of exogenous LPA (Fig. 6b ), suggesting LPAR4 was sufficient to activate CREB. Importantly, treating +R4 cells with CREB short interfering RNA (siRNA) or CREB inhibitor 666-15 resulted in a dramatic decrease of FN1 protein expression (Fig. 6c,d ). Using a CREB chromatin immunoprecipitation (ChIP)–qPCR assay, we detect significantly enriched CREB occupancy on FN1 exon 1 in +R4 cells, relative to +EV cells, in the absence of exogenous LPA (Fig. 6e ). Together, these results demonstrate that LPAR4-mediated activation of CREB directly drives the transcription of FN1 in pancreatic cancer cells. Fig. 6: FN1, induced by the LPAR4/AKT/CREB signalling, is indispensable for LPAR4 -induced TIC properties. a , Effects of hypoxia and LPAR4 knockdown on protein expressions of FN1, phosphorylated CREB at serine133 (p-CREB-S133) and CREB. b , Protein levels of p-CREB-S133, CREB, p-AKT-S473, AKT and vinculin in +EV versus +R4 cells, cultured in charcoal-stripped FBS-containing medium for 72 h. c , d , Effects of CREB inhibition by using siRNA knockdown or 666-15 on the protein levels of FN1, p-CREB-S133 and CREB in 34E +EV versus 34E +R4 cells cultured in charcoal-stripped FBS-containing medium for 72 h. e , ChIP–qPCR assay showing CREB binding occupancy on FN1 exon 1 in +EV versus +R4 cells cultured in charcoal-stripped FBS-containing medium and IgG antibody serves as a negative control. The ChIP–qPCR signal was normalized to 2% input. f , Effect of AKT knockdown by using siRNA on FN1 protein expression in +EV versus +R4 cells cultured in charcoal-stripped FBS-containing medium. g , Effects of AKT inhibition by ipatasertib for 72 h on the protein levels of FN1, AKT, p-GSK-3β-S9, GSK-3β, p-CREB-S133 and CREB in 34E +EV versus 34E +R4 cells cultured in charcoal-stripped FBS-containing medium. h , Relative number of tumourspheres formed by cells transfected with scrambled siRNA (si-CTRL) or FN1 siRNA (si-FN1) in methylcellulose media topped with 10% charcoal-stripped FBS-containing medium (CS-FBS). All numbers were normalized to EV cells transfected with si-CTRL. i , Median fluorescence intensity (MFI) of MitoSOX in +EV and +R4 cells transfected with si-CTRL or si-FN1, grown in serum-free medium for 48 h. j , Relative number of tumourspheres formed by cells transfected with empty vector (EV) or FN1-EDA in methylcellulose medium topped with 10% FBS-containing medium. Immunoblots represent two independent experiments for FN1 protein expression in cells transfected with EV or FN1-EDA at day 3. Western blots shown in a – d , f and g represent three biologically independent experiments. Data are presented as mean ± s.d. in e and h – j ( n = 3 independent experiments for e and j ; n = 5 for h ; n = 4 for Colo-357 cells and n = 3 for 34E cells for i ). Statistical analyses were performed using two-tailed unpaired one-sample t -test ( e and h – j ). Source numerical data and unprocessed blots are available in source data. Source data Full size image Using gene set enrichment analysis to consider kinases known to activate CREB 35 , we found that AKT signalling was upregulated in +R4 cells in the absence of LPA (Extended Data Fig. 6a ). Consistently, all +R4 cells showed relatively high AKT activity, evidenced by p-AKT-S473 (Fig. 6b ), and the AKT inhibitor ipatasertib decreased p-CREB-S133 in a dose/time-dependent manner (Extended Data Fig. 6b ), supporting the notion that CREB is a bona fide substrate of AKT in our cells. Furthermore, blockade of AKT by either siRNA-mediated knockdown or ipatasertib decreased FN1 in +R4 cells (Fig. 6f,g ). In this case, p-GSK-3β-S9 inhibition served as a surrogate indicator for the blockade of AKT kinase activity (Fig. 6g ). As shown in Extended Data Fig. 6c , CREB knockdown resulted in a significant downregulation of several ECM-related genes within the LPAR4 gene signature, including FN1 and versican ( VCAN ), in both +R4 cell lines. Together, our data demonstrate that stress-induced LPAR4 activates AKT, leading to CREB-mediated transcription of FN1 . Not only has fibronectin been linked to cancer progression 32 , but it is critical for the creation of a niche that directs stem cells through different fate changes 36 , 37 . FN1 functions as a scaffold for the deposition of other matrix proteins and anchoring of soluble factors, allowing it to promote angiogenesis, metastasis, chemoresistance and immune evasion in pancreatic cancer 25 , 32 , 38 . Given the abundant FN1 expression in +R4 cells, we asked whether this could account for their phenotype. Indeed, FN1 knockdown significantly reduced LPAR4 -induced 3D tumoursphere formation and growth advantage in the presence, but not absence, of stress (Fig. 6h and Extended Data Fig. 6d ) and significantly increased MitoSOX levels in +R4, but not +EV, cells grown under serum-free conditions (Fig. 6i and Extended Data Fig. 6f ). In addition, ectopic expression of FN1-EDA significantly induced 3D tumoursphere formation (Fig. 6j ). Together, our data indicate that that FN1 is required for LPAR4-dependent stress tolerance and can account for the LPAR4-induced phenotype. Considering that LPAR4 supports cell-autonomous growth under stressed conditions, we asked whether the FN1-containing ECM produced by LPAR4-expressing pancreatic cancer cells could transfer this advantage to cells lacking LPAR4. To do this, +EV or +R4 cells were grown on tissue culture plastic for 72 h to allow matrix deposition, and then removed to create a cell-free ECM (Fig. 7a ). When LPAR4-negative cells were plated in serum-free growth conditions onto ECM deposited by +R4 cells (ECM-R4), they grew significantly faster than when plated on ECM produced by +EV cells (ECM-EV) (Fig. 7b ). Of note, cells plated onto ECM-R4, but not ECM-EV, began to form colonies after several days (Fig. 7c and Extended Data Fig. 7 ), further suggesting that deposited matrix can transfer cell growth advantages from LPAR4-expressing cells to LPAR4-negative cells. Furthermore, LPAR4-negative cells grown on ECM-R4 became more stress tolerant to H 2 O 2 or gemcitabine, and this was negated by FN1 knockdown (Fig. 7d ). Fig. 7: ECM deposited by LPAR4 -positive cells endows stemness features to LPAR4 -negative cells in an FN1-dependent manner. a , Experimental flow showing that +EV or +R4 cells were grown on tissue culture plastic for 72 h in serum-free medium before cells were removed, and LPAR4 -negative cells were plated atop the residual matrix in serum-free medium. LPAR4 -negative cells plated on uncoated tissue culture plate were used as baseline controls in this assay. b , c , Graph showing the number of viable cells, relative to baseline controls, assessed by the CellTiter-Glo assay ( b ). Data are presented as mean ± s.d. for n = 3 independent experiments for Colo-357 and n = 4 independent experiments for 34E cells. c represents three independent experiments for pictures of colonies formed at day 12. d , e , Graphs showing half-maximal inhibitory concentration (IC 50 ) values for 79E cells grown on the matrix deposited by 79E +EV cells or 79E +R4 cells with versus without FN1 knockdown as indicated and then exposed to various doses of H 2 O 2 or gemcitabine, or in combination with 5 µg ml −1 anti-α5 antibody P1D6, or with 5 µg ml −1 anti-αVβ3 antibody LM609. Cell viability was assessed by the CellTiter Glo assay, and IC 50 values were calculated by GraphPad Prism 9. Data are presented as mean ± s.d. for n = 4 independent experiments for d and n = 3 independent experiments for e . f , Diagram showing 300 cells transfected with either scrambled siRNA (si-CTRL) or FN1 siRNA (si-FN1) injected in a nu/nu mouse subcutaneously at four different sites. Graph showing the tumour take rate per mouse ( n = 5) for the four treatment groups analysed at day 20. g , Schematic model of LPAR4 in pancreatic cancer. ‘Isolation stress’ (1) downregulates the expression of miR-139-5p (2), which releases a brake on LPAR4 expression (3 and 4). Even in the absence of microenvironmental LPA, LPAR4 is sufficient to activate AKT/CREB signalling (5), driving the expression of ECM-related genes, including FN1 (6). The deposited FN1 matrix protein ligates with its cell surface integrins α5β1 and/or αVβ3 (7), endowing cells with several self-sufficient traits, including 3D growth and stress tolerance, ultimately promoting tumour initiation and cancer stemness (8). Statistical analyses were performed using two-tailed unpaired one-sample t -test ( b and d – f ). Source numerical data are available in source data. Source data Full size image Finally, we show the ability of ECM-R4 to provide stress tolerance could be eliminated by interfering with fibronectin binding to its cell surface receptors, integrins α5β1 or αvβ3 (Fig. 7e ). Thus, FN1 is a critical effector of LPAR4 that supports LPAR4-induced self-sufficiency, and the ability to transfer this to LPAR4-negative cells requires the function of integrin α5β1 and/or αvβ3. Lastly, to further substantiate the significance of FN1 for the LPAR4-induced tumour initiation advantage, we knocked down FN1 in both +EV and +R4 cells and performed a tumour-initiating assay by injecting 300 cells in mice. As shown in Fig. 7f , FN1 knockdown in +R4 cells, but not +EV cells, resulted in a significantly decreased tumour take rate in vivo, indicating that FN1 was required for LPAR4-induced tumour initiation. Together, a variety of in vitro and in vivo assays consistently indicate that FN1 is a critical effector of LPAR4 in pancreatic cancer cells. In summary, pancreatic cancer cells can overcome isolation stress through the stress-induced suppression of a miRNA that releases a brake on LPAR4 expression, thus activating an AKT/CREB axis allowing a tumour cell to generate a fibronectin-containing ECM niche, and thereby promoting a self-sufficient phenotype (Fig. 7g ). Discussion Evidence is provided that LPAR4 acts as an adaptive response to stress that pancreatic cancer cells exploit to overcome solitary growth conditions encountered during tumour initiation. Established pancreatic tumours feature a dense stroma that supports growth and survival for pancreatic cancer cells in the primary tumour environment by providing cell–cell and cell–matrix adhesion, as well as secreted factors including cytokines, growth factors and ECM 39 . We propose that, at the earliest stages of tumour development or metastatic seeding, solitary tumour cells adapt to isolation stress by downregulating miR-139-5p, which releases the brake on LPAR4 expression. Aside from LPAR4 , we did not find any other previously reported miR-139-5p target genes 40 , 41 , 42 to be stress dependent in our pancreatic cancer cells. This may be because IGF1R and CCNB1 are highly expressed in unstressed conditions, unlike LPAR4 . While IGF1R and CCNB1 were approximately twofold downregulated by the miR-139-5p mimic in 34E cells, this was not the case for Colo-357 cells (Extended Data Fig. 3f ), suggesting potential context-dependent interactions. Upregulating LPAR4 in response to stress allows solitary pancreatic cancer cells to create their own tumour-initiating niche by producing ECM proteins. Fibronectin within this matrix, via integrin ligation, supports a self-sufficient state to enable tumour initiation and can transfer this advantage to nearby LPAR4-negative cells (Fig. 7g ). Importantly, we show that blocking the ability of cells to utilize this fibronectin matrix with integrin antagonists can reverse the stress tolerance benefit of LPAR4 expression. Considering that tumours that develop drug resistance via upregulation of LPAR4 show enhanced stress tolerance and tumour initiation, targeting the LPAR4/AKT/CREB pathway or disrupting the integrin/FN1 interaction might not only reduce tumour drug resistance, but could ultimately delay disease progression by suppressing tumour initiation at metastatic sites. Our gene expression analysis highlights functions for LPAR4 that do not require LPA, suggesting that this mechanism to overcome stress represents a non-canonical role for an LPA receptor that is ligand independent and stress inducible. In ‘unstressed’ states, cells do not require LPAR4–FN1–integrin signalling when sufficient growth support can be obtained from other cell–cell and cell–matrix interactions, along with the growth factors, cytokines, soluble FN1 and vitronectin provided by foetal bovine serum (FBS). The fact that LPAR4 expression is undetectable in cells grown in such states may explain why LPAR4 has not yet been appreciated as a stress-inducible contributor to pancreatic cancer. In a more general context, our work highlights an opportunity to explore additional functions for cell surface proteins with defined roles as ‘receptors’. LPAR4 might represent a larger family of proteins that can be hijacked by tumour cells to overcome the unfavourable growth conditions encountered during phases of tumour progression that require a self-sufficient phenotype. For example, our previous studies revealed that integrin αvβ3 functions in a non-ligated state to promote stemness, tumour initiation and drug resistance in cancer cells 43 , 44 , 45 , 46 . Ultimately, therapeutic agents that block the function of LPAR4 and other receptors enriched on TIC may be overlooked when evaluated by their capacity to halt the growth of well-established primary tumours. Instead, such agents may have an alternative use to limit therapeutic resistance, disease relapse or metastasis by preventing tumour cells from gaining the self-sufficient phenotype that tumour cells exploit to achieve tumour initiation and spread to distant sites. Methods All experiments conform to the relevant regulatory standards and animal studies were conducted under protocols S05018 and S09158 , approved by the University of California San Diego Institutional Animal Care protocol and Use Committee in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Reagents, chemicals and antibodies LPA (18:1) (Avanti Polar Lipids, #857130) was diluted to 5 mM in dH 2 O, sonicated at 37 °C for 15 min, and used fresh. The CellTiter-Glo Kit (Promega, #G7573) was used to measure cell viability. Gemcitabine HCI (#S1149), paclitaxel (NSC 125973) and hydrogen peroxide (H 2 O 2 ) were purchased from Selleck Chem and Thermo Fisher Scientific (H325-100), respectively. Charcoal:dextran-stripped FBS was purchased from GeminiBio (#100-119). Ipatasertib (GDC-0068, HY-15186) and 666-15 (HY-101120) were purchased from MedChemExpress. Primary antibodies used in this study included: LPAR4 (Proteintech, #22165-1-AP, 1:1,000 for western blots), LPAR4 (Thermo Fisher, #PA5-49727, 1:50 for immunohistochemistry), vinculin (Santa Cruz Biotechnology, #sc-25336, H-10, 1:5,000 for western blots), GAPDH (CST, #5174, D16H11, 1:3,000 for western blots), β-actin (CST, #8457, D6A8, 1:5,000 for western blots), fibronectin (CST, #26836, E5H6X, 1:1,000 for western blots and 1:200 for immunohistochemistry and immunofluorescence), phosphor-AKT-Ser473 (CST, #9271, 1:1,000 for western blots), AKT (CST, #9272, 1:1,000 for western blots), phosphor-CREB-Ser133 (CST, #9198, 87G3, 1:1,000 for western blots), CREB (CST, #9197, 48H2, 1:1,000 for western blots), Phospho-GSK-3β-Ser9 (CST, #5558, D85E12, 1:1,000 for western blots), GSK-3β (CST, #12456, D5C5Z, 1:1,000 for western blots) and anti-α5 integrin antibody (Sigma-Aldrich, MAB1956Z, P1D6). Anti-αVβ3 integrin antibody (LM609) was produced as previously described 47 . Secondary antibodies for western blots include anti-rabbit IgG (CST, #7074, 1:3,000 for western blots) and anti-mouse IgG (CST, #7076, 1:3,000 for western blots). Cell lines PDAC cell lines (XPA1 and MiaPACA-2) were obtained from the American Type Culture Collection. 79E and 34E cells were derived from PDX models established by Dr Andrew Lowy (University of California, San Diego). The fast-growing variant of the pancreatic carcinoma cell line Colo-357 was a gift from Dr Shama Kajiji and Vito Quaranta (The Scripps Research Institute). Cells were grown in Dulbecco’s modified Eagle medium containing 10% FBS (Gibco) and 1% penicillin–streptomycin (Thermo Fisher Scientific) and cryopreserved as low-passage stocks. All cell lines were regularly tested for mycoplasma contamination using PCR mycoplasma detection kit (Genlantis). All cell lines tested mycoplasma negative. Genetic knockdown and ectopic expression Cells were transfected with vector control or LPAR4 using a lentiviral system. Lentiviruses were produced in 293T cells co-transfected with lentiviral backbone constructs and packing vectors (ps-PAX2 and VSVG) using Lipofectamine 3000 (Thermo Fisher Scientific). For knockdown experiments, cells were transfected with siRNA (Sigma-Aldrich) using the Lipofectamine RNAiMAX (Thermo Fisher Scientific) or with shRNA (Sigma-Aldrich) using a lentiviral system. Two siRNAs or shRNAs targeting different regions of the gene of interest were applied for the gene knockdown. Ectopic expression and knockdown effects were confirmed by quantitative reverse transcription polymerase chain reaction (qRT–PCR). miRNA inhibitor (Millipore Sigma) or mimic (Millipore Sigma) transfections were conducted using Lipofectamine RNAiMAX. Cells were collected 48 h post transfection for RNA isolation and qPCR analysis or for methylcellulose tumoursphere formation assay. Supplementary Table 5 presents details for constructs, siRNAs and miRNA reagents. Application of cellular stress in vitro Hypoxia: cells at ~50% confluence were grown in a 1% O 2 hypoxia chamber (Coylab) or 21% O 2 normoxia chamber (regular tissue culture chamber) on 2D monolayer culture for 72 h. Oxidative stress: 5 × 10 5 cells were seeded in a 12-well ultralow-attachment plate (Corning, 3D) and treated with different doses of H 2 O 2 for 48 h. Serum deprivation: 5 × 10 5 cells were seeded on 2D (6-well plate) or 12-well ultralow-attachment plate (Corning, 3D) in the presence of 10% regular FBS or 0% FBS-supplemented medium for 48 h. Medium was substituted with 10% regular FBS-supplemented medium for both experimental and control cells for 2 h before cell collection. Chemotherapy stress: 5 × 10 5 cells were seeded in 12-well ultralow-attachment plate a day before treatment of various doses of gemcitabine or paclitaxel for 24 h. Subcutaneous and orthotopic tumour studies Experiments were conducted under protocol S05018 approved by the UC San Diego Institutional Animal Care and Use Committee (IACUC) in accordance with the NIH Guide for the Care and Use of Laboratory Animals. All experiments utilized 6-to-8-week-old female immune-compromised nu/nu mice (Charles River Labs). Mice were kept on a 12 h light/dark cycle, 7:00 to 18:00 at ambient room temperature of ~22 °C, and at humidity of 40–60%, monitored through the building control management system. The maximal tumour volume permitted by IACUC is 2 × 10 3 mm 3 and was not exceeded. A limiting dilution assay was used to estimate the frequency of TICs. Briefly, 1 million, 100,000, 10,000, 1,000, 100 or 10 cells were suspended in a 1:1 mixture of Hank’s Balanced Salt Solution (HBSS) and Phenol Red-free Basement Membrane Matrix (BD Biosciences) and injected subcutaneously. Mice were examined twice weekly for palpable tumours. The results were tabulated as the number of tumours observed (at the specified endpoint) per the number of tumours injected, and used to calculate the frequency of TICs using ELDA software 48 . On the basis of the limiting dilution experiments, the tumour take rate was about 50% for a 300-cell injection and 100% for 1 million cells. As such, we reasoned that tumour formed after injection of 300 cells are enriched for cancer stem or self-renewing cells. Mice received a subcutaneous injection of 300 or 1 million tumour cells to the flank and were checked twice weekly for palpable tumours. Once a tumour was observed, the tumour was collected, cut bluntly into pieces and ground on ice using a glass tissue grinder with 20% Binding Buffer from the High Pure miRNA Isolation Kit (Roche) for tissue homogenization. Supernatants collected after tissue homogenization were utilized for total RNA isolation using the High Pure miRNA Isolation Kit (Roche), following the manufacturer’s protocol. For 300 cells injection xenograft model, +EV or +R4 cells were transfected with either FN1 siRNA or scrambled siRNA for 48 h before collection, then 300 cells mixed with Phenol Red-free Basement Membrane Matrix (1:1 ratio) were subcutaneously injected into mice. At day 20, the number of palpable tumours formed was counted. For tumour initiation in an orthotopic model, Colo-357-sh-CTRL or sh-R4.1 cells were firstly transduced with luciferase lentivirus. Cells with equivalent levels of luciferase activity were implanted into the pancreas of nu/nu mice. Tumour initiation in the pancreas of mice was monitored twice a week by using non-invasive bioluminescence imaging. Note that all mice were imaged 10 min after injected with d -luciferin (L9504, Sigma-Aldrich). The frequency of TICs was analysed using ELDA software 48 . Pancreatic cancer PDX study The study involving patient samples followed UCSD institutional review board-approved protocol (institutional review board number 181755), and all patients were consented and not compensated. The patients’ information (sex and age) was blinded to protect patient confidentiality. The PDX mouse study was conducted under protocol S09158 approved by the UC San Diego IACUC in accordance with the NIH Guide for the Care and Use of Laboratory Animals. Briefly, 1–2 mm 3 PDX fragments ( n = 8 patient tumour samples) were implanted to pancreas of NSG mice. Tumour growth was monitored by weekly ultrasound imaging. Once a tumour size reached 50–100 mm 3 , the mouse was recruited randomly and subsequently treated with vehicle or 100 mg kg −1 gemcitabine (intraperitoneal injection) twice a week for up to 6 weeks. Once the control tumour group reached a diameter of 2.0 cm 3 , all mice were sacrificed for tumour collection and qPCR analysis. qRT–PCR and RNA-seq RNA was collected using the RNeasy RNA Purification kit (Qiagen). Cells were collected at the indicated timepoints and conditions. Adherent (2D) cells were washed twice with 1× HBSS (Gibco, #14025076) and scraped off in the presence of Buffer RLT (Qiagen). Cells grown in ultralow-attachment plates (Corning, #3473) were collected through centrifugation (300 g , 5 min) and washed twice with 1× HBSS before lysing with Buffer RLT. Complementary DNA was synthesized by using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific), and RT–PCR was performed on a LightCycler with SYBR Green (Bio-Rad). Expression of each target gene is normalized by comparing the target Ct value to the geometric mean of the Ct for the five housekeeping genes, RPL37A , ACTB , TUBB , VCL and GAPDH . Primer sequences are listed in Supplementary Table 6 . For RNA-seq experiments, RNA was extracted using the RNeasy RNA Purification Kit (Qiagen) following the manufacturer’s instructions. A total of 1 µg RNA per sample was used to generate RNA-seq libraries using NEBNext Ultra RNA library prep kit (NEB) following the manufacturer’s protocol. PCR products were purified (AMPure XP system) and library quality was assessed on the Agilent Bioanalyzer 2100 system. The clustering of the index-coded samples was performed on Illumina Novaseq sequencer according to the manufacturer’s protocol. Subsequently, the RNA libraries were sequenced on Illumina Novaseq machine and paired-end reads were generated. The Python package HTSeq was used to generate read counts for each gene, which were analysed using the R package DESeq2 (v.1.28.0). Benjamini and Hochberg correction was used to calculate adjusted P values (p.adj) for DEGs. A p.adj <0.01 was required to consider genes as differentially expressed ones. Heat maps were generated using Java Treeview (v.1.1.6r4). Total RNA (including small RNA) was isolated using the High Pure miRNA Isolation Kit (Roche), following the manufacturer’s protocol. cDNA for miRNA was synthesized and the stem-loop RT–PCR method was applied to assess the expression of miRNA of interest following the protocol described in detail by Xie et al. 49 . PCR cycles for amplifying all miRNA were: stage 1: 94 °C, 3 min, 1 cycle; stage 2: 95 °C, 15 s; 60 °C, 45 s; 45 cycles; stage 3: dissociation analysis. Expression of the target miRNA is normalized by comparing the target Ct value to the geometric mean of the Ct for the two housekeeping miRNAs, U6 and miR-16-5p. Primer sequences for miRNA expression are listed in Supplementary Table 6 . LPAR4-3′ UTR luciferase reporter assay The LPAR4-3′ UTR luciferase reporter plasmid (HmiT099310-MT05) and control plasmid (CmiT000001-MT05) were purchased from GeneCopoeia. Briefly, cells were transfected with LPAR4-3′ UTR reporter plasmid or control plasmid using Lipofectamine 3000 (Thermo Fisher Scientific) at day 1, and at day 2 cells were subsequently transfected with 100 nM miR-139-mimic, or anti-miR, or scrambled control mimic, respectively, using lipofectamine RNAiMAX (Thermo Fisher Scientific). At day 4, Gaussia luciferase and secreted alkaline phosphatase luciferase activities for each sample were accessed separately by using Secrete-Pair Dual Luminescence Assay Kit (LF031, GeneCopoeia). ChIP–qPCR assay The ChIP kit was purchased from Cell Signaling (SimpleChIP Plus Chromatin IP Kit, #9005), and the experiments were performed according to the protocol provided by Cell Signaling Technology. Briefly, cells grew in charcoal-stripped FBS-containing medium up to 90% confluence before crosslink using 37% formaldehyde for 10 min at room temperature. Nuclei/chromatin was digested and sonicated after three sets of 20 s pulse and three sets of 30 s on wet ice between pulses (Branson Digital Sonifier 450). Chromatin lysates were incubated with anti-CREB antibody (#4820, Cell Signaling) or anti-IgG (#2729, Cell Signaling) at 4 °C with rotation overnight. The eluted DNA products were subsequently utilized for real-time qPCR assay. Primers for detecting FN1 exon 1 are available in Supplementary Table 6 . The details of the PCR reaction programme are: initial denaturation: 95 °C, 3 min; denature: 95 °C, 15 s; anneal and extension: 60 °C, 60 s; repeat step denature and anneal and extension for a total of 40 cycles. Western blots Western blots were performed as described previously 43 . Briefly, cells were collected at the mentioned timepoints and conditions. Adherent (2D) cells were washed three times with 1× HBSS and scraped off in the presence of 1× RIPA buffer containing protease and phosphatase inhibitors, or 2× Laemmli sample buffer containing reducing agent. Cells grown in suspension (3D) were collected through centrifugation (300 g , 5 min) and washed twice with 1× HBSS before lysing with cell lysis buffer mentioned above. BCA assay (Thermo Fisher Scientific) was performed, and lysates were normalized. Sample buffer containing reducing reagent were added to the 1× RIPA buffer lysate and then heated at 95 °C for 5 min. Twenty micrograms of protein was loaded on an SDS–PAGE gel. Blocking and probing were done in 5% fat-free milk + TBS-T buffer. ECL reagent (Thermo Fisher Scientific) was used to visualize protein bands. Methylcellulose tumoursphere forming assay Methylcellulose tumoursphere forming assay was performed following the manufacturer’s instructions. Briefly, cells were washed with 1× HBSS and centrifuged at 300 g for 5 min, and cell number was counted by Trypan blue exclusion assay. Briefly, 4,000 cells were mixed in 1 ml methylcellulose stock medium (HSC001, R&D Systems) in 24-well non-treated plate (Corning), topped by 1 ml 10% regular FBS- or 10% charcoal-stripped FBS-supplemented medium. After 12 days, one lower magnification image of each well plus three random fields with higher magnification within each well was acquired using an AmScope microscope (MU1603). The number and percentage area of tumourspheres per field was computed using ImageJ (NIH). Flow cytometry detecting MitoSOX MitoSOX Red (Thermo Fisher, M36008 ) staining was carried out according to the manufacturer’s instructions. Briefly, 5 × 10 5 cells were seeded on six-well tissue culture plate in serum-free medium for 48 h. Cells were washed with 1× HBSS twice and were disassociated with 0.5% trypsin buffer, then washed with 1× HBSS and centrifuged with 300 g for 5 min twice. Subsequently, cells were stained with 5 µM MitoSOX reagent for 10 min at 37 °C, protected from light, before flow cytometry analysis. An example of gating strategy for this assay is provided in Supplementary Information. Data were analysed with FlowJo.v10. Immunohistochemistry staining Immunohistochemistry staining of LPAR4 or FN1 in human PDAC tissue array (US Biomax PA484a) or formalin-fixed, paraffin-embedded xenograft tumours were performed using the LPAR4 antibody (Thermo Fisher, #PA5-49727) or FN1 antibody (CST, #26836, E5H6X) following the manufacturer’s protocol. Slides were later imaged on a NanoZoomer Slide Scanning System (Hamamatsu). The LPAR4 antibody was validated for immunohistochemistry application by the manufacturer, as well as in-house validation by staining orthotopic xenograft tumours generated by 79E +EV versus 79E +R4 cells (Extended Data Fig. 8 ). Immunofluorescence staining Briefly, cells were grown on eight-well chamber slide (Nunc Lab-Tek chamber slide, Thermo Scientific) and fixed with 4% formaldehyde for 15 min at room temperature. Cells were then incubated with primary antibodies at a dilution of 1:1,000 (anti-FN1) for overnight at 4 °C, following by secondary antibody at a dilution of 1:2,000 for 1 h and DAPI staining for 5 min at room temperature. Images were acquired by Nikon Eclipse C1 confocal microscope and analysed with NIS-Elements Viewer 5.21. Production and testing of cell-free ECM +EV or +R4 cells were seeded confluently on 6- or 96-well tissue culture plate (Corning) in serum-free medium for 72 h to deposit ECM. Cells were thereafter removed with 20% ammonium hydroxide solution following an established protocol 50 . For the cell growth assay, 1 × 10 5 cells were seeded on six-well plate pre-coated with ECM derived from +EV (ECM-EV) or +R4 (ECM-R4) cells in the absence of serum up to 12 days. For H 2 O 2 and gemcitabine resistance assay, 2 × 10 4 cells were seeded on 96-well plate pre-coated with ECM-EV, ECM-R4 or ECM-R4 with FN1 knockdown in the absence of serum. The next day, cells were treated with various doses of H 2 O 2 or gemcitabine, or in combination with antibody P1D6 (5 µg ml −1 ) or antibody LM609 (5 µg ml −1 ) up to 72 h. The number of viable cells was assessed by the CellTiter-Glo Kit (Promega). Statistical analysis Fisher’s exact tests, one-way analysis of variance tests or Student t -tests were performed using Prism (GraphPad 9.4.1) and Microsoft Excel 365. P < 0.05 was considered significant. Data distribution was assumed to be normal, but this was not formally tested. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications 51 , 52 . No randomization was used except all the mice for the in vivo study were allocated randomized. Single blinding was performed for all in vivo assays and data analysis, as well as for RNA-seq data analysis. Data collection and analysis were not performed blind to the in vitro experiments. No samples, mice or data points were excluded for analysis. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability RNA-seq data that support the findings of this study have been deposited in the Gene Expressing Omnibus (GEO) under accession code GSE198002 . The human PAAD data were derived from the TCGA Research Network ( and ). The TNMplot public dataset that supports the findings of this study is available at or in source data files. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Source data are provided with this paper.

Researchers at University of California San Diego School of Medicine have discovered a molecular pathway critical to the initiation of pancreatic tumors. The mechanism could also contribute to the disease's high resistance to chemotherapy and its propensity for metastasis. The study, published on January 16, 2023, in Nature Cell Biology, found that pancreatic tumor-initiating cells must first overcome local "isolation stress" by creating their own tumor-promoting microenvironment, and then recruit surrounding cells into this network. By targeting this tumor-initiating pathway, new therapeutics could limit the progression, relapse and spread of pancreatic cancer. Pancreatic cancer is one of the most lethal cancers, notoriously resistant to treatment. Almost all patients experience cancer recurrence or metastasis. In the early stages of tumor formation, cancer cells (those with cancerous mutations, called oncogenes) experience a loss of adhesion to other cells and the extracellular matrix—the web of macromolecules that encase and support all cells. This isolation leads to a local lack of oxygen and nutrients. Most cells do not survive such isolation stress, but a certain group of cells can. Tumor-initiating cells (TIC) play a major role in the formation, recurrence and metastatic spread of tumors. What sets them apart from other cancer cells is their resilience to these early substandard conditions. Like cacti in a desert, they can adapt to the harsh environment and set the scene for further tumor progression. "Our goal was to understand what special properties these tumor-initiating cells have and whether we can control the growth and spread of cancer by disrupting them," said senior study author David Cheresh, Ph.D., Distinguished Professor and vice chair of the Department of Pathology at UC San Diego School of Medicine and a member of the UC San Diego Moores Cancer Center. To answer these questions, first author Chengsheng Wu, Ph.D., a postdoctoral fellow in Cheresh's lab, subjected pancreatic cell lines to various forms of stress, including low oxygen and sugar levels. He then identified which cells could adapt to the harsh conditions and observed which genes and molecules were modified in these cells. The stress-tolerant tumor-initiating cells showed reduced levels of a tumor-suppressive microRNA, miR-139-5p. This in turn led to the upregulation of lysophosphatidic acid receptor 4 (LPAR4), a G-protein-coupled receptor on the cell surface. "LPAR4 is not normally found on happy cells, but it gets turned on in stressful environments to help the cells survive, which is particularly advantageous for tumor-initiating cells," said Cheresh. Schematic: When a tumor-initiating cell experiences isolation stress, it begins expressing LPAR4. This leads to the production of a new extracellular matrix full of fibronectin (FN1), which keeps the cell safe and recruits other cells into the niche to begin tumor formation. Credit: UC San Diego Health Sciences The researchers found that LPAR4 expression promoted the production of new extracellular matrix proteins, allowing the solitary cancer cells to start building their own tumor-supporting microenvironment. The new extracellular matrix was particularly rich in fibronectin, a protein that binds to transmembrane receptors called integrins on surrounding cells. Once the integrins on these cells sensed the fibronectin, they began signaling the cells to express their own tumor-initiating genes. Eventually, these other cells were recruited into the fibronectin matrix laid by the tumor-initiating cells and a tumor started to form. "Our findings establish a critical role for LPAR4 in pancreatic tumor initiation, and a likely role in other epithelial cancers, such as lung cancer," said Cheresh. "It is central to tumor-initiating cells' ability to overcome isolation stress and build their own niche in which tumors can form." But isolation stress is not the only way this signaling pathway can be triggered, the researchers said. Chemotherapy drugs are also designed to put cancer cells under stress. Indeed, Cheresh's team found that treating cultured tumor cells and pancreatic tumors in mice with standard-of-care chemotherapeutics also led to the upregulation of LPAR4. The researchers said this might explain how such tumor cells could develop a stress tolerance and resistance to the drugs. Further experiments also showed that using integrin antagonists to block cells' ability to utilize the fibronectin matrix reversed the stress tolerance benefit of LPAR4 expression. Thus, the authors suggest targeting the LPAR4 pathway or disrupting the fibronectin/integrin interaction could be effective in preventing the growth, spread and drug resistance of pancreatic tumors. "We can think of tumor-initiating cells as being in a transient state that can be induced by different stressors, so our clinical goal would be to prevent oncogenic cells from ever entering this state," said Cheresh. "Now that we've identified the pathway, we can assess all the different ways we can intervene." The researchers suggested a new drug targeting this pathway could be used as a prophylactic in patients at high risk of developing the disease, or to prevent new tumors from forming in cancer cases with a high likelihood of metastasis. Pairing the new drug with existing chemotherapeutics that put stress on mature tumor cells could also mitigate the effects of drug resistance and make cancer treatments more effective, authors said. "Treating cancer can feel a little like whack-a-mole," said Cheresh, "but if we have two or three hammers and we know where the moles are going to pop up next, we can beat the game."

10.1038/s41556-022-01055-y

Biology

Study suggests that longer-distance migratory birds may be smarter

The paper, 'Possible linkage between neuronal recruitment and relocation distance in migratory birds', by Shay Barkan, Uri Roll, Yoram Yom-Tov, Leonard I Wassenaar and Anat Barnea will be published in Scientific Reports: www.nature.com/articles/srep21983 Journal information: Scientific Reports

http://www.nature.com/articles/srep21983

https://phys.org/news/2016-02-longer-distance-migratory-birds-smarter.html

Abstract New neuronal recruitment in an adult animal’s brain is presumed to contribute to brain plasticity and increase the animal’s ability to contend with new and changing environments. During long-distance migration, birds migrating greater distances are exposed to more diverse spatial information. Thus, we hypothesized that greater migration distance in birds would correlate with the recruitment of new neurons into the brain regions involved with migratory navigation. We tested this hypothesis on two Palearctic migrants - reed warblers ( Acrocephalus scirpaceus ) and turtle doves ( Streptopelia turtur ), caught in Israel while returning from Africa in spring and summer. Birds were injected with a neuronal birth marker and later inspected for new neurons in brain regions known to play a role in navigation - the hippocampus and nidopallium caudolateral. We calculated the migration distance of each individual by matching feather isotopic values (δ 2 H and δ 13 C) to winter base-maps of these isotopes in Africa. Our findings suggest a positive correlation between migration distance and new neuronal recruitment in two brain regions - the hippocampus in reed warblers and nidopallium caudolateral in turtle doves. This multidisciplinary approach provides new insights into the ability of the avian brain to adapt to different migration challenges. Introduction Exposure to new information in adult animals has been positively correlated with the magnitude of new neuronal recruitment in their brains 1 , 2 . This exposure, linked to new neuronal recruitment, could be manifested in many ways throughout an animal’s life history and the challenges it faces 3 , 4 , 5 . The recruitment of new neurons is considered to be promoted by an elevated need to acquire new information in relevant brain regions 6 . Such recruitment may serve this purpose by facilitating brain plasticity and has been suggested to play an important role in acquiring new memories in both invertebrates 7 , 8 and vertebrates 1 , 9 . Birds have been used as model animals in studies linking new neuronal recruitment with various behaviors, such as social complexity 10 , food hoarding 4 , 11 , song learning 12 , the reproductive cycle 13 and others. Long-distance migration is highly demanding in many aspects of a bird’s life cycle. Beyond the immense physiological effort required in traveling large geographical distances, birds also have to avoid predation, minimize resource competition, overcome adverse weather and accurately navigate and orient 14 . The mortality rate of young avian migrants is far higher than that of their adult counterparts which suggests that learned experience is of great importance for migratory birds 15 . Indeed, several studies have explored the importance of learning and experience for accurate navigation in migratory birds 16 , 17 . Learning could be key to improved navigation ability; identification of suitable stop-over sites en-route; avoidance of harsh weather or potential predators; and for outcompeting both fellow migrants and local birds. Such needs for increased learning ability during migration may be answered by a greater neuronal recruitment in the relevant brain regions of migratory birds. To date, only two studies of passerines 18 , 19 have directly investigated the possible relationship between neuronal recruitment and migratory behavior in birds. Those studies revealed that, for passerines, migrant species or sub-species possessed more new neurons than resident species in the brain regions that play a role in spatial information processing. During the past century, great advances have been made in tracking and monitoring migratory birds. In addition to the increased use of recovery-based ringing and tagging, researchers nowadays have access to radio transmitters and satellite or GPS tracking to monitor individual birds 20 . Recently, the use of stable isotopes has been shown to be effective in quantifying individual or population migratory connectivity 21 , 22 . The use of stable isotopes to track migratory movement is based on the predictable global spatial patterns for hydrogen (H) and carbon (C) isotopes. Wherever a bird grows its feathers (often at its natal or overwintering site), the local isotopic patterns are translated through diet along the trophic cascade and are fixed into the growing feathers, thereby reflecting the region the feathers were grown 23 . When a bird is caught later, its feathers’ 2 H/H and 13 C/ 12 C ratios can be non-lethally analyzed and compared to known H and C isotopic distribution base-maps (e.g. isoscapes). This tissue-to-isoscape comparison is used to obtain estimates of the region where a bird had molted or grown its feathers. One of the primary advantages of the isotope technique is that every bird captured provides intrinsic spatial information about its migration route, without the need for the mark-recapture used in ringing and marking techniques 24 . In this study we hypothesized a possible link between an increase in new neuronal recruitment and the distances traveled by migrant birds. We explored this phenomenon in reed warblers ( Acrocephalus scirpaceus ) and turtle doves ( Streptopelia turtur ), which are summer visitors in Israel and winter and molt in Africa 25 . Migration distance estimates were determined using feather isotopic ratios as an indicator of molt locality in the wintering grounds. New neuronal recruitment was evaluated in two brain regions known to take part in navigation and spatial orientation tasks-the Hippocampal complex (HC) 26 , 27 and the nidopallium caudolateral (NCL) 27 . Results Migration distance and recruitment of new neurons When exploring links between migration distance and new neuronal recruitment in turtle doves we found marginally significant increase in recruitment into the NCL for birds flying longer distances (P = 0.07; ρ = 0.53; N = 12; Fig. 1 ), but not in the HC (P = 0.96; ρ = −0.01; N = 11). In the reed warbler we found a similar but opposite trend, with increased neuronal recruitment into the HC (P = 0.12; ρ = 0.69; N = 6; Fig. 2 ), but not into the NCL (P = 0.8; ρ = −0.13; N = 6) of birds flying longer distances. Notably, these marginally significant results, respectively for the turtle doves and the reed warblers, were based on small sample sizes. For the turtle doves the statistical power for the NCL analysis was 76% at the nominal α = 0.05 and 84.7% for α = 0.1. For the reed warbler analysis of the HC, the statistical power was low (35%), which meant that for our sample size (six birds), at the nominal rejection rate of 0.05 and the effect size we found, it was unlikely that we would obtain significant results. Upon setting our rejection rate to 0.1, the power of the test increased to 47.4%. In order to obtain a power value of 80% with the same effect size we would have needed about 10 birds in our sample at the 0.1 rejection rate or 12 birds for the 0.05 rejection rate. Figure 1 Percentage of new neuronal recruitment into the hippocampus (HC) and nidopallium caudolateral (NCL) of the turtle dove ( Streptopelia turtur ), as a function of the migration distances from their wintering grounds in Africa to Israel. For the HC, Spearman’s rank correlation, P = 0.96; ρ = −0.14 (N = 11) and for NCL P = 0.07; ρ = 0.53 (N = 12). Full size image Figure 2 Percentage of new neuronal recruitment into the hippocampus (HC) and nidopallium caudolateral (NCL) of reed warbler ( Acrocephalus scirpaceus ), as a function of their migration distances from wintering grounds in Africa to Israel. For the HC, Spearman’s rank correlation, P = 0.12; ρ = 0.69 (N = 6) and for NCL P = 0.8; ρ = −0.13 (N = 6). Full size image Wintering molt location Only a few of the individuals that were analyzed for stable isotopes also underwent neuronal recruitment analysis (see below). We used our larger sample size of individuals whose feathers were measured for stable isotopes (25 reed warblers, 14 turtle doves), to estimate probable wintering grounds in Africa for these two species. Figs 3 and 4 display kernel distributions of wintering grounds for reed warblers and turtle doves, respectively. The maps also show data from ringing returns (obtained from the Society for the Protection of Nature in Israel – ), as well as known wintering grounds based on expert drawn maps 28 , 29 . Figure 3 Wintering grounds of turtle doves ( Streptopelia turtur ) in Africa. Kernel percentages display probable wintering grounds based on C and H isotopic values derived from birds caught in Israel. Each dot represents the probable center of a single individual’s wintering range. Black dots represent birds that only provided isotopic information, white dots represent birds that also provided neuronal information. The striped area shows wintering grounds based on Urban et al . 28 (Urban E.K., Fry C.H., Stuart, K. (1986). The birds of Africa. © A&C Black Publishers, used by permission of Bloomsbury Publishing Plc). The green triangle represents a single record of an individual that was ringed in Israel and caught in Africa. The green square represents the location in Israel where the birds were caught. Spatial analysis and map production were conducted in ArcGIS 10.1 56 . Full size image Figure 4 Wintering grounds of reed warblers ( Acrocephalus scirpaceus ) in Africa. Kernel percentages display probable wintering grounds based on isotopic values derived from birds caught in Israel. Each dot represents the probable center of a single individual’s wintering range. Black dots represent birds that only provided isotopic information, white dots represent birds that also provided neuronal information. The striped area shows wintering grounds based on Urban et al . 29 (Urban E.K., Fry C.H., Stuart, K. (1986) The birds of Africa. © A&C Black Publishers, used by permission of Bloomsbury Publishing Plc). The green triangle represents a single record of an individual that was ringed in Israel and caught in Africa. The green square represents the location in Israel where the birds were caught. Spatial analysis and map production were conducted in ArcGIS 10.1 56 . Full size image Discussion In this work we found a tentative link between the migration distance of two bird species and new neuronal requirement into the brain regions that play a role in spatial orientation and navigation. While our sample sizes are inevitably small (see also below), we found in two distinct regions – the HC for reed warblers and the NCL in the turtle doves, an increase in neuronal recruitment corresponding with increased migration distance. Our combined approach, though innovative, employed two commonly used and accepted isotope and neuronal analyses methods 19 , 30 . As noted above, the limitation of our results is that they are based on small sample sizes. Furthermore, just a few data points were responsible for much of the observed effect, both in the HC of reed warblers and the NCL of turtle doves ( Figs 1 and 2 ). However, these data points are not the outcome of any histological failure, because our brains were processed simultaneously in few batches and these data points were in batches that included other brains, which yielded data points with different scores. Moreover, statistically, these data points did not depart from the central tendency of the other values and were not designated as an outlier when fitted to a Poisson distribution 31 . Therefore, despite the small sample sizes and the consequent low statistical power obtained, we suggest that our data might indicate a general phenomenon. This suggestion is manifested in the fairly high coefficients of determination – ρ (0.53 for the turtle doves in the NCL and 0.69 for the reed warblers in the HC), for the two separate tests. Nevertheless, further investigation using larger datasets in different regions and organisms will be critical to obtain additional support for our hypothesis. While a larger sample size would have been desirable in order to confirm our hypothesis, the experimental design and field logistics made this impossible. Initially, we sampled only adult birds caught in the wild within a particular season. Furthermore, these birds had to survive 35 days of captivity, during which new neurons migrated and were incorporated into the designated brain regions. The analysis we conducted targeted neuronal differences within populations of the same species. Such differences probably arose due to evolutionary adaptations to different migration distances. As such, this phenomenon somewhat differs from cases in which new neuronal recruitment is temporally tuned to answer a specific need 10 , 13 , 32 . Our results indicated a variance in the degree of plasticity in the brain, potentially derived from genetic differences between populations migrating longer or shorter distances. A similar mechanism of differences in new neuronal recruitment between populations of the same species has been recorded with respect to food hoarding in black-capped chickadees ( Poecile atricapilla ) experiencing different climatic conditions 33 , 34 . The HC has been shown in previous works on passerines to be important for the processing of long-term memory, navigation and spatial orientation 1 , 26 . Similarly, the NCL has been found to be involved in processing spatial information 27 , 35 and working memory 36 , 37 in pigeons. Thus, our findings that the HC and the NCL are important for navigation and orientation in reed warblers and turtle doves, respectively, are in line with previous results. However, to the best of our knowledge, passerine NCL has never been tested with respect to orientation and navigation. Our results show that, at least for reed warblers, the NCL plays a minor role in these brain functions relative to the HC. Doves were previously shown to depend on the HC for navigation across a familiar space within 10–20 km from the home loft, whereas long-distance navigation over unfamiliar spaces was HC independent 26 . Therefore, turtle dove migration between Africa and Israel is unlikely to be HC-dependent and indeed we did not find any link between neuronal recruitment in the HC and migration distances traveled by this species ( Fig. 1 ). The differences in the brain regions found to be significantly correlated with migration distance in the two species might be explained by differences in their migratory behavior. Reed warblers are lone nocturnal migrants that rely on self-orientation for navigation. Conversely, turtle doves tend to migrate in large flocks, mostly at night but also during the day 38 . The evolutionary advantage of flocking lies in superior group decisions, known as the “many wrong” principle, in which pooling information from many inaccurate compasses yields a single more accurate compass 39 . In addition, group flight may also assist in locating limited or obscured landmarks while flying 40 and allow inexperienced individuals to follow experienced partners 41 , 42 . Such information sharing between flock members may reduce the navigational investment per individual, but requires developed social interaction and communication among migrating individuals 40 , 43 . The NCL has been suggested to take part in the integration of visual and auditory pathways 44 , 45 and may play a primary role in social interactions. Flock size is a plastic social structure which has been found to positively correlate with migration distance in many avian species 46 , 47 . Flocking migrants that fly longer distances could thus need to acquire more information on more complex social structures. This may explain the link we found in doves between neuronal recruitment in the NCL and migration distances. Beyond their behavioral differences, turtle doves and reed warblers belong to clades that are thought to have been separated from one another other ~85 million years ago 48 , that may have resulted in a parallel evolution manifested in similar functions of different brain regions. Using information from two stable isotopes, as well as spatial averaging via use of kernel estimations, we were able to ascribe probable localities of the winter molt for each individual bird. Furthermore, our analysis and methodology enabled us to determine the wintering grounds of the populations of the two species under study - the reed warbler and the turtle dove. Our findings add information about the geographical extent of these species during winter in a region that is lacking in research and surveys. Only 60 returns for ringed reed warblers and six returns for turtle doves from Europe and the Middle East have been recorded in Israel over the past 20 years, with two ringing returns for these two species from all of Africa ( Figs 3 and 4 ). Moreover, the maps we constructed based on isotope analysis have proven similar to the limited published information 28 , 29 and ringing data, thereby supporting our methodology. Wintering ground maps are particularly important for turtle doves as there is a growing concern regarding the global decline of populations of this species 49 . Identifying the wintering grounds of this species could aid in their future conservation measures. Here we present, for the first time, a preliminary indication of a linkage between distances traveled and new neuronal recruitment between different populations of migrating birds. Our evidence suggests a possible important adaptation of migrating birds – those that travel greater distances exhibit greater brain plasticity. Obviously, more work is needed in different species, habitats and brain regions, in order to substantiate our findings. Nonetheless, our methodologies and results can serve as a first important step. This work is also of value in light of the ongoing climatic, enviromental and landuse changes, as an indication that pre-adaptation towards greater brain plasticity could prove advantagous for species or populations in novel enviroments or ecosystems. We hope that our work will promote additional research on the effects of migration on the avian brain. Methods Experimental design Our neuronal analysis was carried out on six adult reed warblers and twelve adult turtle doves, collected under the Israel Nature and National Parks Protection Authority permit (2005/24706). The study was approved by the Tel Aviv University Institutional Animal Care and Use Committee (permit L-06-008) and was performed in accordance with its regulations and guidelines regarding the care and use of animals for experimental procedures. Birds were caught with mist nets in the Jordan Rift Valley, Israel (32°.41′N; 35°.53′ E) during spring (February to April) and summer (June to July) between the years 2005–2009. Birds were aged according to plumage, iris and leg color 25 , 50 . We used only adult birds in order to avoid possible effects of age on new neuronal recruitment and to ensure that the sampled individuals had molted their feathers at least once in Africa (see below). Two tail feathers were collected from each individual and tested for stable isotopes (Carbon: δ 13 C and Deuterium: δ 2 Η). Isotopic data were compared to the respective isoscapes of Africa (see below). Brains of all birds were processed and analyzed for new neuronal recruitment. Stable isotope analysis Two tail feathers were sampled from each individual for the analysis of stable isotope remains (δ 2 H; δ 13 C). Feather H isotopic composition for non-exchangeable H was analyzed following the comparative equilibration method described in Wassenaar and Hobson 51 . All feathers were cleaned of surface oils using a 2:1 chloroform:methanol solution and then dried overnight in a hood. For the δ 13 C analysis, we used 1.2 ± 0.2 mg subsamples of the feathers (weighed by microbalance - Sartorius SE2; Gottingen, Germany) and placed in tin capsules (D1007; Elemental microanalysis; UK). For the δ 2 H analyses, subsamples weighing 0.35 ± 0.01 mg were placed in silver capsules (D1007; Elemental microanalysis; UK). Samples were then analyzed on a Europa 20:20 continuous-flow isotope-ratio mass spectrometer (CF-IRMS) interfaced with a Robo Prep elemental analyzer. δ 13 C measurements were reported in δ-notation relative to the Pee Dee Belemnite standard (PDB) in parts per mil deviations (‰). Measurement error is estimated at 0.1‰ and 0.3‰ for the δ 13 C values. The δ 2 H measurements were reported as parts per mil deviations (‰) relative to Vienna Standard Mean Ocean Water–Standard Light Antarctic precipitation scale (VSMOW-SLAP). The H isotope analysis is more complex than that of δ 13 C due to the problem of uncontrolled isotopic exchange between feathers and ambient water vapor 51 . To correct this effect, accepted keratin standards were used so that the δ 2 H values reported here correspond to non-exchangeable feather hydrogen. This gave us δ 13 C and δ 2 H values for each individual, of the two species. Reed warblers and turtle doves are known to molt during winter in Africa 52 , 53 and we therefore allocated their feathers’ isotope values to this continent. Feather isotope values were matched to precipitation isotope values and a constant isotope fractionation factor of +25‰ was added for δ 2 H feathers values 51 and +1‰ for δ 13 C 54 . Spatial analysis To define the wintering molting regions for all birds we used ‘map lookup’ 55 and spatial averaging approaches. This entailed comparing the two isotopic values measured in the feathers of each bird to their parallel values in the known isoscapes, in order to create a probable molt region for each isotope. Regional δ 2 H isoscapes for the relevant feathers molting months (October–December) were obtained from WaterIsotopes.org web page, while δ 13 C annual isoscape was provided by Dr. C. Still (Department of Geography & Institute for Computational Earth System Science UC Santa Barbara, USA). Initially, we extracted for each bird and each of the two isotopes studied, those values of the isoscapes that corresponded to the measured values in the feathers up to the precision level of the isoscape layer. Thus, if our measured value was X we extracted from the isoscape layer only those values that X lay between – i.e. those values immediately preceding or following X. All of these extracted values for each bird were spatially compared between the two isotopes studies. In cases where there was a distance of 10 km or less between these extractions for the two isotopes a possible location was indicated. The natural spatial spread of both 2 H and 13 C in Africa does not show a clear east-west gradient. Consequently, for some individuals our approach found several possible locations on both the eastern and western sides of the continent that corresponded to their feathers’ isotopic values. However, previous studies have shown that reed warblers and turtle doves migrating across the eastern Mediterranean sea (where the birds were caught), spend the winter in eastern Africa 28 , 29 . Thus we excluded from our analysis those possible locations which were west of the 18 o E longitude. For each bird all possible locations eastern of 18 o E longitude, were spatially averaged using a kernel density smoothing method. This method ultimately averages out those locations further away from the majority of all matches and produces a single potential point (as the centroid of the 10% kernel volume contour) which was then used as the location of the wintering ground for that bird. We next measured the distance between this point and the locality where the bird was captured in Israel (using an equal distance global projection) in order to determine the minimum migration distance traveled by each bird. We later used all the centroid locations of all the birds for each species to produce a map of the wintering grounds of the two species. All spatial analysis was conducted in ArcGIS 10.1 56 . Neuronal recruitment analysis Following capture, the birds were transferred to outdoor aviaries in the Botanical Gardens of Tel Aviv University and injected three times, at 24 hours apart, with the cell birth-date marker 5-bromo-2-deoxyuridine (BrdU). Five weeks post-BrdU treatment birds were killed with an overdose of anesthesia and their brains underwent through histological procedures in which they were embedded in polyethylene glycol, blocked and cut transversely at thickness of 6 μm, along the rosto-caudal axis. Then, brain sections went through immunohistochemistry procedures that stained all neurons with fluorescent green (with anti-HuC/HuD) and nuclei of new neurons with fluorescent red (with anti-BrdU). Therefore, cells with co-localization of green cytoplasm and a red fluorescent nucleus were identified as new neurons ( Fig. 5 ). For full details of the histological and immunohistochemical protocols used for both species, see Barkan et al . 19 . Figure 5 New neurons at three microphotographs of the same field, in the nidopallium caudolateral (NCL; upper row) of turtle dove ( Streptopelia turtur ) and the hippocampus (HC; lower row) of reed warbler ( Acrocephalus scirpaceus ). BrdU-labeled cells were identified with a rhodamine filter ( a ) and Hu-labeled neurons were identified with a FITC filter ( b ). Double-labeled neurons were identified by alternating between these two filters and by using a dual FITC-rhodamine filter to show co-localization of the two markers ( c ). Full size image Brain mapping and quantification In both species, the brain regions we examined were the HC and NCL. In turtle doves, for each brain region, we defined and examined the most rostral and caudal section and examined five additional sections between them, separated by an average distance of 240 μm ( Fig. 6 ). For the HC, the most rostral section was defined by the presence of the commissura anterior (CoA) and corresponded to level A7.75 in the atlas of the pigeon brain 57 and the most caudal section of the HC corresponded to level A6.25 in that atlas ( Fig. 6B ). The ventral, dorsal and medial boundaries in each section were defined according to previously defined criteria 4 . For the NCL, the most caudal section corresponded to level A3.0 in the atlas of the pigeon brain 57 . The most rostral section corresponded to level A4.5 in this atlas and was defined according to the lateral ventricle, along the dorsal part of the brain ( Fig. 6C ). The ventral boundary was indistinguishable by our staining methods and was therefore determined according to the dopaminergic innervation recorded in pigeons 58 . Since brains of laughing doves and turtle doves differ in size from the pigeon brain, we followed the method we previously used 19 to determine the ventral boundary relative to that in the pigeon brain. For the definition of HC and NCL in reed warblers, a similar procedure was performed, as described in our previous publication 19 . Figure 6 Schematic views of the two investigated brain regions in brains of turtle doves ( Streptopelia turtur ). ( A ) Top view of the brain: rostral is to the left, caudal is to the right. We indicate the range within which frontal sections were taken from the hippocampal complex (HC) and nidopallium caudolateral (NCL). Seven sections were sampled along the rostro-caudal axis of each brain region (for details, see text), three of which are shown here: the most rostral, the middle and the most caudal (from left to right), in HC ( B ) and NCL ( C ). Abbreviations: Cerebellum (Cb), Commissura anterior (CoA), Nidopallium (N), Nidopallium caudale (NC), Lateral ventricle (V). Orientations: Dorsal (D), Lateral (L), Ventral (V) and Medial (M). Created from images originally appearing in: Karten, Harvey J. and William Hodos 57 . A Stereotaxic Atlas of the Brain of the Pigeon (Columbia Livia) . pp. 47-54, 61-67. © 1967 The Johns Hopkins Press. Adapted and reprinted with permission of Johns Hopkins University Press. Full size image In both species, we used a computerized brain-mapping system (Stereo Investigator; MicroBrightField Inc.) to draw the boundaries of the HC and NCL in each section sampled, mark the position the new neurons and quantify their number. Total neuronal density was quantified in one section in each brain region, by counting all neurons within 12–18 sampling squares (100 × 100 μm each), randomly chosen by the software, using the fractionator probe. Our measure of new neuronal recruitment was calculated as a percentage of new out of total neurons per mm 3 , in these two brain regions. Statistical analysis Non-parametric Spearman’s rank correlation was conducted between values of new neuronal recruitment and the calculated distances the birds had travelled. These regressions were conducted separately for the two species and each of the brain regions examined. We used the forward discovery rate correction for multiple testing 59 to assign significance, using α = 0.05 as our target for rejection of the null. We then conducted power analyses on these four separate tests. All statistical analyses were conducted in R 60 . Additional Information How to cite this article : Barkan, S. et al . Possible linkage between neuronal recruitment and flight distance in migratory birds. Sci. Rep . 6 , 21983; doi: 10.1038/srep21983 (2016).

Birds that migrate the greatest distances have more new neurons in the regions of the brain responsible for navigation and spatial orientation, suggests a new paper published in Scientific Reports. For some time scholars have widely accepted the view that neurons, the cells that specialise in processing and transmitting information and contribute to brain plasticity, continue to be generated in the brains of animals even when they are adults. After being created in one part of the brain, the neurons then migrate to those regions of the brain that need them most. The international research team, which included scientists from the University of Oxford, focused on the role played by neurons in two species, turtle doves and reed warblers, making their way from Africa to the Middle East or Europe. In both species, the researchers found that the proportion of new neurons increased in line with the migration distance. Interestingly, however, there was a distinct difference between the two species in the areas of the brain that incorporated the new neurons. In reed warblers, birds that migrate as individuals at night, new neurons were found mainly in the hippocampus—a region associated with navigation. In turtle doves, a species that migrates as a group, the new neurons were found mainly in the nidopallium caudolateral, an area associated with communication skills. A Turtle dove. Credit: Guy Barkan The researchers caught 12 turtle doves and six reed warblers in nets in the Jordan Rift Valley in Israel. All the birds were on a migratory flight from Africa, but the researchers were able to estimate the flight distance already taken by each individual by measuring the isotopic signatures on the birds' feathers. The known values of isotopes of carbon and hydrogen, found in the water, soil and plants, differ according to which part of the world they are located. The researchers matched the known local values of isotopes with the particular values found on the feathers when the bird was captured. When birds lose their old feathers (through moulting) and grow new ones, they incorporate the isotopic signature of where they have been through the food and water they have ingested. Then, these migration distances were compared with the amount of new neurons incorporated into the birds' brains. This was done by selectively colouring brain cells in several relevant regions: once—for identifying new cells, and then a few weeks again for identifying neuron cells. Those coloured twice were identified as new neurons. The researchers discovered that both species show a trend of increasing new neurons in line with migration distance and that different brain regions were affected. A Reed warbler. Credit: Asaf Rahamim Researcher Dr Uri Roll, from the School of Geography and the Environment at the University of Oxford, said: "These preliminary findings suggest a potentially exciting, new avenue of research. What we humans do during the day may actually make us more "brainy" as our regular activities may actually determine how our brains adapt and in which areas. In the long term, there are implications for how species evolve. For example, other research already suggests that birds that hoard food in particular periods incorporate new neurons in brain regions responsible for memory and spatial orientation. This latest paper builds on that work, suggesting that birds that need greater navigational help have more new neurons in that part of the brain while those that need to keep up with the flock incorporate new neurons in a different area."

www.nature.com/articles/srep21983

Nano

Cell-sized robots can sense their environment

Volodymyr B. Koman et al, Colloidal nanoelectronic state machines based on 2D materials for aerosolizable electronics, Nature Nanotechnology (2018). DOI: 10.1038/s41565-018-0194-z Journal information: Nature Nanotechnology

http://dx.doi.org/10.1038/s41565-018-0194-z

https://phys.org/news/2018-07-cell-sized-robots-environment.html

Abstract A previously unexplored property of two-dimensional electronic materials is their ability to graft electronic functionality onto colloidal particles to access local hydrodynamics in fluids to impart mobility and enter spaces inaccessible to larger electronic systems. Here, we demonstrate the design and fabrication of fully autonomous state machines built onto SU-8 particles powered by a two-dimensional material-based photodiode. The on-board circuit connects a chemiresistor circuit element and a memristor element, enabling the detection and storage of information after aerosolization, hydrodynamic propulsion to targets over 0.6 m away, and large-area surface sensing of triethylamine, ammonia and aerosolized soot in inaccessible locations. An incorporated retroreflector design allows for facile position location using laser-scanning optical detection. Such state machines may find widespread application as probes in confined environments, such as the human digestive tract, oil and gas conduits, chemical and biosynthetic reactors, and autonomous environmental sensors. Main Two-dimensional (2D) materials, such as graphene and transition metal dichalcogenides, can be fashioned into novel planar circuit elements, including photodiodes 1 , transistors 2 , chemiresistor sensors 3 , memory elements 4 and capacitors 5 , 6 . This has enabled the transfer and construction of nanoelectronic circuits grafted onto unique surfaces. However, one surface that has yet to be explored for this purpose is a fluid dispersible colloid. This could extend nanoelectronics to remote and inaccessible spaces and volumes. As on-board circuit elements increase in number and complexity, access independent sources of power, and manifest input and output information sources, a threshold is reached where the colloid constitutes a finite-state machine or computational device for handling informational flows. Nanofabrication and synthesis methods have allowed researchers to create sophisticated colloidal devices (single particles or assemblies that serve a particular function), but not ones possessing autonomous circuitry, logic manipulation or information storage. In particular, state-of-the-art multifunctional micro/nanoparticles do not process information in an autonomous way when decoupled from their environment 7 , 8 . Similarly, conceptual devices referred to as ‘molecular machines’ rely on the stochastic thermodynamics of their environment 9 (for example, the Brownian ratchet). Directed self-assembly methods are still in their infancy and are only able to create rather simple geometries—and not yet autonomous function 10 , 11 ; on the other hand, nanoelectronics has enabled several state machine implementations, although not in the form of autonomous colloidal particles 12 , 13 , 14 . DNA electronics has been proposed in ref. 15 , but the formation of electronic, DNA-based, colloidal state machines has not been realized. Finally, there has been recent progress in creating so-called ‘biorobots’, or biological state machines (a natural organism whose genome has been modified to manifest diverse, non-natural functions) 16 , 17 , 18 . However, biorobots can only exist under specific biological conditions, requiring limitations to the environment, temperature, pH and salinity, and protection against wild-type organisms. This constrains the envisioned applications compared to purely synthetic state machines, which could bring both digital and analog electronics to harsh environments in a scalable way. For example, the large-area monitoring of environmental bacteria, biological spores, particulates and volatile organic compounds (VOCs) requires significant resources 19 that could be more efficiently deployed using synthetic state machines (Supplementary Note 1 ). The concept of cubic millimetre-sized devices, previously called ‘smart dust’, was not implemented for fluid dispersion, but nevertheless constituted an early attempt towards achieving miniaturized autonomous electronics 20 . However, it is clear that the power density limits further downward scaling. Batteries with typical energy capacities up to 0.1 nW µm − 3 (ref. 21 ) and with power harvesting techniques that can deliver between 0.1 and 10 nW (10 4 µm 2 ) −1 (ref. 22 , 23 , 24 ) remain insufficient for conventional electronics at the microscale (Supplementary Note 2 ). Fortunately, 2D material devices are predicted to bypass these difficulties 25 because they have a number of advantages 26 , such as low-power performance compared to Si (<0.5 V), acceptable gate control with subthreshold swings ≪ 60 mV dec −1 , and large turn-on currents (>10 3 μA μm −1 ). In this work we create a new class of submillimetre colloidal particles with 2D electronic materials (graphene, hexagonal boron nitride (hBN), MoS 2 and WSe 2 ) grafted onto them. Top-down fabrication can pattern them into functional electronic circuits, transistors, memory and sensors, creating what we call colloidal state machines (CSMs), or particles capable of collecting, manipulating and storing information autonomously. Such CSMs can access local fluid hydrodynamics to enter spaces inaccessible to conventional electronics. Our designs include circuits with 2D material p–n heterojunctions of MoS 2 and WSe 2 , powering MoS 2 chemiresistor circuit elements. Such CSMs can be aerosolized and detect VOCs or carbon nanoparticulates along their trajectories, storing the detection in a non-volatile memristor device consisting of a Au–MoS 2 –Ag junction shielded from the environment by an inert hBN monolayer. These state machines exploit a unique property of 2D electronic materials in their ability to impart fundamentally new functions to highly mobile colloids. We demonstrate several specific applications, such as remote sensing of ammonia in constricted pipelines, as well as large-area sensing of aerosolized soot via standoff deployment and retrieval of surface-dispersed state machines. Because of their high aspect ratio, micrometre-scale 2D materials have low mechanical stability (necessary to sustain off-substrate applications). To circumvent this, we first designed a CSM base that plays the role of a carrier substrate for the 2D devices. We chose the photoresponsive polymer SU-8, as it can be processed with micrometre precision via conventional photolithography, becomes mechanically and chemically stable after cross-linking, even during various fabrication stages (Supplementary Fig. 1 ), and has a smooth surface (Supplementary Fig. 2 ). We also designed an electrical state machine with functional elements of combinational logic on this polymer base (Fig. 1a ). Specifically, this has three components—a power source, a switch and a memory element—implemented by a photodiode, chemiresistor and memristor, respectively (Fig. 1b ). The photodiode is built on a p–n heterojunction of MoS 2 and WSe 2 monolayers. Another MoS 2 monolayer serves as a chemiresistor that changes its conductance following binding of external analytes. Finally, the memristor is composed of liquid-exfoliated MoS 2 flakes sandwiched between gold and silver electrodes. The fabricated particle represents a state machine with both light and analyte as inputs and the memory state as an output (Supplementary Fig. 3 ). It operates in the following way. The photodiode generates voltage when it is illuminated with light, the chemiresistor switches its conductance after analyte binding, and the memristor changes its state from off to on when the applied voltage from the photodiode exceeds a threshold voltage. The whole process can be represented as two IF blocks and the logic operator AND on a block diagram. The memory component only changes its state from OFF to ON if both light shines on the photodiode and the analyte binds to the chemiresistor (Fig. 1c ). Fig. 1: CSM fabrication and aerosolization. a , Summary of CSM fabrication steps with side and top view schematics and top view optical micrographs: (1) SU-8 base fabrication; (2) MoS 2 and WSe 2 monolayer transfer with gold evaporation to form photodiodes; (3) MoS 2 flakes transfer to form memristors, while silver evaporation forms electrical contacts for the MoS 2 chemiresistor; (4) CSMs are released from the wafer by liftoff and are stored as a dispersion. Scale bars, 25 µm. b , Electrical circuit diagram of CSM. The photodiode converts light into current (generating voltage ε and with internal resistance R ph ), which turns on the memristor (with a threshold voltage V th and internal resistance R m ) only if the chemiresistor detects an analyte (resistance R ch ). c , Block diagram for the CSM state machine. The initial memory state off changes to on only in the presence of both chemical and light signals. d , CSM dispersion is aerosolized using a nebulizer with 4–10 m s −1 speed across 0.6 m distance in air. e , Digitized positions of aerosolized CSMs collected on the target. f , Angle distribution diagram extracted from e shows that the CSMs have no preferential direction. Full size image The performance of an electronic device changes when it is removed from a native substrate due to the imposed stretch and strain 27 . Remarkably, 2D materials possess higher strain limits than traditional i ii – v materials with similar size 28 , making them more prone to surviving the transfer process. We tested the performance of devices at three stages: (1) as-fabricated on the silicon substrate; (2) after liftoff; and (3) after aerosolization using a nebulizer across a 0.6 m distance in a 0.15-m-diameter tube (Fig. 1d and Supplementary Fig. 4 ). In a typical nebulizer experiment, using the microscope we identified that N = 244 CSMs landed on the target at the opposite end of the tube. Most of the CSMs were concentrated in a 0.06-m-diameter circle and had no preferential angular direction, proving directional aerosolization (Fig. 1e,f and Supplementary Fig. 5 ). Droplets ejected by the nebulizer had an initial speed of 4–10 m s −1 and diameter of <0.3 mm. After liftoff and aerosolization, some CSMs were occasionally bent (37 CSMs in one particular aerosolization experiment) or aggregated (34 CSMs; Supplementary Fig. 6 and Supplementary Note 3 ). Numerical calculations suggest that, after nebulizer propulsion CSMs travel ~3 m (with travel times of <0.3 s) before completely stopping (Supplementary Fig. 7 and Supplementary Note 4 ). The motion equation is given by $$\frac{{{\mathrm{d}}v}}{{{\mathrm{d}}t}} = - \frac{{3\rho _{\mathrm{air}}}}{{4\rho _{\mathrm{p}}D}}v^2C_{\mathrm{d}}$$ (1) where v is the particle speed, ρ air and ρ p are the air and particle densities, respectively, D is the particle diameter, and C d is the particle drag coefficient. Water droplets of this size dry within 25–160 s (Supplementary Fig. 7 ). This hints that the droplets do not dry throughout their flight. Therefore, the presence of a water envelope possibly enhances CSM stability during landing. 2D nanoelectronics for basic state machine components To test individual components, we constructed CSM bases with isolated devices. As a power source, we fabricated a p–n photodiode comprised of a heterostructure of MoS 2 and WSe 2 monolayers with Au contacts in a 90° configuration and ~10 µm channel length (Fig. 2a,b ). Before fabrication, we performed photoluminescence, atomic force microscopy and Raman mapping measurements to confirm the continuity of the MoS 2 and WSe 2 monolayers (Supplementary Figs. 8 and 9 ). Following stacking, we observed the emergence of an additional photoluminescence peak at 800 nm, corresponding to the staggered gap (type II) heterostructure 29 (Supplementary Fig. 10 ). The photodiode fabricated on substrate generates photocurrent under laser illumination, reaching ε = 0.27 ± 0.06 V open-circuit voltage and I sh = 0.15 ± 0.01 µA short-circuit current with a 532 nm laser fluence of P = 7 µW µm − 2 (Fig. 2c ), which is comparable with previous studies 30 , 31 . These characteristics are preserved after liftoff and aerosolization (Fig. 2d and Supplementary Fig. 11 ). Fig. 2: Individual components of CSM. a , b , Diagram ( a ) and optical image ( b ) of CSM with a photodiode fabricated from a continuous MoS 2 monolayer and 25 µm striped monolayer of WSe 2 . c , Typical current–voltage characteristics of a p–n photodiode composed of MoS 2 and WSe 2 monolayers under illumination by a 532 nm laser (black, in the dark; cyan, blue, green and red, under 0.7, 1.75, 3.5 and 7 µW µm − 2 illumination intensities, respectively). d , Master plot for multiple devices, as in c . Both open-circuit voltage and short-circuit current are preserved after liftoff ( ε = 0.27 ± 0.05 V, I sh = 0.20 ± 0.01 µA) and aerosolization ( ε = 0.24 ± 0.05 V, I sh = 0.19 ± 0.01 µA) as compared to the original on-substrate devices ( ε = 0.27 ± 0.06 V, I sh = 0.15 ± 0.01 µA) for N = 20 random devices. e , f , Diagram ( e ) and optical image ( f ) of a CSM with a chemiresistor. g , Current–voltage curves for a monolayer MoS 2 chemiresistor before (red) and after (blue) the addition of 10 mM TEA. h , Master plot for multiple devices, as in g . The red line is a guide for the eye. Conductance increase: from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 19.8 ± 2.3 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 34.7 ± 2.8 nS for devices on the substrate; from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 21.1 ± 5.5 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 36.1 ± 5.5 nS for liftedoff devices from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 20.5 ± 6.6 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 35.0 ± 6.4 nS for aerosolized devices ( N = 20). i , j , Diagram ( i ) and optical image ( j ) of a CSM with a memristor. k , Current–voltage characteristics for a MoS 2 memristor sandwiched between gold and silver electrodes. l , Master plot for multiple devices, as in k . Both the threshold voltage and initial conductance are preserved: V th = 0.16 ± 0.02 V, \(G_{\mathrm{m}}^{\mathrm{off}}\) = 13.4 ± 4.0 nS on the silicon; V th = 0.17 ± 0.02 V, \(G_{\mathrm{m}}^{\mathrm{off}}\) = 18.4 ± 8.4 nS for lifted off device; V th = 0.17 ± 0.02 V, \(G_{\mathrm{m}}^{\mathrm{off}}\) = 14.4 ± 6.9 nS for aerosolized CSMs. In d , h and l , black denotes as-fabricated devices, red denotes lifted off devices, and blue denotes devices dispersed with a nebulizer. Scale bars, 25 µm. Full size image For the sensor element, we chose a chemiresistor design based on a MoS 2 monolayer (Fig. 2e,f ). In the circuit, we designed it as a chemically induced turn-on switch to power the memristor. Mechanistically, MoS 2 increases its conductance with an analyte that demonstrates n-type doping on adsorption. An example is 10 mM triethylamine (TEA) sprayed onto MoS 2 , which results in a conductance increase (Fig. 2g,h ) from \(G_{\mathrm{ch}}^{\mathrm{in}}\) =20.5 ± 6.6 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) =35.0 ± 6.4 nS for the aerosolized elements, with similar values for on-substrate ones. To understand the reaction kinetics, we extracted the MoS 2 –TEA first-order reaction constants by analysing continuous conductance measurements of the MoS 2 chemiresistor 32 : $${\mathrm{MoS}}_2 + {\mathrm{TEA}}\begin{array}{*{20}{c}} {\mathop { \to }\limits^{k_{\mathrm{f}}} } \cr {\mathop { \leftarrow }\limits_{k_{\mathrm{b}}} } \end{array}{\rm{A}}\mathop { \to }\limits_{k_{\mathrm{i}}} {\rm{B}}$$ (2) where binding constant k f = 0.024 ± 0.001 s −1 µM − 1 , dissociation constant k b = 0.23 ± 0.01 s − 1 and irreversible constant k i = 0.06 ± 0.01 s − 1 , A is a physisorbed adduct of TEA and B a chemisorbed adduct (Supplementary Fig. 12 and Supplementary Note 5 ). The process of MoS 2 –TEA charge transfer is limited by the surface coverage, providing the maximum of 95% of MoS 2 resistance change 33 . This signifies that TEA can irreversibly bind to MoS 2 in less than 0.3 s—similar timescales to the flying time of CSMs. To store the detection event, we designed a memristor based on MoS 2 flakes sandwiched between two different-workfunction metals. Most memristors operate in the 1–3 V range, exceeding the photodiode voltage generation of the CSM 34 , 35 , 36 . However, following recent work 37 , a MoS 2 memristor fabricated between Au and Ag electrodes (Fig. 2i,j ) with dimensions of 25 × 25 × 0.1 µm 3 has a turn-on voltage of ~0.15 V, 100 MΩ starting resistance and its on–off ratio can reach millions (Fig. 2k and Supplementary Fig. 13e ). Mechanistically, an ultrathin MoOx layer forms on the MoS 2 surface with subsequent charge trapping/detrapping at the Ag/MoOx interface 37 . These memristors, written to various states during their initial runs, demonstrate excellent retention over a period of 2 h (Supplementary Fig. 13 ). Both the threshold voltage V th = 0.17 ± 0.02 V and initial conductance \(G_{\mathrm{m}}^{\mathrm{off}}\) = 14.4 ± 6.9 nS for aerosolized devices are similar to those for the on-substrate devices (Fig. 2l ). Once assembled, operation of the CSM requires that the photodiode voltage ε exceed the threshold memristor voltage V th . The second criterion is that the memristor should not change its state if there is only analyte detection. To this end, we shielded all circuit elements except for the chemiresistor sensor using hBN monolayers (Supplementary Fig. 3 ). The memristor should also not change state with only light excitation; hence, the chemiresistor design includes an initial conductance \(G_{\mathrm{ch}}^{\mathrm{in}}\) , such that the voltage generated across the memristor does not exceed V th . This sets the lower limit for the initial chemiresistance ( \(R_{\mathrm{ch}}^{\mathrm{in}}\) = 1/ \(G_{\mathrm{ch}}^{\mathrm{in}}\) ), which can be determined by Ohm’s law as $$R_{\mathrm{ch}}^{\mathrm{in}} > R_{\mathrm{m}}^{\mathrm{off}}\left( {\frac{\varepsilon }{{V_{\mathrm{th}}}} - 1} \right) - R_{\mathrm{ph}}$$ (3) where R ph is the photodiode resistance. After chemical detection via reaction with the analyte, the chemiresistor decreases its resistance to \(R_{\mathrm{ch}}^{\mathrm{f}}\) . This allows the memristor to change its state from \(R_{\mathrm{m}}^{\mathrm{off}}\) to \(R_{\mathrm{m}}^{\mathrm{on}}\) , which again can be assessed through the Ohm’s law: $$R_{\mathrm{m}}^{\mathrm{on}} = \frac{{R_{\mathrm{ch}}^{\mathrm{f}} + R_{\mathrm{ph}}}}{{\frac{\varepsilon }{{V_{\mathrm{th}}}} - 1}}$$ (4) For the memristor to change its state, \(R_{\mathrm{m}}^{\mathrm{on}}\) < \(R_{\mathrm{m}}^{\mathrm{off}}\) , has to be satisfied, yielding $$R_{\mathrm{ch}}^{\mathrm{f}} < R_{\mathrm{m}}^{\mathrm{off}}\left( {\frac{\varepsilon }{{V_{\mathrm{th}}}} - 1} \right) - R_{\mathrm{ph}}$$ (5) Equations ( 3 ) and ( 5 ) set the requirements for the circuit design, so we had to carefully tune MoS 2 size (Supplementary Note 6 ). The use of 2D materials keeps the particle aspect ratio α close to that of the bare substrate. This, in turn, leads to higher particle buoyancy, which approaches the values of water droplets in clouds (Supplementary Note 4 ). Sedimentation speed increases with particle size L and with α and ρ p (Supplementary Fig. 14 ). For CSMs made of SU-8 bases, ρ p = 1,200 kg m − 3 , L = 50 µm and α = 0.01, the sedimentation speed is ~20 µm s −1 , allowing them to stay in the air for days. To achieve similar result, silicon particles ( ρ p = 2,300 kg m − 3 ) with conventionally thicker electronics ( α = 0.1) have to be as small as 5 µm. Aerosolizable electronics Large-area monitoring of bacteria, spores, smoke particles, dust components and VOCs is an important objective, which currently requires significant resources 19 . In one implementation, satellite scanning can rapidly cover large areas, but is costly and indirect (with limited applicability) 38 . On-ground sensor installation and networking is labour-intensive and can often be slow in comparison to analyte distribution 38 . Using a fleet of aerial sensors (such as unmanned air vehicle drones) is again associated with high costs. As an alternative, we introduce the concept of aerosolizable electronics: CSMs dispersed in air carry 2D electronic devices that remain operational during and after flight. To demonstrate the state machine operations, we fabricated CSMs with three components assembled in one closed circuit (Fig. 1a ) and tested them (1) as fabricated on the silicon substrate and (2) on spraying in air within a confined chamber (Fig. 3a,b ; see Methods ). On exposure to 10 mM TEA, the sprayed CSMs change their chemiresistor conductance from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 8.9 ± 3.1 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 17.3 ± 4.1 nS (consistent with Fig. 2h ), allowing a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 13.3 ± 4.8 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 17.3 ± 8.9 nS after laser illumination (two-tailed P = 0.0001, N = 100, Fig. 3c ). We obtained similar results for CSMs on substrate: 59 CSMs successfully changed their states, with an average memory conductance ratio of 1.5 (Fig. 3e,g and Supplementary Fig. 15 ). Fig. 3: Aerosolizable electronics. a , Experimental schematic demonstrating remote detection and memory storage in a constrained environment: the left nebulizer injects CSMs (teal squares) across the enclosed tube injected with either TEA or aerosolized carbon nanotube particulates (dark blue droplets) using the top nebulizer. CSMs are collected on the collector, exposed to light, and their memory states are queried afterwards. b , Image of the experimental set-up. c , Chemiresistor conductance changes due to CSM exposure to TEA droplets (10 mM) during spraying, enabling memory conductance change after illumination with 532 nm laser light at 7 µW µm − 2 . CSMs change their chemiresistor conductance from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 8.9 ± 3.1 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 17.3 ± 4.1 nS, allowing a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 13.3 ± 4.8 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 17.3 ± 8.9 nS after laser illumination (two-tailed P = 0.0001). Black squares denote measurements before exposure, and red circles measurements after exposure and illumination. d , As in c , but CSMs were exposed to a carbon nanotube dispersion (0.2 g l −1 ). CSMs change their chemiresistor conductance from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 8.7 ± 2.0 pS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 4.3 ± 2.7 µS, inducing a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 14.7 ± 6.0 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 0.47 ± 0.44 µS. e , f , As in c and d , respectively, but for CSMs on-substrate. e , Chemiresistor conductance changed from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 9.6 ± 1.3 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 19.1 ± 1.9 nS, enabling a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 12.6 ± 3.5 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 16.0 ± 4.0 nS ( P = 0.0001); 41 CSMs did not change their memristor state (red circles). Control experiments with light showed that only 6 CSMs changed their memory state in the absence of TEA, meaning that equation ( 3 ) was not satisfied in their case. f , Chemiresistor conductance changed from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 9.6 ± 1.1 pS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 4.9 ± 2.9 µS, inducing a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 12.9 ± 3.9 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 0.50 ± 0.49 µS; 84 CSMs worked with an average memory conductance ratio of 62. Control experiments with exposure of these CSMs to only light demonstrate that no CSM changed its memristor conductance. Violet triangles denote control measurements after illumination; blue diamonds denote measurements after exposure. Dashed lines are guides for the eye. N = 100 for all experiments. g , h , Ranked memory conductance ratio (ON–OFF) extracted from e and f , respectively. Full size image The chemiresistor sensor works better with larger conductive changes following analyte detection, and this is an area for future improvement. For example, if we use conductive carbon nanotubes dispersed at 0.2 g l −1 as the analyte, the detection efficiency increases substantially. For these experiments we used an insulating substrate (SU-8, sheet resistance ~10 pS) instead of the MoS 2 as the chemiresistor, increasing \(R_{\mathrm{ch}}^{\mathrm{in}}\) . Adsorbed carbon nanotubes form a percolated network with conductivities on the order of µS (Supplementary Fig. 16 ), providing high gain to better satisfy equation ( 5 ). In a typical experiment, sprayed CSMs change their chemiresistor conductance from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 8.7 ± 2.0 pS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 4.3 ± 2.7 µS, inducing a memristor change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 14.7 ± 6.0 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 0.47 ± 0.44 µS with 71 active CSMs (Fig. 3d ), with similar performance for on-substrate CSMs (Fig. 3f,h ). Demonstration of constrained environmental sensing Researchers have identified several important closed systems from which it is difficult to extract information or interface electronics within an inaccessible interior 39 . Examples include oil and gas conduits 40 , 41 , chemical and biosynthetic reactors 42 , 43 , porous geological materials for upstream oil and mining exploration 44 , 45 and the human digestive tract 46 , 47 . Several methods to probe such systems exist, but they are either indirect or very limited in their applicability (Supplementary Note 7 ). At the same time, fully autonomous electronic chips have been limited to the millimetre scale 46 , 48 , which is too large to address the above applications. To this end, we demonstrate that CSMs can be injected into a pipeline system, probe it, and then be successfully retrieved to deliver the captured information. To illustrate, a model pipeline section was fabricated, into which gaseous ammonia was injected. Ammonia is a highly toxic gas used as a fertilizer in agriculture and as a refrigerant in the chemical industry 49 . It is also one of the most dangerous compounds to be transported through pipelines 50 . To probe the pipeline internally, we first injected CSMs within the system (Fig. 4a ) as a nebulized aerosol. A valve was then used to pulse ammonia vapour (~10 kPa) into the system, allowing the CSMs to interact with it for 30 min. Next, the ammonia valve was closed and the CSMs retrieved from the collector (Fig. 4b,c ). The experimental procedure is similar to the previous section, with some CSMs inserted into the tube on-substrate as a control. Ammonia vapour acts as an n-dopant for the MoS 2 layer 51 , changing MoS 2 conductances from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 9.1 ± 2.2 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 18.5 ± 2.5 nS for the sprayed CSMs and, consequently, allowing a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 12.5 ± 3.9 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 14.5 ± 4.3 nS after laser illumination (two-tailed P = 0.0007, N = 100, Fig. 4d ) with similar results for the on-substrate CSMs (Fig. 4e,f and Supplementary Fig. 17 ). Fig. 4: CSMs for monitoring pipeline status. a , b , Schematics of the pipe segment system (22 mm inner diameter), which has two separate valves for metering aerosolized CSMs (teal squares) ( a ) or ammonia ( b ). To allow for retrieval, a layer of cheesecloth served as a collector at the pipe endpoint. c , Image of the experimental set-up with a crucible filled with ammonia. Once the lower valve is open, saturated ammonia vapour (~10 kPa) expands into the rest of the system. d , Chemiresistor conductance changes (from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 9.1 ± 2.2 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 18.5 ± 2.5 nS) due to CSM exposure to ammonia vapour, enabling a memory conductance change (from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 12.5 ± 3.9 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 14.5 ± 4.3 nS) after illumination with a 532 nm laser (7 µW µm − 2 ). Black squares and red circles denote measurements before exposure and after exposure and illumination, respectively. e , As in d , but for control CSMs on-substrate. Chemiresistor conductance changes from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 9.5 ± 1.3 nS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 18.6 ± 2.9 nS, enabling a memristor conductance change from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 11.7 ± 3.8 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 13.7 ± 5.0 nS; P = 0.0017. Violet triangles denote control measurements after illumination, blue diamonds denote measurements after exposure. Dashed lines are guides for the eye. f , Ranked memory conductance ratio extracted from e ; 45 CSMs successfully change their memory conductance with an average ratio of 1.55. N = 100 for all experiments. Full size image State machines for soot exposure monitoring Soot nanoparticles emitted by diesel engines, industrial emissions and power plants pose health, climate and environmental risks 52 . Aerosolized micro- and nanoparticles can travel thousands of kilometres before sedimentation 53 , making it challenging to predict soot distribution and impact. Currently, the large-area monitoring of soot remains an economically inviable task. To this end, CSMs as dispersed, printed devices can potentially cover large areas to successfully detect soot, remaining virtually invisible to the naked eye, but otherwise easily detectable on a surface (see next paragraph). In this case, aerosolization allows CSMs to be rapidly printed over a specific area of interest as intact, functional, autonomously powered devices. To demonstrate the monitoring of undesirable particulates from surface-dispersed CSMs, the nebulized CSMs were deposited over an area of 0.6 × 0.6 m 2 . Next, 2 g l −1 of Printex XE2-B soot was loaded into a separate nebulizer and sprayed over three distinct locations (Fig. 5a,b and Supplementary Fig. 18 ), simulating localized particulate efflux. For these experiments we used an insulating substrate (SU-8, sheet resistance ~10 pS) instead of MoS 2 as the CSM chemiresistor. The remainder of the experimental procedure is similar to previous experiments. By their nature, soot particles are highly conductive 54 ; thus, the chemiresistor element drastically changes its conductance from \(G_{\mathrm{ch}}^{\mathrm{in}}\) = 7.4 ± 2.4 pS to \(G_{\mathrm{ch}}^{\mathrm{f}}\) = 0.80 ± 0.26 µS on exposure to soot. This further translates into memristor conductance changes from \(G_{\mathrm{m}}^{\mathrm{off}}\) = 13.7 ± 4.2 nS to \(G_{\mathrm{m}}^{\mathrm{on}}\) = 127 ± 72 nS following laser illumination (Fig. 5d ). The retrieved CSM positions allow determination of the exposed and unexposed areas (Fig. 5c ). Fig. 5: Large-area sensing. a , Schematic of set-up. Soot particles are sprayed at three locations over an area with previously dispersed CSMs (teal squares). b , Image of experimental set-up. c , Digitized positions of aerosolized CSMs. Dashed circles are guides for the eye, highlighting three areas exposed to soot. d , Chemiresistor conductance changes due to CSM exposure to soot, enabling memory conductance changes after illumination with a 532 nm laser (7 µW µm − 2 ). Full size image Enhancements of state machines to aid standoff detection To efficiently detect the location of CSMs at standoff distances, we fabricated a distinct batch, where the CSM base consisted of a retroreflector design. The design follows that of Switkes et al. 55 with a 100 µm retroreflector for low-intensity laser (10 mW cm − 2 ) reflected light from distances of up to 1 km (ref. 55 ). CSM retroreflectors were fabricated using SU-8 coated with 100 nm Ag, designed in a chequered shape, that allows them to reflect light back to the source from angles up to 60° (Fig. 6b ). We used a custom laser-scanning system (Fig. 6a ) to rapidly scan (<1 ms) and detect reflection from CSMs that landed after spraying over an 8 × 8 mm 2 area at a 5 cm standoff distance (Fig. 6c ). Careful examination under a microscope showed that N = 34 CSMs were detected (Fig. 6d,e ): 25 CSMs (100% refers to the percentage of CSMs detected by laser scanning versus the total number of CSMs identified under the microscope for a given category) had their reflectors facing up and N = 9 (56%) were flipped. The yield becomes slightly lower for CSM detection on an inclined substrate due to imperfect backreflections (Fig. 6f–h ): 23 CSMs (82%) were facing up and 7 CSMs (32%) were flipped. These results suggest that this attribute of CSMs can aid in their tracking and recovery over large areas, and, with refinement, could increase the detection threshold to 100% with a 10 cm 2 s −1 scan. Fig. 6: CSM standoff detection. a , CSMs are sprayed using a nebulizer through 10 mM TEA or 2 mg l −1 soot dispersed in air. A raster-scanning laser system is then used to find CSMs. b , Retroreflector-CSM reflectance as a function of inclination and rotational angles. The dashed line marks the diffuse reflection limit 2 × 10 −3 %. Inset, Top view of 100 µm retroreflectors. c , Laser raster-scanning detection of 100 µm CSMs that landed after spraying. d , CSM positions extracted under a microscope: black circles, CSMs with retroreflectors on top; red circles, CSMs with flipped retroreflectors; open circles, CSMs that were not detected by the laser scan in c . e , Statistics on CSMs. Schematics on the right demonstrate figure labels. f – h , As in c – e , but for 30° substrate inclination angle. Full size image We have demonstrated the capability of grafting autonomous electronic circuits capable of logic operation and information storage on submillimetre-sized particles, forming CSMs. Our particles can undergo aerosolization while carrying functional electronics on-board capable of interaction with the environment. With a thickness of 1.24 µm and weight of ~1.4 g m − 2 , this CSM design represents one of the thinnest and lightest circuits produced so far 27 . In this design, the on-board circuit forms a state machine with two inputs (chemical and optical) and one output composed of a memristor. Due to the use of 2D materials, CSM requires only 30 nW to irreversibly record events, granting it the ability to be powered from the energy harvested by the on-board 2D photodiode (30–100 nW). The results of chemical sensing are irreversibly stored in the memory (inducing memristor conductivity changes of up to 150 times). Furthermore, the addition of integrated retroreflectors allows the dispersed CSMs to be rapidly (<1 ms per frame) detected by a laser-scanning system. CSMs may find application in a wide range of areas, including biosensing (for example, within the human digestive tract), large-area sensing, confined space monitoring (for example, chemical and biosynthetic reactors, oil and gas conduits) and aerospace programmes. Methods 2D materials Large-area MoS 2 films were grown by a chemical vapour deposition process as described elsewhere 56 . Briefly, solid 0.5 g S and 4 mg MoCl 5 were used as precursors, and a 2 × 1 cm 2 300 nm SiO 2 /Si wafer piece was used as a growth substrate in a vacuum tube quartz furnace. The system was filled with 50 s.c.c.m. Ar for 1 h with subsequent growth at 850 °C under 2 torr pressure for 10 min and a 30 min temperature ramp. As-grown MoS 2 films were coated with polystyrene and peeled from the substrate, taking advantage of the surface-energy-assisted method 57 . Continuous films of hBN and WSe 2 monolayers were purchased from Graphene Supermarket and 6Carbon, respectively. Characterization Raman and photoluminescence measurements were performed using a Raman spectrometer HR-800 (Horiba BY) with 532 nm laser. Height profiles were measured using a CCi HD optical profiler. Electrical resistance measurements were performed in an ARS PSF-10-1-4 Cryogenic Probe Station using micromanipulators as probes (7×, Micromanipulator). Conductance measurements were performed by scanning the voltage from −0.1 to 0.1 V. Particle size distribution was measured using a nanoparticle tracking LM-10 Nanosight (Malvern). CSM fabrication To define polymer CSM bases, the first photolithography step was performed using the negative photoresist SU-8 2002 on SiO 2 /Si wafer (Supplementary Fig. 2 ). Next, a monolayer of MoS 2 (patterned in 25-µm-wide stripes using oxygen etching) with a PMMA support layer was transferred onto the structure with subsequent annealing at 80 °C for 1 h and 150 °C for 30 min. The wafer was then washed in acetone and ethanol and dried under nitrogen to remove the PMMA layer. In parallel, a monolayer of WSe 2 was prepared on a separate wafer: 25-µm-wide WSe 2 stripes were defined using the second photolithography step with the positive photoresist Shipley S1805 and subsequent oxygen etching. The photoresist was removed in RemoverPG developer. The patterned WSe 2 monolayer was then transferred onto MoS 2 using PMMA as the support layer, annealed at 80 °C for 1 h and 150 °C for 30 min. The wafer was washed in acetone and ethanol and dried under nitrogen. The third photolithography with a LOR30B sacrificial layer and a positive photoresist Shipley S1805 was used to define 40-nm-thick gold electrical contacts, which were deposited using a Denton electron-beam evaporator. The liftoff process was performed in Remover PG at 80 °C. The fourth photolithography step with a LOR30B sacrificial layer and Shipley S1805 was used to define the structure of a subsequent MoS 2 film. The MoS 2 film was deposited using a modified Langmuir–Blodgett method, where the MoS 2 film was collected at an ethanol–hexane interface 37 . To form the top oxide layer, the structure was annealed at 200 °C for 2 h. The liftoff process was performed in Remover PG at 80 °C. The fifth photolithography with a LOR30B sacrificial layer and Shipley S1805 was used to define 100-nm-thick silver electrical contacts. The liftoff process was performed in Remover PG at 80 °C. Monolayer hBN (patterned in 50-µm-wide stripes) was then transferred on top. Because hBN is transparent, its alignment on the CSM is very challenging. To this end, we transferred hBN with S1850 photoresist to help visualize the structure (the photoresist was removed afterwards). Finally, CSMs were coated with a PMMA layer for support and lifted off the substrate using KOH solution. To disperse CSMs, the PMMA was dissolved in acetone. Retroreflectors were fabricated using SU-8 photolithography with subsequent evaporation of 100 nm silver. Soot experiments We used commercially available Printex XE2-B soot. To make aerosol, we dispersed 2 mg of soot with 4 mg of BSA in 1 ml of deionized water. The mixture was sonicated with a 1/8′′-inch probe tip at 40% amplitude ( ∼ 12 W) for 30 min in an ice bath. For aerosol experiments, the mixture was diluted by 1,000×. Soot aggregates in the absence of the surfactant, inhomogeneously covering CSMs and blocking photodetectors. Aerosol experiments All aerosol experiments were performed with a Master Airbrush G22 nebulizer in a closed tube in the laminar flow hood. CSMs (dispersed in 80% water/20% ethanol) were sprayed under 2–15 psi pressure from a 300 µm nozzle. The second nebulizer was used to spray analyte droplets in the orthogonal direction (Supplementary Fig. 4 ). Aggregated CSMs, CSMs with visual defects, and CSMs that landed upside down were excluded from analysis. In a typical experiment, a portion of dispersed CSMs is pipetted onto the silicon substrate and is left to dry until all water evaporates naturally. Next, we query the initial CSM state with a probe station: both the memristor behaviours and the chemiresistor conductances on N = 100 CSMs are measured. The remainder of the CSMs are sprayed across in an enclosed tube. TEA is continuously sprayed using the second nebulizer from the top of the tube starting 5 s before and ending immediately after the CSMs are sprayed across (to eliminate interference of the two flows). CSMs are then left for 1 h inside to react with TEA and dry. We then place CSMs under the microscope and illuminate every photodiode individually for 5 s (532 nm laser, 7 µW µm − 2 ). Next, the final state of CSMs is assessed ( N = 100), repeating the same measurements as for the initial state. While for sprayed experiments these are random CSMs, we identify CSMs for on-substrate experiments, extracting individual changes on every CSM. Standoff detection Standoff CSM detection was performed using a set-up with galvanized mirrors, a 532 nm laser, and a photodetector (H10330a-25, Hamamatsu). Data availability The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

Researchers at MIT have created what may be the smallest robots yet that can sense their environment, store data, and even carry out computational tasks. These devices, which are about the size of a human egg cell, consist of tiny electronic circuits made of two-dimensional materials, piggybacking on minuscule particles called colloids. Colloids, which insoluble particles or molecules anywhere from a billionth to a millionth of a meter across, are so small they can stay suspended indefinitely in a liquid or even in air. By coupling these tiny objects to complex circuitry, the researchers hope to lay the groundwork for devices that could be dispersed to carry out diagnostic journeys through anything from the human digestive system to oil and gas pipelines, or perhaps to waft through air to measure compounds inside a chemical processor or refinery. "We wanted to figure out methods to graft complete, intact electronic circuits onto colloidal particles," explains Michael Strano, the Carbon C. Dubbs Professor of Chemical Engineering at MIT and senior author of the study, which was published today in the journal Nature Nanotechnology. MIT postdoc Volodymyr Koman is the paper's lead author. "Colloids can access environments and travel in ways that other materials can't," Strano says. Dust particles, for example, can float indefinitely in the air because they are small enough that the random motions imparted by colliding air molecules are stronger than the pull of gravity. Similarly, colloids suspended in liquid will never settle out. Researchers produced tiny electronic circuits, just 100 micrometers across,on a substrate material which was then dissolved away to leave the individual devices floating freely in solution. These were later attached to tiny colloidal particles. Credit: Massachusetts Institute of Technology Strano says that while other groups have worked on the creation of similarly tiny robotic devices, their emphasis has been on developing ways to control movement, for example by replicating the tail-like flagellae that some microbial organisms use to propel themselves. But Strano suggests that may not be the most fruitful approach, since flagellae and other cellular movement systems are primarily used for local-scale positioning, rather than for significant movement. For most purposes, making such devices more functional is more important than making them mobile, he says. Tiny robots made by the MIT team are self-powered, requiring no external power source or even internal batteries. A simple photodiode provides the trickle of electricity that the tiny robots' circuits require to power their computation and memory circuits. That's enough to let them sense information about their environment, store those data in their memory, and then later have the data read out after accomplishing their mission. Such devices could ultimately be a boon for the oil and gas industry, Strano says. Currently, the main way of checking for leaks or other issues in pipelines is to have a crew physically drive along the pipe and inspect it with expensive instruments. In principle, the new devices could be inserted into one end of the pipeline, carried along with the flow, and then removed at the other end, providing a record of the conditions they encountered along the way, including the presence of contaminants that could indicate the location of problem areas. The initial proof-of-concept devices didn't have a timing circuit that would indicate the location of particular data readings, but adding that is part of ongoing work. Similarly, such particles could potentially be used for diagnostic purposes in the body, for example to pass through the digestive tract searching for signs of inflammation or other disease indicators, the researchers say. As a demonstration of how such particles might be used to test biological samples, the team placed a solution containing the devices on a leaf, and then used the devices’ internal reflectors to locate them for testing by shining a laser at the leaf. Credit: Massachusetts Institute of Technology Most conventional microchips, such as silicon-based or CMOS, have a flat, rigid substrate and would not perform properly when attached to colloids that can experience complex mechanical stresses while travelling through the environment. In addition, all such chips are "very energy-thirsty," Strano says. That's why Koman decided to try out two-dimensional electronic materials, including graphene and transition-metal dichalcogenides, which he found could be attached to colloid surfaces, remaining operational even after after being launched into air or water. And such thin-film electronics require only tiny amounts of energy. "They can be powered by nanowatts with subvolt voltages," Koman says. Why not just use the 2-D electronics alone? Without some substrate to carry them, these tiny materials are too fragile to hold together and function. "They can't exist without a substrate," Strano says. "We need to graft them to the particles to give them mechanical rigidity and to make them large enough to get entrained in the flow." But the 2-D materials "are strong enough, robust enough to maintain their functionality even on unconventional substrates" such as the colloids, Koman says. The nanodevices they produced with this method are autonomous particles that contain electronics for power generation, computation, logic, and memory storage. They are powered by light and contain tiny retroreflectors that allow them to be easily located after their travels. They can then be interrogated through probes to deliver their data. In ongoing work, the team hopes to add communications capabilities to allow the particles to deliver their data without the need for physical contact. Other efforts at nanoscale robotics "haven't reached that level" of creating complex electronics that are sufficiently small and energy efficient to be aerosolized or suspended in a colloidal liquid. These are "very smart particles, by current standards," Strano says, adding, "We see this paper as the introduction of a new field" in robotics.

10.1038/s41565-018-0194-z

Medicine

Novel molecules to combat asthma and COVID-related lung diseases discovered

Microbial metabolism of l-tyrosine protects against allergic airway inflammation, Nature Immunology (2021). DOI: 10.1038/s41590-020-00856-3 , www.nature.com/articles/s41590-020-00856-3 Journal information: Nature Immunology

http://dx.doi.org/10.1038/s41590-020-00856-3

https://medicalxpress.com/news/2021-01-molecules-combat-asthma-covid-related-lung.html

Abstract The constituents of the gut microbiome are determined by the local habitat, which itself is shaped by immunological pressures, such as mucosal IgA. Using a mouse model of restricted antibody repertoire, we identified a role for antibody–microbe interactions in shaping a community of bacteria with an enhanced capacity to metabolize l -tyrosine. This model led to increased concentrations of p -cresol sulfate (PCS), which protected the host against allergic airway inflammation. PCS selectively reduced CCL20 production by airway epithelial cells due to an uncoupling of epidermal growth factor receptor (EGFR) and Toll-like receptor 4 (TLR4) signaling. Together, these data reveal a gut microbe–derived metabolite pathway that acts distally on the airway epithelium to reduce allergic airway responses, such as those underpinning asthma. Main Allergic asthma is a chronic airway disease characterized by the production of type 2 cytokines, synthesis of IgE, goblet cell metaplasia, influx of inflammatory cells and, ultimately, airway remodeling. Initiation of allergic asthma is a consequence of a dysregulated interplay between airway epithelium and immune cells, including dendritic cells (DCs), in response to allergen exposure 1 , 2 . In support of this, sensing of house dust mite (HDM) extract via TLR4 expressed on airway epithelial cells was shown to be necessary for the activation of pulmonary DCs and the initiation of allergic sensitization 3 , 4 . However, the immunomodulatory properties of other receptors (and ligands) upon epithelium-driven DC activation that could underpin differences in susceptibility to asthma remain obscure. A previous study identified the involvement of a deubiquitinating enzyme, A20, where its expression was reduced in airway epithelial cells of individuals with asthma 5 . Experimentally, induction of A20 by chronic exposure to endotoxin was sufficient to protect mice against allergic airway disease 5 . Exposure to endotoxins might mimic exposure to a rural environment, a setting associated with protection against asthma. Diet, including consumption of unpasteurized cow’s milk, fibers or breastfeeding, has been reported to confer similar benefits 6 . In the case of fibers, the effect relies on production of short-chain fatty acids in the gut that are released systemically and influence hematopoiesis 7 . Interestingly, both the farm effect and diet involve changes in the microbiome; yet, the impact of microbes or their metabolites on the airway epithelium is currently unknown. In addition to environmental cues, host-intrinsic factors contribute to diversification of the microbiota (a feature usually associated with health benefits 8 ), for example, via the generation of a diverse antibody repertoire 9 . We aimed to explore the impact of a restricted antibody repertoire on the composition of the gut microbiome and assess its capacity to influence airway inflammation. Results MD4 mice do not develop allergic airway disease To model a lack of antibody diversity, we utilized a B cell antigen receptor-transgenic mouse strain, ‘MD4’, with a restricted antibody repertoire such that >99% of its B cells are specific to a single antigen, hen egg lysozyme (HEL). Given the positive correlation between diversification of the antibody repertoire and a diverse microbiota 9 , a characteristic associated with health benefits 8 , we hypothesized that MD4 mice would have reduced microbial diversity and, consequently, an increased susceptibility to inflammation such as allergic airway inflammation, a mouse model of asthma. On the contrary, intranasal exposure of MD4 mice to HDM extract (as indicated in the Methods ) resulted in an almost complete absence of the allergic airway disease seen in wild-type controls, including eosinophilia (Fig. 1a ), recruitment and activation of pulmonary DCs (Fig. 1b ), mucus production (Fig. 1c ), peribronchial and perivascular inflammatory cell infiltrates (Fig. 1d ) and production of the type 2 helper T cell (T H 2)-associated cytokines interleukin (IL)-5 and IL-13 (Fig. 1e ). These immunological differences only manifested following HDM exposure, as no differences were observed in the immune profiles of these mouse strains at steady state. However, because the cellular composition of mediastinal lymph nodes differed markedly between wild-type and MD4 mice exposed to HDM (as a consequence of the lack of B cell proliferation in the latter), we sorted CD4 + T cells from the lymph nodes of both groups and co-cultured them with DCs in the presence of HDM. Consistent with these observations, we failed to detect type 2 cytokines in culture supernatants of MD4 T cells (Extended Data Fig. 1a ). Recruitment of CD4 + T cells was only moderately decreased, with a slight reduction in the proportion of FoxP3 + regulatory T (T reg ) cells (Fig. 1f ). B cell-deficient mice (J H T –/– ) mounted an allergic response similar to that seen in wild-type mice (Extended Data Fig. 1b ), indicating that the lack of an allergic response in the MD4 mouse strain was not due to the absence of antigen-specific B cells. Fig. 1: MD4 mice with a restricted antibody repertoire to HEL fail to mount allergic responses to HDM extract. a , Differential cell counts in the bronchoalveolar lavage fluid (BALF). Mac, macrophages; Neutr, neutrophils; Eos, eosinophils; Lymph, lymphocytes; WT, wild type. b , Total number of DCs in the lungs and their surface expression of PD-L2. GMFI, geometric mean fluorescence intensity. c , Representative periodic acid–Schiff (PAS)-stained lung tissue from WT or MD4 mice and quantification of the frequency of PAS + bronchi in histological sections. Scale bars, 100 μm; P = 0.0074. d , Representative hematoxylin and eosin (H&E)-stained lung tissue from WT or MD4 mice. Scale bars, 100 μm. e , Concentrations of IL-5 and IL-13 in culture supernatants of mediastinal lymph node cells restimulated with the indicated concentrations of HDM for 4 d; *** P = 0.0005 (IL-5), ** P = 0.0012 (IL-13). f , Total number of CD4 + T cells and the frequency of T reg cells (expressed as the percentage of CD4 + T cells) in the lung tissue; ** P = 0.0063 (CD4 + T cells), * P = 0.0495 (T reg cells). g , Principal-coordinate analysis (PCoA) plot (based on Bray–Curtis distance) of the bacterial communities (as determined by sequence analysis of 16S rRNA gene amplicons) in WT and MD4 fecal samples. PC, principal coordinate. All data except in d and g are expressed as the mean ± s.e.m. (error bars shorter than the size of the symbols in e are not depicted). Data in a and f are pooled from four experiments ( n = 19 biologically independent samples per group), data in b are pooled from three experiments ( n = 14 biologically independent samples per group), data in c and d are representative of two experiments ( n = 8 biologically independent samples per group), data in e are pooled from two experiments ( n = 9 biologically independent samples per group) and data in g are pooled from five experiments ( n = 37 MD4 and n = 29 WT biologically independent samples). Statistical significance for a – c and f was evaluated by two-sided unpaired Student’s t -test (in the case of Gaussian distribution) or Mann–Whitney test (non-Gaussian distribution). Statistical significance for e was determined with two-way ANOVA with Šidák’s correction for multiple comparisons. Data distribution was assessed with the D’Agostino–Pearson normality test. Statistical significance for g was evaluated with an ANOSIM controlling for experimental variation. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, **** P ≤ 0.0001. Full size image MD4 microbiota protects against allergic inflammation MD4 mice did not have major alterations in the diversity of the microbiota (Extended Data Fig. 2 ), but substantial differences in composition were evident (ANOSIM, F = 58.28, R 2 = 0.42, P < 0.001) (Fig. 1g ). To evaluate whether the microbiota contributed to the observed protection against allergic inflammation, we co-housed germ-free mice with either wild-type or MD4 mice for 6 weeks, after which they were exposed to HDM. Sequencing of 16S rRNA gene amplicons from fecal DNA confirmed acquisition of the MD4 microbiota by co-housed germ-free mice (GF-MD4) (ANOSIM, F = 19.69, R 2 = 0.45, P < 0.001) (Fig. 2a ). Transfer of the MD4 microbiota ameliorated allergic responses, including airway eosinophilia (Fig. 2b ), lung pathology (Fig. 2c ), mucus production (Fig. 2d ), secretion of T H 2-associated cytokines (Fig. 2e ) and production of HDM-specific antibodies (Fig. 2f ). However, there were no alterations in the recruitment of CD4 + T cells or the proportion of FoxP3 + T reg cells (Fig. 2g ). Collectively, these experiments demonstrate that the MD4 microbiota was sufficient to confer protection against HDM-induced allergic airway inflammation. Fig. 2: Microbiota of MD4 mice confers protection against HDM-induced allergic airway inflammation. a , PCoA plot (Bray–Curtis distance) of the bacterial communities (16S rRNA gene amplicons) in mouse feces. b , Differential cell counts in the BALF; * P = 0.0306. c , Representative H&E-stained lung tissue from GF-WT or GF-MD4 mice. Scale bars, 100 μm. d , Representative PAS-stained lung tissue from GF-WT or GF-MD4 mice and quantification of the frequency of PAS + bronchi in histological sections; * P = 0.0102. Scale bars, 100 μm. e , Cytokine concentration in culture supernatants of mediastinal lymph node cells restimulated with the indicated concentrations of HDM for 4 d; *** P = 0.0004 (IL-13). f , Levels of HDM-specific IgG1 antibodies in the serum; * P = 0.0235. g , Total number of CD4 + T cells and the frequency of T reg cells (percentage of CD4 + T cells) in the lungs. Results are pooled from two experiments ( n = 10 biologically independent samples per group), except for data in d and f , where data are representative of two experiments ( n = 5 biologically independent samples per group). All data except in a and c are presented as mean values ± s.e.m. (error bars shorter than the size of the symbols in e are not depicted). Statistical analysis was performed as described in Fig. 1 . * P ≤ 0.05, *** P ≤ 0.001, **** P ≤ 0.0001. Full size image Changes in the microbiome impact the metabolome of the host Next, we sought to analyze the microbiota composition of the MD4 mice in depth. 16S rRNA gene amplicon sequencing revealed 41 bacterial taxa with increased abundance and 14 with decreased abundance in this mouse strain (Fig. 3a ); 7 of these 41 taxa, and 0 of the 14, were coated with secretory IgA (Fig. 3b , blue nodes), which was found in high abundance in MD4 feces and displayed specificity to HEL (Extended Data Fig. 3 ). MD4 IgA binding showed no overlap with that of wild-type mice (Fig. 3b , black nodes). IgM, which was highly abundant in MD4 feces (Extended Data Fig. 3 ), showed a similar binding pattern, coating five of the seven IgA-coated taxa (Extended Data Fig. 4 ). The exact nature of the antibody–microbe interactions is not clear; they may be facilitated by the cross-reactivity of antigen-binding (Fab) regions of anti-HEL or by non-Fab-dependent affinities (for example, that of a secretory component). Although the mechanisms remain to be fully elucidated, it is clear that restricting the antibody repertoire to HEL alters the microbial community. Because gut microbes can have distal immunomodulatory effects through the release of metabolites into the circulation 10 , 11 , we sought to assess the metabolome of MD4 mice. We performed untargeted plasma metabolomic profiling and identified PCS as the metabolite with the strongest enrichment (limma, log fold change (logFC) = 3.1, adjusted P = 1.42 × 10 –6 ) in MD4 mice (Fig. 3c ). PCS is a sulfation product of p -cresol (Fig. 3d ), the intestinally generated, microbial-derived product of l -tyrosine metabolism 12 . p -Cresol sulfation takes place in the mucosa of the colon and in the liver 13 , 14 , 15 . In line with a microbial origin of PCS, germ-free mice co-housed with MD4 mice also showed increased concentrations of PCS (Fig. 3e ), suggesting that the MD4 microbiota had a superior capacity to utilize l -tyrosine from the diet. Indeed, shotgun metagenomics analyses of fecal samples revealed that 2-iminoacetate synthase ( thiH ) genes, which encode enzymes involved in the direct conversion of l -tyrosine to p -cresol, were more abundant in the MD4 microbiota (logFC = 3.9, adjusted P = 5.54 × 10 –4 and logFC = 3.2, adjusted P = 5.62 × 10 –7 ) (Fig. 3f ). One of these genes mapped to the genome of the Prevotella MGM1 species, which was highly abundant in the MD4 mice (Extended Data Fig. 5 ), while the other was from an unidentified source. An alternative pathway for production of p -cresol from l -tyrosine involves the bacterial genes tyrB , fldH , porA , fldBC , acdA and hpd 16 (Extended Data Fig. 6a ). In our dataset, we did not detect tyrB , fldH or porA , while fldBC and hpd were not differentially abundant between the groups. Sequences mapping to acdA were found to be differentially abundant in both wild-type and MD4 groups. However, differentially abundant acdA genes in wild-type samples related to different putative proteins (AcdA C-terminal domain) than the ones enriched in MD4 feces (Extended Data Fig. 6b ). Given that AcdA is involved in multiple pathways, not just those leading to p -cresol, we conclude that ThiH, which metabolizes l -tyrosine to PCS in a single step, is the metabolic enzyme relevant in our model. The feces of MD4 mice were also found to contain less l -tyrosine, which supports the conclusion that the microbiota of these mice exhibited enhanced metabolism of this amino acid in the gut (Fig. 3g ). Fig. 3: Antibodies in MD4 mice shape the microbiome and the metabolome of the host. a , A heat map representing differentially abundant amplicon sequence variants (ASVs) between MD4 and WT mice using a zero-inflated Gaussian mixture model controlling for experimental variation. MD4 IgA-bound hits analyzed in b are shown in bold. b , Correlation inference network with bacterial taxa bound by anti-HEL IgA (annotated). Blue and black nodes represent taxa differentially abundant in MD4 and WT mice, respectively, while open nodes represent non-differentially abundant hits. Node size is proportional to the IgA binding index calculated from IgA + and IgA – fractions. c , Volcano plot depicting the differential abundance of plasma metabolites between WT and MD4 mice using limma parametric empirical Bayes (eBayes) testing. The y axis represents the –log 10 adjusted P value (with dashed line at α = 0.05), while the x axis represents the log 2 FC (dashed line at twofold change). n.s., not significant. d , Pathway of l -tyrosine conversion to PCS by ThiH. e , Levels of PCS in WT, MD4 and germ-free mice co-housed with WT or MD4 mice; * P = 0.0235 (WT versus MD4), * P = 0.0155 (GF-WT versus GF-MD4). f , Volcano plot representing differences in bacterial gene abundance between WT and MD4 mice. n.s., not significant. g , Levels of l -tyrosine in the feces of WT and MD4 mice; ** P = 0.0041. Data in a are pooled from five experiments ( n = 37 MD4 and n = 27 WT biologically independent samples), IgA binding data in b represent analysis from three independent sorting experiments ( n = 3 biologically independent samples per group), data in c are from one experiment ( n = 8 biologically independent samples per group), data in e are representative of two experiments ( n = 5 biologically independent samples per group), data in f represent samples with the highest-quality DNA from four pooled experiments ( n = 11 MD4 and n = 9 WT biologically independent samples), and data in g are pooled from two independent experiments ( n = 9 MD4 and n = 11 WT biologically independent samples). Data in e and g are presented as mean values ± s.e.m. Statistical analysis was performed as described in Fig. 1 . * P ≤ 0.05, ** P ≤ 0.01. Full size image PCS or l -tyrosine ameliorates allergic airway disease Because the data pointed to the enrichment of the l -tyrosine–PCS pathway in MD4 mice, which were protected against experimental asthma, we sought to investigate the direct influence of PCS or l -tyrosine treatment on allergic airway inflammation. Wild-type C57BL/6J mice received intravenous injection of PCS or saline before HDM sensitization and challenge (as indicated in the Methods ). This treatment ameliorated airway eosinophilia (Fig. 4a ), decreased infiltration of DCs into the lungs (Fig. 4b ) and reduced the production of IL-5 and IL-13 by restimulated mediastinal lymph node cells (Fig. 4c ). As seen in the MD4 mice, there were no major alterations in the numbers of lung CD4 + T cells or FoxP3 + T reg cells (Fig. 4d ). Oral administration of l -tyrosine to wild-type C57BL/6J mice (as indicated in the Methods ) led to an increase of PCS concentrations in the feces and airways (Extended Data Fig. 7 ) and conferred similar effects, including reduced recruitment of eosinophils, neutrophils and DCs (Fig. 4e,f ), and reduced the production of IL-13 (Fig. 4g ). IL-5 concentrations in culture supernatants of mediastinal lymph node cells showed a trend toward a decrease (Fig. 4g ), albeit not reaching statistical significance. The numbers of helper T cells or FoxP3 + T reg cells in the lungs were not altered (Fig. 4h,i ). Antibiotic treatment (as in Extended Data Fig. 8a ) abrogated the protective effect of l -tyrosine (Extended Data Fig. 8b–d ). Fig. 4: Administration of PCS or l -tyrosine confers protection in an HDM model of asthma. a , Differential cell counts in the BALF of vehicle- or PCS-treated mice (as indicated in the Methods ); ** P = 0.0052. b , Total number of DCs in the lungs; *** P = 0.0007. c , Cytokine concentration in culture supernatants of mediastinal lymph node cells restimulated with HDM for 4 d; *** P = 0.0004 (IL-5), ** P = 0.0025 (IL-13). d , Total number of CD4 + T cells and the frequency of T reg cells (percentage of CD4 + T cells) in the lungs. e , Differential cell counts in the BALF of vehicle- or l -tyrosine-treated mice (as indicated in the Methods ); * P = 0.018 (neutrophils), ** P = 0.0019 (eosinophils). f , Total number of DCs in the lungs; *** P = 0.0008. g , Cytokine concentration in culture supernatants of mediastinal lymph node cells restimulated with HDM for 4 d; * P = 0.0185 (IL-13). h , i , Total number of CD4 + T cells ( h ) and the frequency of T reg cells (percentage of CD4 + T cells) ( i ) in the lungs. Results in a and b are pooled from four experiments ( n = 18 biologically independent samples per group), results from c are pooled from three experiments ( n = 14 biologically independent samples per group), data from d – h are pooled from two experiments ( n = 9 biologically independent samples per group), and data in i are representative of one experiment ( n = 5 biologically independent samples per group). All data are presented as mean values ± s.e.m. Statistical analysis was performed as described in Fig. 1 . * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001. Full size image The l -tyrosine–PCS axis inhibits CCL20 production in the lung We aimed to determine the mechanisms behind the protective effects of l -tyrosine and PCS treatment. Intranasal administration of fluorescently labeled HDM into l -tyrosine-treated mice revealed an impairment in the activation of their lung DCs (Fig. 5a ), with only a modest effect on antigen uptake (Fig. 5b ). The migratory capacity of DCs was decreased, as shown by a reduced frequency of HDM + DCs in the draining lymph nodes (Fig. 5c ). In addition, the capacity of DCs to prime naive CD4 + T cells or restimulate in vivo-primed effector helper T cells into an IL-13-producing subset was impaired (Fig. 5d ). To gain insights into how oral supplementation of l -tyrosine influenced DC function, we screened the activity of PCS on chemokine release from HDM-stimulated lung cells isolated using a protocol for epithelial cell enrichment. Briefly, lung cells from naive C57BL/6J mice were isolated using 1% low-melting agarose/dispase II solution, plated on fibronectin-coated plates and stimulated with HDM in the presence or absence of PCS (as indicated in the Methods ). PCS completely abrogated HDM-induced production of the airway epithelial cell–derived DC chemoattractant CCL20 (refs. 17 , 18 ) but did not have an effect on other chemokines (Fig. 5e ). A similar observation was noted in lung cells isolated from MD4 mice (Extended Data Fig. 9a ). Because HDM induced low amounts of CCL20, we next used lipopolysaccharide (LPS), a known potent inducer of CCL20, to further evaluate the efficacy of PCS. LPS induced a 15-fold upregulation of CCL20, which was inhibited by PCS (Fig. 5f ). Consistent with these in vitro data, CCL20 concentrations were reduced in the BALF of l -tyrosine-treated wild-type mice (Fig. 5g ) and MD4 mice (Extended Data Fig. 9b ) exposed to HDM. Because CCL20 function is not restricted to type 2 immunity and may influence a broader range of immune responses, we tested the efficacy of PCS in the context of type 1-mediated lung immunopathology. Mice were administered ovalbumin and LPS intranasally on days 0, 11, 12 and 13 and received PCS intravenously on day –1 and on day 11 4 h before the first ovalbumin and LPS challenge (Extended Data Fig. 10a ). PCS inhibited the infiltration of neutrophils and CD4 + and CD8 + T cells to the airways (Extended Data Fig. 10b ). Next, we sought to delineate the molecular mechanisms of action behind PCS. Molecular simulation and docking analysis indicated that PCS could bind in the interdomain pocket of EGFR, just beneath the EGF-binding site 19 . EGFR is required for optimal signal transduction downstream of TLR4 (refs. 20 , 21 ) by facilitating the recruitment of Lyn to both receptors 20 . This finding led us to hypothesize that PCS inhibits CCL20 production via uncoupling TLR4 and EGFR cross-talk. To test this, we stimulated lung cells with LPS in the presence of EGFR ligands (high affinity (EGF) and low affinity (amphiregulin)) or in the presence of an EGFR inhibitor, gefitinib. As in the case of PCS, all treatments led to a selective reduction in CCL20 production (Fig. 5h ), recapitulating the effect of PCS, albeit with lower efficacy. These data highlighted the importance of unbound EGFR for TLR4-mediated production of CCL20 in response to LPS. Fig. 5: The l -tyrosine–PCS axis modulates DC activation via inhibition of epithelial cell–derived CCL20. a , Surface expression of CD80, CD86 and PD-L2 on HDM + and HDM – populations of lung DCs. b , HDM uptake in vivo by lung DCs from vehicle- or l -tyrosine-treated mice; * P = 0.0367. c , Migration of lung DCs to lung-draining lymph nodes; ** P = 0.0014. MLN, mediastinal lymph node. d , Capacity of pulmonary DCs from l -tyrosine- or vehicle-treated groups to prime OT-II cells from a naive mouse into an IL-13-producing subset (left) or to restimulate effector helper T cells from HDM-treated mice as in Fig. 1 (right). e , Capacity of PCS to modulate HDM-induced secretion of chemokines from lung cells isolated from naive mice. Unstim, unstimulated. f , Capacity of PCS to inhibit CCL20 secretion from LPS-stimulated lung cells from naive mice; ** P = 0.0079. g , Concentration of CCL20 in the BALF of mice treated as described in Fig. 4e ; * P = 0.032. h , CCL20 production by lung cells stimulated with LPS in the presence of PCS (*** P = 0.0005), EGF (* P = 0.043), amphiregulin (AREG; * P = 0.0278) or gefitinib (Gef; ** P = 0.0053). Data from a – c are pooled from two experiments ( a and b , n = 10 biologically independent samples per group; c , n = 9 biologically independent samples per group). Data in d represent technical replicates ( n = 6, left; n = 5, right) from one experiment. Data in e are pooled from three independent experiments ( n = 6 per group). Data in f are pooled from two independent experiments ( n = 5 biologically independent samples per group). Data in g are pooled from two independent experiments ( n = 9 biologically independent samples per group). Data in h are pooled from two independent experiments ( n = 4 biologically independent samples per group). All data are presented as mean values ± s.e.m. Statistical significance for a – d , f and g was evaluated by unpaired Student’s t -test (in the case of Gaussian distribution) or Mann–Whitney test (non-Gaussian distribution). Statistical significance for e and h was determined by ANOVA with Dunnett’s correction for multiple comparisons. Data distribution was assessed by D’Agostino–Pearson normality test. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, **** P ≤ 0.0001. Full size image Discussion Collectively, our data provide new insights into several aspects of host–microbe interactions in the context of lung inflammation. First, we identified l -tyrosine–PCS as a new pathway protecting against airway inflammation. Although these data particularly highlighted the beneficial effects of l -tyrosine and PCS in reducing type 2 inflammation, such as that underpinning atopic asthma, it is noteworthy that they were also efficacious against neutrophilic inflammation, which is a hallmark of type 17 responses. The potential of this pathway for preventing, or treating, inflammatory diseases beyond these models remains to be explored; however, it is tempting to speculate that l -tyrosine and PCS might exhibit efficacy in other inflammatory conditions where TLR4 and/or CCR6–CCL20 signaling is critical. Although PCS has been reported as a uremic toxin in individuals with chronic kidney disease 22 , it is present in the blood of healthy individuals and, notably, is one of the very few metabolites associated with protection against childhood asthma 23 , 24 . In Project Viva, a birth cohort from Massachusetts, involving 237 children (46 children with asthma and 191 controls), PCS was one of six plasma metabolites inversely correlated with asthma (odds ratio (OR) 0.47; 95% confidence interval (CI), 0.23–0.94) 23 . This observation was replicated in a larger cohort involving 411 children, of whom 26% developed asthma by age 3 (VDAART clinical trial); PCS was one of two plasma metabolites inversely associated with asthma (OR 0.83; 95% CI, 0.69–1.00) 23 . A similar observation was made when analyzing fecal samples; PCS was one of five metabolites inversely correlated with asthma (OR 0.56; 95% CI, 0.40–0.79; P = 0.001). In the same study, exclusive breastfeeding for the first 4 months of life inversely correlated with asthma (OR 0.36; 95% CI, 0.18–0.67; P = 0.002) and PCS, which positively correlated with breastfeeding, explained 17.3% of this effect 24 . In mice with normal renal function, intraperitoneally or orally administered PCS is cleared from the blood within 4 h (refs. 25 , 26 ), and chronic administration of PCS (twice a day for 4 weeks) does not lead to its accumulation 26 . Hence, it is likely that PCS has detrimental effects only when kidney function is impaired or that it is primarily a biomarker of this condition. Of note, in our settings we did not observe an elevation of major kidney toxicity markers upon PCS treatment. Thus, PCS and l -tyrosine hold promise to be developed as novel therapeutics against inflammatory diseases, although their safety and efficacy need to be carefully evaluated. Second, although gut microbe–derived metabolites have been reported to influence cell responses beyond the intestine, for example, by influencing hematopoiesis in the bone marrow 7 , 10 , 27 , their capacity to interact with the airway epithelium has not been reported. The mechanism of action of PCS is linked to its capacity to uncouple cross-talk between TLR4 and EGFR, which are known to synergize for optimal signal transduction 20 , 21 . It is intriguing that PCS along with other EGFR ligands (EGF and AREG) and an inhibitor (gefitinib) selectively reduced secretion of CCL20 and not other chemokines. CCL20 is the only known ligand for CCR6, and the CCL20–CCR6 axis was previously identified as an important driver of allergic inflammation in response to cockroach antigens. CCR6-deficient mice had attenuated peribronchial accumulation of eosinophils, exhibited lower pulmonary concentrations of IL-5 and systemic concentrations of IgE, and displayed reduced airway hyperreactivity after challenge with cockroach allergen 28 . This defect was also reported in a subsequent study and linked to reduced infiltration of DCs into lung tissue 29 , which is consistent with our findings. Of note, CCL20 has recently been identified as a ligand for the scavenging atypical chemokine receptor 4 (ACK4) 30 . ACK4 is expressed on epithelial cells, indicative of an autocrine function of the CCL20–ACK4 axis during inflammatory responses 31 , although implications for asthma remain unexplored. Molecular details behind the reduction in CCL20 levels by PCS are yet to be determined, but this might rely on selective inhibition of Akt phosphorylation, as shown for gefitinib 21 . Of note, the superior capacity of PCS to suppress CCL20 might reflect a different mode of binding to EGFR or a capacity to interact with other targets. Third, the enhanced l -tyrosine–PCS pathway observed in MD4 mice is linked to changes in their microbiota, the transfer of which protected germ-free recipients against experimental asthma. This was linked to an enrichment of two variants of a single gene, encoding ThiH, catalyzing the conversion of l -tyrosine to p -cresol 32 . One of these variants was derived from the Prevotella MGM1 species, an uncultured species with as of yet undescribed immunomodulatory functions 33 . Fourth, it is intriguing to note that secretory antibodies in the MD4 mice, although monoclonal, bound some of the microbes found to be enriched in the gut. Secretory IgA binding to bacteria has classically been considered in the context of ‘immune exclusion’, whereby antibody binding to potentially deleterious bacteria prevents their survival, growth or spread within the host 34 , 35 , 36 . However, some studies indicated the capacity of secretory antibodies to coat many non-pathogenic or even beneficial bacteria 37 , 38 , 39 , 40 , 41 , 42 , suggesting their possible role in positive selection of certain species, for example, via embedding them in nutrient-rich mucus 38 . Indeed, a recent report demonstrated the ability of Bacteroides fragilis to modulate its surface antigens to facilitate binding by IgA, which promoted close association of these bacteria with the gut mucosal surface 43 . Similarly, enhanced secretory IgA coating in mice deficient in the ATP-gated ionotropic receptor P2rx7 led to enrichment of intestinal Lactobacillus 44 . These observations indicate that the functional consequences of antibody binding for the selection of different bacterial species may differ, likely depending on various factors, including the targeted epitope (for example, its density on the bacterial cell surface or its function) or the antibody itself (for example, affinity, subclass and glycosylation profile). The exact nature of antibody-mediated positive selection in MD4 mice remains to be explored. It might be facilitated by the cross-reactivity of the Fab region or through non-Fab-mediated interactions (the secretory component or Fc region). In addition to IgA/IgM-coated taxa, MD4 mice had increased abundance of several uncoated taxa, consistent with the idea of IgA playing a role in shaping bacterial community networks 45 . Finally, it is worth noting that not all changes found in MD4 mice after HDM exposure were recapitulated by microbial transfer or the metabolite treatments. These include changes in the CD4 + T cell compartment, affected in the transgenic mouse line but not in germ-free mice recolonized with the MD4 microbiota or wild-type mice treated with PCS or l -tyrosine. The exact reasons underlying this phenomenon are not known. It is possible that changes in the CD4 + T cells of MD4 mice are due to long-term exposure to PCS or might be independent of the microbiome/metabolome and rather due to altered B cell–T cell interactions in these mice. Collectively, our work identifies a new pathway within the gut–lung axis, which could be targeted therapeutically to treat or prevent inflammatory diseases, such as atopic or severe neutrophilic asthma. Methods Experimental animals MD4 B cell-receptor-transgenic mice (on a C57BL/6J background) were originally obtained from the Institute for Research in Biomedicine in Bellinzona, Switzerland, or rederived from frozen sperm originally provided by Australian National University, Canberra, Australia, at Monash Animal Research Platform at Clayton, Victoria, Australia. C57BL/6J wild-type mice were originally obtained from Charles River Laboratories or Monash Animal Research Platform. J H T –/– mice (on a C57BL/6J background) were obtained from École Polytechnique Fédérale de Lausanne (EPFL), Switzerland. OT-II mice were obtained from Monash Animal Research Platform. All mice were bred and maintained under specific-pathogen-free conditions and fed irradiated WEHI Mice Cubes (Barastoc, 8720610). Heterozygous breeding of MD4 mice was set up using non-transgenic (wild-type) C57BL/6 females mated with MD4 males (±tg). Given allelic exclusion in B cells and the strong promotor in the HyHEL10 construct in MD4 cells, this breeding strategy is the only way to generate both MD4 and wild-type littermate controls. Germ-free mice (C57BL/6J background) were obtained from the Clean Mouse Facility, University of Bern, Bern, Switzerland. Six- to twelve-week-old mice were used for all experiments except for l -tyrosine and PCS treatments, which were initiated at the age of 3 weeks. All animal experiments were performed in accordance with institutional guidelines of Lausanne University Hospital and Monash University. l -Tyrosine and p -cresol sulfate treatments in vivo Female mice received water as a control or l -tyrosine (reagent grade, ≥98%; Sigma-Aldrich) resuspended in drinking water under sterile conditions at the concentration of 100 mg kg –1 per day or 500 mg kg –1 per day (based on the assumption of a mouse consuming 4 ml of water daily). All mice received the treatment 2 weeks before and throughout the experiment. PCS (Alsachim) was resuspended in saline under sterile conditions and delivered via injection into the right retro-orbital sinus at a dose of 40 mg kg –1 in a volume of 200 μl 1 d before the first HDM exposure (day –1) and 4 h before the second HDM exposure (day 11). Control mice received 200 μl of saline. For experiments using antibiotic treatment, 3-week-old mice were treated with a combination of enrofloxacin (10 mg kg –1 per day) and amoxicillin with clavulanic acid (1 mg kg –1 per day) for 1 week, followed by 1 week of only amoxicillin with clavulanic acid (1 mg kg –1 per day). Mice were then put on an l -tyrosine diet in drinking water for 2 weeks followed by HDM exposure as before. During this time, mice were maintained on antibiotic treatment with amoxicillin/clavulanic acid until the end of the experiment. Animal model of allergic airway inflammation Mice were anesthetized by inhalation of 4% isoflurane in oxygen for 3–5 min. House dust mite extract (HDM) (Greer Laboratories) was reconstituted in sterile phosphate-buffered saline (PBS) (GibcoTM) at the concentration of 1 μg protein content per μl. 20 μl of this solution was applied intranasally on days 0, 11, 12 and 13. Mice were humanely killed with a lethal dose of pentobarbital (Streuli Pharma) on day 14. Animal model of pulmonary type I inflammation Mice were anesthetized by inhalation of 4% isoflurane in oxygen for 3–5 min. Ovalbumin (100 μg; Invivogen, vac-pova) was mixed with 10 μg of LPS from Escherichia coli (Sigma-Aldrich, L4391) and administered intranasally on days 0, 11, 12 and 13. Mice were killed with a lethal dose of pentobarbital on day 14. Cellular infiltration of the airways BALF was collected by flushing airways with 0.5 ml PBS supplemented with 0.2% BSA (Sigma-Aldrich). Total cell number was determined with Coulter Counter (IG Instrumenten-Gesellschaft), while differential cell staining was performed on cytospins stained with Diff-Quik solution (Dade Behring, Siemens Healthcare Diagnostics). Percentages of neutrophils, macrophages, lymphocytes and eosinophils were assessed by counting 200 cells per sample. ELISA Concentrations of IL-4, IL-5, IL-13, IL-17 and IFN-γ in culture supernatants were analyzed with mouse Ready-Set-Go! ELISA kits (eBioscience) according to the manufacturer’s instructions. To measure the titers of HDM-specific IgG1 and IgE in plasma, half-area 96-well plates (Corning) were coated with HDM (10 μg in PBS) overnight at 4 °C followed by incubation of samples for 2 h at 25 °C and the addition of alkaline phophatase-conjugated goat anti-mouse IgE or IgG1 (SouthernBiotech; 1 μg ml −1 in PBS and 0.2% BSA) for 2 h at 25 °C. 4-Nitrophenyl phosphate sodium salt hexahydrate (Sigma) was used as a substrate, and the colorimetric reaction was read at 405 nm on the Synergy H1 microplate reader (Biotek). To measure the concentrations of total IgA and IgM as well as HEL-specific IgA and IgM in mouse feces, fecal pellets were homogenized in 0.8 ml cold PBS, centrifuged at 400 g for 5 min to remove large debris, filtered through a 40-μm cell strainer and centrifuged at 8,000 g for 10 min to pellet bacteria. The supernatant was collected and loaded for 2 h at 25 °C on 96-well half-area plates coated a day before at 4 °C with anti-IgA (SouthernBiotech; 2 μg ml −1 ), anti-IgM (SouthernBiotech; 2 μg ml −1 ) or HEL protein (Sigma-Aldrich; 10 μg ml –1 ). This step was followed by the addition of alkaline phophatase-conjugated goat anti-mouse IgA or IgM (both at 1 μg ml −1 in PBS and 0.2% BSA) for 2 h at 25 °C. 4-Nitrophenyl phosphate sodium salt hexahydrate (Sigma) was used as a substrate, and the colorimetric reaction was read at 405 nm using SKANIT Software 3.2. Tyrosine assay Fresh feces were collected, homogenized in distilled water and centrifuged at 8,000 g for 5 min at 4 °C. The supernatant was filtered through a 40-μm cell strainer and deproteinized using 10-kDa spin columns (Abcam). l -Tyrosine concentrations were measured with a tyrosine assay kit (Abcam) according to the manufacturer’s instructions. Intestinal permeability assay Mice were water starved overnight and were administered fluorescein isothiocyanate (FITC)–dextran by oral gavage at 0.44 mg per gram of body weight. Six hours later, mice were killed, blood was collected and FITC–dextran concentrations were measured via fluorescence spectrophotometry. Kidney toxicity markers Concentrations of cystatin, clusterin, lipocalin-2 and osteopontin in the serum of vehicle- or PCS-treated mice were determined with a MILLIPLEX MAP Mouse Kidney Injury Magnetic Bead Panel 2, Toxicity Multiplex Assay (Merck) according to the manufacturer’s instructions. Flow cytometry Mediastinal lymph nodes were filtered through a 40-μm cell strainer and washed with PBS supplemented with 1% FBS and 2 mM EDTA (Invitrogen) (MACS buffer). Lungs were finely cut with scissors, digested with collagenase IV (Gibco) in Iscove’s modified Dulbecco’s medium (IMDM; Gibco) for 50 min at 37 °C and processed as the lymph nodes. Cell counts were determined with Coulter Counter, and 10 5 cells were stained with freshly prepared antibody mix in MACS buffer for 20 min at 4 °C in a 96-well round-bottom plate (Costar). Bone marrow was collected from left legs and stained for progenitor cells using antibodies against Ly-6A/E–phycoerythrin(PE)/Cy7 (BioLegend, 108113; 1:800), CD117–Alexa Fluor 647 (AF647) (BioLegend, 105817; 1:400), CD16/32–APC/Cy7 (BioLegend, 101327; 1:400), CD34–PerCP/Cy5.5 (BioLegend, 128607; 1:300), CD135–PE (BioLegend, 135305; 1:400), CD11c–FITC (Thermo Fisher Scientific, 11-0114-81; 1:400), I-A/IE–AF700 (BioLegend, 107622; 1:1,000) and a Lineage Cocktail-Pacific Blue (PB) (BioLegend, 133310; 1:50). DCs were identified using monoclonal antibodies against CD11c–PE/Cy7 (BioLegend, 117318; diluted 1:400 in MACS buffer), SiglecF–AF647 (BD Biosciences, 562680; 1:400) and MHC-II–AF700 (BioLegend, 107622, 1:800). DC activation was assessed using antibodies against PD-L2–PE (BioLegend, 107205; 1:200), CD80–Brilliant Violet 605 (BV605) (BioLegend, 104729; 1:200) and CD86–BV650 (BioLegend, 105035; 1:200). Helper T cells were identified using antibodies against CD3ε–PB (BioLegend, 100214; 1:800) and CD4–PerCP/Cy5.5 (BioLegend, 100434; 1:800). Activated helper T cells were identified using anti-CD44–PE (BD Biosciences, 553134; 1:400). A regulatory subset of helper T cells (T reg ) was identified with the addition of anti-CD25–AF700 (BioLegend, 102024; 1:200) and anti-FoxP3–AF647 (BioLegend, 126408; 1:200). For the latter, intracellular staining was performed where cells were incubated with anti-FoxP3 diluted in 0.5% saponin from Quillaja bark (Sigma-Aldrich) for 40 min at 4 °C. B cells were identified with anti-CD19–PE/Cy7 (eBioscience, 25-0193; 1:200) and anti-B220–FITC (BioLegend, 103206; 1:200). When indicated, HEL and HDM were labeled with an AF647 antibody labelling kit (Invitrogen) and separated from the unlabeled dye with the use of a PD-10 desalting columns (GE Healthcare). Both HEL–AF647 and HDM–AF647 were used for extracellular staining at a dilution of 1:200 in MACS buffer. Cells were acquired on a BD Fortessa (BD Biosciences). Samples were acquired with fluorescence-activated cell sorting (FACS) DiVA v6.3.1 software and analyzed with FlowJo v10.4.2 and v10.6.1 software (Tree Star). Histology Right lung lobes were fixed in 10 ml of 10% buffered formalin at 4 °C and embedded into paraffin. Prepared sections (4 μm) were stained with either H&E or PAS reagents using standardized protocols and analyzed with an Axioskop 2 plus microscope equipped with an Axio-Cam HRc (Carl Zeiss Microimaging). In vivo tracking of dendritic cells Mice were administered 20 μg of HDM–AF647 in 20 μl PBS intranasally. DC antigen uptake, activation and migration to lung-draining lymph nodes were analyzed by flow cytometry. Ex vivo restimulation assay Mediastinal lymph node suspensions were filtered through a 40-μm cell strainer, washed and resuspended in IMDM supplemented with 10% FBS, 1% penicillin/streptomycin (Invitrogen) and 0.05 mM 2-mercaptoethanol (Gibco). Cells were plated in 96-well round-bottom culture plates (Costar) at a density of 10 5 cells per well in the presence of HDM (0–50 μg ml –1 based on protein content) and cultured for 4 d at 37 °C and 5% CO 2 , after which the supernatants were collected for cytokine quantification. DC–T cell co-cultures were set up by sorting CD11c + SiglecF – (DCs) and CD4 + CD44 + T cells from the lungs of HDM-immunized mice on a FACSAria III (BD Biosciences). DCs (5,000 per well) and T cells (10,000 per well) were plated in 96-well round-bottom culture plates and stimulated with HDM (40 μg ml –1 ) for 4 d, after which the supernatants were collected. Bacterial cell sorting Fresh feces was homogenized in ice-cold PBS, filtered through 40-μm cell strainers and centrifuged at 400 g for 5 min at 4 °C. The supernatant was collected, diluted threefold with PBS and 1% BSA and centrifuged at 400 g for 5 min at 4 °C. This step was repeated twice to remove debris and mammalian cells. Bacterial cells were spun down at 8,000 g for 5 min at 4 °C and stained with SYTO BC (1:8,000) for 30 min at 4 °C. They were subsequently blocked with 20% normal rat serum for 20 min at 4 °C and stained with anti-IgA–PE (1:200; eBioscience, 12-4204-82) for 20 min at 4 °C. Cells were then incubated with anti-PE beads (Miltenyi) and enriched by MACS. MACS-sorted cells were stained with anti-IgA–AF647 (1:100; SouthernBiotech, 1040-31) for 20 min at 4 °C. Final bacterial cell populations were sorted by FACSAria III as SYTO BC + PE + AF647 + (IgA positive) or SYTO BC + PE – AF647 – (IgA negative). Bacterial DNA isolation from mouse feces One fecal pellet from each mouse was collected into a sterile 1.5-ml Biopure tube (Eppendorf), put immediately on dry ice and stored at −80 °C until further processing. Total bacterial DNA was isolated using the QiaAMP Fast DNA Stool Mini Kit (QIAGEN) according to the manufacturer’s instructions. DNA was eluted with 100 μl of AE buffer (provided with the kit). DNA was stored at 4 °C until being used for the PCR. Bacterial DNA isolation from fecal pellets Three fresh fecal pellets from the MD4 mice were collected into a 1.5-ml Biopure tube and homogenized. Large debris and cells were removed by centrifugation at 400 g for 5 min at 4 °C. The supernatant was filtered through a 40-μm cell strainer and centrifuged at 400 g for 5 min. This step was repeated until no visible pellet was observed. The supernatant was then centrifuged at 8,000 g for 10 min to pellet bacteria. The pellet was stained with anti-IgA–PE (eBioscience, 12-4204-82; 1:200) followed by anti-PE microbeads (Miltenyi; 1:200) and sorted on LS columns (Miltenyi) using MACS. The positive fraction was subsequently stained with anti-IgA–AF647 (SouthernBiotech, 1040-31; 1:100). IgA + (10 6 ) and IgA – (10 6 ) events were sorted by FACS as PE + AF647 + or PE – AF647 – , respectively, centrifuged at 8,000 g for 10 min and stored at −80 °C until further processing. 16S rRNA gene library preparation and sequencing The V1 and V2 hypervariable regions of the 16S rRNA gene were amplified using modified 27F and 338R universal primers. The following nucleotide sequences were used: 27F-5′- AATGATACGGCGACCACCGAGATCTACAC TATGGTAATTCC AGMGTTYGATYMTGGCTCAG-3′ and 338R-5′- CAAGCAGAAGACGGCATACGAGAT NNNNNN NNNNNN AGTCAGTCAGAA GCTGCCTCCCGTAGGAGT-3′; bold, Illumina adaptor sequences; italic, linkers; NNNNNNNNNNNN, sample-specific MID tag barcodes. PCR reactions were performed in duplicate in a volume of 20 μl each using an AccuPrime Taq DNA polymerase high fidelity kit (Invitrogen), 4 μl of template DNA and 0.44 μl of each primer (10 μM stock concentration). The following PCR program was used: 3 min at 94 °C (initial denaturation), followed by 30 cycles of 30 s at 94 °C (denaturation), 30 s at 56 °C (annealing), 1 min 30 s at 72 °C (extension) and 5 min at 72 °C (final extension). Duplicates were pooled and amplicon quantity and size were determined with the LabChip GX (PerkinElmer). PCR products were pooled in equimolar amounts and purified using Agencourt AMPure XP magnetic beads (Beckman Coulter). Sequencing was performed on an Illumina MiSeq platform with MiSeq reagent kit V2-500 (paired-end, 2 × 250). Shotgun metagenomics library preparation and sequencing Bacterial genomic DNA was processed with the TruSeq DNA PCR-Free Low Throughput Library Prep Kit (Illumina, 20015962). Initial DNA input was 0.5 μg per sample. Shearing was performed using an M220 Covaris according to the manufacturer’s recommendations for 550-bp inserts, except for time of shearing, which was set to 30 s. Sheared DNA was further processed according to the manufacturer’s recommendations for 550-bp inserts. Library sequencing was performed on an Illumina NovaSeq platform using 2 × 250 bp chemistry (SP kit). Shotgun metagenomics data analysis Shotgun metagenomics data were preprocessed using the Sunbeam pipeline 46 for adaptor trimming, quality control and mouse genome decontamination (GRCm38 from Genome Reference Consortium) with default parameters. Taxonomic composition and functional profiling were performed using MetaPhIAn3 and HUMAnN3 pipelines 47 , respectively, with ChocoPhlAn v30 (201901) and the full UniRef90 databases (retrieved 1 October 2020). Gene differential abundance analysis between wild-type and MD4 transgenic mice was performed using limma parametric eBayes testing with the lmfit function of the limma R package (version 3.42.2) on log-transformed data. Differential abundance testing was performed using a zero-inflated Gaussian mixture model (fitZig function) in the metagenomeseq R package (version 1.28.2), and P values were adjusted using the Benjamini–Hochberg method. Computational work was supported by the MASSIVE HPC facility ( ) 48 . 16S rRNA gene sequencing data analysis All 16S rRNA gene sequencing analyses were performed in R statistical software. Raw fastq files were demultiplexed and processed using the custom microbiome-dada2 pipeline ( ) with default parameters. Taxonomic classification and exact sequence matching were performed using the SILVA database (version 123). ASV filtering, normalization, ordination and diversity analyses were performed using the phyloseq R package (version 1.26.1) and visualized using the ggplot2 R package (version 3.1.0). Only samples with >1,000 ASVs were considered for downstream analyses. Unclassified ASVs at the phylum level were removed and filtered based on prevalence (25% of total samples) and counts (100 reads minimum). The ASV count table then was normalized using total sum scaling, where each ASV count is divided by the total read count for each sample. PCoA and ANOSIM were performed using Bray–Curtis distance matrix calculated using the vegan R package (version 2.5.6). Differential ASV abundance testing was performed using a zero-inflated Gaussian mixture model (fitZig function) in the metagenomeseq R package (version 1.28.2). For both ANOSIM and differential abundance testing, a model including the genotype (or recolonization genotype) as an explanatory variable and controlling for experiment variation was implemented. A correlation network was inferred using the CClasso method (version 2.0) ( ). Correlation weights with a P value <0.05 and a correlation coefficient >0.2 were considered significant. IgA binding scores were calculated with the following parameters: for each ASV of each sample, the IgA + to IgA – fraction relative abundance ratio was calculated followed by a mean relative abundance ratio if consistent (minimum 1 and higher than 10) in two of three samples. The network was constructed using igraph R package (version 1.2.5). The following formula was used to calculate the IgA binding index: relative abundance (IgA + )/relative abundance (IgA – ) ≥ 10. Non-targeted metabolite profiling and data analysis Metabolite profiling was performed by Metabolomic Discoveries. Briefly, plasma metabolites from wild-type and MD4 mice were extracted with 90% methanol and 10% water while shaking at 37 °C at 1,000 r.p.m. High-resolution mass spectrometry was combined with modified hydrophilic interaction chromatography, and the samples were randomized on an Agilent 1290 UHPLC system (Agilent) equipped with a ZIC-HILIC column (10 cm per 2.1 mm, 3 μm, Sequant, Merck) coupled to a 6540 QTOF/MS detector (Agilent). The detection range was 50–1,700 m / z (positive and negative ESI mode). Data were analyzed with XCMS, IPO-R package (data conversion, chromatogram peaks extraction), Mzmatch.R (peak filtering and annotation), IDEOM (noise and artifact elimination, putative peak annotation by exact mass ± 10 ppm). Data were normalized applying Normalization using Optimal selection of Multiple Internal Standards and Cross-Contribution compensating Multiple standard Normalization. Differential abundance analysis of metabolites between wild-type and MD4 mice was performed on log-transformed data using limma parametric eBayes testing with the lmfit function of the limma R package (version 3.42.2), and P values were adjusted using the Benjamini–Hochberg method. Targeted metabolomics for p -cresol sulfate Plasma samples (25 μl) were extracted with 100 μl of chilled methanol containing internal standard (PCS-d4 at 500 ng ml –1 plasma concentration), shaken on ice for 30 min and centrifuged at 4 °C for 10 min. Supernatant (100 μl) was diluted with 100 μl of 0.1% formic acid in water. Frozen fecal samples were weighed (3–5 mg) and extracted with 20 μl mg –1 of 80% chilled methanol containing internal standard PCS-d4, vortexed at 4 °C for 15 min, shaken at 25 °C for 60 min and centrifuged at 4 °C at 14,800 g for 30 min. Supernatant was collected and diluted 2.5-fold with 0.1% formic acid. The PCS-d4 concentration was 50 ng ml –1 in samples, which is equal to 2.5 ng per 1 mg of feces. BALF (50 μl) was extracted with 200 μl of cold methanol, mixed on ice for 30 min and centrifuged at 4 °C at 14,800 g for 10 min. Supernatant (200 μl) was transferred to new Eppendorf tubes and evaporated under nitrogen stream for 60 min at 20 °C. Samples were resolubilized in 100 μl of 0.1% formic acid in water, mixed for 15 min at 25 °C, sonicated on ice for 15 min, centrifuged at 4 °C and transferred to vials. Samples were analyzed on the same day as preparation by injecting 6 μl and using the following liquid chromatography–mass spectrometry (LC–MS) acquisition method: LC–MS data were acquired on a Q-Exactive mass spectrometer coupled with a Dionex Ultimate 3000 RSLC separation system (Thermo Scientific Ascentis Express C8 (100 × 2.1 mm, 2.7 μm, Supelco) column protected with a guard column (C8, 2 × 2 mm; Phenomenex) was used for separation). Buffer A was 0.1% formic acid in water and buffer B was 0.1% formic acid in acetonitrile. Gradient elution was achieved starting at 10% B concentration and increased to 95% B in 3.5 min, kept at 95% B until 4.5 min, reduced to 10% B at 5 min and equilibrated at that ratio until 7 min. The autosampler temperature was kept at 4 °C, and the column oven temperature was kept at 40 °C. HESI source spray voltage was set to 4 kV, the capillary temperature was set at 300 °C, the auxiliary gas temperature was set at 120 °C, the sheath gas flow rate was set to 50, the auxiliary gas was set to 20, the sweep gas was set to 2 arbitrary units and the S-lens RF level was set to 50. The mass spectrometer was operated in PRM acquisition mode in negative ion polarity using an inclusion list for PCS and PCS-d4 m / z ( m / z 187.0071 and 191.0321, respectively) with specified HCD collision energy NE = 50 and a retention time between 2 and 3.5 min. The following other PRM parameters were used: 1 microscan, 17,500 resolution, AGC target 2 × 10 5 , maximum IT 100 ms, isolation window 2 m / z , loop count 4 and MSX count 1. Peak integration and quantification were performed using the Tracefinder 4.1 application (Thermo Scientific). Isolation of lung cells enriched for an airway epithelial cell fraction Mice were killed by CO 2 inhalation and injected with 1.5 ml dispase II (Sigma-Aldrich) intratracheally, followed by intratracheal injection of 0.5 ml of 1% low-melting-point agarose (Sigma-Aldrich). Lungs were then covered with ice for 3 min, removed and placed in a 15-ml Falcon tube with 2 ml of dispase II and incubated for 45 min with gentle agitation. This was followed by mechanical disruption of the lung lobes using forceps in DMEM supplemented with DNase I (Sigma-Aldrich; 1 U ml –1 ), filtration through 70- and 40-μm cell strainers (Falcon, Corning) and lysis of erythrocytes using red blood cell lysing buffer (BD Biosciences). Lung cells were plated in flat-bottom 24-well or 96-well plates (Costar) coated with fibronectin (Sigma-Aldrich; 10 μg ml –1 ) at a density of 1 million or 0.2 million cells per well, respectively. In vitro stimulation of lung cells Cells were stimulated with HDM (100 μg ml –1 ; Greer), LPS (10 μg ml –1 ; Sigma-Aldrich), PCS (100 μg ml –1 ; Alsachim), EGF (100 ng ml –1 ; Thermo Fisher Scientific), amphiregulin (500 ng ml –1 ; In Vitro Technologies) or gefitinib (0.16 μM; Sigma-Aldrich) for 24 h at 37 °C and 5% CO 2 , after which the supernatant was collected and stored at −20 °C until further use. Reporting Summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability All raw 16S rRNA amplicons and shotgun metagenomics sequences with corresponding metadata are deposited on the NCBI server under BioProject PRJNA641984 . The metabolomics dataset is deposited on the Mendeley data repository ( ). The SILVA database can be found at . Functional annotation of predicted genes was performed using Uniref90 retrieved on 1 October 2020. The data that support the findings of this study are available from the corresponding author on request.

A study designed to study how the immune system impacts gut bacteria—has led to the extraordinary discovery of two molecules that can not only provide profound protection in experimental models of asthma but can also substantially reduce the severity of an attack. Neither of these molecules, one of which is already commercially available as a dietary supplement, were previously known to have an effect on asthma—and they also appear, from animal studies, to have a role in treating the respiratory illness that is prevalent, and often fatal, in people with serious COVID-19. The researchers aim to test one of the molecules in a clinical trial in 2021 in asthmatics. As further evidence that these two molecules could potentially protect against asthma the Monash University researchers found, through studying the literature, that these metabolites are present in higher amounts in two studies of children without asthma compared to those with the disease, according to Professor Benjamin Marsland from the Monash University Central Clinical School, whose paper is published today in Nature Immunology. Asthma is one of the most common major non-communicable diseases and it impacts 300 million people globally. The global asthma treatment market size stood at over $18 billion US in 2019. The team led by VESKI innovation fellow, Professor Marsland, wanted to understand how the immune system impacts the gut microbiome. While it is known that gut bacteria have an effect on the immune system, "how the immune system influences the gut microbiome has to date been under studied," he said. Studying a mouse that had a limited immune system, consisting of a single type of antibody, the researchers found the gut microbiome was changed. By transferring these gut bacteria into 'normal' mice they could identify which bacteria had an impact on the mouse immune system. In what was an enormous surprise the researchers found that the production of a particular gut bacteria by-product, called p-cresol sulfate (PCS), led to a "profound and striking protection against asthma." Part of the serendipity of the finding is that Professor Marsland's area of expertise is in the immunology of asthma, though he suspects this metabolite may have a role in other inflammatory diseases. The researchers found that the PCS was produced by enhanced bacterial metabolism of L-tyrosine; a well-known amino acid found in dietary supplements aimed at improving attention and alertness. "We found that giving mice either L-tyrosine or PCS, provided significant protection against lung inflammation. PCS travels all the way from the gut, to the lungs, and acts on epithelial cells lining the airways to prevent the allergic asthma response" The researchers also tested the metabolites in animal models of acute respiratory distress syndrome (ARDS), and found it to be protective. ARDS is a common killer of people with serious COVID-19. While L-tyrosine has a long history of use in the clinic, as mentioned in dietary supplements, its potential use as a therapy could be fast tracked into clinical trials because it is known to be safe. Professor Marsland commented "It's very important that a thorough clinical study is performed in order to determine whether L-tyrosine is effective in people with asthma, and for us to determine what is the correct dose and treatment regime." PCS however is known to be in high levels in people with chronic kidney disease and it's suspected to be toxic because of these patient's inability to clear it. The research group has started developing a form of PCS that is a potent protector against asthma without the potential toxic side effects. More importantly the scientists have found that inhaling PCS provides a direct protective effect against lung inflammation, opening the way for a novel inhaled preventive therapy.

10.1038/s41590-020-00856-3

Medicine

New research identifies two drug classes that could be re-purposed for type 1 diabetes treatment

Maikel L. Colli et al, An integrated multi-omics approach identifies the landscape of interferon-α-mediated responses of human pancreatic beta cells, Nature Communications (2020). DOI: 10.1038/s41467-020-16327-0 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-020-16327-0

https://medicalxpress.com/news/2020-05-drug-classes-re-purposed-diabetes-treatment.html

Abstract Interferon-α (IFNα), a type I interferon, is expressed in the islets of type 1 diabetic individuals, and its expression and signaling are regulated by T1D genetic risk variants and viral infections associated with T1D. We presently characterize human beta cell responses to IFNα by combining ATAC-seq, RNA-seq and proteomics assays. The initial response to IFNα is characterized by chromatin remodeling, followed by changes in transcriptional and translational regulation. IFNα induces changes in alternative splicing (AS) and first exon usage, increasing the diversity of transcripts expressed by the beta cells. This, combined with changes observed on protein modification/degradation, ER stress and MHC class I, may expand antigens presented by beta cells to the immune system. Beta cells also up-regulate the checkpoint proteins PDL1 and HLA-E that may exert a protective role against the autoimmune assault. Data mining of the present multi-omics analysis identifies two compound classes that antagonize IFNα effects on human beta cells. Introduction Type 1 diabetes (T1D) is a chronic autoimmune disease leading to pancreatic islet inflammation (insulitis) and progressive beta cell loss 1 . Type I interferons (IFN-I), a class of cytokines involved in antiviral immune responses 2 , are involved in insulitis. Viral infections are a risk factor associated with T1D development 3 and individuals at risk of T1D show a type I interferon signature 4 . The type I interferon, interferon-α (IFNα), is expressed in islets of T1D patients 5 , and antibodies neutralizing different isoforms of IFNα prevent T1D development in individuals with polyglandular autoimmune syndrome type 1 6 . Exposure of human pancreatic beta cells to IFNα recapitulates three key findings observed in human insulitis, namely HLA class I overexpression, endoplasmic reticulum (ER) stress and beta cell apoptosis 7 . Combination of genome-wide association studies (GWAS) 8 and studies using the ImmunoChip 9 have identified around 60 loci associated with the risk of developing T1D. Transcriptomic studies revealed that >70% of the T1D risk genes are expressed in human pancreatic beta cells 10 , and many of these genes regulate innate immunity and type I IFN signaling 11 . Type I IFN signaling is often cell-specific, an effect mediated by differences in cell surface receptor expression, and activation of downstream kinases and transcription factors 12 . Thus, and considering the potential relevance of this cytokine to the pathogenesis of T1D, it is crucial to characterize its effects on human beta cells. To define the global impact of IFNα on human beta cells, we presently performed an integrative multi-omics analysis (ATAC-seq, RNA-seq and proteomics) of IFNα-treated human beta cells to determine the early, intermediate and late responses to the cytokine. The findings obtained indicate that IFNα promotes early changes in chromatin accessibility, activating distant regulatory elements (RE) that control gene expression and protein abundance. IFNα activates key transcription factors (TFs), including IRF1, which act as a mediator of the crosstalk between beta cells and immune cells via the expression of the checkpoint proteins PDL1 and HLA-E. Furthermore, IFNα induces modules of co-expressed mRNA and proteins that physically interact and have relevance to T1D pathogenesis. The integration of high-coverage RNA-seq and ATAC-seq indicates regulatory gene networks and reveals that alternative splicing and different first exon usage are key mechanisms expanding the repertoire of mRNAs and proteins expressed by stressed beta cells. Finally, mining the modules of co-expressed genes and the IFNα beta cell signature against the most recent catalogs of experimental and clinical drugs identifies two potentially interesting therapeutic targets for future trials. Results IFNα modifies beta cell mRNA expression similarly to T1D We performed a time course multi-omics experiment combining ATAC-seq, RNA-seq and proteomics of the human beta cell line EndoC-βH1 exposed or not to IFNα. The data were integrated to determine the dynamics of chromatin accessibility, gene/transcript expression and protein translation, respectively (Fig. 1a ). We also performed RNA-seq of 6 independent human pancreatic islet preparations exposed or not to the cytokine at similar time points (Supplementary Fig. 1a ). To assess whether our in vitro model is relevant for the in vivo islet inflammation (insulitis) in T1D, we took two approaches: (1) Examine whether candidate genes for T1D expressed in human islets are involved in IFN signaling (Supplementary Fig. 2a ); and (2) Compare our in vitro data of IFNα-treated EndoC-βH1 cells and human islets with available RNA-seq data of human beta cells from T1D patients. In line with previous findings suggesting a role for IFNs on the pathogenesis of T1D 13 , we found that T1D risk genes expressed in human islets 10 , 14 are significantly enriched in immune-related pathways, including type I and II interferon regulation/signaling (Supplementary Fig. 2b ). Next, we performed a Rank–Rank Hypergeometric Overlap (RRHO) analysis (which estimates the similarities between two ranked lists 15 ) comparing the log 2 fold-change (FC) ranked list from RNA-seqs of EndoC-βH1 cells and human islets (IFNα-treated vs untreated) against an equally ranked list of genes obtained from RNA-seq of purified primary beta cells 16 from T1D and healthy individuals (Supplementary Fig. 2c and Supplementary Data 1 ). There was a significant intersection between upregulated genes induced by IFNα in both, EndoC-βH1 cells (362 overlapping genes) and human islets (850 overlapping genes), and genes induced by the local pro-inflammatory environment affecting primary beta cells from T1D individuals (Supplementary Fig. 2d, f ). We also compared these two IFNα-treated datasets against beta cells from T2D patients 17 , a condition mostly characterized by metabolic stress 18 . By contrast with the observations made in beta cells from T1D individuals, there was no statistically significant correlation between IFNα-regulated genes in EndoC-βH1 cells and human islets and the gene expression profile present in T2D beta cells (Supplementary Fig. 2e, g ). Fig. 1: Exposure of EndoC-βH1 cells to interferon-α promotes changes in chromatin accessibility, which are correlated with gene transcription and translation. a EndoC-βH1 cells were exposed or not to IFNα (2000 U/ml) for the indicated time points and different high-throughput techniques were performed to study chromatin accessibility (ATAC-seq, n = 4), transcription (RNA-seq, n = 5) and translation (Proteomics, n = 4). b Volcano plot showing changes in chromatin accessibility measured by ATAC-seq. Open chromatin regions indicated as gained (red) or lost (blue) had an absolute log 2 fold-change (|log 2 FC|) > 1, and a false discovery rate (FDR) < 0.05. The regions that did not reach such threshold were considered “stable” (gray). c , d Frequency of upregulated, downregulated or stable transcripts in the vicinity (<20 kb transcription start site (TSS) distance) of one or multiple open chromatin regions (OCRs) as classified in b . e Frequency of differentially abundant proteins in the vicinity (<20 kb TSS distance) of gained or stable open chromatin regions. f Distribution of IFNα-induced changes in protein abundance among upregulated proteins based on the number of linked gained OCRs. g Correlation between RNA-seq and proteomics of EndoC-βH1 cells exposed to INFα. The x axis represents the mRNA log 2 FC. The most upregulated (log 2 FC > 0.58, FDR < 0.05) and downregulated (log 2 FC < −0.58, FDR < 0.05) mRNAs are filled in red and blue, respectively. The y axis indicates the proteomics log 2 FC. The proteins most upregulated (log 2 FC > 0.58, FDR < 0.15) or downregulated (log 2 FC < −0.58, FDR < 0.15) are represented by red and blue borders, respectively. mRNAs and proteins not meeting these criteria were considered equal-regulated (gray fill and border, respectively). Full size image IFNα induces early changes in chromatin accessibility The ATAC-seq experiments demonstrated that INFα induces early changes in chromatin accessibility, with >4400 regions of gained open chromatin regions (OCRs) detected at 2 h, which decreased to 1000 regions by 24 h (Fig. 1b and Supplementary Data 2 ); only nine regions had loss of chromatin accessibility (Fig. 1b ). Most of the OCRs at 24 h were already modified at 2 h (fast response), and only 10% of OCRs were specifically gained at 24 h (late response). The gained OCRs were mostly localized distally to gene transcription starting sites (TSS) (Supplementary Fig. 3a ) acting, therefore, as potential regulatory elements. These regions are evolutionary conserved (Supplementary Fig. 3b ), and enriched for transcription factors (TFs) binding motifs (Supplementary Fig. 3c ), including islet-specific TFs binding sequences. To assess whether changes in chromatin remodeling were associated with variations in gene expression, we first quantified the frequency of ATAC-seq regions gained or stable in the proximity (40 kb window centered on the TSS) of genes with differential mRNA expression (up/down/non-regulated or non-expressed) (Supplementary Data 2 ). There was a higher proportion of upregulated genes associated with gained OCRs in comparison to stable regions at each time point analyzed (Fig. 1c ). Moreover, the number of gained OCRs was associated with changes in both the proportion (Fig. 1d ) and the intensity (Supplementary Fig. 3e ) of transcript induction (Supplementary Fig. 3d , see Methods for more information). There was also a minor association between the number of stable regions and upregulated mRNAs at 2 h (Supplementary Fig. 3e ), likely due to the activation of already nucleosome-depleted regions ahead of cytokine exposure 19 . Consistently with these results, there was an increase in the frequency of upregulated proteins coded by genes proximal to gained OCRs (Fig. 1e ). Likewise, there was a progressive increase in IFNα-induced protein abundance depending on the number of linked gained open chromatin regions (Fig. 1f ). There was a strong correlation between upregulated mRNAs and induced proteins ( r 2 : 0.66 and 0.65 at, respectively, 8 and 24 h, p < 2.2 × 10 −16 ) (Fig. 1g , first column), but a much lower similarity between downregulated mRNAs and proteins (Fig. 1g , second column). Gene ontology analysis of differentially abundant proteins upon IFNα treatment identified several biological processes involved in the pathogenesis of T1D, such as antigen processing and presentation, responses to viruses, apoptosis and NK/T-cell responses (Supplementary Fig. 4a , b); groups of genes associated to protein modification and degradation were also present (Supplementary Fig. 4a, c ). Furthermore, genes related to endoplasmic reticulum (ER) stress, another post-transcriptional mechanism that downregulates translation of many mRNAs 20 , were also upregulated by IFNα at both the mRNA and protein levels (Supplementary Fig. 4d ). These findings are in line with our previous observations 7 and were confirmed here in independent samples for two key ER stress markers, namely the transcription factor ATF3 21 and the ER chaperon HSPA5 (also known as BiP/GRP78) 22 (Supplementary Fig. 4e–h ). ER stress often decreases translation, which may explain the weak association observed between mRNA and protein expression in downregulated mRNAs and proteins (Fig. 1g ). IRF1, STAT1 and STAT2 are key regulators of IFNα signaling To identify the key transcription factors involved, the expression of differentially expressed genes (DEG) from all RNA-seq time points (Supplementary Data 3 ) was analyzed using the dynamic regulatory events miner (DREM) model 23 . This approach identified six patterns of co-expressed genes (Fig. 2a ); 5 out of 6 pathways had an early peak of induction (2 or 8 h), which then decreased or remained stable until 24 h (Fig. 2a ). The model compared the frequency of TF binding sites in the gene promoters between divergent branches of co-expressed genes, assuming that these TFs are responsible for the observed differences in gene expression profiles (Fig. 2a ). This was compared with the TF occupancy determined by assaying the protection of the bound sequence to ATAC-seq transposase cleavage (footprint) (Supplementary Fig. 5a and Methods). There were footprints for the transcription factors IRF1, STAT1 and STAT2, which were deepened upon IFNα exposure in pathway B (which had the highest transcriptional upregulation at 2 h) and for IRF1 in two independent pathways, namely B and D at 24 h (Fig. 2b ). Western blot analysis confirmed the activation of these TFs (Fig. 2c ). STAT1 and STAT2 phosphorylation peaked between 0.5 and 1 h and then returned to near-basal levels at 24 h, while IRF1 peaked later, at 4–8 h decreased by 24 h (Fig. 2c ); these findings support the observed TF footprint profiles (Fig. 2b ). There was also a close correlation between DEGs induced by IFNα in RNA-seq of EndoC-βH1 cells and in human pancreatic islets (Supplementary Fig. 1b ; p < 2.2 ×10 −22 at 2, 8 and 24 h), which resulted in a similar pattern of gene activation under the control of analogous TFs (Supplementary Fig. 1c and Supplementary Data 4). Fig. 2: IRF1, STAT1 and STAT2 regulate IFNα-induced transcription and the expression of checkpoint proteins. a The regulatory paths summarize the temporal patterns of the differentially expressed genes (DEG) detected by RNA-seq (|log 2 FC|> 0.58 and FDR < 0.05, n = 5) (evaluated by DREM 25 ). The x axis represents the time and the y axis the mRNA log 2 FC. Each path corresponds to a set of co-expressed genes. Split nodes (circles) represent a temporal event where co-expressed genes diverge in expression. In blue are the TFs upregulated at the respective time points of the RNA-seq that may regulate the pathways. b IFNα promoted TFs footprint deepening in open chromatin regions (OCR) associated to genes from the indicated DREM pathways. OCRs were associated to the nearest gene TSS with a maximum distance of 1 Mb. Previously annotated TF matrices 79 were used to identify differential DNA-footprints induced by IFNα (blue lines = untreated cells, red lines = IFNα (24 h), dashed lines = reverse strand, continuous line = forward strand, Methods, n = 4). c Time course profile of STAT1, STAT2 and IRF1 protein activation in EndoC-βH1 cells exposed to IFNα (representative of four independent experiments). d – m EndoC-βH1 cells were transfected with an inactive control siRNA (siCT) or previously validated 7 , 24 siRNAs targeting IRF1 (siIRF1), STAT1 (siSTAT1), STAT2 (siSTAT2) or STAT1 plus STAT2 (siSTAT1 + 2). After 48 h the cells were exposed to IFNα The values were normalized by the housekeeping gene β-actin (mRNA) and then by the highest value of each experiment considered as 1 (for h and m ( n = 3); for e – g , i , j and l ( n = 4); for d , k ( n = 5)), ANOVA with Bonferroni correction for multiple comparisons ( d – m ). Values are mean ± SEM ( d – m ). Source data are provided as Source Data file. Full size image Individual DREM pathways usually regulate specific biological processes (GO) (Supplementary Fig. 5b , 1 d ). Among them, was the term “Regulation of immune responses” (Supplementary Fig. 5b ). This pathway comprises several genes involved in the crosstalk between beta cells and the immune system, such as PDL1 ( CD274 ), an immune checkpoint protein expressed in the islets of T1D individuals 24 , and a second co-inhibitory molecule, HLA-E , recently identified as potential target for cancer immunotherapy 25 (Fig. 2d–m ). By using a previously validated siRNA targeting IRF1 24 , we obtained around 60% knockdown (KD) of INFα-induced IRF1 protein and mRNA expression at 2 and 24 h (Supplementary Fig. 5c–f ). IRF1 silencing led to a significant decrease in IFNα-induced PDL1 and HLA-E mRNA expression (Fig. 2 d, f, i, k). Silencing of IRF1 also decreased IFNα-induced upregulation of the chemokines CXCL1 and CXCL10 , the HLA-I component beta-2-microglobulin ( B2M ) and the suppressor of cytokine signaling 3 ( SOCS3 ) (Supplementary Fig. 5d, f ). Small interference RNAs targeting STAT1 (siSTAT1) or STAT2 (siSTAT2) promoted >70% KD of their respective proteins and mRNAs, (Supplementary Fig. 5g–j ). Inhibiting STAT1 or STAT2 alone partially blocked the induction of PDL1 and HLA-E at 2 h (Fig. 2e, j ), but led to a paradoxical increase in PDL1 and HLA-E expression at 24 h (Fig. 2g, i ), which is probably due to a compensatory increase in expression of the non-targeted STAT 24 . In line with this, double KD of STAT1 + STAT2 led to downregulation of both PDL1 and HLA-E (Fig. 2h, m ). STAT2 inhibition decreased the 2 h expression of IFNα-induced CXCL1/10, SOCS1 and MX1 , whereas STAT1 KD only prevented CXCL10 induction (Supplementary Fig. 5h ). At 24 h only 2 out of 4 genes remained partially inhibited by siSTAT2 (Supplementary Fig. 5j ), whereas double KD of STAT1 + STAT2 prevented IFNα-induced gene upregulation at 24 h in most cases (Supplementary Fig. 5k ). Exposure of FACS-purified human beta cells (Supplementary Fig. 6a–c ) to IFNα confirmed the upregulation of genes related to antigen presentation ( HLA-I ), antiviral responses ( MX1, MDA5 ), ER stress ( CHOP ), immune cells recruitment ( CXCL10 ) and checkpoint regulators ( PDL1 ) (Supplementary Fig. 6d ). The checkpoint protein PDL1 is overexpressed in beta cells from people with T1D 24 , and we presently evaluated the expression of another checkpoint protein, i.e. HLA-E 25 . IFNα upregulated HLA-E mRNA expression in EndoC-βH1 cells (Fig. 3a ), dispersed human islets (Fig. 3b ) and FACS-purified human beta cells (Fig. 3c ) and augmented HLA-E protein expression in both EndoC-βH1 cells (Fig. 3d ) and human islets (Fig. 3e ), with peak at 24 h. The inhibitory effects of HLA-E on immune cells require its expression on the cell surface or its secretion 26 . Flow cytometry confirmed that IFNα increases surface HLA-E expression (Fig. 3f, g , Supplementary Fig. 5l ), but there was no HLA-E release to the supernatant (Supplementary Fig. 5m ). HLA-E mRNA expression was upregulated by 8-fold in human islets of donors with recent-onset T1D in the DiViD study 27 and HLA-E protein expression was significantly increased in insulin-containing islets, but not in insulin-deficient islets, of T1D individuals in comparison to healthy individuals (Fig. 3h, i ). HLA-E expression was present in both beta and alpha cells (but not delta cells; Supplementary Fig. 5n ) in the islets of people with T1D, with a predominance of expression among alpha cells as compared to beta cells (Fig. 3j ). This may help to explain why alpha cells are more resistant to the immune assault in T1D. Fig. 3: HLA-E is overexpressed in pancreatic islets of T1D individuals. EndoC-βH1 cells ( a , d ), human islets ( b , e ) or FACS-purified human beta cells ( c ) were exposed (gray bars) or not (black bars) to IFNα for the indicated time points and HLA-E mRNA ( a – c ) and protein ( d , e ) evaluated. The values were normalized by the housekeeping gene β-actin (mRNA) or α-tubulin (protein) and then by the highest value of each experiment considered as 1 (for a ( n = 4); b ( n = 3 (8 h), n = 5 (24 h)); c ( n = 4); d ( n = 4) and e ( n = 2 (8 h), n = 4 (24 h)), ANOVA with Bonferroni correction for multiple comparisons ( a – e )). f , g HLA-E cell surface expression was quantified in EndoC-βH1 cells by flow cytometry. Histograms ( f ) represent changes in mean fluorescence intensity (MFI). The MFI values ( g ) were quantified at baseline and after 24 h exposure to IFNα ( n = 4, two-sided paired t- test). Values are mean ± SEM ( a – g ). h Immunostaining of HLA-E (green), glucagon (red) and insulin (light blue) in representative islets from individuals with or without diabetes. The top and middle panels represent an insulin-containing islet (ICI) and insulin-deficient islet (IDI) from T1D sample DiViD 3, and the lower panel represents an islet from a control donor (EADB sample 333/66). DAPI (dark blue). Scale bar 20 μm. i The MFI analysis of HLA-E expression. 30 ICIs from 6 independent individuals with T1D (5 islets per individual), 20 IDIs from 4 independent individuals with T1D (5 islets per individual), and 30 ICIs from 6 independent individuals without diabetes (5 islets per individual) were analyzed. Values are median ± interquartile range; ANOVA with Bonferroni correction for multiple comparisons, AU (arbitrary units), ns = (non-significant). j Higher magnification image demonstrating that HLA-E (green) localizes predominantly to alpha cells in a T1D donor islet (glucagon (red); insulin (light blue)) but is also expressed in beta cells, as indicated in h and j . Scale bar 30 μm. Source data are provided as Source Data file. Full size image mRNA and protein modules regulated by interferon-α We integrated the RNA-seq and proteomics data (using all the samples from both 8 and 24 h) using the weighted correlation network analysis package (WGCNA) 28 . The heatmaps of the topological overlap matrix from each dataset with module assignment are shown in Fig. 4a . There were initially 32 eigengene modules of mRNAs and 27 of proteins, which were merged (considering a dissimilarity threshold of 0.25) reducing the numbers of mRNA and protein modules to 8 and 7, respectively (Supplementary Fig. 7a–c ). The quality of these modules was determined using a combined score of density and separability measures (Methods) 29 , which indicated that they were well-defined ( Z summary > 10) (Supplementary Fig. 7d ). WGCNA analysis of the RNA-seq of human islets exposed to IFNα identified well-defined modules of mRNAs (Supplementary Fig. 8a–d ), similar to the ones identified in EndoC-βH1 cells exposed to the cytokine (Supplementary Fig. 8e ). To focus on central modules induced by IFNα exposure, we selected only the differentially expressed genes (DEG) (Supplementary Data 3 ) and abundant proteins (DAP) (Supplementary Data 5 ) in each eigengene module, representing 49% of the protein-coding DEGs and 89% of the DAPs, and then examined the overlap between these datasets. There was a significant overlap between five modules of mRNAs and proteins (minimum of 10 elements in common, FDR < 0.05) (Fig. 4b ). The two main new modules, called #1 and #2 (Fig. 4c ), were composed of highly correlated mRNAs and proteins (Supplementary Fig. 7 e, g) predominantly upregulated by IFNα at both 8 and 24 h (Supplementary Fig. 7 f, h). Module #5 also had significantly correlated members (Supplementary Fig. 7 i), but enriched in downregulated mRNAs/proteins at both 8 and 24 h (Supplementary Fig. 7 j). Interestingly, there was significant enrichment of ATAC-seq gained OCRs in module #2 (Fig. 4d ). They were enriched for TF binding motifs including both the pro-inflammatory motifs ISRE / IRF and the islet-specific transcription factor FOXA2 (Fig. 4e ). Fig. 4: Weighted correlation network analysis (WGCNA) identifies IFNα-regulated mRNA and protein modules. a Heatmap representation of the topological overlap matrix. Rows and columns correspond to single genes/proteins, light colors represent low topological overlap, and progressively darker colors represent higher topological overlap. The corresponding gene dendrograms and initial module assignment are also displayed. b Identification of modules presenting significant overlap (FDR < 0.05 and a minimum of 10 members in common) (green border) between differentially expressed genes (DEG) and their translated differentially abundant proteins (DAP). c Composition, number of elements and type of DEG and DAP present in each of the significantly overlapping modules. d ATAC-seq-identified open chromatin regions at 2 h were linked to gene transcription start sites (TSSs) in a 40 kb window. These genes and their open chromatin regions were associated to the modules of DEG and DAP. The enrichment for gained open chromatin regions was then evaluated in each module. (** represents a p -value = 0.002343, one-sided χ 2 test). e De novo HOMER motifs present in the ATAC-seq regions overlapping module #2 as described in Methods. The unadjusted p -values were obtained using the hypergeometric test from the HOMER package 77 . f The protein–protein interaction (PPI) network of module #2 was done using the InWeb InBio Map database 31 . Enriched proteins (FDR < 0.05 and minimum number of connections = 5, represented as squares) were identified and added to the network if they were not already present. Red fill identifies upregulated proteins, blue fills downregulated proteins and gray fill equal-regulated. Colored regions delimitate communities of proteins, as described in Methods. The wordcloud next to each community presents their enriched geneRIFs terms. g The biological processes (GO) overrepresented in module #2 summarize the main findings observed in IFNα-treated human beta cells. The present results were based on RNA-seq ( n = 5) and proteomics ( n = 4) data of EndoC-βH1 cells. Full size image To identify the gene regulatory network (GRN) of module #2, we integrated information from two sources: (1) literature-based collection of TF-target interactions 30 , and (2) the present de novo TF binding motifs and their predicted targets (Supplementary Fig. 9a ). This allowed us to add information from cis -regulatory elements (in orange) acting on the IFNα-induced GRN in human beta cells (Supplementary Fig. 9b ). A similar approach was used for modules #1 and #5, but considering only data from the literature (Supplementary Fig. 10a, c ). The PPI network InWeb InBio Map 31 was used to assess the presence of protein–protein interaction (PPI) networks in the different modules. This generated networks of interacting proteins for modules #1, #2 and #5 (Fig. 4f and Supplementary Fig. 10b, d ) and allowed the recognition of protein communities (grouped by colors) that regulate specific and common biological functions (Fig. 4f and Supplementary Fig. 10b, d ). Module #2, which presents the higher number of connections, showed an enrichment for several key biological processes activated by IFNα and relevant for the pathogenesis of T1D, including cellular response to viruses, antigen processing and presentation via MHC class I, inflammatory and acute phase responses (Fig. 4g ). Interferon-α changes the alternative splicing landscape The present high-coverage RNA-sequencing (>200 million reads) allowed the detection of ~47,000 splicing variants, with IFNα-induced 343 differentially expressed transcripts (DETs) at 2 h, and 1690 and 1669, respectively at 8 and 24 h, with predominance of upregulated transcripts (Fig. 5a and Supplementary Data 6 and 7 ). Considering all the DETs, 4%, 32% and 32% were exclusively modified at 2, 8 and 24 h, respectively, indicating a predominance of intermediary to late transcriptional changes induced by IFNα. Next, we evaluated the frequency of each individual splicing events (with an absolute difference in percent spliced-in (|ΔPSI|) > 0.2) regulated by IFNα at 8 and 24 h. There were 3140 events at 8 h and 2344 events at 24 h (FDR < 0.05) (Fig. 5b ). The most frequent AS event modified by IFNα was cassette exons (CEx), with predominantly increased exon inclusion (represented by ΔPSI > 0.2, FDR < 0.05) (Fig. 5c ). An example of a cassette exon showing increased inclusion upon IFNα treatment is the gene OASL (Fig. 5d, e ), an antiviral factor targeting single-stranded RNA viruses such as picornaviruses 32 . Exposure to IFNα for 24 h increased exon 4 inclusion in both EndoC-βH1 cells and human islets (Fig. 5d ). In line with this, the protein encoded by the isoform OASL −001 (which retains exon 4) displayed a higher IFNα-induced upregulation in comparison with the protein encoded by the isoform OASL −002, which has exon 4 exclusion (Fig. 5e ). Interestingly, the isoform OASL −001 has antiviral activity, whereas the isoform 002 lacks the ubiquitin-like domain required for this response (Supplementary Fig. 11A ) 33 . Fig. 5: Interferon-α changes the alternative splicing landscape. a EndoC-βH1 cells were exposed to IFNα for the indicated time points. The significantly upregulated (red) and downregulated (blue) transcripts were identified using Flux Capacitor ( n = 5, |log 2 FC| > 0.58 and FDR < 0.05). b Frequency of individual alternative splicing events regulated by IFNα ( n = 5, |ΔPSI| > 0.2, minimum 5 reads, FDR < 0.05). c Frequency distribution of alternative cassette exon (CEx) events altered by IFNα (( n = 5, ΔPSI) > |0.2| and FDR < 0.05). d Confirmation of the increased exon 4 inclusion in the antiviral gene OASL by IFNα (24 h). cDNA was amplified by RT-PCR using primers located in the upstream and downstream exons of the splicing event and the product evaluated using a Bioanalyzer 2100 ( n = 4 (EndoC) and n = 7 (human islets), two-sided paired t -test). e The log 2 FCs of the proteins coding for OASL-001 and −002 isoforms from IFNα-treated EndoC-βH1 cells proteomics (24 h) ( n = 4). f Frequency distribution of retained intron (RI) events altered by IFNα (n = 5, |ΔPSI| > 0.2 and FDR < 0.05). g The protein log 2 FC values obtained by proteomics analysis of EndoC-βH1 cells exposed to IFNα for 24 h were classified in three categories according to the levels of retained intron ΔPSI ( n = 5, mean ± SEM, ANOVA with Bonferroni correction). h Expression of RNA-binding proteins (left) that are significantly modified at mRNA level (FDR < 0.05) after exposure to IFNα and their respective proteins (right) in the indicated time points ( n = 4–5). i Positional enrichment of motifs from significantly modified RBPs among regions involved in the regulation of modified cassette exons (CEx) after exposure to IFNα for 24 h. ( n = 5, |ΔPSI| > 0.2, FDR < 0.05). j Comparison between the log 2 FC of a curated list (Supplementary Table 1) of known FMR1 target proteins against the log 2 FC of the remaining proteins detected by the proteomics of EndoC-βH1 cells exposed to IFNα for 24 h ( n = 4, mean ± SEM; two-sided unpaired t -test). Source data are provided as a Source Data file. Full size image Intron retention is an important mechanism of gene expression regulation, promoting nuclear sequestration of transcripts or cytoplasmatic degradation via nonsense-mediated decay 34 . There was a predominance for intron removal after 24 h (represented by ΔPSI < −0.2, FDR < 0.05), but not at 8 h (Fig. 5f ). To understand how this impacts protein translation, we compared changes in protein abundance among three categories of ΔPSI. Genes presenting intron removal had a significant increase in protein expression after IFNα exposure for 24 h in comparison to those with intron retention (ΔPSI > 0.2, FDR < 0.05) or with non-significant intron changes (ΔPSI −0.2–0.2 or FDR > 0.05) (Fig. 5g ). There were clear variations in the mRNA expression of several well-known RNA-binding proteins (RBPs) 35 upon IFNα exposure (Fig. 5h , left panel), but the impact on the respective proteins was less pronounced (Fig. 5h , right panel). We focused on a group of IFNα-modified RBPs at both mRNA and protein levels after 24 h, and mapped their RNA-binding motifs among upregulated and downregulated alternative exons. In support of a biological role for these RBPs on alternative exon splicing, there was an enrichment of their binding motifs in regions controlling alternative cassette exon inclusion/exclusion (Fig. 5 i). To further study some of these findings, we first reproduced the IFNα-induced downregulation of two RBPs, ELAV-like protein 1 ( ELAVL1 ) and heterogeneous nuclear ribonucleoprotein (HNRNPA1) , by using specifics siRNAs (Supplementary Fig. 12a, e ). Next, we evaluated whether this inhibition reproduced the changes induced by IFNα in the exon usage of four-and-a-half LIM domain protein 1 ( FHL1 ) and Caprin Family Member 2 ( CAPRIN2 ) (Supplementary Fig. 12b, f ) two potential targets of, respectively, ELAVL1 36 and HNRNPA1 37 . Silencing these RBPs promoted changes on exon usage (Supplementary Fig. 12c, g ) that were similar to the ones observed after IFNα treatment (Supplementary Fig. 12b, f ). This is especially relevant in the context of the IFNα-induced exon exclusion FHL1, which decreases the expression of transcripts coding for the protein FHL1A (Supplementary Fig. 12d ), an isoform described as a key host factor for the replication of the RNA virus Chikungunya 38 . RBPs can also control gene expression by blocking RNA translation, as described for the Fragile X Mental Retardation 1 ( FMR1 ) gene 39 . Indeed, there was a significant downregulation of previously validated bona fide targets of FMR1 (Supplementary Table 1 ) 40 in IFNα-treated EndoC-βH1 cells as compared to the remaining proteins (Fig. 5j ). IFNα induces increased alternative transcription start sites The usage of alternative transcription start (TSS) sites is another mechanism that generates different transcripts from the same gene 41 . We used the SEASTAR pipeline 42 for the computational identification and quantitative analysis of first exon usage. This approach recognized >250 events of alternative first exon (AFE) usage occurring in 166 different genes at 8 h, and >130 events of AFE usage in 88 genes at 24 h (Fig. 6a ). In agreement with this, 118 and 64 alternative promoters (±2 kb around FE TSS) detected by SEASTAR at 8 and 24 h, respectively, overlapped peaks of TSS identified by the FAMTOM5 Consortium 43 . Among these genes was the 5′-nucleotidase cytosolic IIIA ( NT5C3A ), a negative regulator of IFN-I signaling 44 . This gene had two AFEs identified by the SEASTAR modeling. In untreated condition (controls), there was a higher usage of the proximal first exon (FE), present in the isoforms NT5C3A −001 and 002 in beta cells (Fig. 6b , upper panel). After INFα exposure, however, there was increased usage of the distal FE from the transcript NT5C3A −004 (ΔPSI: 0.71 (8 h)/0.65 (24 h), both FDR < 0.001), which is supported by the cap analysis of gene expression (CAGE) of TSSs 45 (Fig. 6b , upper panel). This was confirmed in independent samples of EndoC-βH1 cells and human islets using specific primers (Fig. 6b , lower panel). Exon Ontology analysis 46 indicated that this FE shift probably has functional impact, since the distal FE lacks both the endoplasmic reticulum (ER) retention signal and the transmembrane helix (Supplementary Fig. 11b ), enabling its encoded protein to remain in the cytosol where NT5C3A acts 44 . Fig. 6: Changes in the alternative transcription start site (TSS) initiation increase the repertoire of IFNα-regulated transcripts. a The tool SEASTAR 42 was used to estimate the frequency of differential alternative first exon (AFE) usage induced by IFNα in human beta cells. The total number of IFNα-dependent AFEs events (left) and number of genes with AFEs (right) in the indicated time point are shown ( n = 5, ΔPSI > |0.2|, FDR < 0.05). b View of the NT5C3A locus showing the transcripts with AFE usage, the RNA-seq (red) signals of EndoC-βH1 cells exposed or not to IFNα and the CAGE TSSs information (black scale) 45 (upper panel). Confirmation of the AFE usage identified by SEASTAR in the gene NT5C3A (lower panel). cDNA was amplified by RT-PCR using primers located in the AFE and in its downstream exon. The PCR products were analyzed by automated electrophoresis using a Bioanalyzer 2100 and quantified by comparison with a loading control. The values were then corrected by the housekeeping gene β-actin. ( n = 4 (EndoC) and n = 6 (human islets), two-sided paired t -test). c View of the RMI2 locus showing all the transcripts in this region, the ATAC-seq (blue) and the RNA-seq (red) signals of EndoC-βH1 cells exposed or not to IFNα for 24 h, the CAGE TSSs information (black scale) 45 and RNA polymerase II ChIP-seq signal of human K562 cells exposed to IFNα (black) 48 . A higher magnification of the RMI2-004 locus is presented below (image representative of 4–5 independent experiments). d Confirmation of the AFE usage in the gene RMI2. Genome mapping (upper part) showing the genomic regions used to design-specific primers located in the AFE of the transcript RMI2-004 and in its downstream exon. The PCR product was analyzed by automated electrophoresis using a Bioanalyzer 2100 and quantified by comparison with a loading control. The values were then corrected by the housekeeping gene β-actin. ( n = 4 (EndoC) and n = 6 (human islets), two-sided paired t -test). Source data are provided as a Source Data file. Full size image Next, we compared the frequency of gained OCRs among alternative promoters. As the SEASTAR pipeline mainly recognizes non-redundant FEs, we evaluated alternative promoters identified by both the SEASTAR pipeline and the FAMTON5 database of alternative TSSs 45 (Supplementary Methods). We thus identified 198 and 51 gained OCRs present in alternative promoter regions at 2 and 24 h, respectively. Characterization of the IFNα-induced alternative promoters presenting a major gain in chromatin accessibility pointed to the T1D risk gene RMI2 47 . At gene level, there was only a ~1.4-fold upregulation of RMI2 expression, but at the transcript level there was a >60-fold increase in two isoforms, RMI2 −002 and −004. Visualization of the RMI2 locus combined with ATAC-seq and RNA-seq peaks indicated that the isoform RMI2 −004 gained chromatin accessibility in its promoter leading then an increase in mRNA expression (Fig. 6c ). Data from CAGE analysis 45 and RNA polymerase II ChIP-seq of another human cell type exposed to IFNα 48 (Fig. 6c , lower part) confirms the presence of the RMI2 alternative promoter. The IFNα-induced RMI2 −004 upregulation was confirmed using specific primers in both EndoC-βH1 cells and human islets (independent samples) (Fig. 6d ). These findings support a double mechanism by which IFNα affects human beta cells, i.e. first a massive change in open chromatin regions followed by later changes in gene expression and AS (see above) and also AFE usage. Mining IFNα signatures to identify T1D therapeutic targets Considering the significant overlap observed between gene profiles of IFNα-exposed EndoC-βH1 cells and beta cells from T1D individuals (Supplementary Fig. 2d ), mining these common signatures might identify relevant T1D therapeutic targets. First, the top 150 commonly upregulated genes detected by the RRHO analysis of both IFNα-exposed EndoC-βH1 cells and beta cells from T1D individuals were selected (Supplementary Fig. 2d and Fig. 7a ) to query the Connectivity Map database 49 . We focused in opposite signatures of perturbagens that may reverse the effects of IFNα. To decrease off-target findings based on individual compounds, the analysis was performed considering only the classes of perturbagens. Four main classes, including bromodomain inhibitors, potentially reversed the signature from our query (tau score < −90) (Fig. 7b ). Comparable results were obtained when analyzing the intersection of IFNα-exposed pancreatic human islets and beta cells from T1D individuals (Supplementary Fig. 13a ). Bromodomain inhibitors have been shown to prevent autoimmune diabetes in animal models 50 and the KD of the bromodomain containing 2 gene ( BRD2 ) induced an opposite signature to our model (Supplementary Fig. 13b ). Pre-treatment of EndoC-βH1 cells with two bromodomain inhibitors decreased both IFNα-induced HLA-I and CXCL10 induction, with no changes in CHOP (DDIT3) expression (Fig. 7c, e ) or in apoptosis induced by IL1β + IFNα (Fig. 7d, f ). In human islets, these inhibitors induced a ~30% decrease in IFNα-induced HLA-I expression and a 90% reduction in CXCL10 expression; at least in the context of I-BET-151, there was a 60% reduction of the ER stress marker CHOP (DDIT3) (Supplementary Fig. 13c, d ). Fig. 7: Mining the type I interferon signature of pancreatic beta cells for identification of potentially T1D therapeutic targets. a The top 150 upregulated genes identified in Supplementary Fig. 2d were used to query the Connectivity MAP database of cellular signatures 49 . b Connectivity map classes of perturbagens that promote an opposite signature to the one shared between beta cells of T1D individuals and EndoC-βH1 cells exposed to IFNα (Supplementary Fig. 2d ). c , e EndoC-βH1 cells were pretreated for 2 h with the bromodomain inhibitors I-BET-151 (1 μM) ( c ) or JQ1 + (0.4 μM) ( e ) and then exposed to IFNα for 24 h. Cells were collected and the mRNA expression for HLA class I (ABC), the chemokine CXCL10 and the ER stress marker CHOP (DDIT3) evaluated. Ethanol (vehicle) and an inactive enantiomer (JQ1−) were used as respective controls for I-BET-151 and JQ1+. ( n = 5, mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). d , f Cell viability after exposure to the combination of cytokines IFNα (2000 U/ml) + IL1β (50 U/ml) in the presence or not of the bromodomain inhibitors ( n = 3, mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). Full size image Next, we searched for clinically approved drugs (DrugBank 5.1 51 ) among the PPI network of the WGCNA module #2 (Fig. 4f ), with a view to possible drug repurposing. Module #2 is particularly interesting in this context as it recapitulates many of the key biological processes induced by IFNα (Fig. 4g ), and because ~50% of its members were also present among the most upregulated genes from the RRHO analysis (Supplementary Fig. 2d ). An interesting target recognized as a hub for different drugs was the kinase JAK1 (Fig. 8a ) and its inhibitor baricitinib, which has shown promising effects in the treatment of human rheumatoid arthritis 52 . Baricitinib prevented IFNα-induced mRNA expression of HLA-I , CXCL10 and CHOP (DDIT3) in EndoC-βH1 cells (Fig. 8b ) and human islets (Fig. 8c ) and it completely protected EndoC-βH1 cells (Fig. 8d ) and human islets (Fig. 8 e) against the pro-apoptotic effects of IFNα + IL1β. Furthermore, baricitinib decreased the cell surface protein expression of MHC class I by >90% in EndoC-βH1 cells (Fig. 9a ) and human islets (Fig. 9b, c ). Fig. 8: Establishing JAK1 inhibition as protective mechanism against IFNα-mediated inflammation and apoptosis. a The PPI network of module #2 was integrated with the DrugBank repository 51 using the CyTargetLinker app 78 in Cytoscape. A higher magnification on JAK1 is shown. b EndoC-βH1 cells were pretreated with DMSO (NT) or baricitinib at the indicated concentrations for 2 h. Cells were then left untreated (black bars), or treated with IFNα alone (white bars) without or with the presence of different concentrations of baricitinib (purple scale bars) for 24 h and mRNA expression of HLA class I (ABC), CXCL10 and CHOP (DDIT3) analyzed. The values were normalized by the housekeeping gene β-actin and then by the highest value of each experiment considered as 1 ( n = 4, mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). c Human islets were pretreated with baricitinib (4 μM) or DMSO (vehicle) and then exposed or not to IFNα for 24 h in the presence or not of baricitinib. mRNA expression of HLA class I (ABC), CXCL10 and CHOP (DDIT3) was analyzed and values normalized by the housekeeping gene β-actin and then by the highest value of each experiment considered as 1. ( n = 3, mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). d , e EndoC-βH1 cells ( d ) and human islets ( e ) were pretreated with DMSO or baricitinib (4 µM) for 2 h. Subsequently, cells were left untreated or treated with IFNα (2000 U/ml) + IL1β (50 U/ml) in the absence or presence of baricitinib for 24 h. Cell viability was evaluated using nuclear dyes by two independent observers. ( d ( n = 5), e ( n = 4), mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). Source data are provided as a Source Data file. Full size image Fig. 9: Baricitinib decreases IFNα-mediated MHC class I protein expression in beta cells. a EndoC-βH1 cells were pretreated with baricitinib (4 μM) or DMSO and then exposed or not to IFNα for 24 h in the presence or not of baricitinib. MHC class I (ABC) protein expression was measured by flow cytometry. The percentage of positive cells was quantified. ( n = 6, mean ± SEM, ANOVA with Bonferroni correction for multiple comparisons). b , c Dispersed human islets were pretreated with baricitinib (4 μM) or DMSO (vehicle). Next, cells were left untreated, treated with IFNα alone or with IFNα in the presence of baricitinib for 24 h. MHC class I intensity was quantified in each condition ( b ) using Fiji software 80 and normalized by the HOECHST intensity to correct for the number of cell per area ( n = 3, ANOVA with Bonferroni correction for multiple comparisons, RFU (relative fluorescence units)). Immunocytochemistry (ICC) analysis ( c ) of MHC class I (ABC) (red), insulin (green) and HO (blue) was performed to confirm MHC class I expression in three independent human islet preparations. Scale bar 10 μm. Full size image Discussion We presently modeled the initial changes observed in the islets of Langerhans during T1D by performing an integrated multi-omics approach in EndoC-βH1 cells exposed to the early cytokine IFNα. The model was validated using human islets RNA-seq and independent experiments using the same human beta cell line, pancreatic human islets and FACS-purified human beta cells. Of relevance, taking into account the major differences between human and rodent beta cell responses to stressful stimuli 53 , 54 , all experiments were performed in clonal or primary human beta cells/islets. This approach identified very rapid and broad beta cell responses to IFNα including: (1) major early modifications in chromatin remodeling, which activates regulatory elements; (2) the key TFs regulating signaling, and the crosstalk between beta cells and immune cells; (3) the functional modules of genes and their regulatory networks; and (4) alternative splicing and first exon usage as important drivers of transcript diversity. Finally, an integrative analysis led to the identification of two compound classes that reverse all or part of these alterations in EndoC-βH1 cells and human islets and may be potential therapeutic targets for future trials in T1D prevention/treatment. During viral infection a prompt innate immune response, mediated to a largest extent via type I interferons, is critical to control virus replication and spreading 55 . In line with this, exposure of human beta cells to IFNα leads to changes in chromatin accessibility already at 2 h, which correlates with subsequent changes in mRNA and protein expression at 8 and 24 h. The majority of these regions are localized distally to TSSs, indicating that they may act primarily as distal regulatory elements. Interestingly, these regions were enriched in motifs of islets-specific TFs, suggesting that tissue-restricted characteristics regulate the local responses during insulitis, as we have recently described for the cytokines IL1β + IFNγ 19 . This could explain the preferential expression of HLA class I (both the classical ABC members and the presently described inhibitory HLA-E) by pancreatic islets in comparison to the surrounding exocrine pancreas. Islet HLA class I overexpression is a key finding during T1D development 56 , contributing for the recruitment of autoreactive CD8 + T cells that selectively attack beta cells 1 . IFNα also induces pathways involved in protein modification (ubiquitination, sumoylation, etc), degradation (proteasome, etc) and ER stress, which can generate neoantigens 14 . The IRF and STAT family members are master TFs involved in IFN-I signaling 2 . Viruses have developed several species-specific mechanisms to antagonize STAT1 and STAT2 activation 55 . For instances, the NS5 protein of Zika virus degrades human but not mouse STAT2 57 . In the present work, we confirmed the importance of both STAT1 and 2 for INFα signaling in beta cells, and observed that their individual KD is compensated in most cases by the remaining member, as a possible backup mechanism to protect against pathogens 58 . Interestingly, IRF1 seems to be a critical regulator of the IFNα-mediated “defense” responses in beta cells, including induction of checkpoint proteins such as PDL1 and HLA-E (present data), and the suppressors of cytokine signaling 1 and 3 (SOCS3) (ref. 59 and present data). This stands in contrast to its pro-inflammatory effects in immune cells 60 . In line with the possible role for IRF1 in dampening islet inflammation, systemic knockout of IRF1 prevents autoimmune diabetes in NOD mice 61 , whereas IRF1 deletion in islets is associated with shorter mouse allograft graft function and survival 62 . Alternative splicing (AS) is a species, tissue and context-specific post-transcriptional mechanism that expands the number of transcripts originated from the same gene thus increasing protein diversity 63 . Pancreatic beta cells share many characteristics with neuronal cells, including analogous signal transduction, developmental steps and splicing networks 64 . Both T1D risk genes 65 and the cytokines IL1β + IFNγ 10 modify AS in beta cells. We presently identified a preferential alternative exon inclusion after IFNα exposure and mapped the potentially involved RBPs, which included the upregulated protein Quaking ( QKI ). QKI activation in monocytes promotes extensive changes in AS, favoring their differentiation into pro-inflammatory macrophages 66 . Furthermore, QKI binds to the genome of RNA viruses and inhibits their replication 67 . A similar mechanism was recently described for FMR1 68 , another RBP induced by IFNα, which controls protein translation in beta cell (present data). Several other RBPs were observed as downregulated by IFNα and identified as potential regulators of IFNα-induced AS events. Thus, inhibition of ELAVL1 and HNRNPA1 reproduced IFNα-mediated changes in exon usage. Different RNA viruses can use both ELAVL1 69 and HNRNPA1 70 to support their replication, indicating that the decreased expression of these proteins may provide an additional IFN-triggered antiviral mechanism. These findings suggest that during potentially diabetogenic viral infections, RBPs may have a dual role: first as splicing regulators and second as regulators of viral replication. In order to identify novel approaches to protect beta cells in T1D, we analyzed the similarities between beta cell signatures from T1D donors and those following IFNα exposure, and compared the top identified genes/pathways with the Connectivity Map 49 and the DrugBank 51 database. This identified two groups of potential therapeutic agents, namely bromodomain and JAK inhibitors. Bromodomain (BRD) proteins are components of chromatin-remodeling complexes that promote chromatin decompaction and transcriptional activation. BET inhibitors have shown protective effects in different animal models of autoimmunity 71 , including the diabetes-prone NOD mice 50 . We have now expanded these findings to human beta cells, showing that two distinctive BET inhibitors (JQ1+ and I-BET-151) decrease IFNα-induced responses, including HLA class I and chemokine overexpression. After binding to its receptor, IFNα promotes phosphorylation of two tyrosine kinases, JAK1 and TYK2, which then trigger the downstream signaling cascade. Chemical inhibition of JAK1 + JAK2 prevents autoimmune diabetes in NOD mice 72 and polymorphisms associated with decreased TYK2 function are protective against human T1D 73 . We presently observed that baricitinib, a JAK1/2 inhibitor recently approved for use in rheumatoid arthritis by the FDA 52 , decreased all the three hallmarks previously identified in islets of T1D individuals and initiated by IFNα in human beta cells, namely HLA class I overexpression, ER stress and beta cell apoptosis, supporting its future testing in T1D. In conclusion, we have applied a multi-omics approach to study the different levels of gene regulation induced by IFNα in EndoC-βH1 cells and pancreatic human islets. This in vitro modeling showed strong correlation with the mRNA profile from beta cells of T1D individuals. At the genomic level, early chromatin remodeling activated cis -regulatory elements, many of them presenting motifs for islets-specific TFs, providing a possible mechanism by which tissue-restricted autoimmune diseases might arise. Post-translational modifications, alternative splicing and first exon usage were induced by IFNα, likely expanding the repertoire of proteins and transcripts generated by beta cells in response to this inflammatory stimuli. This can also be a source of potential neoantigens. Interestingly, IFNα-exposed human beta cells upregulate co-inhibitory proteins such as PDL1 and HLA-E, which may attenuate or delay the autoimmune assault. Finally, the present results provide a useful resource for the discovery of compounds that may be used to reverse the effects of IFNα on human pancreatic beta cells, paving the way for potential T1D interventional trials. Methods Culture of EndoC-βH1 cells and human islets, cell treatment The human pancreatic beta cell line EndoC-βH1 was kindly provided by Dr. R. Scharfmann, University of Paris, France 74 . Human islet isolation from 20 non-diabetic organ donors (Supplementary Table 2 ) was performed in accordance with the local Ethical Committee in Pisa, Italy. The use of pancreatic human islets for this project was approved by the Comité d’Ethique hospitalo-facultaire Erasme-ULB. These cells were maintained in culture and treated as described in Supplementary Methods. FACS-purified human beta cells isolation and treatment Whole pancreatic human islets were exposed or not to IFNα for 24 h. After this period, the islets were dispersed into single cells and surface staining was carried out in FACS buffer (PBS with 0.5% BSA and EDTA 2 mM final concentration). Indirect antibody labeling was performed with two sequential incubation at 4 °C and one wash in FACS buffer followed each step. Cells were resuspended in FACS buffer, viability dye was added (DAPI) and cells were sorted on a FACSAria III cell sorter (BD Biosciences). Primary (mouse anti-human NTPDase3, hN3-B3S, ) and secondary (Alexa Fluor 546 conjugated donkey anti-mouse IgG (A10036, Thermo-Fisher Scientific)) antibodies were used with the dilutions described in Supplementary Table 5 . Data analysis was carried out with FlowJo software (Version 10). ATAC sequencing processing and analysis ATAC sequencing was performed in four independent experiments for each time point (2 and 24 h) 75 . For ATAC-seq 50,000 EndoC-βH1 cells were exposed or not to IFNα for 2 or 24 h. After that, the cells were harvested, and the nuclei isolated by using 300 μl of cold lysis buffer (10 mM Tris–HCl pH 7.4, 10 mM NaCl, 3 mM MgCl 2 , 0.1% Igepal CA-630). The nuclei pellet was resuspended in a 25 μl transposase reaction mix containing 2 μl of Tn5 transposase per reaction and incubated at 37 °C for 1 h. The tagmented DNA was isolated using SPRI cleanup beads (Agencourt AMPure XP, Beckman Coulter). For library amplification two sequential 9-cycle PCR were performed (72 °C for 5 min; 98 °C for 30 s; 9 cycles of 98 °C for 10 s, 63 °C for 30 s; and 72 °C for 1 min; and at 4 °C hold). Finally, the DNA library was purified using the MinElute PCR Purification Kit (Qiagen, Venlo, Netherlands). TapeStation and semi-quantitative PCR assays at target positive and negative controls were performed to ensure the quality and estimate the efficiency of the experiment before sequencing. Libraries were sequenced single-end on an Illumina HiSeq 2500. Data processing and analysis is described in Supplementary Methods. RNA-sequencing processing and analysis Total RNA of five independent experiments with EndoC-βH1 cells and six independent preparation of pancreatic human islets exposed or not to IFNα for different time points was obtained using the RNeasy Mini Kit (Qiagen, Venlo, Netherlands). RNA integrity number (RIN) values were evaluated using the 2100 Bioanalyzer System (Agilent Technologies, Wokingham, UK). All the samples analyzed had RIN values >9. mRNA was obtained from 500 ng of total RNA using oligo (dT)beads, before it was fragmented and randomly primed for reverse transcription followed by second-strand synthesis to generate double-stranded cDNA fragments. The cDNA undergone paired-end repair to convert overhangs into blunt ends. After 3′-monoadenylation and adaptor ligation, cDNAs were purified. Next, cDNA was amplified by PCR using primers specific for the ligated adaptors. (Illumina, Eindhoven, Netherlands). The generated libraries were submitted to quality control before being sequenced on an Illumina HiSeq 2500. RNA-seq data processing and analysis is described in Supplementary Methods. Proteomics processing and analysis EndoC-βH1 cells exposed or not to IFNα were extracted using the Metabolite, Protein and Lipid Extraction (MPLEx) approach. A detailed description of the method used for proteomics processing and analysis is provided in Supplementary Methods. Rank–rank hypergeometric overlap (RRHO) analysis To compare the signature induced by IFNα with the one present during insulitis in T1D individuals, we performed the RRHO mapping 15 . For this goal, a full list of log 2 FC ranked genes from our RNA-seq of EndoC-βH1 cells and human islets (IFNα vs Control, 24 h) were compared against similarly ranked lists of purified primary beta cells obtained from individuals with T1D 16 and T2D 17 (T1D/T2D vs non-diabetic). In a RRHO map, the hypergeometric p -value for enrichment of k overlapping genes is calculated for all possible threshold pairs for each experiment, generating a matrix where the indices are the current rank in each experiment. The log-transformed hypergeometric p -values are then plotted in a heatmap indicating the degree of statistically significant overlap between the two ranked lists in that position of the map. Multiple correction was applied using the Benjamini–Yekutieli FDR correction. Dynamic regulatory events miner (DREM) modeling For reconstructing dynamic regulatory networks, we have used the DREM method 23 , which integrates times series and static data using an Input-Output Hidden Markov Model (IOHMM), where the TF-DNA interaction information obtained from ChIP-seq experiments 48 was used as the input and our RNA-seq time series expression data as the output. A detailed description of DREM-based modeling is provided in Supplementary Methods. Weighted gene co-expression network analysis (WGCNA) On each dataset (RNA-seq and proteomics), we obtained modules of genes/proteins of similar expression profiles using WGCNA 28 . The soft threshold parameter for the RNA-seq dataset was chosen to be 10 (value to approximate a scale-free topology). Similar parameters were used for the analysis of RNA-seq of pancreatic human islets exposed or not to IFNα. Regarding the proteomics dataset, in order to achieve an approximated scale-free topology, we first normalized each protein expression in each temporal group (subtraction by mean and division by standard deviation), and then selected the soft threshold parameter as 14. After merging the modules using a dissimilarity threshold of 0.25, we identified 8 modules in the RNA-seq dataset and 7 modules in the proteomics dataset. To analyze module quality, we have used a set of statistics (density and separability metrics) from the modulePreservation function of the R package WGCNA 29 . For this purpose, we resampled the dataset 1000-times to create reference and test sets from the original data and evaluate module preservation, represented as the Z summary for each module across the resulting networks. Z summary > 2 indicates moderate preservation and Z > 10 high quality/preservation for each module 29 . To evaluate WGCNA module preservation in independent samples, we used the same R function, but in this case applying metrics based on module density and intramodular connectivity to give a composite statistic Z summary . To evaluate the overlap of RNA-seq and proteomics modules, we considered a mRNA to be differentially expressed at 8 or 24 h if its absolute fold-change was >1.5 and its FDR < 0.05. Regarding the proteomics dataset, we considered a protein to be differentially abundant at 8 or 24 h if the t -test p -value was <0.05. We selected only the differentially expressed genes/abundant proteins in the identified WGCNA modules. We then searched for the overlap between the elements of the RNA-seq and proteomics modules and obtained an overlap p -value (hypergeometric probability). We retained overlapping modules with a FDR < 0.05 and a minimum of 10 common elements. Protein–protein interaction network analysis The inBio Map protein–protein interaction (PPI) network database 31 was obtained from . We first restricted the network to contain only the elements expressed in human beta cells based on our RNA-seq database (mean RPKM > 0.5 in at least one condition). For each WGCNA overlapping module, we identified the proteins in the PPI network with a significantly high number of protein-to-protein connections to the set of elements in the module (FDR < 0.01, and minimum number of connections equal to 5). We considered only networks obtained for the overlapping modules #1, #2 and #5, as the other overlapping modules returned empty PPI networks. We then obtained PPI networks for each WGCNA overlapping modules, involving the original set of module elements, plus the respective identified connecting proteins. Communities of interacting proteins were identified using the EAGLE algorithm 76 with the following parameters: CliqueSize threshold: 6 and ComplexSize threshold: 2. Wordclouds of each community were generated using information from geneRIFs terms. Gene regulatory network analysis A network of regulatory interactions was obtained from RegNetwoks 30 ( ). As in the PPI network, we first restricted the network to contain only the elements we found to be expressed in the RNA-seq dataset. Similarly to the PPI network analysis, for each WGCNA overlapping modules, we identified regulators with a significantly high number of regulatory connections to the set of elements in the module (FDR < 0.01, and minimum number of connections equal to 4). We then obtained regulatory networks for each WGCNA overlapping modules, involving the original set of module elements, plus the respective identified regulators. To create a non-redundant dataset of motifs from regions of gained open chromatin, we used the compareMotifs.pl script from the package HOMER 77 to merge motifs with a similarity score threshold of 0.7. The remaining motifs were mapped to the gain open chromatin regions using the annotatePeaks.pl script. Transcription factor motif analysis Sequence composition analysis of de novo motifs was performed using findMotifGenome.pl from the package HOMER 77 with parameters ‘-size given -bits -mask’. The motifs having a p ≤ 10 −12 and observed in >3.5% of the targets were chosen for subsequent analysis. All de novo matches having a similarity score to known TF motifs higher than 0.7 are shown in the tables (Fig. 4e and Supplementary Fig. 3c ), or when no match was present over this threshold, the first hit was elected and its score is presented. Alternative splicing changes validation Alternative splicing changes identified from RNA-seq were validated by RT-PCR using specifically designed primers (Supplementary Table 3 ). To confirm cassette exons, the primers were adjacent to the predicted splicing event. This approach allowed us to discriminate between variants based on their fragment sizes. For alternative first exon usage (AFE) validation, we have designed primers spanning regions that are unique to the isoform of interest (Fig. 6g ), and then normalized the results by the housekeeping gene β-actin. cDNA was amplified using MyTaq Red DNA polymerase (Bioline, London, UK), and PCR products were analyzed using an Agilent 2100 Bioanalyzer system (Agilent Technologies, Wokingham, U.K.). The molarity of each PCR band corresponding to a specific splice variant was quantified using the 2100 Expert Software (Agilent Technologies, Diegem, Belgium), and used to calculate the ratio inclusion/exclusion (SE) or isoform-X/β -actin (AFE). Small-RNA interference Transfection was performed using Lipofectamine RNAiMAX (Invitrogen) as described in Supplementary Methods. After that, the cells were kept in culture for a 48 h recovery period and subsequently exposed or not to IFNα as indicated. Supplementary Table 3 describes the sequences of siRNAs used in the present study. Real-time PCR analysis After harvesting of the cells, Poly(A) + mRNA was obtained using the Dynabeads mRNA DIRECT kit (Invitrogen) and reverse transcribed. Detailed description is provided in Supplementary Methods. Western blot, immunocytochemistry and flow cytometry Detailed description together with additional information on western blot, immunocytochemistry and flow cytometry analysis is provided in Supplementary Methods. Immunofluorescence After dewaxing and rehydration, samples were subjected to heat-induced epitope retrieval (HIER) in 10 mM citrate buffer pH 6.0, then probed in a sequential manner with appropriate antibodies as indicated in Supplementary Table 4 . The relevant antigen–antibody complexes were detected using secondary antibodies conjugated with fluorescent dyes (Invitrogen, Paisley, U.K). Cell nuclei were stained with DAPI. After mounting, images were captured with a Leica AF6000 microscope (Leica, Milton Keynes, UK) and processed using the standard LASX Leica software platform (Version 1.9.013747). For quantification studies, randomly selected insulin-containing islets (ICIs) from individuals with or without diabetes were imaged, in addition to insulin-deficient islets (IDIs) from individuals with diabetes. Thirty ICIs were analyzed from 6 independent individuals (5 islets per individual), 20 IDIs were analyzed from 4 independent individuals (5 islets per individual) and 30 ICIs were analyzed from 6 independent control individuals (5 islets per individual). The mean fluorescence intensity (MFI) arising from detection of HLA-E was measured using LASX Leica quantification software. Therapeutic targets identification The top 150 upregulated genes shared among the RNA-seq of EndoC-βH1 cells and human islets exposed to IFNα for 24 h and the RNA-seq of beta cells 16 from T1D individuals were identified by the RRHO analysis. This list of genes was used to query the Connectivity Map dataset of L1000 cellular signatures, which has transcriptional responses of human cells to different chemical and genetic perturbations, using the CLUE platform ( ) 49 . To identify compounds potentially reverting the effects induced by interferons in beta cells, we have focused on perturbagens promoting signatures that were opposite (negative tau score) to our query list. Only perturbagens having a median tau score < −90 were considered for further evaluation. Additionally, aiming at potential repurposing of drugs under clinical investigation for treatment of other pathologies, we have integrated the DrugBank database v5.1 51 with the PPI network obtained from WGCNA module #2 using the CyTargetLinker v4.0.0 78 within Cytoscape v3.6 to build a biological network annotated with drugs. The small molecules and drugs pointed out by these two approaches were then validated in vitro as described above to verify their impact on IFNα-induced upregulation of cytokines/chemokines, ER stress markers, HLA class I and beta cell apoptosis. Cell viability assessment The cell viability is described in details in Supplementary Methods. Statistical analysis Data of the confirmatory experiments are expressed as means ± SEM. A significant difference between experimental conditions was assessed by paired t -test, unpaired t- test, one-way or two-ways ANOVA followed by Bonferroni correction for multiple comparisons as indicated using the GraphPad Prism program version 6.0 ( ). Results with p ≤ 0.05 were considered statistically significant. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability All raw and processed ATAC and RNA-sequencing data that support the findings of this study have been deposited in NCBI Gene Expression Omnibus (GEO) with the primary accession code GSE133221 (subseries are GSE133218 : RNA-seq of EndoC-βH1 cells, GSE148058 : RNA-seq of human islets, GSE133219 : ATAC-seq of EndoC-βH1 cells). The proteomics datasets have been submitted to Pride under identifier number PXD014244 ( ). The network of regulatory interactions can be obtained from RegNetwoks ( ). The DrugBank database v5.1 can be downloaded from: . The inBio Map protein–protein interaction (PPI) network database can be obtained from: . The CAGE peaks from FANTOM5 database can be obtained on: . The Connectivity Map database can be accessed using the CLUE platform ( ). The RNA polymerase II (POLR2A) ChIP-seq of human K562 cells can be obtained from the ENCODE project ( GSM935474 , ). The Exon Ontology database can be accessed from: . The information about T1D risk genes can be found on immunobase ( ) and GWAS catalog ( ). The source data underlying Figs. 2 c–m, 3 a–e, g, i, 5 d, g, j, 6 b, d, 7 c–f, 8 b–e, 9a, b and Supplementary Figs. 4e –h, 5c –m, 6b , d , 12a –c, 12e –g, 13c–d are provided as a Source data file.

Researchers from the Indiana Biosciences Research Institute (IBRI), a leading independent, industry-inspired applied research institute, and Université Libre de Bruxelles (ULB) Center for Diabetes Research, identified two classes of compounds that prevent most of the effects of interferon-α (IFNα) on human beta cells, paving the way for potential future clinical trials of treatments for type 1 diabetes (T1D). Dr. Decio Eizirik, scientific director of the IBRI Diabetes Center and Professor at the ULB Center for Diabetes Research, and Dr. Maikel Colli, a researcher at the ULB Center for Diabetes Research, used an innovative "multi-omics" approach, funded by JDRF, the European IMI consortium INNODIA and the Belgian Funding Agency Welbio, that combined genomic, transcriptomic and proteomic techniques with advanced bioinformatic tools to analyze the initial changes present in human beta cells exposed to IFNα. "This is a beautiful example of international collaboration and translational research led by scientists at the IBRI and the ULB," said Eizirik. "Indeed, it started with use of complex omics technology and bioinformatics and ended up with the identification of two agents that may be one day re-purposed for the early treatment of T1D." T1D is a chronic autoimmune disease resulting in the destruction of the insulin-producing beta cells. Nearly 1.6 million Americans are affected by T1D. The early steps of the disease involve local release of pro-inflammatory mediators (cytokines) at the pancreatic islet level. One of these mediators is IFNα. In line with this, pancreatic islets obtained from living donors with recent onset T1D have a significant increase in the expression of IFN-stimulated genes, while inhibition of IFNα signaling prevents the development of T1D in animal models. There are currently no effective treatments to prevent T1D. "JDRF's mission is to cure, treat and prevent T1D. To do so, we must validate key biological pathways as integral in T1D disease processes. Dr. Eizirik's work provides excellent justification for IFNα as a contributing factor in the development and progression of T1D," said Sanjoy Dutta, Ph.D., JDRF vice president of research. "Importantly, his work identifies two drug candidates that, based on his findings, should be beneficial to those at risk of or living with T1D. We look forward to clinical testing of such drugs in people with T1D." The study, which was recently published in Nature Communications ("An integrated multi-omics approach identifies the landscape of interferon-α mediated responses of human pancreatic beta cells"), is based on collaborations with colleagues from Belgium, Spain, UK, Italy and USA. The changes induced by IFNα were similar to observations made in beta cells obtained from patients affected by T1D. IFNα promotes rapid changes in chromatin (the complex DNA+protein present in the nucleus) accessibility. These changes are probably required to enable gene expression to fight local viral infections but may contribute to trigger autoimmunity and T1D in genetically susceptible individuals. Furthermore, beta cells exposed to IFNα increased expression of proteins inhibiting the immune system, such as PDL1 and HLA-E, which may help to decrease and/or delay the autoimmune assault. This latest finding may explain why cancer immunotherapies using PDL1 blockers causes T1D in some patients.

10.1038/s41467-020-16327-0

Medicine

Liver fix thyself—How some liver cells switch identities to build missing plumbing

De novo formation of the biliary system by TGFβ-mediated hepatocyte transdifferentiation, Nature (2018). nature.com/articles/doi:10.1038/s41586-018-0075-5 Journal information: Nature

http://nature.com/articles/doi:10.1038/s41586-018-0075-5

https://medicalxpress.com/news/2018-05-liver-thyselfhow-cells-identities-plumbing.html

Abstract Transdifferentiation is a complete and stable change in cell identity that serves as an alternative to stem-cell-mediated organ regeneration. In adult mammals, findings of transdifferentiation have been limited to the replenishment of cells lost from preexisting structures, in the presence of a fully developed scaffold and niche 1 . Here we show that transdifferentiation of hepatocytes in the mouse liver can build a structure that failed to form in development—the biliary system in a mouse model that mimics the hepatic phenotype of human Alagille syndrome (ALGS) 2 . In these mice, hepatocytes convert into mature cholangiocytes and form bile ducts that are effective in draining bile and persist after the cholestatic liver injury is reversed, consistent with transdifferentiation. These findings redefine hepatocyte plasticity, which appeared to be limited to metaplasia, that is, incomplete and transient biliary differentiation as an adaptation to cell injury, based on previous studies in mice with a fully developed biliary system 3 , 4 , 5 , 6 . In contrast to bile duct development 7 , 8 , 9 , we show that de novo bile duct formation by hepatocyte transdifferentiation is independent of NOTCH signalling. We identify TGFβ signalling as the driver of this compensatory mechanism and show that it is active in some patients with ALGS. Furthermore, we show that TGFβ signalling can be targeted to enhance the formation of the biliary system from hepatocytes, and that the transdifferentiation-inducing signals and remodelling capacity of the bile-duct-deficient liver can be harnessed with transplanted hepatocytes. Our results define the regenerative potential of mammalian transdifferentiation and reveal opportunities for the treatment of ALGS and other cholestatic liver diseases. Main In regenerating organs of adult mammals, differentiated cells can replenish other types of differentiated cells by transdifferentiation, as in the pancreatic islet 10 , gastric gland 11 , lung alveolus 12 and intestinal crypt 13 . Whether mammalian transdifferentiation can build these or other structures de novo is unknown. In the liver, hepatocytes can undergo biliary differentiation to form reactive ductules both in humans 14 , 15 , 16 , 17 and animals with cholestatic liver injury 3 , 4 , 5 , 6 , 18 , 19 , 20 , 21 . However, hepatocyte-derived biliary cells exhibit incomplete biliary and residual hepatocyte differentiation, that is, they are not mature cholangiocytes, and revert back to their original identity after the injury is reversed 3 , 4 , 5 , 6 . Moreover, hepatocyte-derived ductules do not contribute to bile drainage 5 . These findings are consistent with metaplasia, but not with transdifferentiation, and call into question the functional importance of hepatocyte plasticity. However, all studies of hepatocyte plasticity published so far used animals with a fully developed biliary system in which residual cholangiocytes are available to regenerate injured bile ducts, probably leading to insufficient pressure for transdifferentiation. To determine the full extent of hepatocyte plasticity, we used mice that lack the intrahepatic biliary system. Specifically, we used mice that mimic the hepatic phenotype of ALGS—a human disease caused by impaired NOTCH signalling 22 , 23 , 24 —generated by deletion of floxed alleles of the NOTCH signalling effector RBPJ and, to prevent compensation 25 , the transcription factor HNF6 (also known as ONECUT1) in embryonic liver progenitors using Cre expressed from an albumin ( Alb ) promoter 2 . These Alb-cre +/− Rbpj f/f Hnf6 f/f mice are severely cholestatic because they lack peripheral bile ducts (pBDs), the branches of the intrahepatic biliary tree, at birth; however, more than 90% of the mice survive and form pBDs and a functional biliary tree by postnatal day (P) 120 26 (Fig. 1a ). Although Alb-cre +/− Rbpj f/f Hnf6 f/f mice have hilar bile ducts (hBDs), the trunk of the biliary tree, we hypothesized that the new pBDs originate from hepatocytes because the livers of Alb-cre +/− Rbpj f/f Hnf6 f/f mice contain hybrid cells that express hepatocyte and biliary markers 2 , 26 . To test this hypothesis, we developed Cre-independent hepatocyte fate tracing. For this, we activated the Flp-reporter (GFP) in R26ZG +/+ mice specifically in hepatocytes by intravenous injection of an adeno-associated virus (AAV) serotype 8 vector expressing Flp from the transthyretin ( Ttr ) promoter (AAV8-Ttr-Flp) (Fig. 1a and Extended Data Fig. 1a–g ). Fig. 1: Hepatocytes can convert into peripheral cholangiocytes and form pBDs contiguous with preexisting hBDs. a , De novo pBD formation and hepatocyte fate tracing in Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice. Cells identified by dolichos biflorus agglutinin (DBA) lectin labelling and wsCK and GFP immunofluorescence. b , Immunofluorescence of hepatocyte-fate-traced P120 Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mouse liver ( n = 7). c , Biliary tree visualized by retrograde ink injection into the common bile duct of P30 ( n = 6), P120–P138 ( n = 6) and ≥P334 ( n = 6) Alb-cre +/− Rbpj f/f Hnf6 f/f mice, and P30 ( n = 3) and P120–P138 ( n = 5) Rbpj f/f Hnf6 f/f mice. d , Maximum projection (top) and 3D reconstruction (bottom) of z -stack image of hepatocyte-fate-traced P120 Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mouse liver ( n = 2). e , Immunofluorescence and bright-field images of hepatocyte-fate-traced P468 Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mouse liver after retrograde ink injection into the common bile duct (≥P334, n = 3). Scale bars, 2 mm ( c , P30, P395 left), 500 µm ( c , P138 left), 250 µm ( c , P138 right), 100 µm ( b , c , P395 right, d , e , left), 25 µm ( e , middle). Full size image To trace hepatocytes during pBD formation, we intravenously injected Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice with AAV8-Ttr-Flp at P17, which is before cells expressing biliary markers emerge in the liver periphery 2 , 26 . At P120, the newly formed pBDs, identified by wide-spectrum (ws) cytokeratin (CK) immunofluorescence, were GFP positive, indicating hepatocyte origin (Fig. 1a, b ). By contrast, pBDs that formed normally during liver development in Cre-negative littermates were GFP negative (Extended Data Fig. 2a, b ). In many pBDs of Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice, all cells were GFP positive, whereas the overall labelling efficiency of peripheral cholangiocytes was 39.2 ± 7.2% (mean ± s.e.m.); however, this number correlated well with the hepatocyte labelling efficiency (36.6 ± 4.8%; P = 0.48, paired two-sided Student’s t -test) (Extended Data Fig. 2c ), indicating that all peripheral cholangiocytes originated from hepatocytes. We also observed hepatocyte-derived peripheral cholangiocytes in mice in which hepatocytes were labelled after weaning (P39) (Extended Data Fig. 2d, e ). These results show that hepatocytes can form pBDs de novo. To investigate the function of the hepatocyte-derived pBDs (HpBDs), we determined whether they are contiguous with the extrahepatic biliary system. At P30, retrograde ink injection into the common bile duct filled only hBDs, reflecting a severely truncated biliary tree that lacked pBDs; however, ink injection at ≥P120 revealed a biliary tree of normal dimensions, demonstrating that HpBDs are connected to hBDs (Fig. 1c ). Three-dimensional (3D) reconstruction of confocal imaging confirmed that HpBDs form contiguous lumens with dolichos biflorus agglutinin (DBA)-labelled hBDs (Fig. 1d and Supplementary Video 1 ). Accordingly, HpBDs were effective in draining bile, as evidenced by normalization of serum markers of cholestasis (total bilirubin and alkaline phosphatase levels; Extended Data Fig. 3a–c ) and hepatocyte injury (alanine aminotransferase and aspartate aminotransferase levels; Extended Data Fig. 3d, e ) and resolution of cholestasis-induced liver fibrosis (Sirius red staining; Extended Data Fig. 3f ). A few mice that continued to have increased levels of serum total bilirubin and liver fibrosis showed abundant wsCK-positive hepatocytes but no HpBDs (Extended Data Fig. 3a, b, g ). In contrast to the transient hepatocyte-derived ductules observed in other mouse models of cholestatic liver injury 3 , 4 , 5 , 6 , HpBDs were stable beyond the time when cholestasis and liver injury were resolved and were maintained for life (≥P334) (Fig. 1c, e ). These results show that HpBDs provide normal and stable biliary function. We also investigated the authenticity and maturity of the cholangiocytes that form the HpBDs. The cells displayed acetylated tubulin-marked primary cilia (Fig. 2a ), indicating a switch from hepatocyte to biliary fate 19 , and somatostatin receptor 2 (SSTR2) (Fig. 2b ), a marker of biliary function 27 . They also expressed the mature biliary markers epithelial cell adhesion molecule (EPCAM) and CK19 (Fig. 2c ), which are lacking or expressed at low levels in hepatocyte-derived metaplastic biliary cells 3 , 5 . To substantiate these results, we performed RNA-sequencing (RNA-seq) analysis on hepatocyte-derived peripheral cholangiocytes isolated as EPCAM-positive DBA-negative cells by FACS from >P115 Alb-cre +/− Rbpj f/f Hnf6 f/f mice (Extended Data Fig. 4a ). Rbpj and Hnf6 were deleted in these cells (Extended Data Fig. 4b, c ), which was expected because Alb-cre is active in embryonic liver progenitors before they commit to hepatocyte or biliary fate 2 , 9 . Principal component analysis showed that hepatocyte-derived peripheral cholangiocytes cluster closely with normal peripheral cholangiocytes isolated from Cre-negative littermates, but not with parental hepatocytes (Fig. 2d, e and Supplementary Table 1 ). Accordingly, hepatocyte-derived peripheral cholangiocytes expressed normal levels of mature biliary markers that are virtually undetectable in 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC) diet-induced hepatocyte-derived metaplastic biliary cells 3 , except for CFTR, which is less enriched in cholangiocytes than the other markers (Fig. 2f ). Hepatocyte-derived peripheral cholangiocytes also expressed other commonly used markers of biliary differentiation 3 , 5 , 20 , 25 , 27 , 28 at normal or near-normal levels (Fig. 2f ), and showed virtually no memory of hepatocyte differentiation 3 , 29 , 30 (Fig. 2g and Extended Data Fig. 4d ). These results show that hepatocyte-derived peripheral cholangiocytes are authentic and mature peripheral cholangiocytes. Fig. 2: Hepatocyte-derived peripheral cholangiocytes are equivalent to normal mature peripheral cholangiocytes. a – c , Immunofluorescence of hepatocyte-fate-traced P120 Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mouse liver ( n = 3 each). acTUB, acetylated tubulin. Scale bars, 100 µm ( c ), 20 µm ( a , b ). d – g , RNA-seq analysis of normal peripheral cholangiocytes (pC; n = 3 mice), hepatocyte-derived peripheral cholangiocytes (HpC; n = 4 mice) and RBPJ- and HNF6-deficient hepatocytes (H; n = 3 mice). d , Principal component analysis. e , Venn diagram showing the number of genes significantly differentially up- and downregulated in pC or HpC versus H. f , Heat maps of genes reflecting cholangiocyte differentiation, including genes lacking in DDC diet-induced hepatocyte-derived metaplastic biliary cells (top) and other marker genes (bottom). g , Heat maps of genes reflecting hepatocyte differentiation, including all differentially expressed cytochrome P450 (CYP) genes enriched in adult mouse liver (top) and other marker genes (bottom). One-way ANOVA, false-discovery rate (FDR)-corrected P < 0.05; fold change > 3 ( e – g ). Two-sided Student’s t -test; bold genes P < 0.05 for HpC versus pC ( f , g ). Full size image Next, we investigated the contribution of proliferation to HpBD formation, and sought to determine whether a few hepatocytes proliferate extensively after transdifferentiation, or whether many hepatocytes transdifferentiate. For this, we measured clonal expansion in HpBDs by sparsely labelling hepatocytes in Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice with low-dose AAV8-Ttr-Flp at P18 and quantifying the cells in GFP-positive clones at P120 (Fig. 3a ). Surprisingly, we found only 1.56 ± 0.05 cells per clone in HpBDs, with one-cell clones accounting for 63.8 ± 3.6% of clones. To exclude the possibility that the nonintegrating AAV vector missed proliferating hepatocytes, we analysed clone size in Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− mice in which hepatocytes and hilar cholangiocytes are stochastically labelled with one of four fluorescent proteins (Fig. 3b and Supplementary Video 2 ). We found 1.91 ± 0.06 cells per clone and 45.6 ± 1.1% one-cell clones in HpBDs at P150. In addition, the number of cells per clone in hBDs was similar between these mice and Alb-cre +/− R26R-Confetti +/− control mice at P90 (2.08 ± 0.08 and 2.41 ± 0.01; P = 0.060, two-sided Student’s t -test) (Extended Data Fig. 5a ). These results show that HpBDs form with little proliferation and are thus polyclonal, and confirm that hilar cholangiocytes do not contribute to HpBD formation. Fig. 3: HpBD formation entails little proliferation and is driven by TGFβ signalling. a , Possible outcomes, maximum projection image and size distribution of clones in hepatocyte-fate-traced P120 Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice ( n = 3). b , Possible outcomes, image stack volume projection and size distribution of clones in P150 Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− ( n = 4) and Alb-cre +/− R26R-Confetti +/− ( n = 3) mice. c , Ink visualization of biliary tree of >P120 Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f ( n = 16), Rbpj f/f Hnf6 f/f Tgfbr2 f/f ( n = 4) and Alb-cre +/− Rbpj f/f Hnf6 f/f ( n = 1) mice. d , Sirius red staining with quantification in >P120 Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f ( n = 4), Rbpj f/f Hnf6 f/f Tgfbr2 f/f ( n = 2) and Alb-cre +/− Rbpj f/f Hnf6 f/f ( n = 2) mice. Dotted lines represent the mean values of the indicated P120 mice from Extended Data Fig. 3f . e , f , Ink visualization of biliary tree and Sirius red staining with quantification in P100 Alb-cre +/− Rbpj f/f Hnf6 f/f mice that did ( n = 9) or did not ( n = 8) receive AAV8-Eef1a1-caTgfbr1. * P = 0.045, two-sided Student’s t -test. g , Serum total bilirubin in P36 Alb-cre +/− Rbpj f/f Hnf6 f/f mice that did ( n = 6) or did not ( n = 8) receive AAV8-Eef1a1-caTgfbr1. * P = 0.047, two-sided Welch’s t -test. Horizontal lines in a , b , d , f and g denote mean values. Scale bars, 2 mm ( c ), 500 µm ( e ), 100 µm ( d , f ), 50 µm ( a ), 20 µm ( b ). Source Data Full size image Unlike in HpBDs, we found significant proliferation in hepatocyte-derived reactive ductules, which are detectable in Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice during cholestasis (Extended Data Fig. 5b ). Clonal analysis showed 2.70 ± 0.08 cells per clone in reactive ductules in P90 Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− mice, with more three- and four-cell clones than in pBDs in Alb-cre +/− R26R-Confetti +/− mice (Extended Data Fig. 5c ). Because HpBDs form with little proliferation, we reasoned that these proliferating cells are hepatocyte-derived metaplastic biliary cells 3 , 4 , 5 , 6 . Indeed, we found that 93.51 ± 0.81% of the proliferating biliary cells identified by KI67 and osteopontin (OPN) immunofluorescence in Alb-cre +/− Rbpj f/f Hnf6 f/f mice near the peak of cholestasis (P54) were CK19 negative (Extended Data Fig. 5d ). We also investigated proliferation in established HpBDs by inducing cholestatic liver injury with a DDC diet in >P120 Alb-cre +/− Rbpj f/f Hnf6 f/f mice. Reactive ductules formed 2 weeks later than in Cre-negative littermates, and consisted mainly of OPN-positive EPCAM-negative cells, indicative of hepatocyte metaplasia (Extended Data Fig. 5e ). Indeed, after feeding a DDC diet to Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice in which hepatocyte fate tracing was induced at >P120, we found that 84.5 ± 5.6% of the cells in reactive ductules originated from hepatocytes, in contrast to 20.1 ± 3.0% in Cre-negative littermates (Extended Data Fig. 5f ). These results confirm that NOTCH signalling is important for cholangiocyte proliferation 31 , which explains our finding of a limited role of proliferation in HpBD formation and underscores the authenticity of hepatocyte-derived peripheral cholangiocytes. As illustrated by a lack of pBDs in Alb-cre +/− Rbpj f/f Hnf6 f/f mice (Fig. 1c ), NOTCH signalling is needed for bile duct development 7 , 8 , 9 , raising the question of which signalling pathway drives HpBD formation in its absence (Extended Data Fig. 4b, c ). We focused on TGFβ signalling because it promotes biliary differentiation and morphogenesis in development 32 , although it is not essential, as shown by the development of a normal biliary tree in mice that lack the TGFβ type II receptor (TGFBR2) in embryonic liver progenitors (Extended Data Fig. 6a ). We reasoned that TGFβ signalling is induced in hepatocytes of Alb-cre +/− Rbpj f/f Hnf6 f/f mice because of the paradoxical role of HNF6 in liver development—it activates the biliary transcription factors HNF1β and SOX9 25 , 33 , but inhibits TGFBR2 32 , 34 . Indeed, Gene Ontology (GO) term enrichment analysis of our RNA-seq data suggested active TGFβ signalling in hepatocyte-derived peripheral cholangiocytes, but not in normal peripheral cholangiocytes (Supplementary Table 1 , down pC versus HpC). Moreover, we found high levels of phosphorylated SMAD3 (pSMAD3) in the nuclei of periportal HNF1-positive epithelial liver cells and in whole-liver nuclear extracts in Alb-cre +/− Rbpj f/f Hnf6 f/f mice at P60 (Extended Data Fig. 6b, c ). We functionally validated these findings by showing that the TGFβ inhibitor SB-431542 blocks biliary differentiation and morphogenesis of hepatocytes that lack RBPJ and HNF6 in vitro (Extended Data Fig. 6d–f ). Moreover, at >P120, HpBDs were still absent or severely truncated, and liver fibrosis persisted in 14 out of 16 Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f mice that, like Alb-cre +/− Rbpj f/f Hnf6 f/f mice, were cholestatic at P34–P53 (Fig. 3c, d and Extended Data Fig. 6c, g ). These findings led us to investigate whether activating TGFβ signalling in hepatocytes enhances HpBD formation. We intravenously injected P19–P24 Alb-cre +/− Rbpj f/f Hnf6 f/f mice with an AAV serotype 8 vector expressing constitutively active TGFBR1 from the eukaryotic translation elongation factor 1α1 ( Eef1a1 ) promoter (AAV8-Eef1a1-caTgfbr1). At P80–P100, 9 out of 11 treated mice had a mature hierarchical biliary network, whereas 9 out of 9 untreated mice still had an immature homogeneous biliary network 35 (Fig. 3e ). Reflecting improved bile drainage, cholestasis and liver fibrosis resolved faster in treated mice (Fig. 3f, g ). We ruled out the possibility that AAV8-Eef1a1-caTgfbr1 is fibrogenic in Rbpj f/f Hnf6 f/f mice (Extended Data Fig. 6h ). These results identify TGFβ signalling as the driver of transdifferentiation and morphogenesis in HpBD formation. To exclude the possibility that transdifferentiation is limited to immature hepatocytes, we investigated whether adult hepatocytes can form HpBDs. To bypass potential adaptive processes in development, we deleted Rbpj and Hnf6 and activated GFP in hepatocytes of P75 Rbpj f/f Hnf6 f/f R26ZG +/+ mice by co-injecting AAV8-Ttr-Cre and AAV8-Ttr-Flp and transplanted the cells 1 week later into P31 Alb-cre +/− Rbpj f/f Hnf6 f/f Rag2 −/− Il2rg −/− mice (Fig. 4a ). After P120, we found donor hepatocyte-derived mature peripheral cholangiocytes in 5.2 ± 0.8% of the portal areas containing donor cells (Fig. 4b ). We also transplanted NOTCH signalling-competent wild-type hepatocytes, which produced donor-derived HpBDs in 78.3 ± 13.9% of such portal areas because 31.6 ± 7.9% of the transplanted cells proliferated after transdifferentiation into cholangiocytes (Fig. 4a, c, d ). These results show that adult hepatocytes, and transplanted hepatocytes, respond to the transdifferentiation-inducing and morphogenetic signals in the bile-duct-deficient liver and form HpBDs. Fig. 4: Clinical relevance and therapeutic potential of HpBD formation. a , Experimental design for hepatocyte transplantation. b , Immunofluorescence of P127 mouse ( n = 5) transplanted with adult GFP-expressing RBPJ- and HNF6-deficient hepatocytes. c , Immunofluorescence of P152 mouse ( n = 4) transplanted with adult RFP-expressing hepatocytes. d , Immunofluorescence of liver of P72 mouse ( n = 2) transplanted at P43 with hepatocytes isolated from P287 Alb-cre +/− R26R-ZsGreen +/+ mouse. e , f , Immunohistochemistry ( e ) and immunofluorescence ( f ) of ALGS ( n = 2) and normal ( n = 1) human livers. Arrowheads indicate nuclear pSMAD3 in pBDs. Scale bars, 100 µm ( b , c , e , f ), 50 µm ( d ). Full size image To determine whether our findings are relevant for human ALGS, we obtained liver samples from two patients who developed regenerative nodules containing pBDs 36 (Extended Data Table 1 ). The regenerative nodules contained CK7-positive pBDs, whereas nonregenerated liver tissue from the same patients showed abundant CK7-positive cells with hepatocyte morphology, indicative of metaplasia 14 , 15 , 16 , 17 (Fig. 4e ). To determine whether TGFβ signalling is active in the new pBDs, we used pSMAD3 immunofluorescence. We found nuclear localization of pSMAD3 in 56.1 ± 6.1% of the pBDs in regenerative nodules, but not in pBDs in normal human liver (Fig. 4f ). These results suggest that the TGFβ-mediated mechanism of HpBD formation identified in Alb-cre +/− Rbpj f/f Hnf6 f/f mice is also active in some patients with ALGS. In conclusion, by showing that hepatocytes can convert into mature cholangiocytes and form a functional and stable biliary system, our findings establish that hepatocyte plasticity extends beyond metaplasia to transdifferentiation, and provide the first example, to our knowledge, of mammalian transdifferentiation building an organ structure de novo. Although hilar cholangiocytes are present in our mouse model, they fail to expand, resulting in severe cholestatic liver injury and strong pressure for hepatocytes to transdifferentiate into peripheral cholangiocytes. Analogously, cholangiocytes transdifferentiate into hepatocytes only when hepatocyte proliferation is completely suppressed 37 . The failure of hilar cholangiocytes to proliferate is probably caused by high TGFβ signalling in the cholestatic liver 38 . Accordingly, we identified TGFβ signalling as the driver of hepatocyte transdifferentiation and HpBD formation in our mouse model, and potentially also in patients with ALGS. Unlike for bile duct development 7 , 8 , 9 , NOTCH signalling is not needed. Using clinically established AAV vectors and hepatocyte transplantation, we show that our findings are potentially translatable into therapies for ALGS and other diseases associated with a lack of bile ducts. Methods Mice Alb-cre +/− Rbpj f/f Hnf6 f/f mice 2 , 26 (mixed background), R26R-RFP +/+ mice 39 (C57BL/6) and R26NZG +/+ mice 40 (FVB) were previously reported. Flp-reporter mice were generated by crossing R26NZG +/+ mice with EIIa-cre +/+ mice 41 (C57BL/6) to remove the Cre-reporter element and then crossing out the EIIa-cre . These R26ZG +/+ mice were crossed with Alb-cre +/− Rbpj f/f Hnf6 f/f mice to generate Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice. R26R-Confetti +/+ mice 42 (C57BL/6) were crossed with Alb-cre +/− Rbpj f/f Hnf6 f/f mice to generate Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− mice. Tgfbr2 f/f mice 43 (C57BL/6) were crossed with Alb-cre +/− Rbpj f/f Hnf6 f/f mice to generate Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f mice. Because Tgfbr2 (68.39 cM) and Hnf6 (41.93 cM) are both on chromosome 9, recombinants were generated at 0.1356 (35 out of 258 mice) observed frequency (0.26 expected frequency). Different founder recombinants were intercrossed to generate Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f mice. R26R-ZsGreen +/+ mice 44 (C57BL/6) were crossed with Alb-cre +/− mice to generate Alb-cre +/− R26R-ZsGreen +/+ mice. Alb-cre +/− Rbpj f/f Hnf6 f/f mice were crossed with Rag2 −/− Il2rg −/− mice 45 , 46 (mixed background) to generate Alb-cre +/− Rbpj f/f Hnf6 f/f Rag2 −/− Il2rg −/− mice. Male and female mice of the indicated age and genotype were chosen randomly for inclusion in experiments. The mouse used as a positive control for biliary gene expression was a C57BL/6 wild-type mouse fed choline-deficient diet (MP Biomedicals) and given 0.15% (w/v) ethionine (Sigma-Aldrich) in the drinking water (CDE diet) for 3 weeks. All mice were kept under barrier conditions. All procedures were approved by the Institutional Animal Care and Use Committee at UCSF or CCHMC. Adeno-associated virus The AAV-Ttr-Flp plasmid was generated by removing Cre from AAV-Ttr-Cre 47 and replacing it with Flpo 48 from pPGKFlpobpA (Addgene 13793); viruses were produced by Vector Biolabs and used at a high dose of 1 × 10 12 –3 × 10 12 vg or low dose of 4 × 10 11 vg. The AAV-Eef1a1-caTgfbr1 plasmid was built by VectorBuilder to contain the activating T204D 49 mutation; virus was produced by Vector Biolabs and used at a dose of 1 × 10 11 vg. Titres were determined using qPCR. Viruses were delivered by tail vein injection in volumes ≤100 µl to prevent hydrodynamic effects. Tissue collection Livers from Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ , R26R-Confetti +/− and transplanted mice were perfused with ice-cold PBS followed by 4% paraformaldehyde (PFA). Samples were cut into slices and fixed overnight in 4% PFA at 4 °C. Livers from R26R-RFP +/+ , Alb-cre +/− Rbpj f/f Hnf6 f/f , Rbpj f/f Hnf6 f/f and Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice were fixed overnight in 4% PFA at 4 °C. For thin sectioning, samples were moved to 30% sucrose overnight at 4 °C to cryopreserve and then embedded in OCT (Tissue-Tek) or processed for paraffin embedding and sectioning. A Leica CM3050 S cryostat was used to cut 6 µm cryosections for staining and imaging. For 3D and sparse-labelling clonal analysis, liver samples were embedded in 4% agarose and sectioned on a Leica VT100 S vibratome. For clonal analysis in R26R-Confetti +/− mice, liver samples were cut into 1 mm slivers, immunostained, washed in PBS, equilibrated in 30% sucrose and embedded in OCT. The 1 mm slivers were then cryosectioned at 100 µm, stained with Hoechst and mounted in PBS for confocal imaging. For analysis of Flp-based hepatocyte fate tracing in R26ZG +/+ mice, liver samples were fixed in neutral-buffered formalin containing zinc (Z-Fix, Anatech), embedded in paraffin and sectioned to 5 µm. Human liver tissue Explant samples from the regenerative nodule and nonregenerated liver tissue of a 3-year-old male patient with ALGS were previously described; regenerative nodule and nonregenerated liver tissue contained the same heterozygous JAG1 exon 1–26 deletion 36 . Samples were obtained with patient consent and approval from the Commission Cantonale d’Ethique de la Recherché CCER. Biopsy samples from the regenerative nodule and nonregenerated liver tissue of a 15-year-old male patient with ALGS caused by a heterozygous c.499 T > A (p.W167R) JAG1 mutation and resection samples from a histologically normal region of the liver of a 35-year-old male undergoing surgery for metastasis of rectal adenocarcinoma were obtained with patient consent and approval from the UCSF Institutional Review Board. Immunostaining and histology Cryosections were blocked in 10% normal serum and permeabilized in 0.3% Triton-X before staining with primary and secondary antibodies listed in Supplementary Tables 2 and 3 . Paraffin-embedded samples were deparaffinized and underwent antigen retrieval in sodium citrate buffer (Bio-Genex or Vector Labs) or Tris EDTA buffer (10 mM Tris Base, 1 mM EDTA, 0.05% Tween 20, pH 9.0) in a steamer or pressure cooker for 15 min before blocking and permeabilization. Thin sections were mounted in FluorSave (MilliporeSigma) or 50% glycerol. Vibratome sections were stained free-floating in 12-well dishes and cleared in Focus Clear (Cedarlane Laboratories) before being mounted in Mount Clear (Cedarlane Laboratories) for confocal imaging. All samples stained with anti-wsCK antibody underwent antigen retrieval in 100 mM Tris buffer (pH 9.5) at 55 °C for 2 h before blocking. Staining with anti-pSMAD3 antibody used a biotinylated secondary antibody followed by avidin/biotin-based peroxidase and tyramide amplification. Staining with anti-SSTR2 antibody used an HRP secondary antibody followed by tyramide-based amplification. M.O.M. Kit (Vector Laboratories) was used for antibodies raised in mice. DAPI or Hoechst was used to stain nuclear DNA. Chromogenic detection used a horseradish peroxidase (HRP) secondary antibody followed by avidin/biotin-based peroxidase amplification and DAB substrate exposure. Sirius red staining was performed on deparaffinized and rehydrated samples using a 0.1% Picrosirius red solution (Dudley Corporation and Newcomer Supply) and 0.5% acetic acid water washes or was done at Peninsula Histopathology Laboratory. Imaging Thin sections were imaged on an Olympus BX51 upright microscope and cultured cells were imaged on an Olympus IX71 inverted microscope using Openlab software (PerkinElmer). Chromogenic stains were imaged on a Leica DM 1000 LED. Fiji 50 and/or Photoshop (Adobe) were used to process (brightness, contrast and gamma) and merge channels. Thick sections for 3D analysis of connectivity and clonal analysis were imaged on a Leica upright AOBS confocal microscope and processed and analysed using Imaris (Bitplane) or Volocity (PerkinElmer) software. For R26R-Confetti +/− mouse and pSMAD3 analysis, images were acquired using a Nikon A1R GaAsP inverted SP confocal microscope and NIS elements software and processed and analysed using Imaris software. Sirius red-stained sections were imaged using a Cytation 5 cell imaging multi-mode reader (BioTek). Ink injection A catheter was inserted in a retrograde fashion into the common bile duct of post-mortem mice and waterproof ink (Higgins) was slowly injected. Left liver lobes were dehydrated in 1:1 methanol:water followed by 100% methanol. Ink was visualized by tissue clearing in 1:2 benzyl alcohol:benzyl benzoate (BABB) solution and imaged on a Leica M205A or Nikon SMZ800 stereoscope. Cholangiocyte isolation Nonparenchymal liver cells were isolated from >P115 Alb-cre +/− Rbpj f/f Hnf6 f/f and Rbpj f/f Hnf6 f/f mice as previously described 51 . Cells were resuspended at 1 × 10 7 cells ml −1 in DMEM/2% FBS and blocked with Mouse Fc Block (BD Biosciences) for 30 min. Cells were incubated with fluorochrome-conjugated antibodies (Supplementary Table 2 ) and DBA-FITC (Vector Laboratories) for 30 min, washed with cold Dulbecco’s PBS (DPBS) three times and resuspended in DMEM/2% FBS. Sytox Red (Thermo Fisher Scientific) was added to label dead cells before sorting. Unstained and single-stained cells were used for compensation. Specificity of DBA binding was verified with a GalNAc (Sigma)-blocked control as previously described 52 . Cells were analysed and sorted on a FACSAria III using FACSDiva software (BD Biosciences). From the CD11b – CD31 – CD45 – population, EPCAM + DBA – cells were collected as peripheral cholangiocytes and EPCAM + DBA + as hilar cholangiocytes. FlowJo (FlowJo, LLC) was used to analyse data and generate charts. Cells were either sorted into DMEM/2% FBS, pelleted and snap frozen, or sorted directly into extraction buffer for RNA purification. Hepatocyte isolation Hepatocytes were isolated from >P115 Alb-cre +/− Rbpj f/f Hnf6 f/f mice by two-step collagenase (Worthington) perfusion followed by purification through a Percoll gradient. Cells were resuspended at 1 × 10 6 cells per 100 µl in Hanks buffer with 10% FBS and incubated with OC2-2F8 antibody (Supplementary Table 2 ) for 1 h on ice. Cells were washed with cold DPBS twice and resuspended in Hanks/10% FBS. Fluorochrome-conjugated secondary antibody (Supplementary Table 3 ) was added and cells were incubated for 30 min on ice followed by two washes with cold DPBS. Cells were blocked with 5% normal rat serum (Jackson Immuno) in Hanks buffer for 10 min on ice. Cells were then incubated with fluorochrome-conjugated antibodies (Supplementary Table 2 ) for 30 min on ice. Cells were then washed in cold DPBS three times and resuspended in Williams E medium/2% FBS. Sytox Red (Thermo Fisher Scientific) was added to label dead cells before sorting. Unstained and single-stained cells were used for compensation. Cells were analysed and sorted on a FACSAria III using FACSDiva software (BD Biosciences). From the CD11b – CD31 – CD45 – EPCAM – population, OC2-2F8 + cells were collected as hepatocytes (Supplementary Fig. 2 ). Cells were either sorted into DMEM/2% FBS, pelleted and snap frozen, or sorted directly into extraction buffer for RNA purification. Hepatocyte transplantation Hepatocytes were isolated from donor mice by two-step collagenase perfusion followed by purification through a Percoll gradient. 1 × 10 6 viable cells were resuspended in 80–100 µl of Williams E medium with glutamine. Transplantation was performed by transdermal intrasplenic injection of the cell suspension under isoflurane anaesthesia. DDC diet feeding Mice received PicoLab Mouse Diet 20, 5058 (LabDiet) with 0.1% 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC; Sigma-Aldrich) for the indicated durations. qPCR and gene expression analysis Genomic DNA was isolated from cells using QIAamp DNA Micro Kit or DNeasy Blood & Tissue Kit (Qiagen). RNA was extracted using Trizol Reagent (Thermo Fisher Scientific) and purified by isopropanol precipitation (cells in collagen gel) or purified using RNAeasy Mini Kit (Qiagen) (liver tissue). Reverse transcription was performed using qScript cDNA Supermix (Quanta Biosciences). qPCR was performed using SYBR green reagent (Thermo Fisher Scientific) in a ViiA 7 system (Thermo Fisher Scientific). Reactions were performed in triplicate, and expression was normalized to an Rbpj (genotyping) or Gapdh (gene expression) reference and quantified using the ΔΔ C t method. Primers are listed in Supplementary Table 4 . RNA-seq RNA was purified from FACS-isolated cells using PicoPure Kit (Thermo Fisher Scientific). RNA quality was assessed using RNA 6000 Pico Kit on a 2100 Bioanalyzer (Agilent). Samples with an RNA integrity number (RIN) ≥ 7.7 and at least 15 ng of RNA were used to construct sequencing libraries using Clontech Low Input Library Prep Kit v2. Libraries were sequenced on a HiSeq 3000, 10 samples per lane, with single-end 50 bp reads. Raw reads were aligned to the mm10 mouse genome with annotations provided by UCSC using CobWEB, a proprietary Burrows-Wheeler Transform method. Reads per kilobase per million (RPKM) were calculated from aligned reads using the expectation-maximization algorithm. RPKM was thresholded at 1, log 2 transformed, normalized using the DESeq algorithm and baselined to the median of all samples. Analyses were performed on transcripts with RPKM > 5 in all samples of at least 1 experimental condition ( n = 17,793 transcripts). These reasonably expressed transcripts were used in principle component analysis. All transcripts with fold change > 3 in at least 1 of the 3 possible pairwise comparisons ( n = 6,464) were selected, and a one-way ANOVA was performed to identify significantly differential genes with FDR-corrected P < 0.05 ( n = 4,997). Venn diagrams were used to identify unique and shared gene signatures. Gene sets were submitted to for identification of pathway and biological process enrichments. Western blot Nuclear and cytoplasmic extracts were generated from whole liver using previously described buffers with protease and phosphatase inhibitors 53 . Samples were run on SDS–PAGE 4–20% Tris-glycine gradient gels, electrophoretically transferred to nitrocellulose membrane and probed with antibodies. Signals were detected by ECL western blotting substrate (GE Healthcare). The membrane was stripped and reprobed with anti-actin antibody to verify and normalize protein loading using densitometry. Quantification was performed using ImageJ software. Serum chemistry Blood was collected from post-mortem Alb-cre +/− Rbpj f/f Hnf6 f/f mice and controls of ages P20 to ≥P150 and tested for serum total bilirubin (TecoDiagnostics) and alkaline phosphatase, alanine aminotransferase and aspartate aminotransferase (ADVIA XPT clinical chemistry system, Siemens). For serial bilirubin measurements, blood was collected by retro-orbital venipuncture and tested for total bilirubin every 2 weeks from P60 to P189 in four different litters. Blood was collected from Alb-cre +/− Rbpj f/f Hnf6 f/f Tgfbr2 f/f mice and controls by retro-orbital venipuncture and tested for serum total bilirubin (TecoDiagnostics). Blood was collected from Alb-cre +/− Rbpj f/f Hnf6 f/f mice intravenously injected with AAV8-Eef1a1-caTgfbr1 and controls by retro-orbital venipuncture and serum total bilirubin was measured as previously reported 54 using a Synergy 2 microplate reader (BioTek). Serum absorbance at 540 nm was subtracted from serum absorbance at 450 nm. A linear trendline equation of serum absorbance versus serum total bilirubin was determined by independently measuring absorbance and total bilirubin (TecoDiagnostics) of known cholestatic and noncholestatic serum samples. This equation was used to convert absorbance readings to serum total bilirubin levels. In vitro conversion assay In vitro 3D culture of hepatocytes to induce biliary conversion was carried out as previously reported 21 with the following modifications. Hepatocytes were isolated from 8–10-week-old Rbpj f/f Hnf6 f/f mice injected with 1 × 10 12 vg of AAV8-Ttr-Cre 2 weeks before to delete Rbpj and Hnf6 . Hepatocytes were isolated by two-step collagenase perfusion followed by purification through a Percoll gradient, all in the absence of serum. Cells were grown on Primaria plates for 6 days to form spheroids. Spheroids were cultured in collagen gels (Cultrex Rat Collagen I, Lower Viscosity, Trevigen) in the presence of 10 µM SB-431542 (Selleckchem) or vehicle (DMSO). The medium was changed every other day. Quantification and statistics For sparse-labelling clonal analysis, eight liver regions from each of three Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice were analysed. GFP-positive cells in all clones within 100 µm-thick sections were manually counted using Imaris software for visualization. GFP-positive cells in direct contact were considered clones. Clones that extended to and potentially beyond the x , y or z boundaries were excluded. For clonal analysis in R26R-Confetti +/− mice, 30–40 µm z -stack images of wsCK-positive DBA-negative pBDs and wsCK-positive DBA-positive hBDs were visualized in 3D with Imaris software. The module ‘surfaces’ within Imaris was used to render a 3D surface created on an intensity value on a per channel basis. The rendered surface per channel was masked to the Hoechst nuclear stain. 3D pBD masks were used to manually count cells per clone. Multiple portal regions were analysed for clones at P90 (14 and 17 per Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− and 8 and 9 per Alb-cre +/− R26R-Confetti +/− mouse) and P150 (7–10 per Alb-cre +/− Rbpj f/f Hnf6 f/f R26R-Confetti +/− and 8–10 per Alb-cre +/− R26R-Confetti +/− mouse). To determine labelling efficiencies in Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ mice, for each mouse, 20 random fields were analysed for wsCK-positive DBA-negative peripheral cholangiocytes (~350 cells) and 5 random fields were analysed for hepatocytes (~1,500 cells). For proliferation analysis, OPN-positive cells in four random portal fields from each of four mice were analysed for KI67 and CK19 staining. For hepatocyte-fate-tracing analysis in DDC diet-fed mice, ten random portal fields from four Alb-cre +/− Rbpj f/f Hnf6 f/f R26ZG +/+ and three Rbpj f/f Hnf6 f/f R26ZG +/+ mice were analysed. For quantification in transplantation experiments, all portal areas with donor-derived cells in a section from at least three lobes per mouse were examined and scored for the presence or absence of donor hepatocyte-derived EPCAM-positive cholangiocytes. For human samples, 63 bile ducts between the 2 patient samples and 45 bile ducts in the control sample were scored for the presence of nuclear pSMAD3. For quantification of Sirius red staining, liver samples were stained in batches and sections of whole lobes were imaged at equal exposure. Using Fiji software, a threshold was set for Sirius red-positive area within each lobe and the percentage of the total area that was Sirius red positive was measured and is reported. Charts were generated in Prism 6 or 7 (GraphPad). Researchers were not blinded when analysing results. P < 0.05 was considered statistically significant. Experiments were replicated independently once (Fig. 2a, b, d–g , 3a, b, d, g , Extended Data Figs. 1c–g , 2c, e , 3a–g , 4b–d , 5a, c, d, f , 6c, g ; Supplementary Table 1 ), at least twice (Fig. 1d, e , 2c , 3f , 4b–f , Extended Data Figs. 5b, e , 6a, b, e, f, h ; Supplementary Video 1 ) or at least three times (Fig. 1b, c , 3c, e , Extended Data Figs. 2b , 4a ; Supplementary Video 2 ). No statistical methods were used to predetermine sample size. Reporting summary Further information on experimental design is available in the Nature Research Reporting Summary linked to this paper. Data availability The Gene Expression Omnibus (GEO) accession number for the RNA-seq data is GSE108315 . Additional data are available from the corresponding authors on reasonable request.

By studying a rare liver disease called Alagille syndrome, scientists from Cincinnati Children's and the University of California San Francisco (UCSF) have discovered the mechanism behind an unusual form of tissue regeneration that may someday reduce the need for expensive and difficult-to-obtain organ transplants. The team's findings, published in the journal Nature, show that when disease or injury causes a shortage in one critical type of liver cell, the organ can instruct another type of liver cell to change identities to provide replacement supplies. This discovery was made in mice but in years to come may lead to a viable treatment for human disease. If ongoing follow-up studies succeed, the medical world may gain an alternative method for repairing tissue damage that does not require manipulating stem cells to grow organs from scratch in a lab dish. "We have known for a long time that the liver has more ability to regenerate than other organs. Only recently have we had the tools to study this ability in depth. Now we have a high-level understanding," says Stacey Huppert, PhD, a developmental biologist in the Division of Gastroenterology, Hepatology and Nutrition at Cincinnati Children's, and one of two leading co-authors of the paper. "Our study shows that the form and function of hepatocytes—the cell type that provides most of the liver's functions—are remarkably flexible. This flexibility provides an opportunity for therapy for a large group of liver diseases," says Holger Willenbring, MD, PhD, the study's other senior co-author and a member of the Department of Surgery, the Eli and Edythe Broad Center of Regeneration Medicine and Stem Cell Research, and the Liver Center at UCSF. What is Alagille syndrome? Alagille syndrome (ALGS) is a rare, inherited genetic disorder due to deficiencies in the Notch pathway, which plays an important role in cell development. ALGS is best known for disrupting the liver's plumbing system, which consists of tubes called bile ducts that deliver bile made in the liver to the intestine. The disorder occurs in about 1 in 30,000 people, and in most cases, the problems caused by the condition emerge during infancy or early childhood. The extent of the condition can range from having too few or too narrow bile ducts to missing all bile ducts in the liver. As a result, the bile that normally helps the body digest fat and carry away toxins backs up inside the liver where it causes severe damage. In many patients with ALGS, bile duct function can be managed and sustained. However, up to 50 percent of patients eventually need a liver transplant, often during childhood. A Quicker Method of Self-Repair Every day, the liver takes a beating as it processes everything from medications to alcohol consumption. All these "insults" have prompted the liver to develop a rapid healing ability that does not rely on stem cells, the co-authors say. "In addition to making more of themselves, liver cells can switch their identity to produce a liver cell type that is lost or, in the case of severe ALGS, never formed," Willenbring says. "Previous research has detected adaptive reprogramming in other organs, but it typically involves only a few cells at a time. Our study shows that cells switch their identity at a massive rate in the liver," Huppert says. Discovering this phenomenon and learning how it works took nearly five years. The team included co-first authors Johanna Schaub and Simone Kurial from UCSF and Kari Huppert from Cincinnati Children's. The researchers generated mice that lack cholangiocytes, the type of liver cell that forms bile ducts. Like patients with severe ALGS, these mice quickly developed signs of liver injury. However, over time the mice's symptoms improved because hepatocytes converted into cholangiocytes and formed fully functional bile ducts. In July 2017, another study published in Nature reported that cholangiocytes can become hepatocytes if their ability to regenerate is impaired. Viewed together, the two studies suggest that switching of cell identity is the main backup mechanism for liver repair. The new study further shows that the Notch pathway, which is essential for forming bile ducts but defective in patients with ALGS, can be replaced by another pathway. This process is regulated in the injured liver by transforming growth factor beta (TGFβ), a protein that controls cell growth. This discovery is a vital step in identifying targets for therapies that might control this process. Next Steps Now the research team is working to determine the precise set of proteins, called transcription factors, that work together to carry out the identity-switching process. "Using transcription factors to make bile ducts from hepatocytes has potential as a safe and effective therapy," Willenbring says. "With our finding that an entire biliary system can be 'retrofitted' in the mouse liver, I am encouraged that this eventually will work in patients." In addition to developing a therapy for ALGS, the team hopes to determine whether liver cell switching can benefit other types of liver disease.

nature.com/articles/doi:10.1038/s41586-018-0075-5

Space

Drastic chemical change occurring in birth of planetary system: Has the solar system also experienced it?

Nami Sakai, Takeshi Sakai, Tomoya Hirota, Yoshimasa Watanabe, Cecilia Ceccarelli, Claudine Kahane, Sandrine Bottinelli, Emmanuel Caux, Karine Demyk, Charlotte Vastel, Audrey Coutens, Vianney Taquet, Nagayoshi Ohashi, Shigehisa Takakuwa, Hsi-Wei Yen, Yuri Aikawa & Satoshi Yamamoto, "Change in the chemical composition of infalling gas forming a disk around a protostar", Nature, DOI: 10.1038/nature13000. Journal information: Nature

http://dx.doi.org/10.1038/nature13000

https://phys.org/news/2014-02-drastic-chemical-birth-planetary-solar.html

Abstract IRAS 04368+2557 is a solar-type (low-mass) protostar embedded in a protostellar core (L1527) in the Taurus molecular cloud 1 , 2 , which is only 140 parsecs away from Earth, making it the closest large star-forming region. The protostellar envelope has a flattened shape with a diameter of a thousand astronomical units (1 au is the distance from Earth to the Sun), and is infalling and rotating 3 , 4 , 5 . It also has a protostellar disk with a radius of 90 au (ref. 6 ), from which a planetary system is expected to form 7 , 8 . The interstellar gas, mainly consisting of hydrogen molecules, undergoes a change in density of about three orders of magnitude as it collapses from the envelope into the disk, while being heated from 10 kelvin to over 100 kelvin in the mid-plane, but it has hitherto not been possible to explore changes in chemical composition associated with this collapse. Here we report that the unsaturated hydrocarbon molecule cyclic-C 3 H 2 resides in the infalling rotating envelope, whereas sulphur monoxide (SO) is enhanced in the transition zone at the radius of the centrifugal barrier (100 ± 20 au ), which is the radius at which the kinetic energy of the infalling gas is converted to rotational energy. Such a drastic change in chemistry at the centrifugal barrier was not anticipated, but is probably caused by the discontinuous infalling motion at the centrifugal barrier and local heating processes there. Main We have conducted high-spatial-resolution observations of the millimetre and submillimetre wave lines of cyclic-C 3 H 2 and SO in the 500 au region around IRAS 04368+2557 (Methods). Figure 1a shows the observed integrated-intensity map of the 5 23 –4 32 rotational line of cyclic-C 3 H 2 (where J KK′ represents the rotational energy level of an asymmetric top molecule; see the footnote of Extended Data Table 1 ). The intensity distribution shows a double-peaked structure, where the southern and northern peaks are separated from the protostar position (the continuum peak position: Extended Data Fig. 1 ) by 1′′ (140 au ). Figure 1b shows the position–velocity (PV) diagram along the north–south line centred at the protostar’s position. A rotation signature is clearly seen outward of 100 au from the protostar, where the brightest velocity component is redshifted and blueshifted for the northern and southern parts, respectively. This situation is schematically illustrated in Fig. 1c . Such a rotation signature abruptly disappears inward of 100 au from the protostar. In addition to this signature, weak blueshifted and redshifted components can also be recognized in the northern and southern parts, respectively ( Fig. 1b ). The existence of such counter-velocity components indicates that the gas motion is not simple rotation but accompanies infalling motion, as shown in Fig. 2a . Most importantly, only the infalling (counter-velocity) components are seen towards the protostar position. This means that cyclic-C 3 H 2 is almost absent inward of 100 au . The other cyclic-C 3 H 2 lines (5 51– 4 40 , 9 18 –8 27 /9 28 –8 17 and 10 0,10– 9 19 /10 1,10 –9 09 ) also show similar distributions ( Extended Data Fig. 2 ). Figure 1: IRAS 04368+2557 in the cyclic-C 3 H 2 (5 23– 4 32 ) and SO ( J N = 7 8 –6 7 ) lines. a , Integrated intensity distributions of cyclic-C 3 H 2 (colour) and SO (contours). Contours are every 10 σ (90 mJy beam −1 km s −1 ), starting from 5 σ (the outermost contour). The red and white ellipses represent the synthesized beam sizes for cyclic-C 3 H 2 and SO, respectively. b , The PV diagram of cyclic-C 3 H 2 along a north–south line passing through the protostar position. Contours are every 3 σ (12 mJy beam −1 ), starting from 3 σ . The white rectangle represents the resolution. c , A schematic of the envelope geometry. d , The PV diagram of SO along the same line as b . Contours are every 4 σ (28 mJy beam −1 ), starting from 2 σ . White contours represent negative values. V LSR denotes the velocity with respect to the local standard of rest. The white rectangle represents the resolution. PowerPoint slide Full size image Figure 2: A model of an infalling rotating envelope. a , A schematic illustration of the infalling rotating envelope. The observer is on the left-hand side, looking at the envelope in an edge-on configuration. b , The highest and lowest velocities calculated with the toy model (Methods). The emissions of the coloured closed circles come from the corresponding coloured closed circles in a . c , Here the highest and lowest velocities shown in b (thick blue lines) are superposed on the PV diagram of cyclic-C 3 H 2 . The thick black lines indicate the Keplerian velocity expected for the disk. They trace the broad line profile of SO (thin grey contours) well. White and red rectangles represent the resolutions for SO and cyclic-C 3 H 2 , respectively. PowerPoint slide Full size image To account for this kinematic feature, we developed a toy model describing an infalling rotating envelope, where the gas motion is approximated by the motion of a particle for simplicity ( Fig. 2a ) (Methods and Extended Data Fig. 3 ). According to this model, the maximum infalling velocity towards the protostar position is half of the maximum rotation velocity at the centrifugal barrier of the infalling gas (that is, half the centrifugal radius). In fact, the observed maximum infalling velocity of cyclic-C 3 H 2 towards the protostar position is just half of the maximum rotation velocity ( Fig. 1b ). Therefore, the radius that gives the maximum rotation velocity in the cyclic-C 3 H 2 emission is the position of the centrifugal barrier. We simulate the PV diagram by using the radius of the centrifugal barrier and the maximum rotation velocity as parameters. In the simulation, we consider only the velocity field, and the highest and lowest velocities at a given offset from the protostar ( Fig. 2b ) are compared with the PV diagram of cyclic-C 3 H 2 . The best-fitting result is shown in Fig. 2c . The overall behaviour of the PV diagram is well explained even by this simple model. From this result, the radius of the centrifugal barrier and the maximum rotation velocity are derived to be 100 ± 20 au and 1.8 ± 0.2 km s −1 , respectively. The radius of the centrifugal barrier is close to the radius of the Keplerian disk (90 au ) 6 . The total mass of the protostar and the inner disk is evaluated to be 0.18 ± 0.05 solar masses, by using the radius of the centrifugal barrier and the maximum rotation velocity derived above. Here, the error is estimated from those of the radius of the centrifugal barrier and the maximum rotation velocity by error propagation. This is consistent with the protostar mass reported in ref. 6 (0.19 ± 0.04 solar masses) from the measurement of the Keplerian rotation. In contrast to the cyclic-C 3 H 2 lines, the SO ( J N = 7 8 –6 7 ) line shows a compact single-peaked distribution centred at the protostar ( Fig. 1a ) ( J N represents the rotational energy level of SO; see the footnote of Extended Data Table 1 ). The SO distribution seems to fill up the dip of the cyclic-C 3 H 2 distribution. The PV diagram of the SO line emission along the north–south line is completely different from that of cyclic-C 3 H 2 ( Fig. 1d ). It consists of two components. One is a component whose velocity is proportional to the position offset from the protostar. The other is a weak and broad component whose velocity width is as broad as 6 km s −1 . The anticorrelation between the distributions of cyclic-C 3 H 2 and SO is, thus, evident. When we compare the PV diagram of SO with that of cyclic-C 3 H 2 , it is most likely that the former component of SO comes from a rotating ring whose radius is close to the innermost radius of the cyclic-C 3 H 2 distribution, corresponding to the radius of the centrifugal barrier ( Fig. 1c ). On the other hand, the latter component seems to come from the inner disk, although it is much fainter than the ring component. These features are essentially the same for the other SO line ( J N = 7 6 –6 5 ). We conclude that a drastic chemical change occurs at the centrifugal barrier. (See the Methods and Extended Data Figs 2 , 4 , 5 , 6 and 7 for details.) We observed four lines of cyclic-C 3 H 2 and two lines of SO in total ( Extended Data Table 1 ). We analysed them to derive physical conditions of emitting regions for both molecules by using a non-local-thermodynamic-equilibrium (non-LTE) large-velocity-gradient (LVG) code (Methods and Extended Data Fig. 8 ). The gas kinetic temperatures of the cyclic-C 3 H 2 emitting region are derived to be 23–33 K, 30 K and 23 K, for the 1′′ × 1′′ region at the centre, 1′′ north of the centre and 1′′ south of the centre, respectively, where the H 2 density ranges from 6 × 10 6 to 1 × 10 8 cm −3 . These values are almost consistent with the reported model 9 (H 2 density of 1.3 × 10 8 cm −3 and the temperature of 30 K at 100 au ). On the other hand, the gas kinetic temperature of the SO-emitting region is constrained to be higher than 60 K, and the H 2 density to be higher than 3 × 10 6 cm −3 for the 1′′ × 2′′ region centred at the protostar position. Given that the sublimation temperature of SO is 50 K, it is most likely that SO is liberated into the gas phase from grain mantle at the centrifugal barrier. Although the iced SO is not identified by infrared observations, it is predicted by the chemical model 10 , 11 . Liberation of the S atom into the gas phase followed by the reaction with OH may also contribute to enhancement of SO (ref. 12 ). Possible mechanisms for the SO liberation are accretion shock in front of the centrifugal barrier, outflow shock on the disk surface, and the protostellar heating. These possibilities are described in the Supplementary Information . Further in the mid-plane of the inner disk, SO is probably depleted again onto dust grains. The depletion timescale is as short as 10–200 years for the H 2 density of 6 × 10 6 cm −3 to reach 1 × 10 8 cm −3 . Therefore, SO mainly exists around the centrifugal barrier. However, the PV diagram of the SO emission also shows a weak broad component inward of the centrifugal barrier. This suggests that SO partly survives even inward of the barrier. The gas infalling directly onto the inner disk surface may also contribute to the broad component 13 . We predict the velocity of the Keplerian rotation as a function of the radius from the estimated protostellar mass derived from the cyclic-C 3 H 2 data ( Fig. 2c ). The expected velocity can roughly explain the PV diagram inward of the centrifugal barrier. Similarly, cyclic-C 3 H 2 is depleted onto dust grains in the mid-plane of the disk as in the case of SO. Cyclic-C 3 H 2 may also be destroyed in the transition region by a gas-phase reaction with the oxygen atom. Although the fresh gas containing cyclic-C 3 H 2 is continuously supplied outside the centrifugal barrier, it is not supplied efficiently within the disk. Hence, the abundance of cyclic-C 3 H 2 in the gas phase becomes much lower inward of the centrifugal barrier than in the infalling rotating envelope, as observed. These results demonstrate a drastic chemical change at the centrifugal barrier in the course of the formation of the inner disk. It does not mean that SO and cyclic-C 3 H 2 are chemically related to each other directly; the change is caused by the discontinuous infalling motion of the gas at the centrifugal barrier and local heating processes there ( Supplementary Information ). When interstellar matter is brought into the disk, it must experience this physical situation and is subject to significant chemical processing including gas–grain interactions in the transition zone. The chemical compositions of interstellar clouds have arbitrarily been assumed as the initial conditions set for the chemical evolution models of the disk 14 , 15 , 16 , 17 . This assumption has now been found to be oversimplified: the chemical processing in the transition zone should be seriously considered. This is also true for the case of tracing the origin of the pre-solar materials found in meteorites back to interstellar matter. Methods Summary Observations were carried out with the Atacama Large Millimeter/submillimetre Array (ALMA) in 2012. We observed the rotational spectral line emissions of cyclic-C 3 H 2 and SO as well as the dust continuum emission in the 249-GHz, 261-GHz, 338-GHz and 351-GHz regions. The synthesized beam is about 0.8′′ × 0.7′′ for 249 GHz and 261 GHz and about 0.7′′ × 0.5′′ for 338 GHz and 351 GHz. The integrated intensity maps and the PV diagrams of the four cyclic-C 3 H 2 lines and the two SO lines are similar to each other for each molecule. For SO, the total flux of the J N = 7 8 –6 7 line emission observed with ALMA recovers 60% of that observed with the single-dish telescope in the Atacama Submillimetre Telescope Experiment (ASTE) (with a beam size of 22′′), indicating that the SO line emission is concentrated around the protostar. The effect of the outflow is confirmed to be negligible by examining the PV diagram of the SO ( J N = 7 8– 6 7 ) line along the outflow axis. To account for the PV diagrams of the C 3 H 2 lines, we use a simple kinematic model for the infalling rotating envelope, revealing that the maximum infalling velocity towards the protostar is just half of the maximum rotation velocity. We evaluated the gas kinetic temperature and the H 2 density using the non-LTE LVG code. Online Methods Observations Our observations of IRAS 04368+2557 in L1527 were carried out with ALMA in its early science operation on 10 August 2012, 26 August 2012 (1.1 mm; band 6) and 29 August 2012 (0.8 mm; band 7) in the extended configuration of the array. The cyclic-C 3 H 2 and SO lines were observed in both bands. The line parameters are listed in Extended Data Table 1 . The number of antennas was 22 to 25 during the observation. The field centre is ( α 2000 , δ 2000 ) = (04 h 39 min 53.89 s, 26° 03′ 09.8′′), which is the 3-mm continuum peak of L1527 reported in ref. 18 . The typical system temperatures in bands 6 and 7 were 80–150 K and 100–200 K, respectively. The backend correlator was tuned to the 469-MHz bandwidth with 122-kHz resolution. The velocity resolution is 0.15 km s −1 at band 6 and 0.1 km s −1 at band 7. The quasar J0510+180 was used for the phase calibration every 25 min. The bandpass calibration was carried out on the quasar J0423-013. The absolute flux density scale was derived from Callisto. The data calibration was performed in the antenna-based manner, and uncertainties are less than 10%. The continuum image was prepared by averaging line-free channels. The images were obtained by CLEANing (CLEAN is the deconvolution algorithm) the dirty images after subtracting the continuum directly from the visibilities with the CASA ( ) software package 19 . The primary beams (half-power beam widths) are 24.5′′ and 17.7′′ at bands 6 and 7, respectively. The total on-source times were 79 min and 35 min for the band 6 and band 7 observations, respectively. The synthesized beam size is about 0.8′′ × 0.7′′ for band 6, whereas it is about 0.7′′ × 0.5′′ for band 7. The 0.8-mm continuum distribution is shown in Extended Data Fig. 1 . The peak and total flux of the continuum are 274 ± 4 mJy beam −1 and 485 ± 7 mJy, respectively, where the errors represent the standard deviation. The peak position is derived to be ( α 2000 , δ 2000 ) = (04 h 39 min 53.87 s, 26° 03′ 09.6′′). Maps of the cyclic-C 3 H 2 lines We observed four transitions of cyclic-C 3 H 2 in total. The distribution and the PV diagram of the cyclic-C 3 H 2 (5 51 –4 40 ) line are almost the same as those of the 5 23 –4 32 line shown in Fig. 1b , although the signal-to-noise ratio of the 5 51 –4 40 line is slightly poorer. The 9 18 –8 27 /9 28 –8 17 and 10 0,10 –9 19 /10 1,10 –9 09 lines of cyclic-C 3 H 2 also show similar distributions. Extended Data Fig. 2a shows the integrated intensity map of the 9 18 –8 27 /9 28 –8 17 line of cyclic-C 3 H 2 , which is the second brightest line of the four observed lines. Extended Data Fig. 2b shows the PV diagram of the 9 18 –8 27 /9 28 –8 17 line. The integrated intensity map and the PV diagram of the 9 18 –8 27 /9 28 –8 17 and 10 0,10 –9 19 /10 1,10 –9 09 lines are essentially similar to the integrated intensity map and the PV diagram of the 5 23 –4 32 line. Because the 9 18 –8 27 /9 28 –8 17 and 10 0,10 –9 19 /10 1,10 –9 09 lines have higher upper-state energies than does the 5 23 –4 32 line ( Extended Data Table 1 ), the centrifugal barrier is highlighted in comparison with the outer envelope. Extended Data Fig. 4 shows the velocity channel map of cyclic-C 3 H 2 (5 23 –4 32 ). The feature seen in the PV diagram can be confirmed. In addition, the weak extended component is seen. However, an outer part of the infalling rotating envelope seen in the previous observation of cyclic-C 3 H 2 with the IRAM Plateau de Bure Interferometer (PdBI) 18 is almost resolved out in the present observation, because of a lack of short spacing data. Maps of the SO lines We observed the J N = 7 6 –6 5 line (261.8 GHz) as well as the J N = 7 8 –6 7 line (340.7 GHz). The synthesized beam size is 0.70′′ × 0.48′′ for the SO ( J N = 7 8 –6 7 ) map, and the deconvolved size of the SO distribution is 1.3′′ × 0.6′′ (90 au 40 au in radius) at a position angle of 180°. Hence, the SO distribution is resolved along the north–south direction. The integrated intensity map and the PV diagram along the north–south cut for the SO ( J N = 7 6 –6 5 ) line are shown in Extended Data Figs 2c and d , respectively. The intensity distribution is single-peaked around the protostar, as in the case of the J N = 7 8 –6 7 line. Given that the synthesized beam is 0.94′′ × 0.60′′ for this line, the distribution is partially resolved in this case. The PV diagram is almost consistent with that for the J N = 7 8 –6 7 line, although the linear component is more dominant than in the J N = 7 8 –6 7 line. The weak high-velocity component in the J N = 7 6 –6 5 line is understandable, because it suffers more from the beam-dilution effect than does the J N = 7 8 –6 7 line. Extended Data Fig. 5 shows the velocity channel map of the SO ( J N = 7 8 –6 7 ) line. The SO emission is centrally concentrated. It should be noted that the extended emission which is thought to be a remnant of resolved-out components is not detected significantly within the field of view both in the J N = 7 6 –6 5 and 7 8 –6 7 lines. Comparison with the single-dish observation of SO In other sources, the SO line emission is often extended over the envelope scale 20 , 21 . However, the total flux of the SO ( J N = 7 8 –6 7 ) line emission observed with ALMA recovers 60% of that observed with the single-dish telescope of ASTE (with beam size 22′′) in the present case, indicating that the SO line emission is strongly concentrated around the protostar. That is, we are not looking at a central portion of widespread emission most of which is resolved out in interferometer observations. Simple model To address the physical meaning of the chemical change, we developed a toy model for an infalling rotating envelope, where the gas motion is approximated by the motion of a particle with mass m , for simplicity ( Extended Data Fig. 3a ). This is a modified version of the model presented in ref. 3 . If the angular momentum L is conserved in the infalling motion, the radial velocity v r and the rotation velocity v θ (where θ is the azimuthal angle) at radius r from the protostar can be written as: and where G and M stand for the gravitational constant and the mass of the protostar (plus the inner disk), respectively. The radius at which all the kinetic energy is converted to the rotational motion is given as: This is the radius of the centrifugal barrier, where the rotation velocity takes its maximum value: The infalling and rotating gas cannot move inward of the radius of the centrifugal barrier, unless it loses kinetic energy and angular momentum. The radius of the centrifugal barrier is half of the centrifugal radius. When we observe the line of sight that is offset from the protostar position by x au on the disk ( Extended Data Fig. 3a ), the velocity of the particle along the line of sight ( y au ) can be written as: In Extended Data Fig. 3b , V ( x , y ) is plotted against y for several values of x . For a positive x , the highest V ( x , y ) originate mainly from the rotational motion, and the lowest mainly from the infalling motion. For a negative x , the behaviour is opposite. This is why we observe the counter-velocity component in the PV diagram of cyclic-C 3 H 2 ( Figs 1b and 2 ). Existence of this component unambiguously means that the gas is infalling with rotation. Most importantly, only the infalling component can be seen towards the protostar position. The maximum infalling velocity towards this direction is readily evaluated as: It is just half of the maximum rotation velocity at the centrifugal barrier, . Using this relation together with the observed PV diagram, the radius giving the maximum rotation velocity in Fig. 1b is definitively identified as the radius of the centrifugal barrier. Evaluation of the total mass of the protostar and the disk Once r 0 and have been determined, we can evaluate the protostar mass (plus the disk mass) from them. Figure 2c shows the highest and lowest velocities expected from the toy model superposed on the PV diagram of cyclic-C 3 H 2 . Here, the model curves are not convolved with the telescope beam. The radius of the centrifugal barrier and the maximum rotation velocity are derived to be 100 ± 20 au and 1.8 ± 0.2 km s −1 , respectively. If they are out of the above ranges, the simulated PV diagram does not reproduce the observed one well. Hence, the quoted errors are the estimated limits of errors. The total mass of the protostar and the inner disk is thus evaluated to be 0.18 ± 0.05 solar masses using equations (3) and (4). This mass-evaluation method is useful, because it can be an independent test of the mass derived from the velocity structure of the Keplerian disk 22 , 23 . Derivation of temperature and density We used the four lines of cyclic-C 3 H 2 and two lines of SO (see Extended Data Table 1 ) to derive the physical conditions of the emitting regions for both molecules by using the non-LTE LVG code, which was originally developed by ref. 24 and modified to include the collisional coefficients retrieved by the BASECOL database 25 . In particular, we used the collisional coefficients of cyclic-C 3 H 2 and He computed by ref. 26 after scaling them for the mass of H 2 , and of SO with para-H 2 as reported in ref. 27 . We ran a large grid of models to cover a wide parameter space: column density from 1 × 10 13 cm −2 to 1 × 10 15 cm −2 , H 2 density from 3 × 10 5 cm −3 to 1 × 10 8 cm −3 , and temperature from 25 K to 120 K. The observed line fluxes were compared with the LVG model predictions and the best-fitting solution was found by minimizing the χ 2 value with respect to the above three parameters ( Extended Data Fig. 6 ). Then we constrained the range of the physical parameters as shown in the text. As for cyclic-C 3 H 2 , the component around the systemic velocity (about 6 km s −1 ; ref. 28 ) is observed to be weak or missing owing to the foreground absorption. Continuum emission is also absorbed by infalling foreground gas with a slightly redshifted velocity. Therefore, we used the integrated intensity excluding the self-absorption feature ( Extended Data Table 1 ). We did not consider the temperature and density variation within the beam in this calculation for simplicity. Furthermore, we did not consider the contributions of the resolving-out components. Because we discuss only the compact component around the centrifugal barrier, the extended component is not important. Estimation of the SO abundance With the above LVG model, the column density of SO is well constrained to be 4 × 10 14 cm −2 for the central 1′′ × 2′′ region. Given that the rotational levels are almost populated in LTE, we can constrain the SO column density as well as temperature. The optical depths of the SO lines are approximately 0.3, which is moderately thick. It is rather difficult to evaluate the fractional abundance of SO relative to H 2 , because of the uncertainty of the column density of the H 2 molecules. We estimate it roughly from the size (about 200 au ) and the H 2 density (about 10 7 cm −3 ) to be 3 × 10 22 cm −3 . Hence, the fractional abundance of SO relative to H 2 is about 10 −8 . This can be compared with the SO abundance expected in the model ( Supplementary Information ). We note that the column density of cyclic-C 3 H 2 is difficult to estimate because of the absorption features described above.

A new star is formed by gravitational contraction of an interstellar molecular cloud consisting of gas and dust. In the course of this process, a gas disk (protoplanetary disk), whose size is on the order of 100 AU, forms around the protostar and evolves into a planetary system. The solar system was also formed in this way about 4.6 billion years ago, and life was eventually born on the Earth. How unique in the universe is the situation which happened for the solar system? In order to answer this question, understanding the formation of protoplanetary disks as well as the associated chemical evolution in various star forming regions is essential. There have been extensive observational efforts made toward this goal. So far, most of them have focused on changes in the physical structure and the kinematics during the formation process. However, it was very difficult to distinguish the protoplanetary disk from the infalling envelope clearly with such conventional approaches. On the other hand, the chemical evolution associated with disk formation has scarcely been studied observationally because of the insufficient sensitivity and spatial resolution of previous radio telescopes. As a result, a chemical model calculation with many assumptions is the only approach. Naturally, the physical and chemical changes in the disk formation should be coupled with each other. The disk formation around a young protostar has been explored from a novel point of view looking at physics and chemistry simultaneously. L1527 in the Taurus molecular cloud is a molecular cloud core which harbors a young protostar. A global team led by Dr. Nami Sakai, the University of Tokyo, conducted high-sensitivity, high-spatial-resolution observations of L1527 with ALMA (Atacama Large Millimeter/submillimeter Array) newly constructed in the Atacama desert in Chile, and investigated the disk formation process by using the spectral lines of several molecules. As a result, Sakai al. have found that carbon-chain molecules and their related species such as cyclic-C5H2 almost completely disappear from the gas phase inside a raius of 100 AU around the protostar (Figure1, top left; top right). Figure 1. An infrared image of the protostar L1527 taken by the Spitzer Space Telescope. Credit: J. Tobin/NASA/JPL-Caltech Precise measurements of the motion of the gas using the Doppler shift in the spectral lines of the gas components revealed that 100 AU corresponds to the radius of the centrifugal barrier (Figure 2). Figure 2. L1527 observed by Spitzer (Left) and the distributions of cyclic-C3H2 (center) and SO (right) observed by ALMA. ALMA reveals the gas distribution just close to the protostar. Emission from cyclic-C3H2 is weak toward the protostar but strong at the northern and southern parts. Meanwhile, SO has its emission peak near the protostar. Credit: J. Tobin/NASA/JPL-Caltech, N. Sakai/The University of Tokyo At this radius, infalling gas is stopped and accumulated due to the centrifugal force, and then is gradually transferred to the inner disk. Namely, this is the edge of the disk forming region. It has clearly been indentified with the spectral line of cycle-C5H2. On the other hand, the distribution of sulfur monoxide molecules (SO) is found to be localized in a ring structure located at the radius of the centrifugal barrier (100 AU) (Figure 1, bottom left; bottom right). Furthermore, the temperature of the SO molecules is found to be higher than that of the infalling gas. This means that the infalling gas probably causes a weak shock when in rushes into the outer edge of the disk at the centrifugal barrier. The gas temperature is raised around this radius, and the SO molecules frozen on dust grains are liberated into the gas phase. Hence, the spectral line of SO also highlights the disk-formation front. Since the density of the disk is 108 cm-5 or higher, most of the molecules are frozen out onto dust grains in the disk after they pass through the front. It is not at all anticipated that such a drastic chemical change occurs in the transition zone between the infalling envelope and the inner disk. The disk formation and the associated chemical change have successfully been detected by observations of the two chemical species, cyclic-C5H2 and SO. This study has demonstrated a drastic change in chemical composition associated with disk formation around the young protostar (cf; Figure 3). With a coupled view of physics and chemistry, it has also succeeded in highlighting the outermost part of the disk where the gas is still accreting. This success was realized by high-sensitivity and high-spatial-resolution observations with ALMA, and such a study will be extended to other various star-forming regions. In particular, it is very interesting to examine how widely applicable the picture seen in L1527 is to other star-forming regions. Although many observational efforts aimed at understanding planetary-system formation have been made, this study is novel in focusing on the chemical change. By extending this new method to various solar-type protostars using ALMA, the diversity and generality of the chemical evolution from interstellar matter to planetary matter will be unveiled within the next few years. Then, we can critically examine whether the solar system experienced this drastic chemical change. In parallel to the astronomical approach, the origin of the solar system is being investigated by exploring the solar system itself through microanalyses of meteorites, spectroscopy of comets, sample return missions to the asteroids, and so on. The present study will also have a strong impact on these studies by tracing the origins back to interstellar clouds.

10.1038/nature13000

Medicine

Treatment in hospital by older doctors linked to higher death rates

Physician age and outcomes in elderly patients in hospital in the US: observational study , BMJ (2017). www.bmj.com/content/357/bmj.j1797 Editorial: Physician age and patient outcomes, BMJ (2017). www.bmj.com/content/357/bmj.j2286 Journal information: British Medical Journal (BMJ)

http://www.bmj.com/content/357/bmj.j1797

https://medicalxpress.com/news/2017-05-treatment-hospital-older-doctors-linked.html

Abstract Objectives To investigate whether outcomes of patients who were admitted to hospital differ between those treated by younger and older physicians. Design Observational study. Setting US acute care hospitals. Participants 20% random sample of Medicare fee-for-service beneficiaries aged ≥65 admitted to hospital with a medical condition in 2011-14 and treated by hospitalist physicians to whom they were assigned based on scheduled work shifts. To assess the generalizability of findings, analyses also included patients treated by general internists including both hospitalists and non-hospitalists. Main outcome measures 30 day mortality and readmissions and costs of care. Results 736 537 admissions managed by 18 854 hospitalist physicians (median age 41) were included. Patients’ characteristics were similar across physician ages. After adjustment for characteristics of patients and physicians and hospital fixed effects (effectively comparing physicians within the same hospital), patients’ adjusted 30 day mortality rates were 10.8% for physicians aged <40 (95% confidence interval 10.7% to 10.9%), 11.1% for physicians aged 40-49 (11.0% to 11.3%), 11.3% for physicians aged 50-59 (11.1% to 11.5%), and 12.1% for physicians aged ≥60 (11.6% to 12.5%). Among physicians with a high volume of patients, however, there was no association between physician age and patient mortality. Readmissions did not vary with physician age, while costs of care were slightly higher among older physicians. Similar patterns were observed among general internists and in several sensitivity analyses. Conclusions Within the same hospital, patients treated by older physicians had higher mortality than patients cared for by younger physicians, except those physicians treating high volumes of patients. Introduction The relation between physician age and performance remains largely unknown, particularly with respect to patient outcomes. Clinical skills and knowledge accumulated by more experienced physicians can lead to improved quality of care. Physicians’ skills, however, can also become outdated as scientific knowledge, technology, and clinical guidelines change. Incorporating these changes into clinical practice is time consuming and can at times be overwhelming. 1 2 3 Interest in how quality of care evolves over a physician’s career has revived in recent years, with debates over how best to structure programs for continuing medical education, including recent controversy in the US regarding maintenance of certification programs. A systematic review of the relation between physician experience and quality of care found that older physicians might perform worse—older physicians have decreased clinical knowledge, adhere less often to standards of appropriate treatment, and perform worse on process measures of quality with respect to diagnosis, screening, and preventive care. 4 Data on patient outcomes, which arguably are most important, have been scarce. 4 Existing studies have also been limited in size or disease scope and have not been nationally representative. 5 6 7 As a result, whether physician age is associated with patient outcomes remains largely unknown. Using nationally representative data on Medicare beneficiaries admitted to hospital with a medical condition during 2011-14, we sought answers to three questions. First, what is the association between age of the treating physician and patient mortality after admission? Second, does this association vary with the volume of patients a physician treats? Finally, given national efforts to improve the efficiency of healthcare, is physician age associated with readmissions and costs of care? Methods Data We linked multiple data sources: the 20% Medicare Inpatient Carrier and Medicare Beneficiary Summary Files (2011-14); physician data collected by Doximity (an online professional network for physicians); and the American Hospital Association (AHA) annual survey of hospital characteristics (2012). Doximity has assembled data on all US physicians (both those who are registered members of the service as well as those who are not) from multiple sources and data partnerships, including the national plan and provider enumeration system national provider identifier registry, state medical boards, specialty societies such as the American Board of Medical Specialties, and collaborating hospitals and medical schools. The database includes information on physician age, sex, year of completion and name of medical school, residency, and board certification. 8 9 10 11 12 Previous studies have validated data for a random sample of physicians in the Doximity database by using manual audits. 8 9 We were able to match about 95% of physicians in the Medicare database to the Doximity database. Patients We identified beneficiaries of Medicare fee-for-service aged ≥65 who were admitted to hospital with a medical condition (as defined by the presence of a medical diagnosis related group on admission) from 1 January 2011 to 31 December 2014. We restricted our sample to patients treated in acute care hospitals and excluded elective admissions and those in which a patient left against medical advice. To allow sufficient follow-up, we excluded patients admitted in December 2014 from 30 day mortality analyses and patients discharged in December 2014 from readmission analyses. Medicare hospital spending and method of assigning physicians to patients In the US, Medicare spending on patients in hospital mainly consists of two components: parts A and B. Part A spending is a fixed payment to a hospital per patient that is determined by the final diagnosis or diagnoses of the patient (categorized into diagnosis related groups) and broadly reflects hospital costs other than professional services. Within each hospital the part A payment does not vary for patients within the same diagnosis related group (with a few exceptions). Part B uses fee-for-service payment, and spending varies with the intensity of services delivered, including visits, procedures, and interpretation of tests and images. Based on previous studies, 10 11 12 we defined the responsible physician for a given admission as the physician who billed the largest share of part B costs during that admission. 13 In a sensitivity analysis, we used alternative assignment methods to assess the robustness of our findings to this attribution rule. We restricted our analyses to admissions for which the highest spending physicians were hospitalists (described below) or general internists. For patients transferred to other acute care hospitals (1.2% of admissions), we attributed the multi-hospital episode of care and associated outcomes to the assigned physician of the initial admissions. 14 15 On average, 51%, 22%, and 11% of total part B spending was accounted for by the first, second, and third highest spending physicians, respectively. Our primary analysis focused on patients treated by hospitalists to examine the possibility that older physicians might treat patients with greater or lesser unmeasured severity of illness. Hospitalists are physicians whose clinical focus is caring for patients admitted to hospital. 16 17 They are typically trained in internal or family medicine. Some complete subspecialty training as well (such as infectious disease or nephrology) but decide to practice general inpatient medicine. The hospitalist specialty began in the 1990s in the US and is the most rapidly growing medical specialty there. Before the introduction of hospitalists, a patient admitted for a general medical condition was cared for by that patient’s primary care physician (equivalent to general practitioner in the UK), who, on any given day, would typically visit his/her inpatients when time permitted in the outpatient schedule. In 2016, it was estimated that more than 50 000 hospitalists were practicing in the US, and about 75% of US hospitals now have hospitalists. 18 Hospitalists typically work in scheduled shifts or blocks (such as one week on and one week off) and do not treat patients in the outpatient setting. Therefore, within the same hospital, patients treated by hospitalists are plausibly quasi-randomized to a particular hospitalist based only on the time of the patient’s admission and the hospitalist’s work schedule. 10 11 19 We assessed the validity of this assumption by testing the balance of a broad range of patient characteristics across categories of age of hospitalist. We defined hospitalists as general internists who filed at least 90% of their total evaluation and management billings in an inpatient setting, a claims based approach that a previous study validated by calling physicians to confirm that they were indeed hospitalists (sensitivity of 84.2%, specificity of 96.5%, and a positive predictive value of 88.9%). 20 Physician age Physician age was defined as the age on the date of admission of patients. Data on physician age were available for 93.5% of physicians. Physician age was modeled both as a continuous linear variable and as a categorical variable (in categories of <40, 40-49, 50-59, and ≥60) to allow for a potential non-linear relation with patient outcomes. We also used linear spline models. Patient outcomes The primary outcome was the 30 day mortality rate in patients (death within 30 days of admission); secondary outcomes were 30 day readmission rates (readmission within 30 days of discharge) and costs of care. Information on dates of death, including deaths out of hospital, was available in the Medicare Beneficiary summary files. Over 99% of dates of death in these files have been verified by death certificate. 21 For mortality analyses, we excluded patients whose death dates were not validated. We defined costs of care as total part B spending per admission. Adjustment variables We adjusted for patient characteristics, physician characteristics, and hospital fixed effects. Patient characteristics included age in five year increments, sex, race or ethnic group (non-Hispanic white, non-Hispanic black, Hispanic, other), primary diagnosis (diagnosis related group), 27 comorbidities (Elixhauser comorbidity index 22 ), median household income of zip code (in 10ths), an indicator for dual Medicare-Medicaid coverage, day of the week of the admission date (to account for the possibility that severity of illness of patients could be higher on specific days of the week), and year indicators. Physician characteristics (other than age) included sex, indicator variables for medical school from which a physician graduated (all foreign schools were grouped into a single category), and whether they graduated from allopathic (MD) or osteopathic (DO) medical schools (allopathic and osteopathic schools both teach the same basic curriculums necessary to become a qualified physician, but osteopathic schools emphasize prevention and other techniques as well). We included indicator variables for each hospital, which allowed each hospital to have its own intercept in the regression analyses, a statistical method known as hospital fixed effects. Hospital fixed effects account for both measured and unmeasured characteristics of hospitals that do not vary over time, including unmeasured differences in patient populations, thereby effectively comparing patient outcomes among hospitalists of varying age within the same hospital. 23 24 25 Statistical analysis First, we examined the association between physician age and 30 day mortality using a multivariable logistic regression model treating age as both a continuous variable and a categorical variable to allow for a non-linear relation, adjusting for patient and physician characteristics and hospital fixed effects. We also used linear age splines. To evaluate whether splines improve goodness of fit compared with modeling a linear relation between physician age and patient mortality, we performed a Wald test adjusted for clustering (to approximate a likelihood ratio test because standard likelihood based tests are unavailable with clustered data). To account for potential correlations of patient outcomes within the same physicians, we clustered standard errors at the physician level. 26 To overcome complete or quasi-complete separation problems (perfect or nearly perfect prediction of the outcome by the model), we combined diagnosis related group codes with no outcome event (30 day mortality or readmission) into clinically similar categories. 27 We calculated adjusted 30 day mortality rates using margins of responses (also known as predictive margins); for each admission we calculated predicted probabilities of outcome with physician age group fixed at each level and then averaged over the distribution of covariates in our national sample. 28 Second, because physicians with high volumes of patients might better maintain clinical knowledge and skills, 29 30 31 32 we examined whether the association between physician age and patient mortality was modified by volume. We classified physicians into thirds of patient volume: low (estimated number of total admissions <90 per year), medium (91-200 admissions), and high (>201 admissions). Within each group, we examined the association between physician age and patient mortality, adjusting for patient and physician characteristics and hospital fixed effects. We used a Wald test to formally test the interaction between physician age and patient volume. Finally, we evaluated the association between physician age and 30 day readmissions and costs of care. We used multivariable logistic regression models for readmission analyses. Because cost data were right skewed, we used a generalized linear model (GLM) with a log link and gamma distribution. 33 Secondary analyses We conducted several secondary analyses. First, to test the generalizability of our findings, we repeated our analyses among general internists overall, including both hospitalists and non-hospitalists. Second, to evaluate whether our findings were sensitive to how we attributed patients to physicians, we tested two alternative attribution rules: attributing patients to physicians with the largest number of evaluation and management claims and attributing patients to physicians who billed the first claim for a given admission (“admitting physician”). Third, because the association between physician age and mortality could be confounded by unobserved care preferences of patients, such as do-not-resuscitate directives, we excluded patients with cancer and those discharged to a hospice. Fourth, to assess the relation between physician age and patient outcomes in a relatively young population whose probability of death is lower, we restricted our analysis to patients aged 65-75. Fifth, an increasing number of young subspecialists in specialties like nephrology and infectious disease work as hospitalists but were excluded from our primary analyses. To investigate this, we reanalyzed the data including hospitalists with medical subspecialties and adjusted for their specialty. Sixth, patients who are admitted multiple times might not be randomly assigned to a given hospitalist but instead to the hospitalist who treated the patient previously. To deal with this, we reanalyzed the data after restricting our sample to the first admission. Seventh, we also evaluated in hospital, 60 day, and 90 day mortality rates to assess if any survival gains were short lived. Eighth, we used generalized estimating equations (GEE) with an independent covariance matrix to account for the hierarchical structure of the data because of the grouping of patients within hospitals, adjusting for patient and physician characteristics and hospital fixed effects. 34 Ninth, to focus on more homogenous patient populations, we separately analyzed the four most common conditions treated by hospitalists in our data (sepsis, pneumonia, congestive heart failure, and chronic obstructive pulmonary disease) (see table A in the appendix for diagnosis codes). Tenth, we used years since completion of residency, instead of physician age, as a measure of physician experience. We did not use this variable for our primary analyses because data on year of residency completion were missing for 35.5% of physicians, and we were concerned that missingness might not be at random. Eleventh, we conducted a formal sensitivity analysis to assess the extent to which an unmeasured confounder might explain our results. 35 Twelfth, we conducted cost analysis using different model specifications: a GLM model with a log link and a negative binomial distribution, a GLM model with a log link and a Poisson distribution, and an ordinary least squares model after winsorizing the top 1% of observations with largest residuals (replacing outlier costs by the most extreme retained values). Finally, we conducted analyses among subgroups including Medicare beneficiaries aged ≥65 who were admitted to hospital with an emergency medical condition (as opposed to our baseline analysis of “non-elective” conditions, which included both emergency and urgent admissions), Medicare beneficiaries aged ≥65 who were admitted with an elective medical condition, and Medicare beneficiaries aged 20-64. The latter group qualified for Medicare through disability and has generally worse health status than the general US population aged below 65, but nonetheless the generalizability of our findings to populations of younger patients is of interest. Data preparation was conducted with SAS, version 9.4 (SAS Institute), and analyses were performed with Stata, version 14 (Stata-Corp, College Station, TX). Patient involvement No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for the design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. There are no plans to disseminate the results of the research to study participants or the relevant patient community. Patient consent was not required for the study. Results Physician and patient characteristics The median and mean age among 18 854 hospitalist physicians in our sample in 2014 was 41.0 and 42.9, respectively. A broad range of patient characteristics, including the number of Elixhauser comorbidities and composite Elixhauser comorbidity scores, 36 were balanced across physicians with different ages (table 1 ⇓ ). Table 1 Study population of Medicare beneficiaries treated in hospital in 2011-14 by age of treating physician. Figures are percentage unless otherwise specified View this table: View popup View inline Physician age and patient mortality The overall 30 day mortality rate in our final sample of 736 537 hospital admissions was 11.1%. Figure 1 ⇓ shows the results of a logistic regression model with linear splines. After adjustment for patient and physician characteristics and hospital fixed effects, older physicians had significantly higher patient mortality than younger physicians. We could reject the null of linearity (P=0.02, fig 1 ⇓ ). Because we observed a non-linear relation between physician age and patient mortality, we also fitted a linear spline logistic model allowing for different slopes for physicians aged <60 and aged ≥60 and found that an additional 10 years increase in physician age was associated with an adjusted odds ratio of 30 day mortality of 1.03 (95% confidence interval 1.02 to 1.05; P<0.001) for physicians aged <60, and 1.22 (1.08 to 1.37; P=0.01) for physicians aged ≥60 (table 2 ⇓ ). When age was modeled as a continuous linear variable, an additional 10 year increase in physician age was associated with an adjusted odds ratio of 30 day mortality of 1.04 (1.03 to 1.06; P<0.001), interpreted as the average odds ratio across all physician age groups (table 2 ⇓ ). Fig 1 Adjusted association between physician age and patient mortality with linear spline model. Multivariable logistic regression model with linear splines was used with knots placed at physician age of 40, 50, and 60, adjusted for patient and physician characteristics and hospital fixed effects. Solid line represents point estimates, and shaded area represents 95% CI around these estimates Download figure Open in new tab Download powerpoint Table 2 Association between physician age and 30 day patient mortality in patients admitted to hospital. Table includes results of three analyses: modeling age as a continuous variable, modeling age as a continuous variable with separate splines at ages <60 and ≥60, and modeling age as categorical variable. All models adjusted for patient and physician characteristics and hospital fixed effects. Standard errors were clustered at physician level. Estimates should be interpreted as average odds ratio across all physician age categories View this table: View popup View inline Treating physician age as a categorical variable showed a monotonic relation between physician age and patient mortality. Physicians aged <40 had the lowest patient mortality rate (adjusted 30 day mortality rate 10.8%, 95% confidence interval 10.7% to 10.9%), followed by physicians aged 40-49 (11.1%, 11.0% to 11.3%), 50-59 (11.3%, 11.1% to 11.5%), and ≥60 (12.1%, 11.6% to 12.5%) (table 2 ⇑ ). Physician age and patient mortality by volume Physician age was positively associated with patient mortality among physicians with low and medium volumes of patients but not among those with high volumes of patients (table 3 ⇓ ), who also had the lowest overall mortality rate. For instance, each 10 year increase in physician age was associated with adjusted odds ratios of 30 day mortality of 1.19 (95% confidence interval 1.14 to 1.23; P<0.001) and 1.06 (1.03 to 1.09; P<0.001) among low and medium volume physicians, respectively. In contrast, despite the larger sample size among high volume physicians, we observed no association between physician age and patient mortality (adjusted odds ratio for additional 10 years, 1.01, 0.99 to 1.03; P=0.29). The interaction between physician age and patient volume was significant (P<0.001). Table 3 Physician age and 30 day patient mortality in patients admitted to hospital, stratified by patient volume View this table: View popup View inline Patient readmissions and part B spending We found no association between physician age and the patient 30 day readmission rate (adjusted odds ratio for additional 10 years, 1.00, 95% confidence interval 0.99 to 1.01; P=0.82) (table 4 ⇓ ). Although differences in part B spending between physicians of varying age were significant, they were small. Each 10 year increase in physician age was associated with a 2.4% increase (2.0% to 2.8%; P<0.001) in part B spending. Table 4 Association between physician age, 30 day readmission rate, and costs of care in patients admitted to hospital View this table: View popup View inline Secondary analyses Our overall findings were qualitatively unaffected by including non-hospitalist general internists, although the observed relation between patient mortality and physician age was smaller (table B in appendix). The smaller relation between physician age and patient outcomes might be because of unobserved differences in severity of illness between patients treated by young versus old physicians or actual differences in how physician age relates to patient mortality among hospitalists versus non-hospitalist general internists. Our findings were also not sensitive to using alternative methods for attributing physicians (table C in appendix); excluding patients with cancer or discharged to hospice (table D in appendix); restricting analysis to patients aged 65-75 (table E in appendix); including hospitalists with medical subspecialty boards (table F in appendix); restricting to the first admission for patients with multiple admissions (table G in appendix); using in hospital, 60 day, and 90 day mortality rates instead of 30 day mortality (tables H and I in appendix); and the use of GEE instead of cluster robust standard errors (table J in appendix). When we stratified by primary diagnosis, older hospitalists had higher patient mortality for sepsis, congestive heart failure, and chronic obstructive pulmonary disease, but not for pneumonia (table K in appendix). When using years in practice instead of age we found similar results (table L in appendix). A formal test for an unmeasured confounder showed that it is unlikely that this could explain the observed association between physician age and patient mortality (table M in appendix). Use of different model specifications for cost analyses did not qualitatively affect our findings (table N in appendix). Finally, we observed similar relations between physician age and 30 day patient mortality in subgroup analyses of Medicare beneficiaries aged ≥65 who were admitted with an emergency medical condition, Medicare beneficiaries aged ≥65 who were admitted with an elective medical condition, and admitted Medicare beneficiaries aged 20-64 (tables O and P in appendix). Discussion Principal findings In a national sample of elderly Medicare beneficiaries admitted to hospital with medical conditions, we found that patients treated by older physicians had higher 30 day mortality than those cared for by younger physicians, despite similar patient characteristics. These associations were found among physicians with low and medium volumes of patients but not among those with high volumes. Readmission rates and costs of care did not meaningfully vary with physician age. Taken together, our findings suggest that differences in practice patterns or process measures of quality between physicians with varying years of experience reported in previous studies 1 2 3 4 37 might have a meaningful impact on patient outcomes. Mechanisms that could explain our current findings can be broadly categorized into effects of age (“age effects”) versus effects arising from differences in how younger and older physicians trained (“cohort effects”). It is possible that physicians further from training are less likely to adhere to evidence based guidelines, might use newly proved treatments less often, and might more often rely on clinical evidence that is not up to date. 38 Moreover, while intense exposure to a large number of patients during residency training might enable physicians shortly out of such training to provide high quality care, the benefits of this training experience could wane if physicians care for fewer inpatients after residency. The lack of association between physician age and patient mortality among physicians with higher volume of patients supports this age related hypothesis. Our findings might just as likely reflect cohort effects rather than declining clinical performance associated with greater age, which has important implications for interpretation of our findings. Hospital medicine is among the most rapidly evolving specialties within medicine, with dramatic changes in the training of recent cohorts of physicians who now work as hospitalists, including greater emphasis on multi-professional team based practice, adherence to clinical guidelines, training on patient handoffs, familiarity with shift work during residency training, and an improved hospital safety culture. Because the specialty of hospital medicine was first recognized in the 1990s, our study might have compared younger physicians who began their careers as hospitalists with older physicians who began their careers as primary care physicians and later became hospitalists. Thus, cohort differences in physician training, as well as declines in skill with aging, could explain our findings. Under this hypothesis, the cohort of physicians entering hospital medicine today might experience no reduction in patient outcomes with aging or possibly improved outcomes. Nonetheless, from the perspective of policymakers and administrators, current outcomes of older versus younger hospitalists might still be important to know irrespective of the path by which younger versus older physicians entered the specialty. Our findings suggest that within the same hospital, patients treated by physicians aged <40 have 0.85 times the odds of dying (1.00/1.17) or an 11% lower probability of dying (10.8/12.1), compared with patients cared for by physicians aged ≥60 (table 2 ⇑ ). This difference in mortality is comparable with the impact of statins for the primary prevention of cardiovascular mortality on all cause mortality (odds ratio of 0.86) 39 or the impact of β blockers on mortality among patients with myocardial infarction (incidence rate ratio of 0.86), 40 indicating that our observed difference in mortality is not only statistically significant but arguably clinically significant. In addition, if our results are causal, an adjusted risk difference of 1.3 percentage points suggests that for every 77 patients treated by doctors aged ≥60, one fewer patient would die within 30 days of admission if those patients were cared for by physicians aged <40. Policy implications Our findings should be regarded as exploratory. Nonetheless, they highlight the importance of patient outcomes as one component of an assessment of how physician practices change over a career. The purpose of continuing medical education is to ensure that physicians provide high quality care over the course of their careers. Although continuing medical education can take multiple forms that vary across specialties and across countries, the issue of ensuring that physicians keep up with current standards of care is applicable across all specialties and countries. In the US, for example, there are ongoing debates about the requirements for maintenance of certification, with many physicians arguing that current requirements could be burdensome and unneeded. Although our study did not analyze the effects of current such policies in the US, it suggests that continuing medical education of physicians could be important and that continual assessment of outcomes might be useful. In addition, although quality of care initiatives have largely focused on system level measures (such as hospital 30 day mortality and readmissions), there is increasing policy emphasis on the role of individual physicians in influencing costs and quality of care. 41 42 43 For example, in the US, the Centers for Medicare and Medicaid Services has just promulgated draft final regulations for a new approach to pay individual clinicians for the value and quality of care they provide under the Medicare Access and CHIP Reauthorization Act (MACRA). 44 Strengths and limitations of study Our study has several limitations. First, our findings would be confounded if older physicians, on average, treat patients at higher risk of 30 day mortality because of factors unmeasured by our analysis. We specifically chose our within hospital study design to deal with this concern, hypothesizing that patients are essentially randomized to hospitalist physicians of various ages within the same hospital, an assumption supported by the largely similar demographic and clinical characteristics across patients that older and younger physicians treat. Second, we found that the positive association between physician age and patient mortality was driven primarily by physicians treating a low to medium volumes of patients, suggesting that high volumes could be “protective” of clinical skills. The association between practice volume and skills, however, could be bidirectional—physicians whose skills are declining might either self select, or be encouraged by others to leave, positions in which they are responsible for clinical management of large numbers of patients and could, therefore, treat fewer patients over time. Nonetheless, it is still important to know that older physicians with low and medium volumes of patients have worse patient outcomes because this information could suggest that specific interventions could be targeted towards these physicians. Third, the cross sectional nature of our study did not allow us to distinguish the degree to which our findings were attributable to declines in clinical performance with physician age versus cohort effects associated with secular changes in training. In the latter case, although older physicians could now be associated with higher patient mortality, as the current cohort of younger hospitalists age they might retain their superior patient outcomes even without individual maintenance of certification-type interventions. Fourth, physician age is only one of several factors associated with physician performance; physicians of varying skill level can be found within every age category. Finally, our findings might not generalize to the non-Medicare population, to patients cared for by surgeons or other specialists, or to physicians practicing in other countries (particularly as rates of hospitalist use might differ across countries). Further studies are warranted to understand whether similar patterns are observed in these other settings. Conclusions Patients in hospital treated by older hospitalists have higher mortality than patients cared for by younger hospitalists, except for hospitalist physicians with high volumes of patients. We found similar associations among patients treated by general internists. Readmission rates and costs of care did not meaningfully vary with physician age. What is already known on this topic Whether quality of care differs between younger and older physicians remains largely unknown Though clinical skills and knowledge accumulated by more experienced physicians could lead to improved quality of care, physicians’ skills might become outdated as scientific knowledge, technology, and clinical guidelines change Older physicians might have decreased clinical knowledge, adhere less often to standards of appropriate treatment, and perform worse on process measures of quality with respect to diagnosis, screening, and preventive care What this study adds This study examined patient outcomes, including 30 day mortality, readmissions, and costs of care, in a nationally representative sample of US Medicare beneficiaries admitted to hospital with a medical condition in 2011-14 Patients were treated by hospitalists (physicians whose clinical focus is caring of patients in hospital), to whom they are typically assigned based on scheduled work shifts Within the same hospital, patients treated by older hospitalists had similar characteristics to patients treated by younger hospitalists but had higher mortality rates, with the exception of those hospitalists who treated high volumes of patients Readmissions did not vary with physician age, while costs of care were slightly higher among older physicians Footnotes Contributors: All authors contributed to the design and conduct of the study, data collection and management, analysis interpretation of the data; and preparation, review, or approval of the manuscript. ABJ supervised the study and is guarantor. Funding: YT is supported in part by the Abe Fellowship (Social Science Research Council and the Japan Foundation Center for Global Partnership). ABJ is supported by the Office of the Director, National Institutes of Health (NIH Early Independence Award, grant 1DP5OD017897-01). The research conducted was independent of any involvement from the sponsors of the study. Study sponsors were not involved in study design, data interpretation, writing, or the decision to submit the article for publication. Competing interests: All authors have completed the ICMJE uniform disclosure form at (available on request from the corresponding author) and declare: ABJ has received consulting fees unrelated to this work from Pfizer, Hill Rom Services, Bristol Myers Squibb, Novartis Pharmaceuticals, Vertex Pharmaceuticals, and Precision Health Economics, a company providing consulting services to the life sciences industry. JPN is a director of Aetna and has received consulting fees from EMD Serono; no other relationships or activities that could appear to have influenced the submitted work. Ethical approval: The study was approved by the institutional review board at Harvard Medical School. Data sharing: No additional data available. Transparency statement: The corresponding author affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies are disclosed. This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: .

Patients in US hospitals treated by older physicians have higher mortality than patients cared for by younger physicians, except those physicians treating high volumes of patients, finds a study published by The BMJ today. If the results are causal, they suggest that for every 77 patients treated by doctors aged 60 or over, one fewer patient would die within 30 days of admission if those patients were cared for by physicians aged less than 40. However, the researchers stress that their findings should be regarded as exploratory. Clinical skills and knowledge accumulated by more experienced physicians can lead to improved quality of care. It is possible, however, that physicians' skills may become outdated as scientific knowledge, technology, and clinical guidelines evolve. Whether quality of care differs between younger and older physicians remains largely unknown, so a team led by Yusuke Tsugawa at Harvard T H Chan School of Public Health in Boston, set out to investigate whether outcomes of patients who were admitted to hospital differ between those treated by younger and older physicians. They analysed 30 day mortality, readmissions and costs of care for a random sample of 736,537 elderly Medicare patients (aged 65 or over) managed by 18,854 hospital physicians (average age 41) at US acute care hospitals from 2011 to 2014. Patients were assigned a physician based on scheduled work shifts and patients'characteristics were similar across physician ages. After adjusting for patient, physician, and hospital characteristics that could have affected the results, patients' 30 day mortality rates were 10.8% for physicians aged less than 40, 11.1% for physicians aged 40-49, 11.3% for physicians aged 50-59, and 12.1% for physicians aged 60 or over. Among physicians with a high volume of patients, however, there was no associationbetween physician age and patient mortality, suggesting that high volumes could be "protective" of clinical skills, say the authors. Readmissions did not vary with physician age, while costs of care were slightly higher among older physicians. And similar patterns were observed after further analyses to test the strength of the results. The researchers say this is an observational study, so no firm conclusions can be drawn about cause and effect, and they outline some limitations could have introduced bias. Nevertheless, they conclude that "within the same hospital, patients treated by older physicians had higher mortality than patients cared for by younger physicians, except those physicians treating high volumes of patient volumes." In a linked editorial, researchers at the University of Pennsylvania ask what are the options for ensuring that quality and safety of care is optimized for patients. They point out that patient outcomes research "is providing much needed evidence to inform clinical practice, educational innovation, organizational redesign, and healthcare policy." The challenge, they say, "is to integrate findings across multiple studies within an overarching framework of health system responsibility, as recommended by the Institute of Medicine, which holds promise of safe care and good patient outcomes despite diversity of performance by individuals."

www.bmj.com/content/357/bmj.j1797

Physics

Devil in the defect detail of quantum emissions unravelled

Identifying carbon as the source of visible single-photon emission from hexagonal boron nitride, Nature Materials (2020). DOI: 10.1038/s41563-020-00850-y , www.nature.com/articles/s41563-020-00850-y Journal information: Nature Materials

http://dx.doi.org/10.1038/s41563-020-00850-y

https://phys.org/news/2020-11-devil-defect-quantum-emissions-unravelled.html

Abstract Single-photon emitters (SPEs) in hexagonal boron nitride (hBN) have garnered increasing attention over the last few years due to their superior optical properties. However, despite the vast range of experimental results and theoretical calculations, the defect structure responsible for the observed emission has remained elusive. Here, by controlling the incorporation of impurities into hBN via various bottom-up synthesis methods and directly through ion implantation, we provide direct evidence that the visible SPEs are carbon related. Room-temperature optically detected magnetic resonance is demonstrated on ensembles of these defects. We perform ion-implantation experiments and confirm that only carbon implantation creates SPEs in the visible spectral range. Computational analysis of the simplest 12 carbon-containing defect species suggest the negatively charged \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) defect as a viable candidate and predict that out-of-plane deformations make the defect environmentally sensitive. Our results resolve a long-standing debate about the origin of single emitters at the visible range in hBN and will be key to the deterministic engineering of these defects for quantum photonic devices. Main Single defects in solids have become some of the most promising front-runner hardware constituents of applications in quantum information technologies and integrated quantum photonics 1 . Substantial effort has been devoted to isolate and deterministically engineer such defects in wide-band-gap materials such as diamond and silicon carbide 2 , 3 . This collective effort resulted in spectacular proof-of-principle demonstrations ranging from quantum networks to spin–photon interfaces 3 , while simultaneously and steadily leading to understanding the fundamental-level structures of these defects. Recently, hexagonal boron nitride (hBN) has emerged as a promising host material for defects which display ultrabright single-photon emission (SPE) 4 , 5 , 6 , 7 , 8 . They exhibit remarkable properties: a strong response to applied strain and electric fields (Stark shifts) 9 , 10 , 11 , stability under high pressure and elevated temperatures 12 , 13 , potential for resonant excitation above cryogenic temperatures 14 , 15 , 16 and addressability via spin-selective optical transitions 17 , 18 . Yet, despite the numerous experimental characterizations and in-depth theoretical attempts to model their possible crystalline structure 6 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , the nature of these defects remains unknown. Part of the challenge stems from standard hBN bulk crystal synthesis via high pressure and high temperature not being amenable to the deterministic control of impurity incorporation. This is aggravated by the induced impurities often segregating and forming regions of inhomogeneous defect concentration 26 . In addition, the two-dimensional, layered nature of hBN makes ion implantation difficult to control. These limitations have precluded identifying the exact origin of the single-photon emission in the material. Here we address this problem by carrying out a detailed study surveying various hBN samples grown in different laboratories by metal–organic vapour-phase epitaxy (MOVPE) and molecular beam epitaxy (MBE). We find compelling evidence that to observe photoluminescence from SPEs the inclusion of carbon atoms in hBN is required. By systematically growing samples with different carbon concentrations, we show that the carbon content determines whether the photoluminescence signal originates from an ensemble of emitters (high carbon concentration) or isolated defects (low carbon concentration). Defect ensembles are demonstrated to display room-temperature optically detected magnetic resonance (ODMR). We carry out multispecies ion-implantation experiments on both MOVPE films and exfoliated hBN, showing that only carbon implantation creates SPEs and that the density of emitters scales directly with the implantation dose of carbon. Our results are supported by rigorous modelling analysis of carbon-related defects. Optical properties of carbon-doped hBN Table 1 summarizes the materials analysed. They are epitaxial hBN samples grown by different methods and under various conditions. The rationale was to understand whether the single defects are intrinsic to hBN (for example, substitutional or interstitial nitrogen or boron complexes) or involve foreign atoms (for example, carbon). We investigated hBN samples grown by four methods: (1) MOVPE with varying flow rates of the precursor triethylboron (TEB)—a parameter known to systematically alter the levels of incorporated carbon; (2) high-temperature MBE on sapphire with and without a source of carbon; (3) high-temperature MBE on SiC with a varying orientation of the Si face to explore the possibility of carbon incorporation occurring from the substrate; (4) growth by the conversion of highly oriented pyrolytic graphite (HOPG) into hBN. Note that in the current work we focus on bottom-up growth of hBN as it offers an opportunity for large (centimetre) scale films of desired thickness (down to ~1 nm), as well as better control over the inclusions of impurities. Table 1 Epitaxial hBN samples with varying carbon concentrations Full size table We first explore the photoluminescence (PL) from a series of hBN samples grown by MOVPE 27 as the TEB flow rate is increased and the ammonia flow is kept constant (Fig. 1a ). The aim of this measurement is to engineer an ensemble of hBN emitters, and to compare their properties with isolated SPEs grown using the same growth technique. A region of the TEB 10 (µmol min −1 ) sample with the lowest percentage of carbon shows negligible fluorescence. Increasing the flow rate to TEB 20 is accompanied by the appearance of a bright fluorescence signal with two clear peaks appearing at ~585 and ~635 nm. Further increasing the flow rate to TEB 30 and 60 provides a similarly structured PL signature, with higher fluorescence intensity, confirming that higher PL intensity directly correlates with higher TEB flux. Moderate fluctuations in the peak positions and the intensity ratio of the 585 and 635 nm peaks at different sample locations are consistent with emission from dense ensembles of hBN emitters. This also confirms previous findings showing that hBN emitters possess zero-phonon line (ZPL) wavelengths clustered at ~585 nm when the sample is grown epitaxially 28 , 29 . The energy detuning between the ZPL of the ensemble and phonon sideband (PSB) peak is ~176 meV on average (Extended Data Fig. 1 ) 30 , 31 . Fig. 1: Photoluminescence from MOVPE hBN samples. a , MOVPE hBN grown with increasing flow rates of TEB. As TEB flow increases, the fluorescence of SPE ensembles increases. b , Percentage of B–C bonding with increasing TEB flow evaluated by XPS. c , Percentage of N–C bonding with increasing TEB flow evaluated by XPS. d , Room-temperature ODMR displayed as relative contrast, spin-dependent variation in photoluminescence (∆PL/PL), observed from the ~585 nm ensemble emission of MOVPE hBN (TEB 60) at applied fields of 19, 24 and 29 mT, respectively. e , Spectrum of a representative SPE found in MOVPE hBN TEB 10. Inset: corresponding autocorrelation measurements from the spectrum. Source data Full size image X-ray photoelectron spectroscopy (XPS) was used to quantify the incorporation of carbon (Extended Data Fig. 2 ). Figure 1b,c demonstrates a near-linear correlation between C–B (C–N) bonding and increasing TEB flux, with C–B bonding being roughly an order of magnitude more prevalent than C–N bonding. Preferential formation of C–B bonds follows logically from noting the boron species are introduced with three pre-existing bonds to carbon. The PL intensity of the resulting ensemble emission likewise displays a linear correlation with carbon concentration (Extended Data Fig. 3 ). Based on these results, we advance that the SPE emission at ~580 nm in hBN probably originates from a carbon-related defect complex. Figure 1d shows the ODMR spectra recorded from the TEB 60 ensemble. The highly symmetric shape of the signal does not reveal a structure that would allow a clear assignment to a specific intrinsic or extrinsic defect. However, it is clearly a spin-carrying defect, likely with a spin state higher than S = 1/2. By varying the static magnetic field B , we measure resonances at ~523, ~668.5 and ~815.4 MHz for B = 19, 24 and 29 mT, respectively. A value for g e of ~2.09 is extracted (Extended Data Fig. 4 ). However, we observe no splitting of the signal with the magnetic field, which means that the zero-field splitting, D , should be small. In previous experiments, a similar ODMR signal was observed at low temperature ( T ≈ 8.5 K) 18 , while our measurements show that ODMR is also feasible at room temperature. We observe no narrowing of the resonance upon cooling, suggesting the linewidth is not governed by the coherence time (Extended Data Fig. 5 ). The line-broadening may be due to dipole–dipole coupling; for example, by hyperfine interaction with nearby nuclei. This is consistent with the preliminary assignment that the ODMR signal measured on the heavily carbon-doped sample and shown in Fig. 2d is associated with a carbon-related defect 18 . Fig. 2: Photoluminescence from hBN samples fabricated by MBE and HOPG conversion. a , Undoped MBE hBN on sapphire displays no SPEs. b , Carbon-doped MBE hBN on sapphire displaying a number of isolated peaks spanning the visible range. Typically, many SPEs are found within the laser excitation spot. c , Polarization-resolved PL of a single peak in carbon-doped MBE hBN on sapphire, demonstrating the polarized nature of the emission. d , Undoped MBE hBN on SiC, with the Si face oriented at 0° (blue) and 8° off (red). While growth of the Si 0° face SiC shows no SPEs, growth on the Si 8°-off face effectively incorporates SPEs via diffusion of carbon from the SiC substrate. e , Raman spectra of the HOPG to hBN conversion sample. f , Converted hBN displays an SPE ensemble emission centred around ~585 nm. Source data Full size image We next employ a lab-built confocal PL set-up with a 532 nm excitation source to study in detail the TEB 10 sample. The level of carbon doping is such that we can isolate single quantum emitters; a representative spectrum for one such emitter is shown in Fig. 1d . The quantum nature of the emission was confirmed by measuring the second-order autocorrelation function; the value of g (2) ( τ = 0) < 0.5 (Fig. 1d inset) is conventionally attributed to a single photon source with sub-Poissonian emission statistics. We measured the ZPL wavelength of 77 SPEs in the MOVPE hBN (TEB 10) sample, finding that ~78% of the emitters are located at 585 ± 10 nm and 95% at wavelengths <600 nm (Extended Data Fig. 6 ), consistent with previous studies on epitaxially grown hBN 28 , 29 . The typical line shape of these emitters at room temperature is also consistent with previous studies, including the ZPL and a PSB centred at ~177 meV from the ZPL energy. This suggests that when the carbon concentration is sufficiently low, individual quantum emitters can be isolated. Their optical properties and spectral distribution are consistent with those observed in samples with higher carbon doping, with the difference merely being due to the density of emitters. To further confirm that carbon-based defects are responsible for SPE emission from hBN we analysed a series of hBN samples grown by a different method: MBE 32 , 33 . Fig. 2a displays the PL spectrum observed from undoped MBE hBN grown on sapphire substrate. The resulting PL signal was relatively low; no SPEs could be found despite the material being of good quality as shown by a clear hBN E 2 g Raman line. However, when the elemental boron source was placed inside a carbon crucible—with otherwise identical growth conditions—we observed the appearance of sharp spectral lines, shown in Fig. 2b . The carbon crucible used for e-beam evaporation of the boron shows clear signs of sidewall etching, which suggests that carbon was present in the gas phase during growth. The carbon-doped MBE growth resulted in a high density of emitters such that we could not isolate at single sites. We instead probed the polarization dependence of particular emission peaks by placing a polarizer in the collection path. Figure 2c shows one such collection, where emission from a ZPL at ~577 nm is linearly polarized, with the PL intensity dropping to the background level when the polarizer is perpendicular to the polarization direction of the probed emitter. We next explored MBE growth of hBN on silicon carbide (SiC), investigating different crystal orientations: specifically, with the top Si face-on (0°) and slightly off (8°). Representative spectra from both sample types (Si at 0° and at 8°) are displayed in Fig. 2d . When growth was performed with the Si face at 0°, only a single SPE peak was located across a 40 µm 2 scan. In contrast, when the Si face is oriented at 8° we again find a high density of SPE incorporation. The incorporated SPEs display a similar ZPL distribution to the carbon-doped MBE hBN on sapphire (Extended Data Fig. 6 ). We attribute the incorporation of these SPEs during hBN growth on SiC to carbon diffusion from the substrate. At the growth temperature of 1,390 °C, some sublimation of Si from the surface of the SiC substrates is expected, with the subsequent formation of an extra carbon layer on the surface of SiC 32 , 34 . While these temperatures are sufficient to sublime Si, they are not sufficient to evaporate carbon from the SiC surface 32 . Interestingly, carbon incorporation into hBN appears dramatically enhanced when the Si face is oriented 8° out of plane. The observed dependence of SPE incorporation during MBE growth further supports the role of carbon in the origin of hBN SPEs in the visible spectral range. Finally, we analyse a third technique for hBN growth, the conversion of HOPG to hBN, known to yield high-quality porous hBN 35 . Conversion was confirmed by Raman spectroscopy (Fig. 2e ) 36 . The conversion from graphite, proceeding via atomic substitutions, provides a high availability of carbon for incorporation as defects in the resulting hBN. Figure 2f displays a typical PL spectrum from the sample. We observe a bright SPE ensemble, displaying a structured emission profile with ZPL and PSB peaks displaying similar transition energies as observed for high-carbon MOVPE ensembles. Ion implantation We now turn our attention to using ion implantation for defect creation, in an attempt to confirm the role of carbon. We performed a series of implantation experiments (dose, 10 13 ions cm −2 ; energy, 10 keV) with carbon as well as silicon and oxygen used as controls to rule out the possibility of the photoemission being due to native vacancy defects. The implantation experiments were performed on MOVPE hBN (TEB 10) films to compare the relevant results to those for the samples synthesized while increasing carbon content during growth. Figure 3a shows the confocal scan of the TEB 10 sample after carbon implantation, but prior to annealing, where a TEM grid with 50 µm 2 square apertures was used as a mask. The implanted region is labelled I, while the masked region is labelled II. Figure 3b displays spectra collected from emitters within the implanted region (I), and a representative g (2) ( τ = 0) < 0.5, confirming the quantum nature of the emission from these centres. Figure 3c displays a representative emitter from the masked region (II), showing the typical line shape of the ZPL and the PSB peaks found in TEB 10 films, with the corresponding g (2) ( τ = 0) shown to the right. Fig. 3: MOVPE hBN (TEB 10) samples implanted with carbon. Implantations were done at a dose of 10 13 ions cm − 2 and an energy of 10 keV, using a TEM grid with 50 µm 2 square apertures as a mask. a , Confocal scan of carbon-implanted sample, where square areas marked I were implanted, and those marked II were masked. The recorded PL intensity is displayed in avalanche photodiode (APD) counts. b , Spectra from implanted area I displays three distinct SPEs with narrow ZPLs—coloured green, blue and purple for clarity—each with almost no PSB and are attributed to emitters created via implantation. Inset: displays the same three SPEs plotted for the reduced wavelength range 570–595 nm. A representative g (2) ( τ ) is shown to the right. c , Spectrum and g (2) ( τ ) from the masked area II display borderer ZPLs, and prominent PSBs. d , Confocal scan of the carbon-implanted sample, post-annealing, where areas marked I and II were implanted, while area III was masked. The recorded PL intensity is displayed in avalanche photodiode (APD) counts. e , A representative spectrum and g (2) ( τ ) from area I, showing an ensemble of hBN emitters, and a corresponding g (2) ( τ ) measurement showing no dip as expected for ensemble emission. f , A representative spectrum and g (2) ( τ ) from area II showing evidence of quantum emission but with appriciable spectral contributions from nearby SPEs resulting in a g (2) ( τ ) value of ~0.75. g , A representative spectrum and g (2) ( τ ) from the masked area III post-annealing showing a well-resolved SPE and PSB and a g (2) ( τ ) confirming a single emission centre. Source data Full size image Inside the carbon-implanted region, most emitters (~80%) display narrow ZPL peaks (~5 nm full-width at half-maximum (FWHM)) and extremely weak PSBs compared with the typical ZPL/PSB found in these TEB 10 films (Extended Data Fig. 7 ). The remaining ~20% of SPEs within the implanted region display similar line shapes and phonon coupling to those for the emitter in Fig. 3c and are attributed to pre-existing SPEs in the region. Our results indicate that the sharp emission lines belong to SPEs created via implantation of carbon ions. The reasons for the observed narrow line shape and the minimal phonon coupling are explored further via computational modelling below. The samples were then annealed in high vacuum (1,000 °C, <10 −6 torr, 2 h), and the same set of measurements was performed. As shown in Fig. 3d , the implanted regions are still visible; however, they show variations in PL intensity. This effect is likely due to ion scattering around the mask edges and vacancy diffusion—which have been observed for implantation in diamond 37 . The PL spectra from three different areas are shown in Fig. 3e–g , and correspond to (I) the implanted region of high PL intensity, (II) the implanted region of lower PL intensity and (III) the masked region of the film. Figure 3e displays a representative spectrum from inside region I, where we found broad emission similar to those observed in the high TEB flux growths. This emission is confirmed to be due to an ensemble of SPEs as the corresponding g (2) ( τ ) measurements show no anti-bunching despite the associated ZPL/PSB structure. A similar spectral signature is observed consistently throughout region I, again implying the creation of an ensemble of carbon-based SPEs. Figure 3f displays a representative spectrum from the implanted region II, where we again observe luminescence with a similar line shape. The overall ensemble signal remains homogeneous in this region, although appears less dense and bright, and the g (2) ( τ ) measurement shows a value of ~0.75, confirming the presence of fewer emitters within a confocal spot. Note that in both implanted areas (I and II) we no longer observe the narrow emission lines with low phonon coupling found before annealing. Finally, Fig. 3g displays a representative spectrum from region III (masked area), showing a typical ZPL and PSB profile with a g (2) (0) < 0.5. Control experiments implanting silicon and oxygen with otherwise identical conditions were also performed, but the emitters, either singles or ensembles, were not observed (Extended Data Fig. 8 ). To further study SPE formation via ion implantation we performed dose-dependent experiments with carbon fluences over the range 1 × 10 11 –10 14 ions cm − 2 , while oxygen and silicon implantation at 1 × 10 13 ions cm − 2 served as a control. Both MOVPE (TEB 10) hBN and exfoliated pristine hBN flakes (HQ Graphene) were used. The samples were analysed via wide-field imaging, allowing for the direct visualization of the resulting SPE density. Figure 4a shows representative images from the exfoliated hBN flakes before and after annealing. The results demonstrate unambiguously that emitter creation scales with the dose of carbon implantation in both cases, which directly confirms the creation of SPEs. Only a few emitters are formed pre-annealing, even at higher doses, but a direct correlation between SPE formation and implantation fluence is clearly evident post-annealing. Fig. 4: Wide-field imaging of ion-implanted MOVPE and exfoliated hBN. a , Exfoliated hBN samples for a series of carbon-implanted samples with increasing fluences from 10 11 to 10 14 ions cm − 2 . The samples were analysed before and after annealing in high vacuum (1,000 °C, <10 −6 torr, 2 h). Isolated bright spots (corresponding to SPEs) increase with the dose of carbon implantation. b , MOVPE (TEB 10) and exfoliated hBN samples implanted with carbon-, oxygen- and silicon-implanted samples at a fluence of 10 13 ions cm − 2 . For both material types SPE density is not increased by oxygen and silicon implantation but increases upon carbon implantation. Red and green circles in a identify the position at which the spectrum was recorded. Scale bars in a and b , 2 µm. c , Spectra from carbon-implanted exfoliated hBN before and after annealing. Source data Full size image Figure 4b shows a direct comparison for carbon, oxygen and silicon ion implantations at a dose of 1 × 10 13 ions cm − 2 . For the MOVPE samples implanted with oxygen and silicon we observe a similar SPE density to pristine TEB 10, while carbon implantation considerably increases the density. In the exfoliated samples, only carbon implantation results in the direct formation of single emitters at a high density. Figure 4c shows two spectra recorded from the localized emission spots in the carbon-implanted exfoliated samples, before and after annealing. Green and red circles in Fig. 4a mark the position of the recorded spectra in each case. Additional wide-field imaging and spectral characterization for exfoliated and MOVPE hBN are displayed in Extended Data Figs. 9 and 10. In light of the implantation results, we briefly consider a potential ancillary role of carbon. This could occur through the stabilization or charge state modification of alternative defects, as well as modification of the material Fermi level. Critically, our implantation results allow us to rule out these possibilities. The creation of SPEs prior to annealing occurs only for carbon implantation (that is, not with silicon and oxygen implantation), despite clear evidence of increased vacancy creation excluding the secondary role of carbon as that of simply activating native vacancy complexes. Furthermore, complex native defects (for example, V N N B ) or non-carbon heteroatom impurities involving oxygen and silicon are similarly inconsistent with our results. Note also that the dose-dependent implantation experiments conclusively demonstrate that the density of created SPEs scales with the fluence of carbon ions in high-purity hBN materials. Electronic structure calculations To gain further insight into the structure of the carbon defect, we searched for defect transitions from which the observed photoemission could originate. To do so, time-dependent density-functional theory 38 (TD-DFT) calculations were performed using the CAM-B3LYP 39 density functional (see Supplementary Information for extensive details). These are supported by calculations using the HSE06 density functional 40 and the advanced equation-of-motion coupled cluster singles and doubles (EOM-CCSD) 41 methods. Four main defect candidates were considered: C B , C N , V N C B and V B C N (Fig. 5a–d ) in their neutral, negative (−1) and positive (+1) charged states. Two spin manifolds were considered for each (either singlet and triplet or else doublet and quartet), as well as at least ten excited states of each type. Calculations were performed using three-ring, five-ring and ten-ring model compounds containing one or three hBN layers, to account for the effects of the host matrix on the defect. Figure 5e displays the three-ring, one-layer and three-layer model for \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) . Calculations on ten-ring systems were performed using a mixed quantum-mechanics/molecular-mechanics (QM/MM) scheme, utilizing an AMBER 42 potential fitted to mimic CAM-B3LYP results on five-ring hBN. Fig. 5: Computational modelling. Properties were determined for defects in their neutral and ±1 charged states. a – d , Indicative high-symmetry defect structures (N, blue; B, peach; C, cyan) for \({\rm{C}}_{\rm{B}}^ +\) , \({\rm{C}}_{\rm{B}}^ -\) and \({\rm{C}}_{\rm{B}}\) ( a ), \({\rm{C}}_{\rm{N}}^ +\) , \({\rm{C}}_{\rm{N}}^ -\) and \({\rm{C}}_{\rm{N}}\) ( b ), \({\rm{V}}_{\rm{N}}\rm{C}_{\rm{B}}^ -\) , \({\rm{V}}_{\rm{N}}{\rm{C}}_{\rm{B}}^ +\) and \({\rm{V}}_{\rm{N}}{\rm{C}}_{\rm{B}}\) ( c ) and \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) , \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ +\) and V B C N ( d ). The only feasible emission source is the (1) 4 B 1 → (1) 4 A 2 transition in \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) , with fully optimized three-ring, one-layer and three-ring, three-layer structures shown in e , along with the allowed in-plane-perpendicular electric polarization vector. f , predicted band shape (black dashed line, basic CAM-B3LYP three-ring, one-layer model; black solid line, QM/MM EOM-CCSD ten-ring, three-layer out-of-plane distorted model) compared with observed ones from region I (blue, Fig. 3b C-implanted) and region II (red, Fig. 3c masked). The observed spectra are shown after correction for instrument response functions and converted from raw emission E ( λ ) to band shape λ 5 E ( λ ) plotted versus energy hν = hc / λ , displaying broadening to a resolution of 0.01 eV. The predicted spectra are both too low in energy and too broad compared with the experimental ones, but the assignment is within computational uncertainty. Source data Full size image Given the large number of possible defect candidates considered, we proceeded by eliminating unsuitable ones by benchmarking our calculations to known experimental properties. We focused on three well-established experimental features of the SPEs, a ZPL energy transition of ~2.1 eV 28 , 29 , a fast excited state lifetime of ~2–6 ns 7 and a high quantum efficiency 43 . Accordingly, computational results were filtered to reproduce, first, a CAM-B3LYP-calculated lowest-energy transition of 1.6–2.6 eV (based on the expected worst-case computational error, calibrated for this method to be ±0.5 eV) 22 , and second, an oscillator strength exceeding 0.1, compatible with the observed short photoluminescence lifetime and high quantum yield. Few defects have lowest transition energies in this range, and most transitions are predicted to have oscillator strengths 100th of this or much less. Based on these considerations, only two candidates remain of interest amongst the options considered: the (1) 4 B 1 → (1) 4 A 2 transition in \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) and the (2) 3 B 1 → (1) 3 B 1 transition in V N C B . Of these, V N C B is immediately eliminated as its spectral band shape and most other properties are highly inconsistent with observed features; hence we focus on \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) . The ground state of \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) is predicted to be (1) 4 A 2 , with unpaired electrons in the a 1 ( σ ), b 1 ( π ) and b 2 ( σ ) orbitals. Four low-energy excited states are predicted, of which the lowest-energy one would need to be (1) 4 B 1 . One-layer models predict that this state undergoes out-of-plane distortion which lowers the energy. The distortion, however, can be either removed or enhanced once multilayer models are considered. This transition has dominant a 1 ( σ ) → b 2 ( σ ) character, polarized in-plane and perpendicular to the defect’s C 2 v axis (see Fig. 5e ), with an oscillator strength exceeding 0.1. Figure 5f compares calculated emission bandshapes E ( ν )/ ν 5 (obtained as the raw emission scaled by wavelength to the fifth power) with experimentally observed spectra from Fig. 3 . The (1) 4 B 1 → (1) 4 A 2 emission is predicted to be slightly lower in energy and much broader. The calculated width is environment dependent (Fig. 5f ) and dominated by how the calculations perceive torsional changes at the defect associated with light emission that generate low-frequency phonons. The observed spectra in region I are indicative of such effects, but their magnitude is reduced to one-third. The observed spectra in region II are very different, primarily manifesting the effects of activation of BN-stretch phonons instead. It could be that the perceived sensitivity of \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) photoemission to the local environment can account for the stark contrast in the observed spectra (Fig. 5f ). The most important shortcoming of the proposal of this defect as the dominant hBN SPE would be that intense absorption is predicted in only one polarization, whereas experiments suggest that higher-energy absorptions exist with alternate polarization 7 , 9 . However, of the 24 defect manifolds considered herein, it is the only one to remain of interest. More complex carbon-cluster defects, including, for instance, C 2 C N and C 2 C V , have been considered as alternatives 44 , 45 . In summary, we have presented rigorous experimental results to confirm the central role of carbon in hBN quantum emitters in the visible spectral range. We compared samples grown by MOVPE, MBE and HOPG conversion. All methods exhibited a direct correlation between the introduction of carbon as a precursor/substance and the formation of SPEs. Furthermore, MOVPE growth enabled us to deterministically control carbon incorporation and vary the density of the quantum emitters from singles to ensembles and observe room-temperature ODMR. We have also generated SPEs using direct ion implantation of carbon and showed that their density scales with the implantation dose. Employing a TD-DFT method, we proposed the negatively charged \({\rm{V}}_{\rm{B}}{\rm{C}}_{\rm{N}}^ -\) as a suitable transition to explain the observed results. Our results will accelerate the deployment of visible quantum emitters in hBN into quantum photonic devices and will advance potential strategies for the controlled engineering of quantum emitters in van der Waals crystals. Methods MOVPE hBN layers were grown on commercially available 2 inch sapphire substrates using MOVPE, as described in ref. 27 . TEB and ammonia were used as the boron and nitrogen precursors, respectively, while hydrogen was the carrier gas. The precursors were introduced into the reactor as short alternating pulses, in order to minimize parasitic reactions between TEB and ammonia. hBN growth was carried out at a reduced pressure of 85 mbar and the growth temperature was set to 1,350 °C. In the present study, the TEB flux was varied from 10 to 60 µmol min −1 to study the effect on carbon incorporation on sub-band-gap luminescence from the hBN films. For ion implantation, PL and SPE measurements, centimetre-sized hBN films were transferred from sapphire on to SiO 2 /Si substrates, using water-assisted self-delamination 27 . The thickness of the hBN films was also measured using atomic force microscopy (AFM), as described in the Supplementary Information . XPS was used to determine the impurity levels in the as-grown MOVPE hBN films, as shown in the Supplementary Information . A gentle etching using an argon beam was performed in situ to remove adventitious carbon and impurities from the surface; all spectra were collected from the bulk of hBN films. MBE BN epilayers were grown using a custom-designed Veeco GENxplor MBE system capable of achieving growth temperatures as high as 1,850 °C under ultra-high-vacuum conditions, on rotating substrates with diameters of up to 3 inches. Details of the MBE growth have been previously published 32 . In all our studies, we relied on thermocouple readings to measure the growth temperature of the substrate. For all samples discussed in the current paper the growth temperature was in the range 1,250–1,390 °C. We used a high-temperature Knudsen effusion cell (Veeco) or electron beam evaporator (Dr Eberl MBE-Komponenten) for evaporation of boron. High-purity (5 N) elemental boron contains the natural mixture of 11 B and 10 B isotopes. To contain boron in the e-beam evaporator we used boron nitride and vitreous carbon crucibles. We used a conventional Veeco RF plasma source to provide the active nitrogen flux. The hBN epilayers were grown using a fixed RF power of 550 W and a N 2 flow rate of 2 sccm. We used 10 × 10 mm 2 (0001) sapphire and on- and 8°-off oriented Si-face SiC substrates. Variable-angle spectroscopic ellipsometry provided the thickness of the hBN layers. HOPG to hBN conversion The conversion takes place in a graphite crucible. A HOPG crystal is placed in the centre of the crucible on a separate graphite holder. Small holes in the stage holding the HOPG allow vapours from the boron–oxide powder, placed at the bottom of the crucible, to flow to the HOPG crystal. A radiofrequency induction furnace is then heated to 2,000 °C, and N 2 gas is introduced as the nitrogen precursor. A central tube mixing the N 2 gas with the boron–oxide vapour premixes the precursors before these reach the HOPG crystal. Further details can be found in ref. 35 . Ion implantation Ion implantation was carried out on 40-nm-thick MOVPE hBN films, grown using a TEB flux of 10 µmol min −1 . For this, the hBN films were first transferred on to SiO 2 /Si substrates. A copper grid with a square mesh (GCu300, ProSciTech) was used as the implantation mask. Carbon, silicon and oxygen were separately implanted into the hBN films. During implantation, the ion energy and fluence were 10 keV and 10 13 ion cm − 2 , respectively. Confocal microscopy The optical measurements were carried out using a confocal microscope equipped with a 532 nm excitation source. The collected signal was sent either to a spectrometer or to an avalanche photodiode for photon statistics measurements. The g (2) ( τ ) measurements were analysed and fitted without background correction unless specified otherwise. More details on the set-up can be found in ref. 28 . ODMR The ODMR spectra were measured with a confocal microscope set-up. A ×100 objective (Olympus MPLN100X) was used to focus a 532 nm laser (LaserQuantum opus 532) onto the sample and collect the PL signal. The PL signal is collected back through a 650 nm short-pass dichroic mirror for separation from scattered laser light. Additionally, 532 and 550 nm long-pass filters were used before the PL was detected by a silicon avalanche photodiode (Thorlabs APD440A) to filter out the laser light. The microwave field was applied through a signal generator-plus-amplifier system (Stanford Research Systems SG384, VectaWave VBA1000-18 Amplifier); the sample was placed on a 0.5-mm-wide copper stripline. In order to detect the ODMR signal (that is, the relative ∆PL/PL contrast) by lock-in technique (Signal Recovery 7230), the microwaves were driven with an on–off modulation. The resonant condition was changed with the external magnetic field by mounting a permanent magnet below the sample. Raman spectroscopy Raman spectroscopy measurements were carried out on an In-Via confocal Raman (Renishaw) system using a 633 nm excitation source. Calibration of the spectrometer was carried out using a Si substrate to 520 cm −1 . The peaks were then fitted to a Lorentzian line profile, from which the corresponding peak centre position and FWHM were extracted. Samples were analysed after transfer to SiO 2 . AFM The thickness of the hBN films transferred on to SiO 2 /Si substrates was measured using tapping mode AFM (ScanAsyst-Air, MultiMode 8, Bruker). The AFM scan was performed at the film boundary to facilitate thickness measurement across the step-edge, as shown in Supplementary Fig. 8 . Computational Many innovative approaches are used in order to model defects with large three-dimensional spatial deformations, as described in detail in the Supplementary Information . The core elements are the use of Gaussian 16 46 to perform TD-DFT calculations within a QM/MM model utilizing CAM-B3LYP and an AMBER h-BN force field fitted to mimic CAM-B3LYP. Spectra are simulated with the Huang–Rhys model based on analytically obtained second derivatives for both ground and excited states. Data availability Source data for most experimental and theoretical data for this work are provided. Confocal maps and wide-field images are available from the corresponding author upon request due to their size. Source data are provided with this paper.

Systems which can emit a stream of single photons, referred to as quantum light sources, are critical hardware components for emerging technologies such as quantum computing, the quantum internet, and quantum communications. In many cases the ability to generate quantum light on-demand requires the manipulation and control of single atoms or molecules, pushing the limit of modern fabrication techniques, and making the development of these systems a cross-disciplinary challenge. In new research, published in Nature Materials, an international multidisciplinary collaboration led by the University of Technology Sydney (UTS), has uncovered the chemical structure behind defects in white graphene (hexagonal boron nitride, hBN), a two dimensional nanomaterial that shows great promise as a platform for generating quantum light. The defects, or crystal imperfections, can act as single photon sources and an understanding of their chemical structure is critical to being able to fabricate them in a controlled way. "hBN single photon emitters display outstanding optical properties, among the best from any solid state material system, however, to make practical use of them we need to understand the nature of the defect and we have finally started to unravel this riddle," says UTS Ph.D. candidate Noah Mendelson and first author of the study. "Unfortunately, we cannot simply combine powerful techniques to visualize single atoms directly with quantum optics measurements, so obtaining this structural information is very challenging. Instead we attacked this problem from a different angle, by controlling the incorporation of dopants, like carbon, into hBN during growth and then directly comparing the optical properties for each, " he said. To realise this comprehensive study, the team, led by Professor Igor Aharonovich, chief investigator of the UTS node of the ARC Centre of Excellence for Transformative Meta-Optical Materials (TMOS), turned to collaborators in Australia and around the world to provide the array of samples needed. The researchers were able to observe, for the first time, a direct link between carbon incorporation into the hBN lattice and quantum emission. "Determining the structure of material defects is an incredibly challenging problem and requires experts from many disciplines. This is not something we could have done within our group alone. Only by teaming up with collaborators from across the world whose expertise lies in different materials growth techniques could we study this issue comprehensively. Working together were we finally able to provide the clarity needed for the research community as a whole," said Professor Aharonovich. "It was particularly exciting as this study was enabled by the new collaborative efforts with collaborators Dipankar Chugh, Hark Hoe Tan and Chennupati Jagadish from the TMOS node at the Australian National University, " he said. The scientists also identified another intriguing feature in their study, that the defects carry spin, a fundamental quantum mechanical property, and a key element to encode and retrieve quantum information stored on single photons. "Confirming these defects carry spin opens up exciting possibilities for future quantum sensing applications, specifically with atomically thin materials." Professor Aharonovich said. The work brings to the forefront a novel research field, 2-D quantum spintronics, and lays the groundwork for further studies into quantum light emission from hBN. The authors anticipate their work will stimulate increased interest in the field and facilitate a range of follow up experiments such as the generation of entangled photon pairs from hBN, detailed studies of the spin properties of the system, and theoretical confirmation of the defect structure. "This is just the beginning, and we anticipate our findings will accelerate the deployment of hBN quantum emitters for a range of emerging technologies," concludes Mr. Mendelson.

10.1038/s41563-020-00850-y

Medicine

Study reveals commercial tool used for epigenome studies is not actually appropriate for population epigenetics

Chathura J. Gunasekara et al, Systemic interindividual epigenetic variation in humans is associated with transposable elements and under strong genetic control, Genome Biology (2023). DOI: 10.1186/s13059-022-02827-3 Journal information: Genome Biology

https://dx.doi.org/10.1186/s13059-022-02827-3

https://medicalxpress.com/news/2023-01-reveals-commercial-tool-epigenome-population.html

Abstract Background Genetic variants can modulate phenotypic outcomes via epigenetic intermediates, for example at methylation quantitative trait loci (mQTL). We present the first large-scale assessment of mQTL at human genomic regions selected for interindividual variation in CpG methylation, which we call correlated regions of systemic interindividual variation (CoRSIVs). These can be assayed in blood DNA and do not reflect interindividual variation in cellular composition. Results We use target-capture bisulfite sequencing to assess DNA methylation at 4086 CoRSIVs in multiple tissues from each of 188 donors in the NIH Gene-Tissue Expression (GTEx) program. At CoRSIVs, DNA methylation in peripheral blood correlates with methylation and gene expression in internal organs. We also discover unprecedented mQTL at these regions. Genetic influences on CoRSIV methylation are extremely strong (median R 2 =0.76), cumulatively comprising over 70-fold more human mQTL than detected in the most powerful previous study. Moreover, mQTL beta coefficients at CoRSIVs are highly skewed (i.e., the major allele predicts higher methylation). Both surprising findings are independently validated in a cohort of 47 non-GTEx individuals. Genomic regions flanking CoRSIVs show long-range enrichments for LINE-1 and LTR transposable elements; the skewed beta coefficients may therefore reflect evolutionary selection of genetic variants that promote their methylation and silencing. Analyses of GWAS summary statistics show that mQTL polymorphisms at CoRSIVs are associated with metabolic and other classes of disease. Conclusions A focus on systemic interindividual epigenetic variants, clearly enhanced in mQTL content, should likewise benefit studies attempting to link human epigenetic variation to the risk of disease. Introduction Genome-wide association studies (GWAS) have revolutionized the field of genetics by identifying genetic variants associated with a range of diseases and phenotypes [ 1 , 2 , 3 ]. Nearly 20 years into the GWAS era, however, most human disease risk and phenotypic variation remain unexplained by common genetic variants [ 2 ], fueling interest in the possibility that individual epigenetic variation is an important determinant of phenotype [ 4 , 5 ]. To test this, over the last decade myriad studies have performed genome-scale screens to identify genomic regions at which epigenetic variation is associated with disease. Nearly all these epigenome-wide association studies (EWAS) used commercial arrays manufactured by Illumina (predominantly the HM450 and subsequently the scaled-up EPIC850 array) to assess methylation at CpG dinucleotides (a highly stable epigenetic mark) in peripheral blood DNA [ 6 , 7 ]. EWAS have uncovered associations between blood DNA methylation and neurological outcomes including Alzheimer’s disease [ 8 ], neurodegenerative disorders [ 9 ], educational attainment [ 10 ], and psychiatric diseases [ 11 ]. The HM450 and EPIC arrays were instrumental in discoveries in epigenetic aging [ 12 , 13 , 14 ], smoking-induced DNA methylation alterations [ 15 ], and understanding how maternal smoking [ 16 ] and alcohol consumption [ 17 ] affect DNA methylation in newborns. Peripheral blood DNA methylation has been associated with birthweight [ 18 ] and body mass index [ 19 ]. The Illumina methylation arrays have also played a central role in advancing our understanding of genetic influences on CpG methylation. Genetic variants that correlate with methylation at a specific CpG site (usually in cis) are known as methylation quantitative trait loci (mQTL). Seminal observations of familial clustering of CpG methylation levels [ 20 ] led to the first formal study of mQTL [ 21 ], which utilized an early version of the Illumina methylation platform. Now, hundreds of studies, nearly all using Illumina methylation arrays, have investigated mQTL in humans [ 22 ], enabling estimates of methylation heritability and insights into how genetic effects on disease risk may be mediated by DNA methylation [ 23 ] and mechanisms of trans (inter-chromosomal) mQTL effects [ 24 ]. Despite these successes, existing and legacy Illumina methylation platforms are not ideal for population epigenetics. The success of GWAS was built upon the HapMap [ 25 ] and 1,000 Genomes [ 26 ] projects, which systematically mapped out human genome sequence variants so they could be assessed at the population level. So far, however, no “EpiHapMap” project has been conducted. Several large consortium projects, including the Roadmap Epigenome Project [ 27 ], the Blueprint Epigenome Project [ 28 ], and the International Human Epigenome Consortium [ 29 ], focused primarily on characterizing tissue- and cell type-specific epigenetic variation rather than mapping out human genomic regions of interindividual epigenetic variation. The EWAS field therefore relied almost exclusively on Illumina arrays [ 30 ] which were designed without consideration of interindividual variation in DNA methylation [ 31 , 32 ] and generally target CpGs that show little [ 33 , 34 , 35 , 36 ]. To address this lacuna, we recently conducted an unbiased screen for correlated regions of systemic (i.e., not tissue-specific) interindividual epigenetic variation (CoRSIVs) in the human genome [ 37 ]. Because that screen was based on only ten individuals, we set out to assess these regions in a larger cohort to characterize associations among interindividual genetic, epigenetic, and transcriptional variation. In addition to validating CoRSIVs as systemic epigenetic variants, assessing correlations with gene expression, and characterizing associations with transposable elements, we discovered that CoRSIVs exhibit much stronger mQTL than previously observed. Because interindividual variation is essential not just for mQTL detection but also for epigenetic epidemiology, our results have important implications for the EWAS field. Results Target-capture bisulfite sequencing confirms systemic interindividual variation in DNA methylation In collaboration with the NIH Genotype-Tissue Expression (GTEx) program [ 38 ], we conducted target-capture bisulfite sequencing to quantify DNA methylation at 4641 gene-associated CoRSIVs in multiple tissues representing the three embryonic germ layers from each of 188 GTEx donors (807 samples total) (Fig. 1 A, B). For donor and sample information and regions targeted, see Additional file 2 : Table S1 and S2, respectively. The raw data have been deposited in a controlled-access public repository (dbGaP accession phs001746.v2.p1) linked to GTEx identifiers. We achieved high capture efficiency (Additional file 1 : Fig. S1A, B, C); over 90% of targeted regions were covered at 30x sequencing depth in nearly all 807 samples (Fig. 1 C, D, Additional file 1 : Fig. S1B). Data on read counts, alignment efficiency, bisulfite conversion efficiency, and duplication rate are provided (Additional file 2 : Table S3). A small subset of difficult-to-capture regions failed to meet coverage criteria in all libraries (Additional file 1 : Fig. S1C, Additional file 2 : Table S4). A set of Y-chromosome regions included in the capture enabled us to confirm that all 807 samples are of the correct sex (Additional file 1 : Fig. S1D), indicating reliable sample handling. Fig. 1 Target-capture bisulfite sequencing in 807 GTEx samples confirms systemic interindividual epigenetic variation at CoRSIVs. A DNA samples were obtained from multiple tissues (representing the three embryonic germ layers) from each of 188 GTEx donors. B CoRSIV capture process using Agilent reagents. C Percentage of CoRSIVs for which target-capture bisulfite sequencing achieved various read depths; each point represents one of 807 samples. D Plots of read depth at two target regions illustrate specificity of targeting across all six tissues. The Y -axis scales are same for each region and indicated for thyroid. E Scatter plots between all possible tissue pairs illustrate high inter-tissue correlations at a CoRSIV within HPCAL1 . F Heat map of inter-tissue correlations across 4086 CoRSIVs shows generally high correlation coefficients between all possible tissue pairs. G For the 232 tissue samples from 53 donors with data on at least 4 tissues (excluding cerebellum), unsupervised hierarchical clustering of methylation data at 2349 fully informative CoRSIVs groups perfectly by donor Full size image CoRSIVs were identified based on unbiased genome-wide assessment of DNA methylation in thyroid, heart, and brain [ 37 ]. Our first goal, therefore, was to examine additional tissues to confirm systemic interindividual variation (SIV) at these regions. High inter-tissue correlation in DNA methylation is the hallmark of SIV (Fig. 1 E). Of the 4641 genic CoRSIVs targeted, the 4086 that satisfied coverage criteria in at least 10 donors in every possible pair of tissues were evaluated. Most of these showed high positive inter-tissue correlations (Pearson R >0.6) across all possible tissue pairs (Fig. 1 F, Additional file 1 : Fig. S1E, Additional file 2 : Table S5), confirming SIV. Accordingly, unsupervised clustering of methylation data at the 2349 CoRSIVs covered in all 5 tissues (except cerebellum) across 53 donors grouped perfectly by the donor (Fig. 1 G, Additional file 2 : Table S6). This clustering was not associated with sample-level variation in capture efficiency (Additional file 2 : Table S7). As DNA methylation in the cerebellum often differs from that in other brain regions [ 39 ], including cerebellum in this analysis resulted in a minor cerebellum cluster (Additional file 1 : Fig. S1F); nonetheless, high inter-tissue correlations were maintained (Additional file 1 : Fig. S1G). Of greatest relevance to epigenetic epidemiology, CoRSIV-specific scatter plots of methylation in brain, thyroid, skin, lung, and nerve versus that in blood show that methylation in blood generally serves as a proxy for methylation in other tissues ( five tissues vs. blood ). By comparison, in an HM450 study of 122 individuals [ 39 ], correlations between methylation in 4 brain regions vs. blood averaged only 0.2 and rarely exceeded 0.5. Although the inter-tissue scatter plots at CoRSIVs commonly show either a uniform distribution or three clusters (suggesting a single-genotype effect) (Additional file 1 : Fig. S2), other patterns observed include 2, 4, and 5 distinct clusters (Additional file 1 : Fig. S3). Consistent with our earlier study [ 37 ], in all six tissues almost every CoRSIV displayed an interindividual methylation range >20% (median range 40–42%) (Additional file 1 : Fig. S4). Together, these results validate these CoRSIVs as systemic individual variants, essentially epigenetic polymorphisms. Gene expression in internal organs correlates with CoRSIV methylation in blood Compared to genetic epidemiology, epigenetic epidemiology is complicated by the inherent tissue-specificity of epigenetic regulation [ 5 ]. Because nearly all EWAS are based on measuring methylation in peripheral blood DNA, attempts to discover associations with, for example, Alzheimer’s disease [ 9 ] or schizophrenia [ 40 ] are implicitly predicated on the assumption that methylation variants in blood associate with epigenetic regulation in the brain. Of those on the Illumina arrays, however, such probes are the exception [ 39 , 41 ]. We therefore used our target capture bisulfite sequencing data and transcriptional profiling (RNA-seq) data from GTEx to test for cross-tissue correlations between CoRSIV methylation and expression of associated genes. Of 3768 CoRSIV-associated genes, over half showed appreciable expression in at least 5 of the six tissues under consideration (Additional file 1 : Fig. S5A, B). Tibial nerve was excluded from this analysis due to low sample size; for each other tissue, both CoRSIV methylation and gene expression data were available for at least 60 individuals (Additional file 1 : Fig. S5C). Tissues that are difficult to sample non-invasively (thyroid, lung, and cerebellum) were considered “target” tissues. Within each of these, we identified all CoRSIV-gene pairs for which gene expression is associated with CoRSIV methylation (FDR<0.05) (Additional file 1 : Fig. S6A, B show two examples). Relative to those within a gene body, CoRSIVs located within 3 kb of either the 5′ or 3′ end of a gene showed predominantly negative correlations between methylation and gene expression (OR=2.84, P = 0.002) (Additional file 1 : Fig. S6C). For each CoRSIV-gene pair showing an expression vs. methylation association in a target tissue, we next asked whether methylation measured in easily accessible “surrogate” tissues (blood or skin) is associated with expression in the target tissue. Of 156 genes for which expression was correlated with CoRSIV methylation in the thyroid, for example, 122 (75%) showed a significant correlation and in the same direction when methylation in blood was used as the independent variable (Additional file 1 : Fig. S6D). Likewise, in the lung and cerebellum, at least 75% of all methylation-expression correlations were detected when methylation in blood was used to infer expression (Additional file 1 : Fig. S6D). In the other surrogate tissue, skin, this figure was slightly lower (60%). These data demonstrate that, at gene-associated CoRSIVs, methylation measurements in easily accessible tissues like blood can be used to draw inferences about epigenetic regulation in internal organs, a major advantage for epigenetic epidemiology. Genetic influences on methylation at genic CoRSIVs are strong and biased The Genetics of DNA Methylation Consortium (GoDMC) recently analyzed HM450 and genotyping data on nearly 33,000 people in 36 cohorts [ 42 ] and documented mostly modest effects; for 75% of the cis mQTL associations, the genetic variant explained less than 5% of the variance in methylation. In the largest unbiased study of human mQTL, Busche et al. [ 43 ] performed whole-genome bisulfite sequencing in 43 female twins and concluded that environment, not genetics, is the main source of interindividual variation in DNA methylation. We wondered to what extent individual variation in CoRSIV methylation is explained by genetic variation in cis . Within each CoRSIV, methylation of multiple CpGs is highly correlated [ 37 ]; we therefore tested for genetic associations with average CoRSIV methylation, rather than at the CpG level. Also, given the multiplicity of mQTL associations at each CoRSIV (median 22 SNVs with P <10 −10 per CoRSIV, Additional file 1 : Fig. S7), rather than attempt to detect all possible SNV-CoRSIV associations, we employed the Simes correction [ 44 ] to identify the single SNV most strongly associated with methylation at each CoRSIV (lowest p value, adjusted for multiple testing) (Fig. 2 A, B, Additional file 1 : Fig. S8, Additional file 2 : Table S8; listed p values are adjusted for multiple testing). This approach conservatively tests each CoRSIV for evidence of genetic influence on its methylation and is much more powerful than those we were able to employ in our earlier study [ 37 ] based on just 10 individuals. Fig. 2 Genetic influences on CoRSIV methylation are strong and biased. A , B Representative plots of mQTL associations at individual CoRSIVs on chromosomes 1 and 2, respectively. Significant associations are shown for all SNVs within 1Mb of each CoRSIV; positive and negative beta coefficients are plotted in blue and red, respectively. The most significant SNV (Simes SNV) is circled. Insets show average CoRSIV methylation vs. Simes SNV genotype. C Distribution of distances between CoRSIVs and corresponding Simes SNVs. D For each of 4086 CoRSIVs, heat map depicts the number of tissues in which the Simes SNV falls within the same haplotype block, illustrating the largely systemic nature of mQTL at CoRSIVs. E Distribution of beta coefficients of significant Simes mQTL associations for the GoDMC blood mQTL data [ 42 ]. F Distribution of beta coefficients of significant Simes mQTL associations at 3723 CoRSIVs in blood DNA from 188 GTEx donors. G Distribution of beta coefficients of significant Simes mQTL associations across 2939 CoRSIVs in blood DNA from 47 newborns (USC). H Distribution of Simes mQTL R 2 (goodness of fit) for the GoDMC data. I Distribution of Simes mQTL R 2 at CoRSIVs (GTEx, blood). J Distribution of Simes mQTL R 2 at CoRSIVs (USC samples) Full size image Although we tested all SNVs within 1 Mb, “Simes SNVs” were generally proximal to the CoRSIV, 72% within 10 kb (Fig. 2 C, Additional file 1 : Fig. S9). Remarkably, although the Simes procedure was carried out independently in each tissue, at each CoRSIV the exact same SNV in many cases yielded the strongest mQTL association in all or most of the tissues (Additional file 1 : Fig. S10A, B). When we asked how often the Simes SNV was within the same haplotype block in all or most tissues, concordance was even stronger (Fig. 2 D), indicating the systemic nature of genetic influences on methylation at genic CoRSIVs. Previous studies of mQTL using the HM450 array [ 22 , 42 ] consistently report beta coefficients balanced on both sides of zero, as we found by employing the Simes procedure to the GoDMC data (Fig. 2 E). Conversely, most cis mQTL associations at genic CoRSIVs show a negative beta coefficient (i.e., the major allele is associated with higher methylation) (Fig. 2 F). This imbalance held not just for Simes SNVs, but for all mQTL SNVs (Additional file 1 : Fig. S11). The strength of mQTL associations at genic CoRSIVs also appears to be without precedent [ 22 , 42 ]. In the GoDMC data, for example, few Simes mQTL associations show an R 2 > 0.2 (Fig. 2 H); at CoRSIVs, the median R 2 = 0.76 (Fig. 2 I, Additional file 1 : Fig. S12). This tendency for high- R 2 mQTL was largely independent of the distance between CoRSIV and SNV (Additional file 1 : Fig. S13). We made several attempts to disprove these surprising findings. Though unlikely (because each CoRSIV contains at least 5 CpGs [ 37 ]), we first asked whether the strong mQTL effects could be caused by SNVs abrogating CpG sites within CoRSIVs. Of SNVs present in our sample of 188 individuals, at least one did overlap a CpG within most of the CoRSIVs we surveyed. The distributions of beta coefficient and R 2 values of Simes mQTL associations for the 1155 CoRSIVs without any such overlaps, however, were nearly identical to those of the 2759 with SNV-CpG overlaps (Additional file 1 : Fig. S14). We next asked whether, instead of affecting CpG sites, SNVs within CoRSIVs might introduce an artifact by compromising the binding of the baits used for target capture. Despite their small size (median 200 bp), most CoRSIVs contain 2 or more SNVs (Additional file 1 : Fig. S15A); however, neither the beta coefficients nor the R 2 values of the Simes mQTL associations were strongly associated with the number of SNVs per CoRSIV (Additional file 1 : Fig. S15B, C). Together, these data indicate that the strong and biased mQTL effects we detected are not due to SNVs within CoRSIVs. For a complementary analysis, we employed a haplotype-based approach to assess genetic influences on CoRSIV methylation. We used phased genotype data from GTEx to infer each individual’s haplotype within the haplotype block overlapping each CoRSIV and assessed correlations between CoRSIV methylation and haplotype allele sum (sum of minor alleles in each individual) (Additional file 1 : Fig. S16A). This analysis yielded a preponderance of negative coefficients, and local haplotype explained much of the variance in methylation (median R 2 = 0.43) (Additional file 1 : Fig. S16B, Additional file 2 : Table S9), consistent with the mQTL analysis. Lastly, to independently validate genetic effects on CoRSIV methylation, we performed CoRSIV-capture bisulfite-sequencing and SNV genotyping in 47 individuals from a different (non-GTEx) population (USC cohort). To ensure computational independence, a separate member of our laboratory wrote new code for the Simes mQTL analysis. The USC results corroborated the negative bias and high R 2 of mQTL effects at CoRSIVs (Fig. 2 G, J, Additional file 2 : Table S10). An independently performed haplotype-based analysis likewise corroborated the results obtained on the GTEx samples (Additional file 1 : Fig. S16C, Additional file 2 : Table S11). Together, these additional analyses and data indicate that the strong and biased genetic influences on methylation at CoRSIVs are genuine. We wondered how the total amount of mQTL we detected at genic CoRSIVs compares with that reported by the GoDMC [ 42 ], which used HM450 arrays to study 33,000 people. With 3 genotype calls possible at each SNV, the average methylation difference (delta) associated with each SNV can be calculated from the mQTL beta coefficient (Additional file 1 : Fig. S17A). And, since the mQTL R 2 measures what proportion of this delta is explained by SNV genotype, the product (delta)x( R 2 ) measures the absolute methylation variation explained by SNV genotype. To make our results interpretable, we initially assessed (delta)x( R 2 ) based on beta values (rather than using the M-value transformation). Across all CoRSIV mQTLs ( P < 10 −10 ), median (delta)x( R 2 ) was 24.6% methylation (Additional file 1 : Fig. S17B); for a CoRSIV with an R 2 near the median (0.76), this equates to an interindividual range of 32.4% methylation, within the normal range for CoRSIVs (Additional file 1 : Fig. S4). To compare our results with those of GoDMC [ 42 ], whose coefficients were provided based on M values, we repeated our analysis after applying the M value transformation. At the CoRSIVs we assayed, the total methylation variance explained by genetics (sum of (delta)x( R 2 )) was 72-fold greater than that detected by GoDMC [ 42 ] (Additional file 1 : Fig. S17C, D, E), the largest study of human mQTL ever reported. Genetic influences on tissue-specific expression (eQTL) can be mediated by mQTL [ 23 , 45 ]. Given the strong mQTL effects at genic CoRSIVs, we used data from GTEx [ 46 ] to ask whether Simes SNVs are enriched for eQTL. Consistent with the analysis of GTEx data overall [ 46 ], many eQTL effects were shared among non-brain tissues, whereas eQTL associations in the brain and blood were more distinct (Additional file 1 : Fig. S18A). Relative to all common variants, which have a 50% chance of being associated with expression of a nearby gene [ 46 ], a bootstrapping analysis indicated that Simes SNVs are 3.4-fold more likely to show eQTL effects (Additional file 1 : Fig. S18B). The distributions of magnitude, slope, and SNV-eGene distance for eQTL effects at Simes SNVs were similar to those of all common variants (Additional file 1 : Fig. S18C, D). Future studies will be required to determine if the enriched eQTL at Simes SNVs is in some cases mediated by CoRSIV mQTL. CoRSIVs occur in genomic regions with far-reaching enrichments in transposable elements The earliest known examples of systemic interindividual epigenetic variants in mammals are mouse metastable epialleles such as agouti viable yellow and axin fused , both of which resulted from retrotransposition of an intracisternal-A particle (an LTR-retrotransposon) [ 47 , 48 ]. We previously showed that CoRSIVs are enriched for direct overlaps with LINE, SINE, and ERV retrotransposons [ 37 ]; we provide a more granular analysis of those overlaps here (Additional file 1 : Fig. S19). Given the ability of transposable elements for long-range regulation of transcriptional and epigenetic dynamics in the early embryo [ 49 , 50 ], we asked whether the exceptional behavior of CoRSIVs might be associated with specific classes of repetitive elements working over long genomic distances. Relative to a set of control regions matched to genic CoRSIVs by chromosome, size, and CpG density [ 37 ], in regions flanking genic CoRSIVs we detected long-range depletion of CpG islands and enrichments of specific classes of LINE and LTR retrotransposons (Fig. 3 A, Additional file 2 : Table S12). Similar and stronger enrichments were detected in comparison with size-matched tissue-differentially methylated regions (tDMRs) [ 37 ] (Additional file 1 : Fig. S20). Interestingly, enrichments relative to control regions (Fig. 3 A) were strongest among the evolutionarily youngest subclasses, the LINE1-PA elements [ 51 ] among LINEs, and ERV-K elements [ 50 ] among LTRs. Fig. 3 Genic CoRSIV-flanking regions show long-range enrichments and depletions for specific classes of transposable elements. A Using 1 kb step sizes, each plot shows significant enrichments or depletions for CpG islands (CGI) and subclasses within each of 8 classes of transposable element within 50 kb of genic CoRSIVs. Compared to control regions, CoRSIV-flanking regions show long-range depletion of CpG islands and enrichment of specific classes of LINEs and LTRs. B Compared to CoRSIVs showing a positive mQTL beta coefficient, those with negative coefficients are depleted for CpG islands and show long-range depletion of specific LINE1s and all subclasses of Alus. C The strength of mQTL associations at CoRSIVs (R 2 in 4 th vs. 1 st quartile) is not associated with widespread differences in genomic content of transposable elements. D Compared to regions in which HM450 probes are located, CoRSIVs show short- and long-range enrichments for many subclasses of LINE1 and LTR retrotransposons Full size image We next asked whether either the negative bias (i.e., the major allele associating with higher methylation) or the strength of mQTL associations at CoRSIVs might be associated with transposable elements in flanking genomic regions. Compared to genic CoRSIVs showing a positive mQTL beta coefficient, those characterized by negative coefficients were depleted for CpG islands (Fig. 3 B). There were no robust short-range associations of transposable elements with “negative mQTL” CoRSIVs; rather, at distances > 5–10kb from the origin, they show extensive long-range depletion of specific LINE1 and all classes of Alu elements (Fig. 3 B, Additional file 2 : Table S13). Surprisingly, the strength of mQTL at genic CoRSIVs was not associated with widespread differences in genomic content of transposable elements. Relative to those in the bottom quartile for R 2 , mQTL effects in the top quartile showed proximal and long-range depletion in just CpG islands and G-rich low-complexity repeats (Fig. 3 C, Additional file 2 : Table S14). As most human mQTL data are based on the HM450 array, we next compared genomic regions flanking genic CoRSIVs with those flanking genic HM450 probes, finding striking differences. Although the HM450 array specifically targets CpG islands, these are more strongly enriched within 1 kb of genic CoRSIVs (Fig. 3 D, Additional file 2 : Table S15); at greater distances, CoRSIV-flanking regions are relatively depleted of CpG islands. Compared to genomic regions containing genic HM450 probes, those housing genic CoRSIVs show strong short-range (1–2 kb) enrichments in LINE1, LTR, and Alu elements (Fig. 3 D). The LINE1 and LTR enrichments gradually weaken but extend to at least 50 kb from the origin. Enrichments for Alu extend only to ~5 kb; at greater distances, regions flanking genic CoRSIVs are relatively depleted (Fig. 3 D). These enrichments were not unique to genic CoRSIVs; the full set of 9926 CoRSIVs showed similar patterns of enrichment relative to matched control regions, tDMRs, and HM450 probes (Additional file 1 : Fig. S21). These observations suggest a straightforward explanation for the strong and biased mQTL effects at CoRSIVs. To limit hybridization artifacts, the Illumina methylation arrays avoided genomic regions rich in transposable elements. But these are the same regions in which SIV tends to occur. Given the potentially deleterious consequences of transcriptional activation of retrotransposons, the strong and negative mQTL beta coefficients at CoRSIVs could reflect evolutionary selection for genetic variants favoring their methylation and silencing. In support of this, values of Tajima’s D (a test statistic assessing evidence of evolutionary selection) are higher in CoRSIVs compared to control, tDMR, or HM450 probe regions (Additional file 1 : Fig. S22, Additional file 2 : Table S16). CoRSIV flanking regions are enriched for heritability of disease Across diverse outcomes including Alzheimer’s [ 23 ], chronic obstructive pulmonary disease [ 52 ], obsessive-compulsive disorder [ 53 ], and cardiovascular disease [ 54 ], integrative analyses of GWAS and DNA methylation profiling data increasingly indicate that mQTL mediates associations between genetic variation and risk of disease. We therefore asked whether the strong mQTL effects identified at genic CoRSIVs are associated with genetic variants identified by GWAS. Indeed, permutation testing indicates that SNVs identified in our mQTL analysis are enriched for SNVs implicated in metabolic, hematological, anthropometric, cardiovascular, immune, neurological, and various other diseases (Fig. 4 A, B, Additional file 2 : Table S17). By contrast, despite an abundance of CoRSIV-associated genes linked to cancer [ 37 ], no enrichment was found relative to cancer GWAS SNVs (Fig. 4 A, B). Notably, a recent HM450 analysis employing these same categories [ 24 ] found nearly opposite categorical enrichments with trans -mQTL loci. With the caveat that 90% of GWAS alleles impact multiple traits [ 55 ], it is interesting that cancer traits are not enriched. This may indicate that CoRSIV methylation plays no role in this maladaptive phenotype, or reflect dilution of effects across multiple cancer subtypes and various genetic pathways leading to cancer [ 56 ]. Overall, and particularly considering that Simes SNVs are enriched for eQTL, these results are consistent with the possibility that human genetic variants influence disease risk via mQTL effects at CoRSIVs. Fig. 4 CoRSIV mQTL SNVs are enriched for GWAS associations. A Within each of 8 disease/phenotype categories, the histogram shows the null distribution obtained by permutation testing for overlap of GWAS SNVs with SNVs randomly sampled within 1Mb of each CoRSIV. The red diamond shows the actual number of overlaps between CoRSIV mQTL SNVs and GWAS SNVs. Numbers of GWAS SNVs considered in each category are anthropometric: 8106, cancer: 3163, cardiovascular: 4816, hematological: 7461, immune: 5263, metabolic: 10,121, neurological: 14,741, and various: 14,573. B Statistical significance (Bonferroni-adjusted p -value) vs. fold enrichments for the analysis in A . Strong and statistically significant enrichments were found for all outcomes except cancer. C Statistical significance (Bonferroni-adjusted p -value) vs. fold enrichments for 8 metabolic traits and 4 cancer outcomes from the LDSC analysis confirms that the vicinity of CoRSIVs is enriched for heritability of metabolic traits Full size image As a complementary analysis, we used LD score regression (LDSC) [ 57 ] to determine if, in the vicinity of genic CoRSIVs, there is enrichment for heritability of metabolic phenotypes and cancer. GWAS summary statistics from the UK Biobank representing 12 metabolic traits and 4 cancer outcomes were downloaded [ 58 ]. As nearly all Simes SNVs are within 20 kb of their associated CoRSIV (Fig. 2 C), we evaluated genomic regions encompassing genic CoRSIVs ± 20 kb. Consistent with our results based on direct overlap with Simes SNVs, individual LDSC models focused on each outcome detected significant enrichment for 3 metabolic outcomes (HbA1c, HDL cholesterol, and glucose) but none for cancer (Fig. 4 C). As suggested by Finucane et al. [ 57 ], we repeated these analyses including in each a full “baseline” model comprising 53 sequence and epigenomic features. Enrichment for heritability of two of the metabolic traits, HbA1c and HDL cholesterol, was attenuated but remained significant (Additional file 1 : Fig. S23A). The baseline-adjusted analysis (Additional file 1 : Fig. S23B) confirmed strong evolutionary conservation in the vicinity of genic CoRSIVs. Also, significant enrichments for coding regions and transcription start sites may explain the attenuated associations with metabolic outcomes. Regardless, we would argue that because CoRSIVs were identified based solely on SIV in DNA methylation, it is inappropriate to penalize them for association with genic and regulatory features. Hence, the LDSC results corroborate that CoRSIV-flanking regions are enriched for heritability of metabolic disease. Discussion Following up on our previous screen for human CoRSIVs [ 37 ], here we have, for the first time, demonstrated the feasibility of studying these regions at the population level using target-capture bisulfite sequencing. Performing these analyses on donors from GTEx allowed us to integrate our methylation data with genome sequence and gene expression data on these same individuals. As expected, our results validated SIV at the CoRSIVs we analyzed and indicate the ability to use methylation profiling in peripheral blood to draw inferences about epigenetic regulation in various organs of the body. More surprisingly, our analyses of genetic influences on CoRSIV methylation indicate an unprecedented level of mQTL at these regions. Also unlike previous reports, our mQTL analysis showed strongly biased beta coefficients (i.e., the major allele associated with higher methylation). Lastly, we found evidence that genomic regions encompassing CoRSIVs are enriched for the heritability of human disease traits. Though unprecedented, the extremely strong mQTL effects at the CoRSIVs we surveyed are unsurprising. Because variation at each SNV is fixed (ranging from 0 to 2 copies of the minor allele), the best way to increase the power of mQTL detection is to focus on CpG sites with the greatest interindividual range of DNA methylation. Other than our work [ 37 , 59 , 60 ], we are not aware of previous studies that took this approach. Instead, nearly all investigations of human mQTL have employed Illumina arrays [ 22 ], which do not target interindividual variants. One may question the validity of quantitatively comparing our mQTL results with those of GoDMC [ 42 ]. After all, GoDMC analyzed HM450 data on 420,000 CpG sites across nearly 33,000 individuals, whereas we analyzed target-capture bisulfite sequencing data on 4086 CoRSIVs in just 188 individuals. But although the targeted regions and studied populations differ, both analyses employed the same statistical method for mQTL detection. Because GoDMC performed their mQTL analyses using M values (a transformation of the Beta value intended to improve normality), we also transformed our percent methylation data to M values for this comparison. Therefore, despite the different approaches and vastly dissimilar numbers of subjects considered, our analysis is quantitatively comparable to that of Min et al. [ 42 ]. Our ability to detect more mQTL than ever before despite surveying a much smaller number of CpG sites speaks to the importance of targeting the right CpGs. Known human CoRSIVs comprise just 0.1% of the genome; whilst some may question the wisdom of focusing on such a small fraction of genomic CpG sites, common human sequence variants comprise only ~0.3% of the genome [ 26 ] but have been a major focus of the GWAS field for the last 20 years. In addition to the extremely strong mQTL effects at genic CoRSIVs, we are not aware of previous studies showing a bias in mQTL regression coefficients (Fig. 2 F, G). The mQTL bias at genic CoRSIVs reflects that the major allele is generally associated with higher methylation. This is consistent with the enrichment of L1 and LTR transposable elements in the vicinity of CoRSIVs (Fig. 3 ), because these tend to locate in heterochromatic regions [ 61 ]. During human pre-implantation development, when methylation at CoRSIVs is thought to be established [ 37 , 62 ], widespread genomic de-methylation leads to transient transcriptional activation of transposable elements, prior to their re-methylation and silencing in differentiated tissues [ 63 ]. The high density of L1 and LTR retrotransposons in CoRSIV-flanking regions therefore raises the question of whether mQTL effects at CoRSIVs reflect modulation of the establishment of de novo or early embryonic maintenance of existing zygotic methylation. In this regard, it is striking that, in mice, L1 elements and IAPs (a class of LTR retrotransposons) are preferentially methylated in sperm and not oocytes, whereas Alus show the opposite pattern (methylated in oocytes but not in sperm) [ 64 ]. These observations mirror our data on transposable element enrichments in regions flanking CoRSIVs (Fig. 3 A). The biased mQTL beta coefficients at CoRSIVs lead us to speculate that they could reflect evolutionary selection for genetic variants that maintain methylation marks in the paternal genome, potentiating transgenerational epigenetic inheritance as observed at the murine metastable epiallele axin fused [ 65 ]. As DNA methylation can act as an intermediary molecular mechanism linking genetic variation to tissue-specific transcriptional regulation [ 23 , 45 ], mQTLs may provide mechanistic insights into how genetic variants influence gene expression. In this regard, the dramatically different nature of mQTL effects at genic CoRSIVs, in terms of both strength and allelic bias, indicates that we have uncovered a fundamentally different component of epigenetic regulation compared to CpGs represented on the HM450 and EPIC arrays which have largely been the focus of the field [ 22 ]. Also, our observation that SNVs wielding the strongest mQTL effects at genic CoRSIVs are enriched for eQTL suggests a mechanistic pathway in which genetic effects on CoRSIV methylation modulate tissue-specific gene expression. On the other hand, 16% of CoRSIVs showed weak effects explaining less than half of the interindividual variation (Fig. 2 I). These are candidate metastable epialleles. Future large human studies can better characterize genetic effects on CoRSIV methylation and elucidate true epipolymorphisms (i.e., metastable epialleles) at which a majority of interindividual epigenetic variation is unexplained by genetics, such as the non-coding RNA nc886 (also known as VTRNA2-1 ) [ 17 , 66 ]. Combining data on such regions with those on recently identified murine metastable epialleles [ 67 ] may enable comparative genomic approaches to characterize sequence features that confer epigenetic metastability, informing in silico identification of metastable epialleles in other mammalian species. Many important questions remain unanswered by our study. Our initial identification of CoRSIVs was based on ten White, non-Hispanic individuals. Reflecting the GTEx study overall, 90% of the donors included in this current study are also White, non-Hispanic. Although our previous studies [ 37 , 59 , 60 ] indicate that SIV regions identified in White, non-Hispanics generally also show SIV in other ancestry groups, future studies screening for SIV directly in other populations may identify CoRSIVs specific to those ancestry groups. Also consistent with the GTEx study population overall, most donors studied here were between 50 and 70 years old (Additional file 2 : Table S1). Considering the influence of age on epigenetic marks [ 12 ], one might ask to what extent interindividual variation at CoRSIVs is influenced by age. Notably, the validation studies we performed to corroborate mQTL effects at CoRSIVs (Fig. 2 G, J) were based on peripheral blood of newborns yet showed nearly identical profiles of mQTL slope and variance explained, arguing that age is not a major factor in the regulation of systemic interindividual epigenetic variation. Compared to our initial screen which surveyed thyroid, heart, and cerebellum, here we evaluated SIV in 4 additional tissues, with at least one representing each germ layer lineage (Fig. 1 A). Hence, whereas our results confirm high inter-tissue correlation coefficients across most tissue pairs for ~90% of genic CoRSIVs (Fig. 1 F), many more tissues and cell types remain to be evaluated. The small fraction of genic CoRSIVs with low inter-tissue correlations (Fig. 1 F) may reflect false positives in our original screen, or possibly exhibit interindividual variation across specific tissue lineages not evaluated here. The generally strong mQTL at CoRSIVs is not necessarily due to the systemic nature of their interindividual variation. Most of these same regions would have been detected if, instead of our original three-tissue screen [ 37 ], we had conducted an unbiased genome-wide screen for interindividual variation in, say, peripheral blood leukocytes. In addition to CoRSIVs, such an experiment would detect interindividual variants specific to blood. Rather than interindividual variation intrinsic to leukocytes, however, many of these reflect interindividual variation in leukocyte composition (ratio of B cells to T cells, for example) [ 68 ]. We would argue that such variants are not bona fide interindividual epigenetic variants. Because most human tissues exhibit such cellular heterogeneity, the specific composition of which can differ among individuals and disease states, interindividual variation observed in just one tissue is difficult to interpret. CoRSIVs, on the other hand, are unaffected by individual differences in tissue cellular composition [ 37 ]; like sequence variants, they are stable epigenetic variants intrinsic to essentially all cells in an individual. The CpG methylation profile at CoRSIVs can therefore reasonably be considered a readout of an individual’s epigenome, enabling adoption of concepts and applications developed for genomics, such as GWAS. Given the strong influence of genetics on methylation at CoRSIVs, one might ask whether profiling CoRSIV methylation offers additional information beyond that obtained by genotyping. We anticipate many advantages. First, as multiple genetic variants influence methylation at each CoRSIV (Additional file 1 : Fig. S7), CoRSIV methylation can be viewed as an integrative readout of these influences. Also, GWAS variants may logically be prioritized based on known mQTL effects at CoRSIVs, just as investigators now prioritize GWAS hits based on evidence of eQTL [ 69 ]. In fact, mQTL effects at CoRSIVs may in some cases mediate eQTL. Lastly, whereas our current data on CoRSIV mQTL is based on a mostly White, non-Hispanic cohort in the USA, it is possible that additional sources of variation (for example, due to periconceptional environment [ 37 , 59 , 60 ]) will be uncovered as CoRSIVs are studied in a broader range of ancestral and cultural contexts, providing insights into gene by environment interactions. Conclusions For over 10 years, the Illumina methylation platform has been the predominant tool for population studies of DNA methylation [ 22 , 30 ]. A major reason is that it interrogates a stable subset of CpG sites within the human genome (yielding one quantitative value for each), simplifying data sharing and integration across multiple studies and populations. Nonetheless, the platform has a major and undeniable shortcoming in the context of population epigenetics: most CpGs included do not show appreciable interindividual variation [ 33 , 34 , 35 , 36 ]. Here we have shown that focusing on systemic methylation variants enables the identification of far stronger mQTL than has been detected by the Illumina arrays [ 42 ]. We anticipate that the greater population variance at CoRSIVs will also improve the power of studies aiming to associate epigenetic variation with risk of disease. Generating the data to explore associations between CoRSIV methylation and a wide range of human diseases is beyond the scope of this study. However, though grossly underrepresented on the HM450 and EPIC arrays (less than 1% of EPIC probes overlap known SIV regions; see annotated list in Additional file 2 : Table S18), CoRSIVs are often among the top “hits” in existing EWAS [ 70 ]. Indeed, these stable [ 36 , 60 , 71 ], systemic epigenetic variants are already showing great promise for disease prediction [ 72 , 73 , 74 , 75 , 76 , 77 , 78 ]. We suggest that improving the coverage of CoRSIVs would enhance the utility of the Illumina EPIC array for the study of population epigenetics. Additionally, we wish to make our validated human CoRSIV-capture reagents available to the field to facilitate the study of these systemic variants. The list of known human CoRSIVs is currently incomplete, and screening is underway to identify more, including in various ancestry groups. Materials and methods Study samples We obtained de-identified genomic DNA from multiple tissues of 188 donors in collaboration with NIH Genotype-Tissue Expression (GTEx) program [ 38 ] (total of 807 samples). Informed consent was obtained by GTEx, including authorization to release the patient’s medical records and social history, sequencing of the donor’s genome, and blanket consent for all future research using the donated tissue and resultant data. The donor and tissue information is available in Additional file 2 : Table S1 in the Supplementary Appendix. For the independent mQTL validation (USC cohort), newborn blood spots from pediatric glioblastoma cases and controls (47 samples total) were obtained from the California Biobank, using information from the California Cancer and Vital Statistics registries. Genotype data for the 188 individuals were generated by GTEx, and for the other 47 samples, DNA extraction, preprocessing, and genotyping were performed as previously described [ 79 ] (see Additional file 1 : Materials and Methods for more details). Target capture bisulfite sequencing and data processing Out of 9926 CoRSIVs previously reported [ 37 ], we included only those within 3000 base pairs from the body of a gene present in the PubTator [ 80 ] compendium, using BEDTOOLS [ 81 ] software, yielding 4641 CoRSIVs as targets for capture. The goal of using PubTator was to focus not just on known genes but on those most likely to be associated with a measurable phenotypic outcome. Libraries were made using the Agilent SureSelect Methyl-seq library kit with modifications (Design ID: S3163502 ) . Capture design details and version history are available in Additional file 1 : Materials and Methods. As for the data processing, Bisulfite-sequencing reads were trimmed using Trim Galore, then mapped to the human genome build UCSC hg38 using the Bismark aligner [ 82 ]. Uniquely mapped reads were retained for further analysis (see Additional file 1 : Materials and Methods). Our CoRSIV-capture reagents are commercially available from Agilent Technologies, Inc. Evaluating genetic influences on CoRSIV methylation Analysis of associations between CoRSIV DNA methylation and genetic variation in cis was performed relying on the Simes correction as described previously [ 44 ]. Using the EMatrixQTL R package [ 83 ], Spearman rank correlation was computed for all SNVs within 1Mb of each CoRSIV, and the Simes correction was applied. Simes-adjusted p -values for each CoRSIV were collected, and the false discovery rate (FDR) correction was applied across all CoRSIVs analyzed in each tissue, with significance achieved at FDR-adjusted p <0.05. To compare the summed total of mQTL detected at CoRSIVs vs. that reported by GoDMC [ 42 ], mQTL associations were identified with P < 10 −10 . This conservative P value was selected to avoid false positives, given the relatively small number of individuals in the GTEx CoRSIV analysis. To further evaluate genetic influence on CoRSIV methylation, we used a haplotype-based approach. Phased genotype data from GTEx were used to infer each individual’s haplotype within the haplotype block overlapping each CoRSIV and correlations between CoRSIV methylation and haplotype allele sum were assessed (see Additional file 1 : Materials and Methods). Availability of data and materials The raw target capture bisulfite sequencing data for the 807 GTEx tissues (188 individuals) have been deposited to the AnVIL repository [ 84 ]. Controlled access is administered through dbGaP (accession phs001746.v2.p1) [ 85 ]. The samples used in the mQTL validation analysis (USC cohort) are biospecimens from the California Biobank Program. Any uploading of genomic data and/or sharing of these biospecimens or individual data derived from these biospecimens would violate the statutory scheme of the California Health and Safety Code Sections 124980(j), 124991(b), (g), (h), and 103850 (a) and (d), which protect the confidential nature of biospecimens and individual data derived from biospecimens. Full results of our mQTL and haplotype-based analyses on the USC cohort are available in Additional file 2 : Tables S10 and S11, respectively.

Twenty years ago, after the human genome was first sequenced, geneticists began conducting large genome-wide association studies to identify genomic regions linked to human disease. In addition to DNA sequence, another stable level of molecular information—epigenetic modifications established during development—can affect one's risk of disease. For more than a decade, scientists have studied these epigenetic modifications to test associations with disease. Today more than 1,000 such epigenome-wide association studies have been published. Now, in a study published in the journal Genome Biology, a team led by researchers at Baylor College of Medicine reveals that the commercial tool that has been the workhorse for these studies is actually not appropriate for population epigenetics. "Many people know that each person has a unique DNA sequence or genome. Less well known is that every cell in the body likewise has a unique level of molecular individuality called its epigenome," said co-corresponding author Dr. Robert A. Waterland, professor of pediatrics—nutrition at Baylor's USDA/ARS Children's Nutrition Research Center. The epigenome—meaning "above" the genome—is a system of molecular markings on DNA that tells different cells in the body which genes to turn on or off in that cell type. "Epigenetic differences between people can affect their risk of disease," said Waterland, a member of the Dan L Duncan Comprehensive Cancer Center at Baylor. To look for such differences, epigeneticists study DNA methylation, which occurs at specific locations called CpG sites. The standard tool for population studies of DNA methylation is a commercial array that assays hundreds of thousands of CpG sites distributed throughout the genome. For the last 15 years, the Waterland lab and colleagues have focused on a different set of CpG sites: those at which DNA methylation differs substantially among people but is consistent across the different tissues of each person. They reasoned that these sites would be most useful for population studies, because DNA from a blood sample can be used to investigate epigenetic causes of disease in internal organs like the brain or heart. "Three years ago we reported nearly 10,000 such regions in the human genome (named CoRSIVs for correlated regions of systemic interindividual variation) and proposed that studying them could be a novel way to uncover epigenetic causes of disease," Waterland said. As a step toward this, the current study investigated how DNA methylation at CoRSIVs is affected by genetics. Correlations between a genetic variant and methylation at a specific CpG site are called methylation quantitative trait loci (mQTL). More than 200 studies of human mQTL have been reported, nearly all using the commercial methylation arrays. The team developed an approach to target CoRSIVs and studied their methylation in DNA samples from multiple tissues of nearly 200 individuals. When they compared their results with those of the largest previous study, "what we found was somewhat of a shock," said first author Dr. Chathura J. Gunasekara, a data analyst in the Waterland lab. "Compared to the most powerful previous study including 33,000 people, our much smaller study focused on CoRSIVs discovered 72-times more mQTL." Looking to explain this surprising finding, the team discovered that around 95% of the CpG sites on the commercial methylation arrays do not show appreciable methylation differences among people. Interindividual variation, which scientists call variance, is the foundation for statistical associations. With no population variance, there is no possibility of detecting mQTL. This finding should also shock the field of epigenetic epidemiology. "Population variance is essential not only for mQTL detection, but also for detecting associations between DNA methylation and risk of disease," said co-corresponding author Dr. Cristian Coarfa, associate professor of molecular and cellular biology and in the Dan L Duncan Comprehensive Cancer Center and the Center for Precision Environmental Health at Baylor. "Compared to what the field has been doing, we anticipate that focusing on CoRSIVs will make epigenome-wide association studies about 70 times more powerful." Indeed, CoRSIVs have already been associated with diverse health outcomes including thyroid function, cognition, cleft palate, schizophrenia, childhood obesity and autism spectrum disorder. "It's as if there's been this massive and very expensive fishing expedition for the last 10 years, but everyone's been fishing in the wrong place," Waterland said. "We hope that the new tool we've developed will accelerate progress in understanding epigenetic causality of disease."

10.1186/s13059-022-02827-3

Biology

How rabbits help restore unique habitats for rare species

More information about the Brecks, Shifting Sands, Back from the Brink and a toolkit to help rabbit conservation is available via https://naturebftb.co.uk/wp-content/uploads/2021/09/Shifting-Sands-Techniques-to-encourage-European-rabbit-recovery.pdf

https://naturebftb.co.uk/wp-content/uploads/2021/09/Shifting-Sands-Techniques-to-encourage-European-rabbit-recovery.pdf

https://phys.org/news/2021-09-rabbits-unique-habitats-rare-species.html

Abstract Background How microbes affect host fitness and environmental adaptation has become a fundamental research question in evolutionary biology. To better understand the role of microbial genomic variation for host fitness, we tested for associations of bacterial genomic variation and Drosophila melanogaster offspring number in a microbial Genome Wide Association Study (GWAS). Results We performed a microbial GWAS, leveraging strain variation in the genus Gluconobacter , a genus of bacteria that are commonly associated with Drosophila under natural conditions. We pinpoint the thiamine biosynthesis pathway (TBP) as contributing to differences in fitness conferred to the fly host. While an effect of thiamine on fly development has been described, we show that strain variation in TBP between bacterial isolates from wild-caught D. melanogaster contributes to variation in offspring production by the host. By tracing the evolutionary history of TBP genes in Gluconobacter , we find that TBP genes were most likely lost and reacquired by horizontal gene transfer (HGT). Conclusion Our study emphasizes the importance of strain variation and highlights that HGT can add to microbiome flexibility and potentially to host adaptation. Background Microbes are important drivers of host phenotype and evolution [ 1 ]. Benefits derived from microorganisms can facilitate the occupation of new ecological niches [ 2 , 3 , 4 , 5 ] and microbial effects on host phenotypes and fitness can spur adaptive processes [ 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 ]. Changes in the effects of microbes on host fitness can alter interactions along the parasitism mutualism continuum [ 6 , 15 , 16 , 17 , 18 ], thus affecting the evolutionary trajectories of the partners. The importance of microbes in evolution and health of higher organisms has sparked a search for the molecular underpinnings of how microbes affect host phenotype. In this search, microbial Genome Wide Association Studies (GWAS) are an important tool [ 19 , 20 , 21 , 22 , 23 ]. The principle behind a microbial GWAS is to establish a link between traits and genetic variation of microbes by the means of GWAS. By testing for association between host traits and microbial genomic variation, Chaston et al. [ 24 ] introduced a particularly helpful approach to unravel how microbes affect host phenotypes [ 22 , 24 ]. The authors measured host phenotypes, here from Drosophila melanogaster that were mono-associated with several microbial isolates. Differences in host phenotype were then associated with the presence and absence of genes in the microbial isolates. By applying this approach, it was found that genes that play a role in glucose oxidation in bacteria affect D. melanogaster triglyceride levels [ 24 ] and that bacterial methionine and B vitamins are important for starvation resistance [ 25 ] as well as life span [ 26 ]. For targeting host phenotypes with microbial GWAS, model systems that allow the generation of axenic hosts that can successively be associated with individual microbial isolates are particularly useful [ 24 ]. One such model system is D. melanogaster and its bacterial microbiome. Techniques for the generation of gnotobiotic flies are readily available and standardized measurements of phenotypes exist. Microbe-affected host phenotypes include the life history traits such as development time, fecundity, and life span as well as size of the adults [ 14 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ]. These traits are components of fitness, emphasizing the potential importance of microbes in host evolution and adaptation [ 14 , 32 ]. Microbes often affect fitness-related traits by provisioning nutrients. These nutrients include vitamins, amino acids, lipids, and trace elements [ 24 , 34 , 35 , 36 , 37 , 38 , 39 ]. Nutrient provisioning is a recurring theme in metazoan–microbe interactions that are adaptive for the host [ 40 , 41 , 42 ]. The acquisition of nutrients from microbes need not rely on microbes that live inside the host. Instead, nutrients can also be acquired by harvesting or preying upon microbes that live outside the fly and subsequent digestion [ 35 , 43 , 44 , 45 ]. Furthermore, bacteria have been identified that affect D. melanogaster phenotypes by increasing the ability for nutrient uptake [ 46 ] or metabolizing components of the food substrate, and thus modulating its nutrient content [ 24 ]. Interestingly, the metabolic potential to produce nutrients that affect fly fitness differs between closely related microbes and so do the effects on fly phenotype and fitness [ 26 , 29 , 32 , 47 , 48 , 49 , 50 , 51 ]. These findings contribute to the notion that microbial variation at taxonomically low levels is not only important for human [ 15 ], mouse [ 52 ], and plant [ 18 ] hosts, but also for Drosophila [ 53 ]. Because variation between closely related bacteria is important for the interaction of the host and its microbiome, it is also important to consider closely related microbes in studies that aim at elucidating the molecular underpinnings of host–microbe interaction. At the same time, limiting a GWAS to the pan-genome of closely related microbes might offer particular power to the approach: limitation to a narrower range of genes that vary in their presence-absence patterns in a similar genomic background, microbial genes that affect the host can be more precisely pinpointed. For studies that are aimed at better understanding host–microbe interaction in an evolutionary context, it is also important to consider microbes that are associated with the host under natural conditions and if possible, to measure evolutionary relevant host phenotypes in a natural or near natural environment. Finally, tracing the evolutionary changes of the genomic elements that affect host fitness can help us to gain deeper insights into how host–microbe interaction evolves. We aimed our study at better understanding whether and how fly fitness is affected by its natural microbiome by a microbial GWAS. In order to increase the resolution of the approach and consider variation at low taxonomic levels, we concentrated on variation within a taxonomically restricted group of bacteria. Therefore, we focused our study on Gluconobacter, a bacterial genus that is commonly associated with D. melanogaster under natural conditions [ 54 , 55 , 56 , 57 ]. We assessed offspring number per female fly as a fitness component on grape juice-based fly food as a near natural food source. In order to better understand how microbial effects on host fitness evolve, we traced the evolutionary events that lead to changes in bacteria-mediated host fitness. Results We performed a microbial GWAS for the number of offspring produced by females that were mono-associated with 17 bacterial isolates from genera that co-occur with Drosophila melanogaster in its natural environment. Gluconobacter was represented by 13 isolates. Two additional isolates were from the genus Acetobacter . Species from this genus can benefit Drosophila development [ 28 ]. One isolate was Commensalibacter intestini that might have a probiotic function in D. melanogaster [ 58 ] and is enriched in flies over substrate in wild-caught flies [ 57 ]. As an outgroup and to get a baseline for the fitness effect of an ingested pathogen, we added Providencia sneebia that is highly pathogenic when entering the hemolyph [ 59 ]. All bacterial genomes analyzed were >99% complete with the exception of P. sneebia (>96%, Additional file 1 : Table S1). The mean number of offspring varied significantly between flies mono-associated with different isolates ( P = 4.2 × 10 −15 , Kruskal-Wallis test, Fig. 1 A) up to a 2.8-fold difference between Gluconobacter morbifer and Gluconobacter sp. P5H9_d. Differences between bacterial strains were also a significant covariate of offspring number when we accounted for bacterial loads per fly ( P = 1.4 × 10 −4 , linear model). Furthermore, bacterial load alone was not significantly associated with fly offspring number ( P = 0.11, linear model), suggesting that not only bacterial biomass affects fly fitness. Presence-absence patterns of 11,269 genes were tested for association with the number of offspring that mono-associated females produced using the PA method [ 60 ]. Associations were confirmed by permutation tests and TreeWAS [ 21 ] (Table 1 , Additional file 2 and 3 : Figure S1, Table S2). The six highest PA scores depended strongly on presence-absence patterns between the closely related strains P1C6_b, DSM2003, DSM2343, and DSM3504 (mean ANI = 95.5%) in the branch that includes G. morbifer (Additional file 2 : Figure S1 and accompanying text). Fig. 1 A Left: Bacterial tree based on 134 single-copy orthologs. Bootstrap support is 100% for all nodes. Leaf labels of bacteria that do not carry a complete thiamine biosynthesis pathway are on red background. Right: Number of offspring produced by mono-associated CantonS females; vertical bars: median; ctrl: axenic flies; conventional: flies reinfected with lab microbiota. B Thiamine treatment (1 μg/ml added to the food) increased the relative number of offspring for the strains that do not possess a complete thiamine biosynthesis pathway (TBP−) compared to strains that possess the complete pathway (TBP+). Relative offspring number was determined by dividing the number of offspring for each TBP− strain by average offspring number of the TBP+ strains. A value of one would mean equal offspring number between TBP+ and TBP− strains. P -value was determined with a linear mixed effects model. Error bars indicate standard error of the mean (Additional file 4 : Script S1) Full size image Table 1 List of the ten genes that were most strongly associated with offspring number according to PA-association scores. All associations were confirmed by at least one of three methods from treeWAS [ 21 ] Full size table The bacterial thiamine biosynthesis pathway is associated with increased offspring number Five of the six bacterial genes that were most strongly associated with offspring number were part of the thiamine biosynthesis pathway (TBP, Table 1 ). Females reared on bacteria carrying a complete TBP (TBP+) produced more offspring ( P = 0.0038, Mann-Whitney Test on strain medians, n = 17, Fig. 1 A), suggesting that bacterial thiamine production might increase the number of offspring. Because high numbers of Drosophila offspring on a confined resource like a Drosophila vial can lead to crowding effects, including smaller adults and reduced individual fitness, we weighed the adult flies at the end of the experiment. Weight did not differ significantly between the offspring of females reared on TBP+ and TBP− strains ( P = 0.55, Mann-Whitney Test on strain medians, n = 17, Additional file 2 : Figure S2), providing no evidence for larval crowding or reduced adult size. Significance of all p -values was confirmed in a linear model framework that accounts for bacterial load and also in a phylogenetic ANOVA (Additional file 4 : Script S1). We hypothesized that if bacterial thiamine increased the offspring number of females reared on TBP+ strains, supplementing the diet of flies reared on TBP− strains with thiamine would increase offspring number when compared to TBP+ strains. To test this, we supplemented the diet of females that were mono-associated with TBP− strains ( G. sp. DSM3504, G. morbifer G707, C. intestini A911) with thiamine and applied the same assay for offspring number as for the initial GWAS. Supplementing the diet of flies reared on closely related TBP+ strains served as control ( G. oxydans DSM2343, G. oxydans DSM2003, G. sp. P1C6_b, G. cerevisiae DSM27644). In order to account for variation between experiments, we calculated the relative offspring number between TBP− and TBP+ strains in the initial unsupplemented experiment (“thiamine treatment −”, Fig. 1 B) and the supplementation experiment (“thiamine treatment +”, Fig. 1 B). Indeed, the relative offspring number on TBP− strains increased with thiamine supplementation ( P = 0.025, linear mixed effects model, Fig. 1 B), supporting a role of thiamine production in the number of offspring that flies produced. We found no evidence that the addition of thiamine increased bacterial loads of TBP− strains ( P = 0.85, generalized linear model), suggesting that the increase in offspring number is not due to an increase in bacterial biomass alone. Thiamine biosynthesis genes were most likely lost and reacquired by horizontal gene transfer as an operon on the branch that includes G. morbifer In order to better understand the evolutionary history of the TBP (Fig. 2 A) in Gluconobacter , we analyzed the synteny of the underlying loci in a phylogenetic framework. The strains in the upper two panels of Fig. 2 B possess all genes required for thiamine biosynthesis. A closer inspection of TBP genes on the G. morbifer branch (II in Fig. 2 C) revealed that two strains are TBP−, while the four other strains are TBP+. Inspection of the TBP gene loci revealed that all strains on branch II are missing the operon-like structure thiOSG (Fig. 2 C) at the locus that is syntenic with branch I. The same pattern was found for thiC and thiD (Additional file 2 : Figure S3). thiOSG (Fig. 2 C), thiC, and thiD (Additional file 2 : Figure S4) are present in the closely related bacteria Gluconobacter samuiensis and Neokomagateaa tanensis at syntenic loci, suggesting deletion on branch II. The strains with an intact operon on branch II carried a TBP operon at loci not syntenic with the locus shown in Fig. 2 C as evident from different flanking genes (Fig. 2 D, Additional file 2 : Figure S5), suggesting insertion. Fig. 2 A The thiamine biosynthesis pathway in acetic acid bacteria. B Overview of thiamine biosynthesis genes in the analyzed bacteria. Note that the function of thiF that appears to be missing in the strains of the upper row can be replaced by the function of the homolog MoeB (Rodionov et al., 2002) that we found in all strains analyzed. Genes forming one operon are separated by a hyphen. Genes from different loci are separated by slashes. C Synteny of the flanking regions of thiamine biosynthesis genes in Gluconobacter and Acetobacter . thiOSG are missing on the G. morbifer branch (II) at this locus. Thiamine biosynthesis genes are in blue. The hypothetical protein is of unknown function. D Right: The complete pathway to synthesize Thiamine-P (green) forms an operon on the G. morbifer branch (branch II); left: the phylogeny depicts the inferred evolutionary scenario on branch II. E Phylogeny of thiE. G. oxydans DSM2343, G. oxydans DSM2003, G. sp. P1C6_b, and G. cerevisiae DSM27644 have two copies of thiE, thiE1 (blue) and thiE2 (green). The phylogeny of thiE1 (blue background) is congruent with the core genome phylogeny. ThiE2 (green background) forms a distinct clade that is more distant than thiE from Acetobacter, indicating HGT from a distant clade. Node labels represent posterior probabilities as assessed by MrBayes v 3.2.6 [ 62 ] Full size image Analyzing the sequences of the inserted genes in a phylogenetic framework, we found that the inserted genes form a distant clade. For example, thiE1, the copy that remained at the locus shown in Fig. 2 C, followed the phylogeny based on the core genome, while the potentially newly acquired copy thiE2 that is part of the operon thiCOSGEFD formed a distant clade (Fig. 2 E), supporting HGT. Within this clade, the phylogeny of thiE2 is again congruent with the core genome phylogeny, consistent with a single reacquisition event of thiCOSGEFD. The same phylogenetic patterns were found for the other TBP genes that were shared across branches (thiCOSGD, Additional file 2 : Figure S6), further supporting a single HGT of thiCOSGEFD to the G. morbifer branch. Because TBP genes can occur on plasmids [ 61 ], we blast searched the plasmids of the strains for which the plasmids were resolved for TBP genes, finding no evidence for TBP genes (data not shown). In order to identify a potential donor of the operon, we blast searched the sequence of the entire operon against the ncbi non-redundant nucleotide database (nr). The best matching non- Gluconobacter sequences were from Rhodobacteraceae, a phylogenetically distant bacterial family (Additional file 5 : Table S3). A closer inspection of the non- Gluconobacter blast hits with the highest scores (query coverage 79–82%, ~73% identity, Additional file 5 : Table S3) revealed that gene order within the operon, but not synteny of flanking genes, was conserved (Additional file 2 : Figure S7). Despite a modest difference in GC-content between the potential donors (~66% vs 62% in Gluconobacter oxydans DSM2343), the GC-content of the putatively inserted operon did not differ from that of the genomic background (Additional file 2 : Figure S8), providing no evidence for a recent acquisition from any of the top 3 blast hits. Furthermore, the best non- Gluconobacter blast hits were in marine bacteria. Taken together, this implies that the true donors remain enigmatic. A single reacquisition event of the essential TBP genes in the past, close to the base of clade II, as suggested by the concordance of the inserted operon with the core gene phylogeny, implies that the TBP− strain DSM3504 lost the operon again in an independent event, as depicted in Fig. 2 D (left). Discussion Microbial GWAS for host traits can benefit from strain level variation We applied a microbial GWAS approach that associates bacterial genes with host phenotype focusing on the genus Gluconobacter . Microbial GWAS approaches can be particularly powerful, when pan-genomic variation of closely related bacterial strains can be leveraged, as has been shown for, e.g., virulence genes [ 63 ]. We showed that genetic variation below the species level between the strains P1C6_b, DSM2003, DSM2343, and DSM3504 (mean ANI = 95.5%) empowered us to pinpoint the TBP (Additional file 2 : Figure S1). Variation between bacteria that have ANI > 95% is considered strain level variation [ 64 ]. The only gene that had a higher association score for offspring number than the TBP genes was a transposase. Transposases more frequently produce rare presence-absence patterns because they are mobile and not linked as strongly to the rest of the genome as are non-mobile genetic elements. Therefore, we suspect that the high association score is an artifact of its mobility although we cannot exclude an effect of the transposon on the number of fly offspring. From the other genes with significant associations with offspring number presented in Table 1 , the oxidoreductase, LysR family transcriptional regulator, and the methyltransferase domain protein, a plausible link to fly offpsring number is more difficult to test. Nonetheless, these genes might also affect fly offspring number. The ferric iron siderophore receptor is located close to the inserted thiamine operon in P1C6_b and DSM27644 which also possesses a gene with the same annotation at that locus, as is apparent from Additional file 2 : Figure S5. While it seems possible that this gene contributes to fly fitness, it must be considered that DSM3504 that lacks TBP genes and confers relatively low fitness to the host also carries a ferric iron siderophore receptor that is orthologous to that shown in Additional file 2 : Figure S5 in DSM27644. Given this and the demonstrated fitness effects of the thiamine supplement (Fig. 1 B), we must assume that this gene received a high association score mainly due to linkage to the TBP genes. Variation between closely related microbes is important for host phenotypes We observed significant variation of phenotypes between flies that were associated with closely related microbial strains. This supports the notion that strain level variation is important to consider when studying host–microbe interaction in animals, humans, and plants alike (e.g. [ 15 , 18 , 52 , 65 , 66 ]. In particular, in D. melanogaster , evidence for the importance of variation between closely related bacteria is accumulating for life history of the host [ 26 , 29 , 47 , 48 , 49 , 50 , 51 , 67 ]. Unawareness of strain level variation in bacterial effects on the host might have led to perceived inconsistency between studies [ 53 ]. Furthermore, our study provides an example of the limits of 16S rRNA sequencing in functional inference: The 16S rRNA gene sequences, as assessed by full length Sanger sequencing of strains P1C6_b (TBP+) and DSM3504 (TBP−), were identical (Additional file 2 : Figure S9). The number of offspring as a component of fitness As a fitness component, we assessed the number of adult offspring produced after 16 days. As such, our assay captures developmental rates and fecundity on a time scale that we consider highly relevant for the reproductive success of an organism that is adapted to an ephemeral resource, rotting fruit [ 68 ]. The effect of thiamine on offspring number that we describe is consistent with previously described effects of bacterial thiamine on D. melanogaster development and survival to adulthood [ 37 ]. Yet, a limitation of our study is that other components of life time reproductive success, and thus fitness, were not directly assessed, in particular egg laying and longevity. However, Sannino et al. [ 37 ] also showed that bacterial thiamine neither affects egg laying nor longevity in D. melanogaster. This suggests that the effect of thiamine on the number of offspring that we observed is directly related to lifetime reproductive output, and thus fly fitness. The loss and regain of the TBP by HGT in the context of the evolution of host–microbe interaction Offspring number was strongly associated with genes from the TBP. The acquisition of B vitamins like thiamine (B1) is a typical benefit that insects receive from microbes [ 25 , 69 , 70 ] and falls into the greater context of nutrient provisioning by microbes, which is a recurring theme in the evolution of host–microbe interaction [ 40 , 41 ]. By tracing the genes of the TBP across genomes and the phylogeny, we found that the pathway to produce Thiamine-P was regained most likely via HGT (Fig. 2 D). As such, our study exemplifies that individual events of HGT into a host-associated microbe can alter host fitness outcomes. Other studies that show an effect on host fitness via HGT to a host-associated microbe involve defensive compounds produced by microbial symbionts in plants [ 71 ] and animals [ 17 , 72 ]. In our study, the increase in host fitness with the reacquisition of the TBP is most likely mediated via nutritional benefits. Only a few similar cases have been described so far. The most prominent may be the acquisition of vitamin B7 (biotin) and vitamin B2 (riboflavin) synthesis by planthopper-associated Wolbachia [ 73 ]. Similarly, bed bug-associated Wolbachia [ 74 ] and cat flea-associated Wolbachia seem to have gained the ability to produce biotin via HGT [ 75 ]. In ticks, pabA (and possibly pabB ) required for the synthesis of folic acid was acquired by a Coxiella -like symbiont through horizontal gene transfer (HGT) from an Alphaproteobacterium [ 76 ] and is thought to affect tick fitness. A Rickettsia endosymbiont of deer ticks has acquired the genes necessary for the synthesis of biotin on a plasmid [ 77 ]. In this study, as in ours, a complete operon has been transferred. A difference between our and these previous studies is that although Gluconobacter is frequently associated with D. melanogaster under natural conditions [ 54 , 55 , 56 , 57 ], it is neither an obligate symbiont nor is it restricted to the fly gut. While it is interesting that the natural fly isolates of Gluconobacter are TBP+, there is currently no evidence that the fly host significantly affects Gluconobacter evolution given its occurrence in the environment and the opportunity for horizontal transfer of the bacterium between hosts. Further, taking into account evidence that the abundance of mobile metabolic genes is governed by selection [ 78 , 79 ], we must assume that loss and gain of the TBP must first of all benefit the bacterium to persist in the bacterial genome. Thiamine is considered essential for bacteria [ 80 ], and thus, the TBP can only be lost if enough thiamine is available in the environment. Fruit, the main food substrate of Drosophila under natural conditions [ 68 ] and the basis for the food used in our study, is mostly poor in thiamine [ 81 ]. However, other bacteria that are associated with Drosophila , for example other strains of Gluconobacter (this study), Acetobacter pomorum or Lactobacillus plantarum , can produce thiamine [ 37 , 39 ]. Under these conditions, it might be beneficial for a community member to lose TBP as a result of selection for reduced metabolic expenditure [ 82 ]. This is consistent with TBP-dependent fitness effects on the host being a byproduct of selection on thiamine production in the microbe. Our study suggests that HGT to host-associated microbes could quickly increase host fitness. An increase in microbe-mediated host fitness should also increase selection pressure on the host to favor that particular microbe that provides an increased benefit [ 41 , 83 , 84 ]. Waterworth et al. [ 85 ] suggested that the acquisition of genes to produce a defensive compound via HGT was key to the domestication of a bacterial defensive symbiont in beetles. We speculate that similar scenarios might be plausible for nutritional benefits in Drosophila because (i) mechanisms of host selection work efficiently for environmentally acquired bacteria [ 86 , 87 , 88 , 89 ]; (ii) stable, strain-specific associations of Drosophila with mutualistic bacteria have been reported [ 50 ]; and (iii) evidence for host selection in the fly is accumulating in the laboratory [ 45 , 90 ] as well as under natural conditions [ 55 , 57 ]. Conclusion Because the result of HGT here provides a potential benefit to the host under thiamine poor conditions that are often encountered under natural conditions, e.g., on thiamine poor fruit, our study contributes to a broader view of adaptation that can involve a flexible microbiome [ 4 , 91 ]. Methods Fitness assays Canton-S stocks were kept at 25 °C on a 12:12 light:dark cycle on food prepared following the Bloomington Drosophila Stock Center “Cornmeal Molasses and Yeast Medium” (532-ml water, 40-ml molasses, 6.6-g yeast. 32.6-g cornmeal, 3.2-g agar, 2.2-ml propionic acid, and 7.6-ml Tegosept). To generate axenic flies, embryos were collected and washed in PBS, dechorionated in 50% bleach for 2–3 min, and rinsed in sterile PBS for 1 min. Embryos were placed in sterilized food bottles under a sterile workbench and maintained at 25 °C under a 12:12 light:dark cycle in axenic condition for 3 weeks during which the flies had time to hatch and mate. One axenic female from these bottles was used per vial in the fitness assay. For the fitness assay, bacterial cultures were grown in liquid YPD medium for 48–72 h and normalized to the same optical density (OD600 = 0.6). One hundred fifty microliters of OD normalized medium was added directly on 10-ml sterile grape juice food (667-ml water, 333-ml Jacoby white grape juice, 8-g yeast, 50-g cornmeal, 10-g agar, 3-ml propionic acid). The food was autoclaved without proprionic acid and proprionic acid added after the food had cooled down. Please note that yeast can in principle serve as a thiamine source, but autoclaving might have reduced the thiamine content of the food, as thiamine is heat labile [ 92 , 93 ], such that it became a limiting factor for offspring number. Axenic females were transferred to the vial immediately after addition of the bacterial culture. We prepared two control treatments. First, we added sterile YPD medium to the food as axenic control. Second, we used conventionally reared flies homogenized in YPD as inoculum. On the day 16, flies were counted, collected, and weighed. All offspring were weighed together in one Eppendorf tube for each replicate and weight per fly was calculated. All fitness-related measurements were done blind. That means the vials were given random numbers and only after the measurements were taken, the bacterial strain ID was connected to the result. For the thiamine supplementation experiment, food was prepared as described above but we added 1 μg/ml thiamine to the food after autoclaving. That concentration has proven effective for phenotypic rescue in [ 37 ]. All statistical analyses were performed in R and can be found in Additional file 4 : script S1. Bacterial loads and contamination control Fly offspring from the fitness assays were stored in PBS/glycerol mixture at −80 °C for later contamination control and the counting of colony forming units (CFUs). Please note that glycerol is a standard cryoprotectant that allows to keep bacteria alive at −80 °C for extended time periods [ 94 ]. Effective conservation of live bacteria with this method is supported by CFU counts in the range of 10 2 –10 5 CFUs (Additional file 6 : Table S4) per fly for the majority of our samples, matching expectations from the literature well [ 67 ]. Finally, all samples underwent the freezing procedure in our randomization scheme that should prevent systematic treatment effects. Nonetheless, we cannot fully exclude that strain variation in the response to cryopreservation might have affected colony counts. For CFU enumeration 3–6 replicates per bacterial isolate were picked. To this end, samples of 3 to 5 offspring were homogenized with a pestle in 300 μl of PBS. The homogenates were plated on YPD agar medium. Plates were incubated for 48 h. CFU counts were done visually or with the OpenCFU software [ 95 ] (Additional file 6 : Table S4). Plates for CFU counting were also inspected for colony morphology and colony color that could indicate potential contamination, with negative results. All homogenates were plated on antibiotic YPD agar medium (with 100 μg/ml kanamycin or ampicillin) for assessing yeast contamination. No yeast colonies were observed except in the control replicates in which conventional lab microbiota were used. To further assess potential bacterial contaminants during our experiment, we quantified the relative abundance of target isolates that flies were inoculated with on fly offspring using 16S rRNA gene sequencing (Additional file 2 : Figure S10). In short, DNA was extracted from pools of 3–5 offspring for 3–6 replicates per bacterial isolate after the experiment, including the replicates with the highest and lowest offspring number. The V4 regions of the bacterial 16S rRNA gene were amplified and sequenced on an illumina MiSeq sequencer following [ 56 , 96 ]. Sequencing data were analyzed using mothur [ 97 ] (see Additional file 7 : script S2 for all commands executed). The relative abundance of target 16S rRNA gene sequences for mono-associated isolates was calculated. The average relative abundance of target 16S sequences was over 88% (Additional file 2 : Figure S10A) in the initial experiment. Only in (6 out of 66) replicates the relative abundance was below 75%, including 3 cases of P. sneebia that showed very low bacterial loads. For the thiamine treatment, the target bacteria were significantly enriched in the microbial community, but there was also some evidence for contaminating 16S gene sequences, which were likely introduced during the PCR or sequencing steps (Additional file 2 : Figure S10B). Bacterial isolates, genome sequencing, and assembly We sequenced, assembled, and annotated draft genomes of eleven bacteria and added genome data for six bacteria from public databases (Additional file 1 : Table S1). Nine strains were isolated from wild-caught Drosophila collected in the San Francisco Bay Area (California, USA). Isolates were cultured in YPD for standard phenol-chloroform DNA extraction. Bacterial genomes were sequenced using Illumina MiSeq technology and assembled with the A5 MiSeq assembler [ 98 ]. Completeness and contamination were assessed with checkM v1.1.2 [ 99 ], using standard settings. Assembly statistics were generated with QUAST v5.0.2 [ 100 ]. Annotation was performed with prokka v1.1 [ 101 ] or imported from GenBank. Average nucleotide identity (ANI) was computed with fastANI (v0.1.2). New isolates were taxonomically classified, using GTDBtk (v0.1.4) [ 102 ]. FastANI and GTDBtk were run on the kbase web interface [ 103 ]. Pan-genome clustering and phylogenetic trees Genomes were analyzed using the panX analysis pipeline [ 60 ] with standard parameters (Additional file 8 : script S3). Genes were grouped into 11,269 clusters of homologous sequences, including clusters with a single gene. Thereby, the presence and absence of each gene cluster in the 17 genomes was estimated. Based on the alignments of all 134 inferred single-copy gene clusters that are present in all 17 genomes, panX reconstructs a phylogenetic tree (Fig. 1 ). For this phylogeny, FastTree 2 [ 104 ] and RaxML [ 105 ] were applied to all variable positions from these alignments. To create bootstrap values, we used a separate raxml call with the -b option based on the alignments created by panX (see Additional file 8 : script S3). For the phylogeny of gene clusters, nucleotide sequences were aligned using MUSCLE v3.8.425 [ 106 ] and the tree was built by MrBayes 3.2.6 [ 62 ], using a molecular clock with default parameters in the Geneious software suit v1.1 (Biomatters ltd.). Microbial pan-genome-wide association study We calculated the gene presence absence association score (PA score) between each predicted cluster of homologous genes and fly offspring number. That is, if D g is the difference between the mean fly offspring for strains with and without gene g, σ is the global standard deviation of fly offspring for all strains and n g is the number gene gains and losses as inferred from the phylogeny. The association score is given by n g − − √ D g σ \sqrt{n_{\mathrm{g}}}\frac{D_{\mathrm{g}}}{\sigma } . Three alternative association scores from treeWAS [ 21 ] and the corresponding model-based p -values were calculated. Association scores based on the presence and absence of genes are prone to false positives because genome wide linkage results in strongly correlated presence and absence of genes. PanX and treeWAS reduce this effect by taking the reconstructed ancestral gene gain and loss events into account. Availability of data and materials The sequencing data generated and analyzed during the current study are available in the NCBI SRA repository, [ 107 ]. The bacterial genomes and the assemblies are either available under SRA number SRS7200184 – SRS7200194 [ 107 ] or from the sources described in Additional file 1 : Table S1 [ 108 , 109 , 110 , 111 , 112 , 113 ] with raw data available in [ 114 , 115 , 116 , 117 , 118 , 119 ]. The 16S rRNA gene sequences are available under SRA number SRS7426971 - SRS7427068 [ 107 ] with the sample titles corresponding to the column “name_in_mothur” in Additional file 6_Table S4. Sequences of the closely related species used for the alignment in Additional file 2 : Figure S4 are from [ 120 , 121 ] with raw data available in [ 122 , 123 ].

European wild rabbits are a 'keystone species' that hold together entire ecosystems—according to researchers at the University of East Anglia. Their grazing and digging activity keeps the ground in a condition that is perfect for sustaining other species that would otherwise move on—or die out. But their numbers are declining regionally, nationally and globally. And they are even being classed as endangered in their native region, the Iberian Peninsula. The findings come as efforts to save England's most threatened species from extinction are turning the tide for wildlife in Norfolk and Suffolk thanks to the Shifting Sands project. Shifting Sands is one of 19 projects across England that make up the national Back from the Brink initiative. Together, these projects aim to save 20 species from extinction and benefit over 200 more. Lead partner of the rabbit workstream and rabbit expert Prof Diana Bell, from UEA' School of Biology, said: "The Breckland-based Shifting Sands project was set up to save some of the region's rarest wildlife. "After several years of hard work by this multi-partner project, the fortunes of species classed as declining, rare, near-threatened or endangered are now improving in the Brecks. "The project has seen species recover in record numbers—including endangered beetle and plants, one of which is found nowhere else in the world. "Rabbits are incredibly important because their grazing and digging activity keeps the ground in a condition that is perfect for sustaining other species. "Sadly, rabbit populations have declined dramatically in the UK and across Europe, and the European wild rabbit is now listed as endangered in its ancestral Iberian Peninsula range. Their decline is largely due to a spill-over of new viruses from commercially bred rabbits. "The Shifting Sands project has shown us how important rabbits are to entire ecosystems, and it is vital that these habitats are conserved and protected. "We encouraged a rabbit revolution in the Brecks and we have produced a toolkit in partnership with Natural England to help landowners of similar rabbit-dependent habitats to do the same." "Simple cost-effective ways of encouraging rabbits include creating piles of felled branches, known as brush piles, and banks of soil." Monitoring over the past three years has shown the interventions are working, with evidence of significantly higher amounts of rabbit activity. Prof Bell said: "Our work resulted in evidence of rabbit activity in significantly higher numbers. 91 percent of brush piles showed paw scrapes and 41 percent contained burrows. Even when burrows did not form, the brush piles helped expand the range of rabbit activity." The UEA research team worked in collaboration with Natural England, Forestry England, Plantlife, Breckland Flora Group, Norfolk Wildlife Trust, Suffolk Wildlife Trust, Butterfly Conservation, Buglife, the Elveden Estate and the RSPB to deliver this ambitious partnership project. It has seen five kilometers of 'wildlife highways' created, more than 100 specimens of rare plants re-introduced, habitat created and restored across 12 sites, species encouraged, and landscape-management practices improved. As a result, seven species of plant, bird and insect are increasing in number and many more are benefiting in turn. Among those species recovering are rare plants such as the prostrate perennial knawel that is unique to the Brecks, basil thyme and field wormwood. The endangered wormwood moonshiner beetle, lunar yellow underwing moth and five-banded digger tailed wasp are also increasing. All these species are identified in the UK's Biodiversity Action Plan as being priorities for conservation. The open habitat maintained by rabbits supports two rare plants: the prostrate perennial knawel—found nowhere else in the world—and field wormwood. Pip Mountjoy, Shifting Sands project manager at Natural England, said: "The Brecks were described by Charles Dickens as "barren." They are anything but. Their 370 square miles of sandy heathland, open grassland and forest support almost 13,000 species, making it one of the UK's most important areas for wildlife. "That wildlife is under threat. Felling trees and encouraging a species that is often considered a pest may seem a strange solution. But in this instance, carefully managed 'disturbance' is exactly what this landscape and its biodiversity needs." "The project's interventions have provided a lifeline for this unique landscape, and shown how biodiversity can be promoted by 'disturbing' places—not just by leaving them alone. " "These rare habitats are becoming overgrown and species are declining as a result of changing land management practices and human impacts. It's our responsibility to restore and maintain these spaces for nature. Some of these species exist only here and, if lost, will be lost forever."

https://naturebftb.co.uk/wp-content/uploads/2021/09/Shifting-Sands-Techniques-to-encourage-European-rabbit-recovery.pdf

Biology

A mechanism of color pattern formation in ladybird beetles

Toshiya Ando et al, Repeated inversions within a pannier intron drive diversification of intraspecific colour patterns of ladybird beetles, Nature Communications (2018). DOI: 10.1038/s41467-018-06116-1 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-018-06116-1

https://phys.org/news/2018-09-mechanism-pattern-formation-ladybird-beetles.html

Abstract How genetic information is modified to generate phenotypic variation within a species is one of the central questions in evolutionary biology. Here we focus on the striking intraspecific diversity of >200 aposematic elytral (forewing) colour patterns of the multicoloured Asian ladybird beetle, Harmonia axyridis , which is regulated by a tightly linked genetic locus h . Our loss-of-function analyses, genetic association studies, de novo genome assemblies, and gene expression data reveal that the GATA transcription factor gene pannier is the major regulatory gene located at the h locus, and suggest that repeated inversions and cis -regulatory modifications at pannier led to the expansion of colour pattern variation in H. axyridis . Moreover, we show that the colour-patterning function of pannier is conserved in the seven-spotted ladybird beetle, Coccinella septempunctata , suggesting that H. axyridis ’ extraordinary intraspecific variation may have arisen from ancient modifications in conserved elytral colour-patterning mechanisms in ladybird beetles. Introduction There are approximately 6000 ladybird beetle species described worldwide 1 . Charismatic and popular, ladybird beetles are famous for the red and black spot patterns on their elytra (forewings), thought to be a warning signal to predators that they store bitter alkaloids in their body fluids 2 , 3 and are unpalatable. This red/black warning signal is shared among many ladybird beetle species, and provides a model for colour pattern mimicry by other insect orders. While most ladybird beetle species have only a single spot pattern, a few display remarkable intraspecific diversities, such as the multicoloured Asian ladybird beetle, Harmonia axyridis , which exhibits >200 different elytral colour forms (Fig. 1 a). This striking intraspecific variation prompted us to investigate its genetic and evolutionary basis. Fig. 1 Intraspecific genetic polymorphisms of elytral colour patterns in H. axyridis . a Highly diverse elytral colour patterns of H. ayridis . b Four major alleles of the elytral colour patterns. h C , conspicua ; h Sp , spectabilis ; h A , axyridis ; h, succinea . c An example of inheritance of elytral colour forms. When h C / h C and h / h are crossed (P), all F1 progenies show the colour pattern of h C / h . Note the small black spots within the red spots in the F1 progeny. When the F1 heterozygotes are sibcrossed, F2 progeny shows three phenotypes ( h C / h C , h C / h and h / h ) at the 1:2:1 ratio predicted for Mendelian segregation of a single locus. Inheritance of any combination of colour patterns follows this segregation pattern. Images in Fig. 1a are used under license from Insect DNA Research Society Japan (newsletter, vol. 13, September 2010). All rights reserved Full size image The first predictions regarding the genetics underlying the highly diverse elytral colour patterns of H. axyridis and its relevance to speciation were made by the evolutionary biologist, Theodosius Dobzhansky based on his comprehensive classification of specimens collected from various regions in Asia 4 . Successive genetic analyses 5 , 6 , 7 revealed that many of these colour patterns are actually regulated by a tightly linked genetic locus, h , which segregates either as a single gene, or as strongly linked pseudoallelic genes (a supergene 8 , 9 ) (Fig. 1b, c ). The elytral colour patterns are assumed to be formed by the superposition of combinations of two of the four major allelic patterns and dozens of minor allelic colour patterns (>20 different allelic patterns in total). The major allelic patterns cover more than 95% of colour patterns in the natural population 4 . In the elytral regions where the different colour elements are overlapped in heterozygotes, black colour elements are invariably dominant against red colour elements (mosaic dominance 10 ). Whether all of the supposed alleles linked to the h locus correspond to a single gene or multiple genes is unknown. Elucidating the DNA structure and the mechanisms underlying the evolution of this tightly linked genetic locus that encodes such a strikingly diverse intraspecific colour pattern polymorphism would provide a case-study that bears upon a major evolutionary developmental biology question; how does morphology evolve? Here we show that the gene pannier is responsible for controlling the major four elytral colour patterns of H. axyridis . Moreover, we illustrate how modification to this ancient colour-patterning gene likely contributed to an explosive diversification of colour forms. Results Elytral pigmentation during H. axyridis pupal development To identify the gene regulating elytral colour pattern formation of H. axyridis , we first investigated the pigmentation processes during development. In the developing pupal elytra, red pigment (carotenoids 11 ) was accumulated in the future red-pigmented regions (Fig. 2a , pharate adult elytron). Red pigmentation occurred only in the thick ventral epidermal cells of the two layers of the elytral epidermis (Fig. 2b, c , red), and started at 80 h after pupation (80 h AP). Black pigmentation (melanin accumulation 11 ) occurred only in the dorsal cuticle of black-pigmented regions (Fig. 2d , black), and started approximately 2 h after eclosion. Although pharate adult elytra are not black, we detected a strong upregulation of enzymatic activity related to melanin synthesis 12 in the nascent dorsal cuticle in the future black regions from 80 h AP (Fig. 2a , lower panels; Fig. 2c , black; Supplementary Figure 1 ). Every black-pigmented region was deployed complementary to the red regions. Therefore, we concluded that the developmental programs for both red and black pigmentation started around 80 h AP. Fig. 2 Developmental programs for elytral pigmentation initiate at the late pupal stage. a Adult elytral colour patterns (upper panels), and localisation of carotenoid (middle panels, orange) and phenol oxidase (PO) activity (lower panels, black) in pharate adult elytra at 96 h AP. Proximal is up. Outer rims are to the left. b – d Cross sections of elytra at 96 h AP ( b , c ) and adult ( d ). Dorsal is up. b magenta, nuclei (propidium iodide); green, F-actin (phalloidin). c left, PO staining; right, No staining. Arrowheads indicate pigmented areas. d Haematoxylin and eosin staining. Scale bars, 1 mm in ( a ), 50 μm in ( b – d ) Full size image pannier promotes melanin and represses carotenoids in elytra We hypothesised that some of the conserved genes essential for insect wing/body wall patterning 13 , 14 , 15 , 16 , 17 , 18 are recruited to regulate these elytral pigmentation processes, and tested this possibility using larval RNAi 19 . We performed a small-scale candidate screening focusing on genes involved in wing/body wall patterning (Supplementary Table 1 ), and found that the Harmonia orthologue of Drosophila pannier , which encodes a GATA transcription factor 20 , is essential for formation of all of the black-pigmented regions in the elytra. For all four major h allele backgrounds, larval RNAi targeting pannier resulted in complete loss of black colour elements and alternative emergence of red colour elements in the elytra (Fig. 3a , H. axyridis ), indicating that pannier is essential for inducing black pigmentation in dorsal elytral cells and suppressing red pigmentation in ventral elytral cells. This result was unexpected because pannier is not essential for wing blade patterning in Drosophila , but rather essential for patterning of the dorsal body plate attached to the wings (notum) 21 , 22 , 23 . pannier mRNA was upregulated from 48 h AP to 96 h AP in elytra (Supplementary Figure 2a ), and preferentially in black regions ( h C , Supplementary Figure 2b , b′). Immediately before or after 80 h AP (start of the pigmentation program, 76−84 h AP), pannier seemingly showed higher expression in the future black regions in the dorsal elytral epidermis (Fig. 3b ). These data suggest that region-specific upregulation of pannier during the pupal stage regulates black pigmentation in the ladybird beetle’s dorsal elytral cells, and that regions of expression differ among the major h alleles to form different black patterns in H. axyridis . Fig. 3 pannier expression foreshadowing the adult colour pattern switches red/black pigmentation processes. a The adult phenotypes of RNAi treatments targeting GFP (negative controls, GFP RNAi) and pannier ( pnr RNAi) in H. axyridis ( h C , h Sp , h A , h ) and C. septempunctata . Scores in the lower left corners indicate penetrance of the loss-of-pattern phenotype in surviving animals. b The pattern of pannier expression ( pnr ) in the dorsal elytral epidermal cells immediately before or after pigmentation (76−84 h AP). Left panels indicate the corresponding adult elytral phenotypes ( h C and h ) adapted from Fig. 2a . White arrowhead, the region with a weak signal. Black arrowheads, the regions with intense signals. Scale bars, 1 mm Full size image The genomic basis for the colour pattern polymorphism These data led us to test whether pannier is associated with the classically identified locus h , which regulates elytral colour patterns. To identify DNA sequences near the h locus, we assembled de novo genome sequences (assembly version 1: 423 Mb; contig N50, 63.5 kb; scaffold N50, 1.6 Mb), and performed a genetic association study using the strains with different h alleles. We obtained the scaffold containing pannier and two additional adjacent scaffolds based on the truncated gene structures at the scaffold ends (Fig. 4a , Bgb and pnr ). Restriction-site Associated DNA Sequencing (RAD-seq) analysis of backcrossed progenies (BC1, h A × h C F0 cross, n = 183) revealed that these three scaffolds are included in the five scaffolds that showed complete association with colour patterns (Fig. 4a , the upper left panel). In addition, genotyping of F2 individuals from two other independent genetic sib-crosses ( h C × h F0 cross ( n = 80) and h A × h Sp F0 cross ( n = 273)) indicated that the pannier locus is included in the relevant regions of all of the major four h alleles ( h C , 690 kb; h Sp , 750 kb; h A , 660 kb; h , >2.1 Mb) (Fig. 4a , Supplementary Data 1 ). Fig. 4 pannier is the major elytral colour pattering gene located at the h locus. a Genetic association studies of the h locus. Upper left panel, LOD plot for 4419 RAD tags deduced from genotyping BC1 progeny from the h C − h A F0 cross. RAD tags showing segregation patterns of markers located on the X chromosome were excluded from the analysis. The LOD score peaked at RAD tags in Linkage Group 2. RAD tags with complete association with elytral colour patterns corresponded to five genomic scaffolds including the three scaffolds adjacent to the pannier locus (lower bars). Upper right panel, the candidate genomic regions responsible for the major four h alleles. Lower bars, the three genomic scaffolds adjacent to the pannier locus. Grey, genes predicted from RNA-seq. Green, the pannier locus. Contiguity of the scaffolds was predicted based on the truncated genes at the end of scaffolds ( Bgb and pnr ( pannier )). Arrows indicate the respective candidate genomic regions responsible for each allele. pannier is included in all four relevant regions. The images of ladybird beetles are adapted from Fig. 1b . b Fold changes of gene expression in the presumptive red and black elytral epidermis around the candidate genomic region responsible for h . Grey and red bars on the bottom indicate predicted genes. Red, FDR < 0.01. Only pannier was significantly upregulated in this region at 72 h AP. The samples for RNA-seq were collected as depicted in the right panel in the h C background. c Gene structures of pannier in H. axyridis . Exon number is indicated above each exon. 1A isoform cDNA was cloned by rapid amplification of cDNA ends (RACE). 2A−4B isoforms were predicted from RNA-seq analysis. pannier has at least four transcription start sites. Coding sequences are located from exon 2 to exon 5 (yellow). There are two alterative exons at exon 3 (3A and 3B), one encoding one of the two zinc finger domains of Pannier (A isoforms), and the other skipping that zinc finger domain (B isoforms) Full size image To test contiguity of these three scaffolds, we re-assembled the genome using a novel genome assembler (Platanus2), and performed additional de novo genomic assemblies of h C , h A and h alleles using linked-read and long read sequencing platforms (10× Genomics Chromium system; PacBio system). We obtained contiguous longer genomic scaffolds including the three described above ( h C , 3.13 Mb/2.74 Mb; h A , 1.42 + 1.61 Mb; h , 2.79 Mb) (Supplementary Figure 3 ; Supplementary Data 2 , H. axyridis ) and the genotyping markers showing complete association with colour patterns and incomplete association at both ends ( h C and h ) or one end ( h A ) of each scaffold (Supplementary Figure 4a–c ). These data support the result of our genetic association studies. To further delimit the candidate genes associated with the elytral colour patterns, we performed RNA-seq analysis using epidermal tissues isolated from the developing red or black regions before pigmentation in the h C genetic background (Fig. 4b , 24 and 72 h AP RNA-seq). We found that pannier was the only gene statistically significantly upregulated in the developing black region compared to the red region at 72 h AP within the h locus candidate region (Fig. 4b , red bars, false discovery rate (FDR) < 0.01; Supplementary Data 3 ). These data pinpoint pannier as the major gene regulating the elytral colour pattern variation in H. axyridis . Inversions and high diversification within a pannier intron We next investigated allele-specific polymorphisms at the pannier locus. We found that alleles of the first intronic region of pannier are more diverse than the surrounding genomic regions (Fig. 5a , asterisk, the middle whitish regions in h C vs. h A , h A vs. h , and h vs. h C comparisons), whereas the same allele in different strains shows conserved fragments distributed throughout the region (Fig. 5a , blue bars, h C (F2-3) vs. h C (NT6) comparison). In comparisons between the alleles, we consistently found traces of large inversions in the upstream half of the first intron (Fig. 5a , reddish lines, h C vs. h A , h A vs. h , h vs. h C ; 56 kb–76 kb in size) (Supplementary Figure 5 , dot-plot). However, we found that the coding sequences of pannier only showed a single nonsynonymous substitution in the region not conserved among organisms (G235V , h Sp ) (Supplementary Figure 6 , 7 ), suggesting that cis -regulatory differences in the first intronic region of pannier are the major cause of intraspecific colour variation. Fig. 5 Traces of inversions and high sequence diversification within a pannier intron in ladybird beetles. a Sequence comparison of the genomic region surrounding the pannier locus. 700 kb genomic sequences surrounding the pannier locus were extracted from the genome assembly of each allele in H. axyridis ( h C , h A , h ) and C. septempunctata ( C. sep ). Strain names are given in parentheses. Arrows indicate genes predicted by the exonerate program (Orange, pannier ; Blue, GATA transcription factor genes paralogous to pannier ; Green, other genes). Gene names are listed at the top. Vertical or diagonal bars connecting adjacent genomic structures indicate BLAST 72 hit blocks (bluish, forward hit; reddish, reverse hit) in the comparison between the two adjacent genomic scaffolds. The colour code for colouring the bars is at the bottom. The exon−intron structure and the first intron (*) of pannier (1A isoform) is depicted on the top of the panel. The upper half (first intron) of the pannier locus is diversified (whitish) between different h alleles in H. axyridis , and shows traces of inversions (crossed reddish bars). Several intronic sequences are conserved between H. axyridis and C. septempunctata (bars located in the upper half of pannier in C. sep ). The black arrow indicates the region specifically expanded in H. axyridis . b Overview of the size of the upper noncoding regions (the first intron + the upper intergenic region) at the pannier ( pnr ) locus in holometabolous insects. The topology of the phylogenetic tree of surveyed insects is adapted from ref. 107 (Coleoptera), and TIMETREE 108 (Diptera, Hymenoptera, Lepidoptera). The sizes of the first introns are given in parenthesis if cDNA information was available. H. axyridis has the largest noncoding sequence at the pannier locus. In some species, synteny of the three paralogous GATA genes was broken up by translocation (*) or insertion (**). c , d ML phylogenetic trees constructed with nucleotide sequences of pannier ( pnr ) coding region ( c ), and those of the conserved region in the first intron ( d ). The trees were drawn to scale with branch lengths measured in the number of substitutions per site. Bootstrap values were calculated from 1000 resampling of the alignment data. Bars, 0.01 substitutions/site Full size image Moreover, we found that in H. axyridis , the size of the upstream noncoding sequences of the pannier locus (including the first intron of pannier , and the upstream intergenic region between the 5′ end of pannier 5′ UTR and 3′ end of the GATAe 3′ UTR) are 46–65 kb larger than the currently available corresponding genomic sequences of the other holometabolous insects (Fig. 5b , H. axyridis , 153–172 kb; other holometabolous insects, 13−107 kb). Comparison of the exon−intron structures of H. axyridis to those of some of the holometabolous insects also suggested that especially the first intron of pannier is expanded in H. axyridis ( H. axyridis , 108–118 kb; the other holometabolous insects, 11–44 kb). The expanded region in H. axyridis included at least four transcription initiation sites of pannier transcripts (Fig. 4c , pnr-1A – 4B ). In addition, in this region, several known DNA-binding motifs of transcription factors involved in Drosophila wing formation were more enriched allele-specifically than those in the other genomic regions (Table 1 , allele-specifically enriched motifs; Supplementary Data 4 ). For example, the highly conserved Scalloped (SD) DNA-binding motif of the insect wing selector transcription factor complex Vestigial/Scalloped 24 , 25 occurred frequently in the upstream and the downstream regions of the first intron of pannier specifically in the h C allele (Table 1 , allele-specifically enriched motifs, h C , Sd in the upstream and the downstream regions of the first intron). Furthermore, the RNA-seq data for the h C background also revealed that the sd coactivator gene vestigial was the only transcription factor gene that was significantly upregulated in the future black region from early pupal stages (Supplementary Figure 8 ), implicating Vestigial as one of the upstream trans -regulatory factors acting together with Sd to form the two-spotted elytral colour pattern of h C . It is noteworthy that the noncoding region of pannier in each allele possesses putative DNA-binding motifs that can respond to variety of developmental contexts such as anterior−posterior patterning 26 , 27 , 28 (En, Inv), wing fate specification 17 , 24 , 29 (Sd), hinge-wing blade patterning 17 , 18 , 30 , 31 and wing vein patterning 29 , 32 , 33 , 34 (Ab, Al, B-H1, B-H2, Brk, Exd, H, Hth, Kni, Mad, Med, Nub, Rn, Ss, Vvl), hormonal cues 35 , 36 (EcR, Tai, Usp), and auto-regulation (Pnr) (Table 1 , allele-specifically enriched motifs). These results suggest that allele-specific elytral colour patterns of H. axyridis may be formed by integrating appropriate combinations of developmental contexts of wing formation shared among insects. Table 1 Known DNA-binding motifs enriched in the noncoding regions of the pannier locus Full size table Colour-patterning function of pannier conserved in ladybirds We further tested whether the regulatory function of the red/black colour pattern in elytra is a conserved or a derived aspect of pannier function in ladybird beetles using the seven-spotted ladybird beetle, Coccinella septempunctata , which shows a monomorphic seven-spotted elytral colour pattern. The pannie r mRNA was detected in the larval elytral primordium, was upregulated from 24 h AP to 96 h AP (Supplementary Figure 9a ), and preferentially expressed in the black spots of elytra in C. septempunctata (Supplementary Figure 9b , b′) similarly to that in H. axyridis . The black-to-red switching phenotype was also observed in C. septempunctata adults treated with larval RNAi targeting pannier (Fig. 3a , C. septempunctata ). These data suggest that the elytral colour-patterning function of pannier may be conserved at the inter-genus level in ladybird beetles. To investigate the putative regulatory sequences at the pannier locus, we performed de novo assembly of the C. septempunctata genome using a linked-read sequencing platform (10× Genomics Chromium system), and obtained a contiguous genomic scaffold including the pannier locus (Fig. 5a , C. sep ) (2.41 Mb; Supplementary Figure 4d ; Supplementary Data 2 , C. septempunctata ). Whereas the noncoding sequences of C. septempunctata pannier are enriched with several species-specific DNA-binding motifs (Table 1 , C. sep ), we found DNA-binding motifs commonly enriched between Harmonia and Coccinella , which are associated with wing vein formation and wing/body wall patterning (Exd, Hth and Mad) 29 , 31 , 32 (Table 1 , commonly enriched motifs). Therefore, co-option of such wing developmental modules in the regulatory region may have facilitated acquisition of a novel expression domain of pannier in pupal elytral blades in ladybird beetles. In order to explore the history of the emergence of elytral colour patterns in H. axyridis , we also performed a molecular phylogenetic analysis focusing on the highly conserved pannier intronic sequences shared among H. axyridis alleles and C. septempunctata (three blocks, totalling 1.1 kb in length, Supplementary Data 5 ). The maximum likelihood (ML) phylogenetic tree inferred from nucleotide sequences of the pannier coding region did not resolve the phylogenetic relationship among the alleles in H. axyridis to a satisfactory level (Fig. 5c , bootstrap values <75). However, the ML tree inferred from the conserved intronic sequence suggested that in H. axyridis the contrasting colour patterns of the h allele (black spots in red background) and the other three alleles (red spots in black background) diverged first. The latter three alleles diverged more recently (Fig. 5d , bootstrap values >90). Discussion The pannier locus identified in this study appears to be the key genetic locus responsible for the origin of large-scale intraspecific variation genetically linked to the h locus in ladybird beetles 1 , 2 . Also, it is worth noting that a concurrent study by Prud’homme, Estoup and their colleagues independently identified the same locus in H. axyridis by whole-genome sequencing, population genomics, gene expression and functional genetics approaches 37 . Based on the results presented in this study, we propose an evolutionary model that might underlie the high level of diversification of the intraspecific elytral colour patterns of H. axyridis . In addition, we also discuss the underlying evolutionary developmental backgrounds specific to ladybird beetles. The common ancestor of Harmonia and Coccinella (Coccinellinae) diverged more than 33.9 million years ago, according to molecular phylogenetic analyses and fossil records 38 , 39 . Therefore, the elytral colour-patterning function of pannier shared between H. axyridis and C. septempunctata was most likely acquired before this divergence event. The 1.1 kb sequence blocks in the first intron of pannier conserved between H. axyridis and C. septempunctata are a likely candidate for a regulatory element associated with the ladybird beetle-specific elytral expression of pannier in the pupal elytra. The effects of enhancer activities of these sequence blocks have not yet been experimentally addressed. However, the acquisition of such regulatory sequences during evolution would have coincided with the acquisition of the elytral colour-patterning function of pannier (Fig. 6 , blue diamond). These conserved sequence blocks are located in the expanded intronic region specific to H. axyridis (Fig. 5a , black arrow). Therefore, the expansion of the first intron in the ancestral lineage of H. axyridis (Fig. 6 , intronic expansion) might be one of the events that facilitated diversification of the intraspecific elytral colour patterns. Fig. 6 An evolutionary model for the colour pattern diversification in H. axyridis . See details in the Discussion. The images of ladybird beetles are adapted from Figs. 1b and 3c Full size image In the genus Harmonia , colour patterns similar to those encoded by the h allele and those of C. septempunctata (black spots in red background) are commonly observed. Also, the position of the spots is similar across species (e.g. H. quadripunctata , H. octomaculata , and H. dimidiata ). Therefore, we speculate that the intronic sequence of pannier in the h allele of H. axyridis might retain a repertoire of regulatory sequences acquired in a common ancestor of the genus Harmonia (Fig. 6 , green arrowhead). However, in the ancestral lineage of H. axyridis , the regulatory region of pannier appears to have been modified to generate novel colour patterns of the recently diverged alleles ( h C , h Sp and h A ; red spots in black background; Fig. 6 magenta, red and purple arrowheads). The 70 kb-scale noncoding sequences located at the upstream region of the first intron of pannier that is specifically expanded in H. axyridis (Fig. 6 , Intronic expansion, yellow box) might have facilitated accommodation of the allele-specific regulatory motifs responsible for the diversified colour pattern of elytra. In addition, traces of inversions in this region consistently found in allele comparisons suggest that repeated inversions in this region (Fig. 6 , white arrowheads) created opportunities to diverge the noncoding sequence of pannier to successively generate novel diverse alleles within a species by suppressing recombination within this region. Such inversion events would have occurred in the common ancestor of H. axyridis and its reproductively isolated sister species, H. yedoensis because the major elytral colour patterns are shared between the two species 40 . Large-scale chromosomal inversion is believed to be one of the major driving forces generating and maintaining intraspecific morphological variation within a species 41 , 42 , 43 , 44 . Our study exemplifies that not only a single inversion event but also repeated inversion events at an expanded intron can lead to the acquisition of novel morphological traits within a species. From the viewpoint of evolutionary developmental biology, it is noteworthy that in H. axyridis , of all of the developmental genes known to regulate colour pattern and pigmentation, a single gene, pannier , is responsible for the major classes of intraspecific entire wing colour pattern diversification. This evolutionary pattern contrasts with that of the intensely studied warning signals of Heliconius butterflies. In the case of Heliconius erato and Helconius melpomene , five major loci and several minor loci located on different chromosomes regulate multiple intraspecific wing colour patterns prevailing in the population 45 . This difference in evolutionary mechanisms may stem from a paucity of available options of evolvable genes in the gene regulatory network of elytral colour patterning. Ladybird beetles diverged from ancestral species of Cucujoidea 38 (Fig. 5b , Cucujoidea), leaf-litter or rotten-tree dwelling insects. Thus, the ancestor of ladybird beetles would have had far less colourful and more simply patterned forewings (elytra) than the ancestors of butterflies, moths. Therefore, these ancestors presumably would have possessed far fewer colour pattern regulatory genes. In H. axyridis , this developmental constraint may have led to the selection of pannier as the major evolvable gene to a signal-integrating “input−output” regulatory gene 46 , 47 . This might have generated >200 colour patterns genetically tightly linked to the h locus by utilising the expanded regulatory DNA sequence. Future research aiming to identify specific regulatory inputs to pannier will help clarify the regulatory mechanisms underlying the generation of highly diverse intraspecific polymorphism at the interspecific level. Another important issue to clarify whether pannier is indeed the hotspot of morphological evolution in ladybird beetles is whether pannier is responsible for the remaining >20 minor colour patterns in H. axyridis . Methods Insects Laboratory stocks of H. axyridis and C. septempunctata were derived from field collections in Japan. They were reared at 25 °C and usually fed on artificial diet 48 , or fed on the pea aphid Acyrthosiphon pisum (kindly provided by Dr. T. Miura) for egg collection. Larvae and pupae analysed in this study were not sexed. Phenoloxidase (PO) activity staining Pupa elytral discs were dissected in a potassium phosphate buffer (K-PO 4 buffer; 100 mM KH 2 PO 4 /K 2 PO 4 , 150 mM NaCl, pH 6.3) on ice. PO staining was performed using 0.4 mg/ml dopamine as a substrate in 40% K-PO 4 buffer/60% isopropyl alcohol for 2 h at room temperature as previously described 12 . After washing several times in the potassium phosphate buffer containing 0.3% Triton-X100 and mounted in this solution. Images were captured with a stereoscopic microscope (MZ FLIII, Leica) equipped with a digital camera (DP70, Olympus). Histological analysis To visualise tissue morphology and PO active tissues, pharate adult elytra dissected in ice-cold PBS (137 mM NaCl, 2.68 mM KCl, 10.14 mM Na 2 HPO 4 , pH 7.2) at 96 h AP or those after PO activity staining were fixed with 4% paraformaldehyde (PFA) in PBS for 15 min on ice and for 75 min at room temperature. After fixation, the elytra were washed several times in 100% methanol and stored in 100% methanol at −20 °C until use. After dehydration, the elytra were embedded in 4% carboxymethyl cellulose (FINETEC), and were frozen in hexane cooled with dry ice. The freeze-embedded elytra were stored at −80 °C until use. The 6 μm frozen sections were prepared using an adhesive film (Cryofilm Type 1; FINETEC) 49 . Sections of the PO activity-stained elytra were dried at least 1 h at room temperature, mounted in PBS, and photographed under an inverted microscope (IX70, Olympus). For nuclear and F-actin staining, sections were treated with 2.5 μg/ml propidium iodide, 1 mg/ml RNase A and 5 U/ml AlexaFluor 488 phalloidin (Molecular Probes) for 1 h at 37 °C under a dark condition. After washing three times in PBS, the sections were mounted in an antifade reagent (FluoroGuard TM ; Bio-Rad), and images were captured with a confocal laser-scanning microscope (LSM 510; Carl Zeiss). For localisation of carotenoid, elytra at 96 h AP were embedded and sectioned as described above. All procedures were rapidly performed to prevent diffusion of carotenoids. The sections were dried for 1 min, mounted in PBS and immediately photographed under an inverted microscope (IX70, Olympus). cDNA cloning Larval and pupal elytral discs and pharate adult elytra of H. axyridis ( h C ) and C. septempunctata were dissected in PBS on ice. Soon after dissection, the tissues were frozen in liquid nitrogen and stored at −80 °C until use. Total RNA was extracted from each sample using TRIzol Reagent (Invitrogen) or RNeasy Micro Kit (Qiagen) according to the manufacturer’s instructions, and treated with 2 U DNase I (Ambion) for 30 min at 37 °C. The first-strand cDNA was synthesised with SMARTer PCR cDNA Amplification Kit (Clontech) using 1 μg of total RNA according to the manufacturer’s instructions. H. axyridis and C. septempunctata cDNA fragments were amplified by reverse transcription-polymerase chain reaction (RT-PCR) and rapid amplification of cDNA ends (RACE) with the primers listed in Supplementary Tables 2 , 3 . The PCR product was cloned into the Eco R V site of the pBluescript KS + vector (Stratagene) or pCR4-TOPO vector (TOPO TA Cloning Kit; Invitrogen). The nucleotide sequences of the PCR products were determined using a DNA sequencer 3130 genetic analyser (Applied Biosystems). The SNPs in open reading frame (ORF) of pannier were determined through direct sequencing of the PCR products treated with ExoSAP-IT (Affymetrix). Sequencing was performed by DNA sequencing service (FASMAC) using the primers listed in Supplementary Table 4 . Sequence analysis was carried out using DNASIS (Hitachi Software Engineering) or ApE 50 (version 2.0.45) software. Nucleotide sequences and deduced amino acid sequences were aligned with ClustalW in MEGA 51 software (version 7.0.18). The alignment figures were generated using Boxshade 52 (version 3.21). Gene expression analysis by RT-PCR For the gene expression analysis in each developmental stage, elytral tissues of three individuals of H. axyridis ( h C ) and C. septempunctata were dissected as described above. Six elytral tissues from each sampling stage were pooled in one test tube. Total RNA extractions and the subsequent first-strand cDNA syntheses (using 425 and 267 ng of total RNA for H. axyridis and C. septempunctata samples, respectively) were performed as described above. Three microlitres of 100 and 62.8 times diluted H. axyridis and C. septempunctata first-strand cDNA was used as a template for each PCR, respectively. The PCR cycle number was 35 for all genes. A set of primers #1 and #2 for each gene was used for this analysis (Supplementary Table 5 ). For the gene expression analysis in the future red and black regions, the red and black regions of pharate adult elytra at 84 h AP were collected by boring with injection needles. Internal diameters of the needles were 0.7 and 0.6 mm for H. axyridis and C. septempunctata , respectively. In the case of C. septempunctata , elytra stained with PO activity were used for boring because carotenoid localisation was not observable unlike in H. axyridis . cDNA synthesis was performed as described above, using as much total RNA as we could extract. Twenty microlitres of ten times diluted first-strand cDNA was used as a template for PCR. The PCR cycle numbers were 45 cycles for Ha-pnr , 38 cycles for Ha-rp49 , 47 cycles for Cs-pnr and 40 cycles for Cs-rp49 . A set of primers #3 and #4 for each gene was used for this analysis other than rp49 . Ha-rp49 and Cs-rp49 were used as internal controls. Reactions without reverse transcriptase were performed with cDNA synthesis as negative control samples for the RT-PCR experiments. No band was detected in these reactions for all genes. The primers used for this analysis are described in Supplementary Table 5 . Larval RNAi DsRNA synthesis and microinjection into larvae were performed as described previously 19 . In brief, the cloned cDNA fragments in DNA vectors were amplified by PCR using the primers flanked with the T7 promoter sequence (Supplementary Table 6 ), and used as templates for dsRNA synthesis. Amplified PCR products were separated by electrophoresis on 1% agarose gels and purified using MagExtractor PCR & Gel Clean up kit (Toyobo). DsRNAs were synthesised using the MEGA script T7 kit (Ambion). Approximately 1.4−2.7 μg and 1.4−2.0 μg of the dsRNAs of Ha-pnr and Cs-pnr were injected into 2-day-old forth (final) instar larvae, respectively. Approximately 2.0−2.7 μg and 1.4–2.7 μg of the EGFP dsRNA were injected into H. axyridis and C. septempunctata larvae as negative controls, respectively. For other genes in the initial small screening, approximately 1 µg of dsRNA was injected into each early last instar larva. Different amount of dsRNA for each gene in this range gave no difference in phenotypic effects. In order to give enough time for the completion of pigmentation, images of adults were captured more than 2 days after eclosion using a digital microscope (VHX-900, Keyence). In situ hybridisation Essentially the same protocol for whole mount pupal antennal primordia of the silk moth 53 was used. For sclerotized pupal elytra of ladybird beetles, several procedures were modified as follows: to increase RNA probe penetrance in elytral epidermis covered with sclerotised cuticle at 76−84 h AP, the peripheral edge of an elytron was cut off, and then, ventral and dorsal elytral epidermis layers appressed together were carefully separated with fine forceps after fixation; to reduce nonspecific probe hybridisation, fixed, separated and detergent-permeabilised elytra epidermal samples were stored in 100% methanol for more than 12 h at −30 °C, and prehybridisation treatment was extended to two overnight incubation; the concentration of cRNA probes were reduced to 0.4 ng/µl; the ventral epidermis samples were not used for analysis because of high nonspecific background signals. Sample washing was performed for 10 min three times unless otherwise noted. pannier antisense probes were designed at 5′ and 3′ regions of ORF excluding the two conserved GATA zinc finger coding regions in the middle to prevent cross-hybridisation with other GATA family genes. The PCR primers used to amplify the template DNA for in vitro RNA probe synthesis were listed in Supplementary Table 6 . Briefly, pannier ORF fragment was amplified by RT-PCR using cDNA from 72 h AP ( h C ), and cloned into pCR4-TOPO vector (Invitrogen). Sense and antisense probe templates were amplified from the cloned cDNA. Sense and antisense DIG-labelled riboprobes were transcribed using the flanking T7, T3 or SP6 promoter sequences, and DIG RNA labelling kit (Roche). Mixture of 5′ and 3′ probes was used for hybridisation. The concentrations of RNA probes were quantified by agarose gel electrophoresis. First, pupal elytra were dissected in PBS, and fixed with 4% PFA in PBS for 2 h, and washed with PBS including 0.1% Tween20 (PTw). Fixed elytra were dorsoventrally separated using fine forceps, and permeabilized with 0.5% Triton X-100 for 45 min. After washing with PTw, samples were treated with 20 μg/ml proteinase K in PTw at 37 °C for 30 min. Proteinase K was immediately washed out by quick washes with 2 mg/ml glycine in PBS and following washes with PTw. After postfixation with 4% PFA and 0.1% glutaraldehyde for 30 min, samples were equilibrated to hybridisation solution with five steps of hybridisation solution wash series (50, 75, 87.5, 100, 100%), and prehybridized at 57 °C over two nights (for ca. 40 h), and hybridised with 0.4 ng/μl cRNA probes overnight. After reverse hybridisation solution wash series (50, 25, 12.5, 0% (PTw) 0% (PTw)) and a wash with RNase A reaction buffer (10 mM Tris-HCl, 500 mM NaCl, pH 8.0), single stranded probes not hybridised to mRNA were degraded with 20 μg/ml RNase A at 37 °C for 30 min. After equilibrated to hybridisation solution again, nonspecifically bound degraded probes were washed out with 4 times of 20 min washes with hot hybridisation solution at 57 °C and following hot reverse hybridisation solution wash series (50, 25, 12.5%; 20 min for each). After washing with PBTw and PBSS (PBS with 0.01% Saponin), and blocking for 30 min with PBSS-BSA (PBSS with 0.2% bovine serum albumin), samples were incubated with alkaline phosphatase-conjugated anti-DIG Fab fragment (1:2000, Roche) in PBSS-BSA. After washing with PBSS and NTMT buffer (0.1 M NaCl, 0.1 M Tris-HCl (pH 9.5), 50 mM MgCl 2 , 0.1% Tween20), Colour development was conducted using 170 µg/ml BCIP (Roche) and 340 µg/ml NBT (Roche) diluted in NTMT buffer. After washing with NTMT buffer and PBSS, the solution was exchanged to 100% ethanol in two steps (50, 100%), and the samples were de-coloured for 1 h. After returning to PBSS in two steps, the samples were mounted in 80% glycerol, and observed under a stereomicroscope (Stemi 508, Zeiss). Images of the samples were collected using a digital camera system (NY-D5500 super system, Microscope Network). Brightness and contrast of each image was adjusted using Photoshop CS6 (Adobe). The same image processing was applied to all images. De novo genome assembly of H. axyridis A single female adult from F2-3 strain sibcrossed for three generations ( h C ) was used for the first version of de novo genome assembly of H. axyridis . Genomic DNA was extracted using DNeasy Blood and Tissue Kit (Qiagen). Paired-end (300 and 500 bp) and mate pair (3, 5, 8, 10, 12 and 15 kb) libraries were constructed using TruSeq DNA PCR-Free LT Sample Prep Kit and Nextera Mate Pair Sample Prep Kit (Illumina) following the manufacturer’s protocols. Sequencing libraries were run on Illumina HiSeq2500 sequencers. In total, we generated 133.6 Gb of raw sequence data for de novo genome assembly. Genome assembly was performed using the Platanus v1.2.1.1 assembler 54 after removal of adapter sequences and error correction (SOAPec v2.01) 55 . Reassembly of the genomic scaffold at H. axyridis pannier Adaptor sequences and low-quality regions in paired-end and mate-pair reads were trimmed using Platanus_trim 56 (version 1.0.7) with default parameters. Trimmed reads were assembled by Platanus2 57 (version 2.0.0), which was derived from Platanus 54 to assemble haplotype sequences (i.e. haplotype phasing) instead of consensus sequences. Procedures of Platanus2 are briefly described as follows: (1) De Bruijn graphs and scaffold graphs are constructed without removal of bubble structures caused from heterozygosity. Paths that do not contain junctions correspond to assembly results (scaffolds). Scaffold pairs in bubbles represent heterozygous haplotypes. (2) Paired-ends or mate-pairs are mapped to the graphs to detect links between bubbles, and linked bubbles are fused to extend haplotype sequences. (3) Each haplotype (contig or scaffold) is independently extended by modules of de novo assembly derived from Platanus. (4) Homologous pairs of haplotype scaffolds are detected using bubble information in the initial de Bruijn graph. (5) Steps 1–4 are iterated using various libraries (paired-ends or mate-pairs). (6) Homologous pairs of scaffolds are formatted into bubble structures as output. For each pair, longer and shorter scaffold were called “primary-bubble” and “secondary-bubble”, respectively. Primary-bubbles, secondary-bubbles and nonbubble scaffolds are collectively called “phased-scaffolds”. In addition, Platanus2 can connect primary-bubbles and nonbubble scaffolds to construct long “consensus scaffolds”, which consists of mosaic structure of haplotypes (i.e. paternal and maternal haplotypes are mixed). Employing the strategy of Platanus2, certain highly heterozygous regions were expected to be assembled contiguously compared to Platanus. Using the markers of the responsible region for elytral colour patterns (the h locus), we found that two long bubbles and one short nonbubble scaffold corresponded to the locus. Consequently, one consensus scaffold covering the breakpoint markers at the h locus was constructed from these phased scaffolds. We used that consensus scaffold (3.13 Mb) for the downstream in silico sequence analyses. We assessed the completeness of the genome assembly using BUSCO 58 (version 3.0.2, Insecta dataset (1658 orthologues)). Genome sequencing by long reads and linked-reads High molecular weight (HMW) genomic DNA was extracted using QIAGEN Genomic-tip 100/G (QIAGEN) according to the manufacturer’s instructions. The concentrations and qualities of the extracted HMW genomic DNA were evaluated using Qubit dsDNA, and RNA HS kits (Thermo Fisher). For library preparation for 10× Genomics Chromium system, one pupa ( h C (NT6 strain) and h (NT8 strain)) or one adult ( h A (F2 adult progenies in genetic cross h A × h Sp ) and C. septempunctata (MD-4 strain)) was used. Size selection by BluePippin (range: 50 kb−80 kb, Sage Science) was performed only for h A genomic DNA used in 10× linked-read library preparation. Preparation of gel bead-in-emulsions (GEMs) for each 10× Genomics Chromium library was performed using 0.5–0.6 ng of HMW genomic DNA according to the manufacturer’s instructions. The prepared GEMs were quality-checked using Qubit dsDNA HS kit (Thermo Fisher) and Bioanalyzer (Agilent), and processed with Chromium Controller (10× Genomics). The constructed DNA libraries were quality-checked again in the same way. Sequencing of the libraries was performed in the Hiseq X ten (Illumina) platform (1 library/lane) at Macrogen. In total, we generated 66.9, 64.6, 64.9 and 60.3 Gb of raw reads for linked-read sequencing ( h C , h A , h , and C. septempunctata , respectively). For library preparation for PacBio system, 10−11 pupae ( h C (NT6 strain) and h (NT8 strain)) or adults ( h A (F2 adult progenies in genetic cross h A × h Sp )) were used. The libraries were prepared according to the 20-kb Template Preparation Using BluePippin™ Size-Selection System (Sage Science). Sequencing of the libraries was performed in the PacBio RS II (Pacific Biosciences) platform. In total, 4.31, 4.92 and 4.44 Gb of insert sequences (approximately 10× coverage of the genome, assuming a genome size of 423 Mb) were obtained from 4 to 5 SMRT cells for h A , h C and h , respectively. De novo assembly of 10 ⨯ linked-reads For 10× linked-reads libraries of four samples (three H. axyridis and one C. septempunctata ), Supernova (version 2.0.0) 59 was executed with default parameters except for the maximum number of used reads (the --maxreads option) to obtain the optimum coverage depth for Supernova (56×). For each sample, the value for --maxreads was determined as follows: (1) Barcode sequences in raw linked-reads were excluded using “longranger basic” command of Long Ranger 60 (version 2.1.2), resulting in “barcoded.fastq” file. (2) Adaptor sequences and low-quality regions in “barcoded.fastq” were trimmed using Platanus_trim (version 1.0.7) with default parameters. (3) 32-mers in the trimmed reads were counted by Jellyfish 61 (version 2.2.3) using the following two commands and options: $ jellyfish count -m 32 -s 20M -C -o out.jf barcoded_1.trimmed barcoded_2.trimmed $ jellyfish histo -h 1000000000 -o out.histo out.jf In summary, all 32-mers in both strands (-C) were counted and distribution of the number of occurrences without upper limit of occurrences (-h 1000000000). (4) The haploid genome size was estimated using the custom Perl script. For the distribution of the number of 32-mer occurrences (“out.histo”), the number of occurrences corresponding to a homozygous peak was detected, and the total number of 32-mers was divided by the homozygous-peak-occurrences. Here, 32-mers whose occurrences were small (<the number of occurrences corresponding to the bottom between zero and heterozygous peak) were excluded for the calculation to avoid the effect from sequencing errors. (5) The values for --maxreads were calculated as follow: estimated-haploid-genome-size / mean-read-length-of-barcoded.fastq × 56 As a result, we obtained the scaffolds including the genes surrounding H. axyridis - pannier ( h C (NT6), 2.74 Mb; h A (F2 hybrid), 1.42 + 1.61 Mb; h C (NT8) 2.79 Mb), and homologous regions in C. septempunctata (haplotype 1, 10.16 + 2.41 Mb; haplotype 2, 10.13 + 2.44 Mb). We used those sequences for the downstream in silico analyses. We assessed the completeness of the genome assembly using BUSCO 58 (version 3.0.2, Insecta dataset (1658 orthologues)). Gap filling of the genomic scaffolds at the pannier locus Concerning the genome assemblies of H. axyridis , we used minimap2 62 (ver. 2.9) and PBjelly 63 (ver. PBSuite_15.8.24) software to fill gaps around the pannier locus. In each genome of three strains of H. axyridis , we first mapped PacBio reads to the genome assemblies generated from the 10× linked-reads using minimap2. Then, we chose PacBio reads mapped to the scaffold containing pannier gene. These PacBio reads were subjected to gap-filling of the scaffold with PBjelly. We obtained gap-free nucleotide sequences spanning the entire pannier locus and the upstream intergenic regions. Concerning the genome assembly of C. septempunctata , there was a single gap estimated to be 15 kb long by Supernova program in the first intron of pannier locus. We handled this gap region as repeated N , and included it in the downstream in silico analyses. Validation of the pannier scaffold re-assembled by Platanus2 For the H. axyridis F2-3 sample, trimmed reads of the 15 kbp-mate-pair library were mapped to the consensus scaffold set of Platanus2 using BWA-MEM 64 (version 0.7.12-r1039) with default parameters. Next, a consensus scaffold corresponding to the pannier locus was segmented into 2 kbp-windows, and links between windows (≥3 mate-pairs) were visualised by Circos 65 (version 0.69-6). Preliminary resequencing of H.axyridis genome for RAD-seq Genomic DNA was extracted from each of h C (F6 strain), h A ( NT3 strain) and h Sp (CB-5 strain), and used to create Illumina libraries using TruSeq Nano DNA Sample Preparation Kit (Illumina) with insert size of approximately 400 bp. These libraries were sequenced on the Illumina HiSeq 1500 using a 2 × 106-nt paired-end sequencing protocol, yielding 84.7 M paired-end reads. SNP site identification was conducted basically according to the GATK Best Practice 66 (ver. August 7, 2015). After trimming adaptor sequences with Cutadapt software (ver. 1.9.1) 67 , the sequence data were mapped to the de novo genome assembly data using bwa software (ver.0.7.15, BWA-MEM algorithm) 64 . Sequences and alignments with low quality were filtered using Picard 68 tools (ver. 2.7.1) and GATK 66 software (ver. 3.6 and 3.7), and 734,443 SNP markers in the strains were identified. The most distantly related strains ( h C (F6 strain) and h A (NT3 strain)) were selected for the RAD-seq analysis by performing phylogenetic analysis using SNPhylo 69 (Version: 20140701). Comparison of the genomic scaffolds at pannier in ladybirds For each pair of the entire scaffolds and the extracted pannier region, we constructed dot plots by performing pairwise-alignment using “nucmer” program in the MUMmer package 70 (version 3.1). The options of nucmer were as follows: (1) the entire scaffolds, H. axyridis vs. H. axyridis , Default parameters; (2) the entire scaffolds, H. axyridis vs. C. septempunctata , “-l 8 -c 20”; (3) the pannier region, H. axyridis vs. H. axyridis , “-l 12”. Alignment results (delta files) were input into “mummerplot” program to generate dot plots. Note that resultant gnuplot scripts resulting from mummerplot were edited for visualisation. We also visualised the homology and structural differences between the 700 kb-genomic region including pannier using Easyfig 71 (ver. 2s2.2). Short BLAST 72 hit fragments less than 500 bp, and putative short repeat sequences less than 1250 bp, which showed more than two BLAST hit blocks within the 700 kb region, were filtered using a custom Perl script. Exon−intron structures of putative genes in the 700 kb regions were obtained using Exonerate 73 (ver. 2.2.0) with the options “-m est2genome --showvulgar yes --ryo”>%qi length=%ql alnlen=%qal/n>%ti length=%tl alnlen=%tal/n” --showtargetgff yes --showalignment no --score 2000′. cDNA sequences cloned by RT-PCR or predicted by RNA-seq were used as queries. If a single cDNA unit was split into multiple fragments, we merged the fragments by performing exonerate search again using the cDNA sequences whose subsequences were substituted by the genomic hit fragments in the first exonerate search as a query. Exonerate output files were converted to the GFF3 format using our bug-fixed version of the process_exonerate_gff3.pl 74 Perl script with the option “–t EST”. The GFF3 file and a FASTA format file of each scaffold were converted to a GENBANK format file using EMBOSS Seqret 75 program (ver. 6.6.0.0) with the options “-fformat gff -osformat genbank”. The GENBANK format files corresponding to the 700 kb genomic sequences surrounding pannier , which were used as input files of Easyfig, were extracted using the Genbank_slicer.py 76 Python script (ver. 1.1.0). Flexible ddRAD-seq We newly constructed a flexible ddRAD-seq library preparation protocol to facilitate high-throughput ddRAD-seq analyses at low cost. We designed all enzymatic reactions to be completed sequentially without DNA purification in each step to make the procedures simple. In addition, we designed 96 sets of indexed and forked sequencing adaptors compatible with Illumina platform sequencers (Supplementary Data 6 ). Briefly, 100 ng of genomic DNA was first double-digested with 15 U of Eco RI-HF and 15 U of Hin dIII-HF in 20 µl of NEB CutSmart Buffer (New England Biolabs) at 37 °C for 2 h. Fifteen microlitres of the digested DNA, 4 pmol of adaptor DNA, 10 µmol of ATP, 400 U of T4 DNA ligase were mixed in 20 µl, incubated at 22 °C for 2 h, and denatured at 65 °C for 10 min. Ligated library DNA fragments were purified with Agencourt AMPureXP (Beckman Coulter) according to the manufacturer’s instructions. Library DNA fragments ranging from 300 to 500 bp were size-selected with Pippin Prep (Sage Science). Concentration of each library DNA was quantified using KAPA Library Quantification Kits (Roche) according to the manufacturer’s instructions. Sequence data were obtained by applying 96 DNA libraries to a single lane of Hiseq 1500 (Illumina). Linkage map construction and a genome-wide association study A single h C male (F6 strain) and a single virgin h A female ( NT3 strain) were crossed, and the obtained F1 progenies were backcrossed with the F0 male ( h C , F6 strain). Finally, 183 adult F2 progenies ( h C = 80, h C / h A = 103) and 2 F0 adults were collected for RAD-sequence analysis, and stored at −30 °C until use. Genomic DNAs were extracted individually using an automatic nucleic acid extractor (PI-50α, KURABO). Briefly, each frozen ladybird beetle and a zirconia bead were transferred to 2 ml plastic tube (Eppendorf) on ice. Immediately, 250 µl of cold lysis buffer including Proteinase K and RNase A, but not SDS was added to the sample, and the tubes were vigorously shaken with a tissue grinder (Tissue Lyser LT, Qiagen) at 3000 rpm for 1 min. Then, 250 µl of lysis buffer including SDS was added to each crushed sample, and processed with the automatic program for DNA extraction from mouse tail, according to the manufacturer’s instructions. The extracted genomic DNA was diluted in 30 µl of TE buffer. The DNA concentration of each sample was quantified using Qubit dsDNA BR Kit according to the manufacturer’s instructions (Thermo Fisher Scientific). We performed flexible ddRAD-seq using these genomic DNA samples. 0.6–6.0 million (mean = 2.0 million) of 106 bp paired-end reads per sample were generated using two lanes of Hiseq 1500 (Illumina) following the methods in the User Guide. Mapping and polymorphic site calling were conducted as described in the resequencing analysis above except that the procedure for filtering duplicated reads using Picard was eliminated because we did not amplify DNA library by PCR. Count data at each SNP sites were extracted from the obtained vcf file using vcf_to_rqtl.py script in rtd software 77 with the options “5.0 80”. To avoid program errors, we modified the script to skip the read depth data (DP) including characters in the GATK vcf file. We constructed a linkage map using R/qtl 78 (version 1.42.8) and R/ASMap 79 (version 1.0.2) R 80 packages according to the QTL mapping workflow for BC1 population of Jaltomata 81 . Using the obtained csv file as an input, we eliminated the polymorph sites that behaved as located on the X chromosome. In addition, individuals with low mapping quality, and marker sites with low-quality or highly distorted segregation patterns were eliminated as well. Finally, 4419 markers sites and 177 F2 individuals were used. The linkage map was initially constructed with mstmap program (R/ASMap) with the options “dist.fun = ‘kosambi’, p.value = 1e-25”, and highly linked linkage groups were merged manually. The markers consistently incongruent with neighbouring markers were eliminated using correctGenotypes.py 81 Python script with the options “-i csvr -q 0.1 -t 4.0”. A genome-wide association study (GWAS) was conducted using calc.genoprob program (R/qtl) with the options “step = 1, error.prob = 0.001” and scanone program (R/qtl) with the option “model = ‘binary’”. The result data were visualised with the “plot” program in R/ASMap. Genetic association studies focusing on the pannier locus In addition to the genetic cross in the GWAS ( h C × h A ), two independent crosses (( h × h C ) and ( h A × h Sp )) were performed. In the former cross, a single h male (D-5 strain) and a single virgin h C female ( F2-3-B strain) were crossed, and the obtained F1 progenies were sibcrossed. Finally, 80 F2 adult progenies ( h C = 30, h C / h = 34, and h = 16) were collected for genotyping, and stored at −30 °C until use. In the latter cross, a single h Sp male (CB-5 strain) and a single virgin h A female ( NT3 strain) were crossed, and the obtained F1 progenies were sibcrossed. Finally, 273 F2 adult progenies ( h Sp = 103, h Sp / h A = 80, and h A = 90) were collected for genotyping, and stored at −30 °C until use. Genomic DNA was extracted individually using the automatic nucleic acid extractor (PI-50α, KURABO) as described in the previous section, and diluted to approximately 100 ng/µl. We searched for genotyping markers by amplifying and sequencing the intronic region of the genes surrounding pannier with PCR. The individual PCR was performed using approximately 100 ng of genomic DNA and Q5 DNA polymerase (New England Biolabs) with 45 cycles. The primers used, the markers identified and the typing results are summarised in Supplementary Data 1 . RNA-seq analysis The total RNA extraction procedure for RNA-seq is essentially the same as that for the gene expression analysis in the presumptive red and black regions by RT-PCR. The same strain used for de novo genome sequencing (F2-5 strain, h C ) was used. In total, 12 samples (2 colours [Black/Red] × 2 developmental stages [24 h AP/72 h AP] × 3 biological replicates) were prepared for RNA-seq analysis. Two fragments of bored epidermis from left and right elytra were collected as a single sample in each condition. All total RNA extracted from each sample (12–158 ng) using RNeasy Mini Kit (QIAGEN) and QIAcube (QIAGEN) was used for each cDNA library preparation. RNA-seq library preparation was performed using the SureSelect strand-specific RNA library prep kit (Agilent) according to the manufacturer’s instructions. Briefly, mRNA was purified using Oligo-dT Microparticles. The strand-specific RNA-seq libraries were prepared using dUTP and Uracil-DNA-Glycosylase. The libraries and its intermediates were purified and size-fractionated by AMPure XP (Beckman Coulter). For quality check and quantification of the RNA-seq libraries, we employed 2100 Bioanalyzer and DNA 7500 kit (Agilent). 100 bp paired-end read RNA-seq tags were generated using the Hiseq 2500 (Illumina) following the methods in the User Guide. In advance of reference mapping, adaptor and poly-A sequences were trimmed from raw RNA-seq reads by using Cutadapt (ver. 1.9.1) 67 . Low-quality reads were also filtered out by a custom Perl script as described previously 82 . The preprocessed RNA-seq reads were mapped to the reference H. axyridis genome (assembly version 1) using TopHat2 83 (ver. 2.1.0) with default parameters, and assembled by Cufflinks 84 (ver. 2.2.1) with the -u option in each sample. All predicted transcript units and all loci from different samples were merged by Cuffmerge in the Cufflinks suite. The RNA-seq read pairs (fragments) mapped to each predicted transcript unit and locus were counted using HTSeq 85 (ver. 0.6.1) with the options “-s no -t exon -i transcript” and “-s no -t exon -i locus”, respectively. The downstream statistical analyses were performed using edgeR 86 , 87 package (ver. 3.16.5). The raw RNA-seq fragment counts were normalised by the trimmed mean of M-values (TMM) method. Fold change between black and red regions in each stage and its statistical significance (FDR) were calculated. The mean fold changes of the genes in the scaffolds including the h locus candidate region were visualised with IGV 88 , 89 software (ver. 2.3.88). Comparison of the pannier locus size The holometabolous insects, whose genomic sequences are well assembled at the pannier locus to the extent that at least the two paralogous GATA transcription factor genes, GATAe and serpent , are included in the same scaffold, were selected for comparison. Concerning Coleoptera, genomic sequences were collected from the Genome database at NCBI 90 (GCA_000002335.3, Tcas5.2; GCA_001937115.1, Atum_1.0; GCA_000390285.2, Agla_2.0; GCA_000648695.2, Otau_2.0; GCA_001412225.1, Nicve_v1.0; GCF_000699045.1, Apla_1.0; GCA_002278615.1, Pchal_1.0) and Fireflybase 91 ( Photinus pyralis genome 1.3, Aquatica lateralis genome 1.3). Concerning holometabolous insect other than Coleoptera, genomic information at Hymenoptera Genome Database 92 (Hymenoptera) (GCF_000002195.4, Amel_4.5; GCF_000217595.1, Lhum_UMD_V04; GCF_000002325.3, Nvit_2.1), Lepbase 93 (Lepidoptera, butterfly; Danaus_plexippus_v3_scaffolds), SilkBase 94 (Lepidoptera, silk moth; Genome assembly (Jan. 2017)), Flybase 95 (Diptera, Drosophila , dmel_r6.12_FB2016_04), and the Genome database at NCBI 90 (Diptera, mosquitos; GCA_000005575.1, AgamP3; GCA_002204515.1, AaegL5.0) (Lepidoptera, moth; GCA_002192655.1, ASM219265v1) were utilised. We performed BLAST 72 search (TBLASTN, ver. 2.2.26) using the amino acid sequence of H. axyridis Pannier as a query, and identified pannier orthologues by focusing on the top hits, and the conserved synteny of the three paralogous GATA transcription factor genes, in which serpent , GATAe , and pannier are tandemly aligned in this order from the 5′ to 3′ direction. The sizes of the upper noncoding region of pannier were estimated by calculating the difference between the coordinates of the 3′ end of BLAST hit region of GATAe and the 5′ hit region of pannier . Traces of translocation or insertions between the paralogous GATA genes were surveyed by looking into the annotations between the GATAe and pannier loci. If such a genomic rearrangement was found, we recalculated the size of the upstream region of pannier using the neighbouring gene. If the exon 1 of pannier was being annotated as an exon distinct from that including the initiation codon as in H. axyridis , the size of the first intron was calculated as well. Motif enrichment analysis To search for the DNA-binding sites of known transcription factors at pannier , 1139 DNA-binding motifs of Drosophila transcription factors were retrieved from the JASPAR 96 database using the MotifDb 97 R package. Concerning the SD-binding motif, the position weight matrix (PWM) scores were calculated using the 2557 ChIP-seq peaks in Drosophila genome obtained by Ikmi et al 98 . and the RSAT peak-motifs program 99 . The nucleotide sequences at the three upper noncoding regions of pannier (the upper intergenic region, the upstream half of the first intron, and the downstream half of the first intron) were collected by forging BSgenome 100 data packages using our H. axyridis and C. septempunctata genome sequences, and by retrieving the sequences using coordinate information obtained for annotation in Fig. 5a and the GenomicFeatures 101 R package. We here defined the upstream half of the first intron as the region including all of the traces of inversions or corresponding sequences shared among different h alleles in H. axyridis . GRange objects were generated using the coordinate information obtained for annotation in Fig. 5a . Motif enrichment was quantified using the PWMEnrich 102 R package. As control background genomic regions for the upper intergenic regions of pannier , we used 2 kb promoter sequences of 11,279 genes to which RNA-seq reads were mapped ( H. axyridis genome assembly version 1), but which was not located within the 10 kb from both ends of each genomic scaffold. As control background genomic regions for the upstream and downstream regions of the first intron of pannier , we used 2 kb sequences at the 5′ and 3′ end of the first introns, which are more than 2 kb long, and without gaps (2825 and 2810 sequences, respectively). Since there is no reliable gene annotation data for the C. septempunctata genome, we used the H. axyridis genomic background sequences. Each motif enrichment score, which is related to average time that transcription factors spend in binding to a DNA sequence 103 , was calculated using default parameter of PMWEnrichment. br and da DNA-binding motifs were excluded from the analysis. Molecular phylogenetic analysis Concerning the coding region of pannier , nucleotide sequences of the cloned cDNAs were aligned, and trimmed in the same way (Supplementary Data 5a, b ). Concerning the conserved regions in the upper half of the first intron of pannier , nucleotide sequences were collected from the BLAST hits obtained to construct Fig. 5a . The collected BLAST hit sequences were arranged in the same directions using a custom Perl script, and aligned using MAFFT 104 (ver. 7.222). We concatenated three alignment blocks, and manually excluded GAP sites and seemingly nonhomologous sites in the alignment (Supplementary Data 5c, d ). The ML phylogenetic trees were constructed using RAxML 105 (ver. 8.0.0) with the options “--maxiterate 1000 --localpair --clustalout”. We determined appropriate models of sequence evolution under the AICc4 criterion using Kakusan4 106 . One hundred replicates of shotgun search for the likelihood ratchet were performed. Nodal support was calculated by bootstrap analyses with 1000 replications. Data availability The cloned cDNA sequences were deposited in the DNA Data Bank of Japan (DDBJ) under the accession numbers LC269047 – LC269055 ( Ha-pnr ), LC269056 ( Cs-pnr ), and LC269057 ( Cs-rp49 ). The genomic sequencing data, the genomic resequencing data (short reads, RAD-seq, linked-reads, and long reads), and the RNA-seq data were deposited in DDBJ under the accession numbers DRA002559 , DRA006068 , DRA007003 , DRA007002 , DRA007004 , and DRA005777 , respectively. The assembled genomic sequences were deposited in DDBJ under the accession numbers BHEG02000001−BHEG02018515 ( H. axyridis genome assembly version 1), BHEG02000001−BHEG02018515 ( H. axyridis genome assembly version 2), BHEF01000001−BHEF01044316 ( H. axyridis linked-read genome assembly, h C , NT6 strain), BHEE01000001–BHEE01050762 ( H. axyridis linked-read genome assembly, h A , F2 hybrid), BHED01000001–BHED01050274 ( H. axyridis linked-read genome assembly, h , NT8 strain), BHEC01000001–BHEC01055573 ( C. septempunctata linked-read genome assembly, MD8 strain), and AP018896 – AP018898 (the H. axyridis gap-filled genomic scaffolds including the pannier locus).

Many ladybirds have attractive color patterns consisting of black and red. This prominent color pattern is thought to function as a warning that indicates to predators that they are very bitter and unpalatable. A research team led by Professor Teruyuki Niimi at the National Institute for Basic Biology in Japan focused on the multicolored Asian ladybird beetle Harmonia axyridis (also known as the harlequin ladybird), which lives mainly in Siberia and East Asia, and shows >200 color patterns within a species. The team has identified a single gene that regulates such highly diverse ladybird color patterns. Their genetic and genomic analyses in H. axyridis identified a single gene: pannier. The pannier gene was expressed in the black pigmented regions where the red pigment is not deposited. Functional inhibition of the pannier gene during pupal development resulted in the loss of the black color patterns and ectopic red pattern formation in the forewing. Therefore, the pannier gene has dual functions: promotion of black pigmentation and suppression of red pigmentation in the forewing. A similar result was also obtained in the seven-spot ladybird (Coccinella septempunctata), which lives all over the world. Professor Niimi stated, "The multicolored Asian lady beetle (Harmonia axyridis) shows extraordinary intraspecific variation in wing colour patterns (>200 described colour forms), and seemed to be the best model to elucidate how insects have generated the diverse traits during evolution at the molecular level. In order to investigate the cause of these diverse morphological patterns, we sequenced the genomes of the multicolored Asian ladybeetle (Harmonia axyridis), and the seven-spot ladybird (Coccinella septempunctata) in collaboration with the specialists at the National Institute of Genetics and others." In the genomic analyses of ladybird beetles, they found that the DNA sequences of the pannier gene in the multicolored Asian ladybird beetle was greater compared to other holometabolous insects including the seven-spot ladybird which shows a single color pattern. Genomic sequence comparison among ladybird beetles with different color patterns in H. axyridis revealed that the difference in color pattern is associated with the difference in the DNA sequence of the first intron of the pannier gene. In addition, they found traces of repeated chromosomal inversion within the pannier intron. Assistant Professor Toshiya Ando who performed genomic analysis said "Until now, it has not been reported that morphological diversification within a species was associated with repeated inversions within a single developmental gene. Our findings have shed light on intragenic chromosomal inversions as a driving force for morphological evolution of organisms." A mechanism of color pattern formation in ladybird beetles Credit: NIBB

10.1038/s41467-018-06116-1

Nano

Graphene: The more you bend it, the softer it gets

Edmund Han et al, Ultrasoft slip-mediated bending in few-layer graphene, Nature Materials (2019). DOI: 10.1038/s41563-019-0529-7 Journal information: Nature Materials

http://dx.doi.org/10.1038/s41563-019-0529-7

https://phys.org/news/2019-11-graphene-softer.html

Abstract Continuum scaling laws often break down when materials approach atomic length scales, reflecting changes in their underlying physics and the opportunities to access unconventional properties. These continuum limits are evident in two-dimensional materials, where there is no consensus on their bending stiffnesses or how they scale with thickness. Through combined computational and electron microscopy experiments, we measure the bending stiffness of graphene, obtaining 1.2–1.7 eV for a monolayer. Moreover, we find that the bending stiffness of few-layer graphene decreases sharply as a function of bending angle, tuning by almost 400% for trilayer graphene. This softening results from shear, slip and the onset of superlubricity between the atomic layers and corresponds with a gradual change in scaling power from cubic to linear. Our results provide a unified model for bending in two-dimensional materials and show that their multilayers can be orders of magnitude softer than previously thought, among the most flexible electronic materials currently known. Main Two-dimensional (2D) materials exhibit a host of unusual properties that arise from their anisotropic atomic structure and bonding. For example, the 3D Young’s modulus of few-layer graphene (FLG) is 1 TPa (ref. 1 ), three orders of magnitude larger than its 4.6 GPa shear modulus 2 . Bending, a process that couples in-plane and out-of-plane motion, provides an opportunity to test the effects of anisotropy on the mechanical properties of 2D materials. In particular, interlayer slip has been shown to be a dominant mechanism for relieving stress at van der Waals interfaces 3 , 4 and in multiwalled carbon nanotubes 5 and nanotube bundles 6 . Slip should have similarly important impacts on the bending properties of 2D materials. Bending stiffness takes on practical significance in a new generation of devices where 2D materials are highly curved and shaped into complex 3D architectures 7 , 8 , 9 , 10 , 11 . Highly curved 2D materials have promise across soft robotics and deformable electronics because they combine the high charge carrier mobilities of hard materials with the pliability of soft materials. In these systems, the bending stiffness governs the 3D nanoscale deformations of 2D materials, the structure and onset of folding 12 , rippling 13 , buckling 13 and crumpling 7 , as well as the interfacial mechanics 14 , 15 of deformed structures on surfaces. Yet, after more than a decade, there is still no single model that describes the widely divergent measurements of bending stiffness in monolayer graphene and FLG 2 , 12 , 16 , 17 , 18 . Unlike graphene’s well-known in-plane properties such as its Young’s modulus and breaking stress 1 , 19 , 20 , the small bending stiffness of FLG has proven difficult to characterize. FLG exhibits markedly different mechanical properties from bulk graphite 1 , 21 , 22 , 23 , 24 because structural imperfections in bulk systems overshadow the intrinsic properties of the individual atomic layers. In the few-layer limit, techniques such as nanoindentation, electrostatic actuation, atomistic simulations and measurements of nanoscale fold morphologies 12 , 17 , 18 , 25 , 26 , 27 , 28 , 29 have produced a broad range of bending stiffness for FLG, which appear to be in conflict. For monolayer graphene, literature values for its bending stiffness range from 0.83 to 10,000 eV (refs. 9 , 16 , 27 , 28 , 30 , 31 , 32 ). Furthermore, the reported bending stiffness of bilayer graphene ranges across two orders of magnitude, from 3.4 to 160 eV, while values for trilayer graphene range from 7 to 690 eV (refs. 12 , 17 , 18 , 27 , 31 ). There is also little agreement on the mechanisms and scaling laws that govern the bending of FLG: linear, quadratic and cubic scaling of bending stiffness with thickness have all been reported 2 , 12 , 16 , 17 , 18 . Some inconsistencies arise from the difficulty of measuring the intrinsic bending stiffness of FLG without the contributions from extrinsic stiffening from out-of-plane corrugations or in-plane strain. Another possibility is that these discrepancies reflect real differences in bending stiffness in different physical regimes. When highly bent, 2D materials may be governed by fundamentally different mechanics than in nearly flat geometries, and they may undergo dynamic transitions in bending properties as they are bent and flattened during operation. Yet, the mechanical behaviour of 2D materials spanning from low to high bending regimes is not well understood. In this work, we aim to produce a unified understanding of bending in few-layer 2D materials across curvature regimes. To tackle this challenge, we fabricated heterostructures of FLG draped over atomically sharp steps of hexagonal boron nitride (hBN), as illustrated in Fig. 1a . These structures allow us to systematically vary the thickness and degree of curvature of the graphene, then analyse their bending through cross-sectional imaging with aberration-corrected scanning transmission electron microscopy (STEM). As shown in Fig. 1b , the graphene is placed perpendicularly across hBN steps using established dry aligned transfer techniques (see Methods ). We confirmed the cleanliness and uniformity of the heterostructure with atomic force microscopy (AFM) and Raman spectroscopy (Supplementary Methods 1.1 and Supplementary Figs. 1 and 2 ), then prepared cross-sectional TEM samples using standard focused ion beam lift-out procedures (see Methods ). An example of a resulting sample is shown in the low-magnification STEM image in Fig. 1c ; each sample contains multiple hBN steps with varying heights. Figure 1d–i shows annular dark-field (ADF) STEM images of FLG on hBN steps with varying FLG thickness and hBN step height. Electron microscopy provides a powerful platform for measuring the mechanical properties of nanomaterials by enabling studies of their conformation and strain at atomic resolution 28 , 33 , 34 , 35 , 36 . We imaged 22 individual steps with FLG thicknesses of 1–12 layers and hBN step heights of 1–19 layers (for the raw images of each step see Supplementary Figs. 3–5 ). In these images, the bending profile of FLG and the corresponding mechanics are described by two critical parameters, the radius of curvature R and bending angle θ , as illustrated in Fig. 1g . Throughout the text, we define the bending angle as the angle subtending the two lines perpendicular to the straight sections on either side of the bend. We observe a wide range of bending angles (8.5–63°) and radii of curvature as small as 1.0 nm, comparable to the inner radii of carbon nanotubes. Fig. 1: Fabrication and STEM imaging of curved FLG on hBN steps. a , Schematic of the heterostructure. A graphene bilayer (black) is shown on top of a hBN step (red). b , Optical image of FLG transferred over an exfoliated hBN flake. Colour gradient of hBN indicates varying thickness and presence of steps in the flake. Scale bar, 10 μm. c , Low-magnification ADF-STEM image of the stair-step structure in cross-section. From bottom to top, the layers in cross-section are Si/SiO 2 /hBN/FLG/amorphous carbon/Pt (not all layers are distinguishable). Most prominent are two large hBN steps. Scale bar, 100 nm. d – i , ADF-STEM images of N -layer graphene over H -layer-thick hBN steps. We observe varying graphene bending profiles for different N and H . We parameterize the bending profile according to the radius of curvature R , bending angle θ and step height H , as indicated in g . Throughout the text, we define the bending angle as the angle subtending the two lines perpendicular to the straight sections on either side of the bend. Scale bars, 2 nm ( d – i ). Full size image Here, we study how the conformation of FLG (described by R and θ ) varies with the controlled parameters: FLG thickness ( N ) and hBN step height ( H ). Figure 2a shows the bending angle as a function of hBN step height for two different FLG thicknesses, while Fig. 2b shows the radius of curvature versus the number of graphene layers, colour-coded by hBN step height. These plots demonstrate clear relationships between the geometric parameters: higher bending angles are associated with taller steps and thicker FLG samples have larger radii of curvature. To relate the bending profile of FLG to its mechanics, we apply a simple model in which the conformation is governed by competition between the graphene/hBN interfacial adhesion energy and the FLG bending energy. Solving for the minimum energy, we obtain an equation that relates the bending stiffness of FLG to its equilibrium conformation (see Supplementary Methods 1.2 and Supplementary Fig. 6 ). This model is similar to the elastic shell model used to describe carbon nanotube mechanics 37 , 38 . We use this equation to calculate the bending stiffness from geometric parameters measured in the STEM images: $$B=R\varGamma \left(\frac{H-2R(1-\cos \theta )}{{\sin }^{2}\theta }\right)$$ (1) where \(B\) is the bending stiffness, \(\varGamma =0.126\) J m −2 is the graphene/hBN interfacial adhesion energy 39 , H is the hBN step height, R is the radius of curvature and θ is the bending angle. This equation assumes that the in-plane strain energy is negligible because incommensurate hBN/graphene interfaces are superlubric 3 , which, in the absence of interfacial contamination, prevents the build-up of in-plane stress in the laminated top-layer graphene 40 . See Supplementary Methods 1.3 for an additional discussion of error sources. Fig. 2: Measurement of bending stiffness from STEM images. a , Plot of bending angle versus step height H for two- and seven-layer graphene. For a given number of graphene layers N , the bending angle increases with hBN step height. b , Plot of radius of curvature versus thickness for all step heights and graphene thicknesses. Each point is colour-coded by hBN step height. The radius of curvature generally increases with thickness. c , Experimental measurements of bending stiffness versus thickness for FLG. Experimental values (black) are extracted using geometric parameters from the STEM images. Bending models representing cubic (blue) and linear (red) scaling are shown for comparison. Error bars for geometric parameters and bending stiffness values represent 95% confidence bounds as determined by four independent geometric measurements and their error propagation. Full size image We first use these methods to extract the bending stiffness of monolayer graphene. Analysing the two steps shown in Supplementary Fig. 3a,b we obtain bending stiffnesses of \({B}_{\mathrm{ml}}=1.2\pm 0.11\) eV and \({B}_{\mathrm{ml}}=1.7\pm 0.50\) eV, respectively (the stated error represents the 95% confidence bounds). We also performed density functional theory (DFT) analysis of monolayer graphene and found a deformation-independent bending stiffness of 1.4 eV (Supplementary Fig. 7 ). Although the bending stiffness of monolayer graphene has long been a topic of debate 9 , 16 , 27 , 28 , 30 , both our experimental and theoretical values are consistent with previous DFT results from the literature 30 and the experimental value of 1.2 eV derived from graphite phonon modes 41 . These values are comparable to other experimental values for the intrinsic bending stiffness, but significantly lower than the effective bending stiffness of \(1{0}^{2}\) to \(1{0}^{4}\) eV measured in micrometre-scale suspended graphene 9 , where factors such as buckles and thermal rippling dominate the bending properties. Next we analyse the bending stiffness of FLG. Figure 2c plots the extracted bending stiffness as a function of thickness (black). Our measurements yield low values for B , near or below the lowest FLG bending stiffnesses reported in the literature. For example, we report bending stiffness values between 2.6 and 5.8 eV for bilayer graphene, while previously reported values range from 3.4 to 160 eV (refs. 12 , 17 , 18 , 27 , 31 ). We compare these values to upper and lower bounds of FLG bending stiffness from continuum mechanics plate theory, where 2D materials may be described as a single or series of stacked plates, depending on the interlayer coupling strength. The bending stiffness of a single plate scales with the cube of its thickness, or \(B\propto {Y}_{\mathrm{3D}}{t}^{3}\) where t is the thickness and \({Y}_{\mathrm{3D}}\) is the 3D Young’s modulus. The blue line in Fig. 2c represents the single-plate continuum model, modified for the discrete nature of FLG 17 , 18 : \(B={\frac{{Y}_{{\mathrm{3D}}\times{t_{0}^3}}}{12}}({N}^{3}-N)+{B}_{\mathrm{ml}} N\) , where \({t}_{0}=0.334\) nm is the interlayer separation 13 , \({B}_{\mathrm{ml}}=1.4\) eV is the intrinsic monolayer graphene bending stiffness 30 and \(N\) is the number of layers. In contrast, for a stack of frictionless plates, the bending stiffness scales linearly with layer number, \(B={B}_{\mathrm{ml}} N\) (red line, Fig. 2c ); this lowered scaling power reflects the ability of the layers to move independently. Our experimental data are close to the lower limit given by the linear model, indicating weak interlayer interactions in the regimes measured. Intriguingly, we also observe a spread of \(B\) for each FLG thickness. As we show below, these variations indicate an angle dependence of the bending stiffness and its scaling laws. In Fig. 3 , we model the bending of FLG using DFT 42 , 43 . Figure 3a plots the bending stiffness of one- to five-layer graphene as a function of bending angle. In these simulations, graphene is bent along the zigzag \(\langle \bar{2}110\rangle\) direction (see Supplementary Methods 1.4 and Supplementary Figs. 7–9 for simulation details and results for the armchair \(\langle \bar{1}100\rangle\) direction). In Fig. 3a , the bending stiffness for each FLG thickness \(N>1\) decreases sharply with increasing angle and gradually levels off above a threshold angle around 40°. The variations in Fig. 3a are significant; for example, the bending stiffness of five-layer graphene decreases from 51 eV at 4° to 8.5 eV at 80°. Figure 3b directly compares experimental (filled symbols) and DFT measurements (empty symbols) of bending stiffness for one- to four-layer graphene, colour-coded by bending angle. Supplementary Fig. 10 also plots the experimental bending stiffness versus bending angle and curvature. We obtain remarkable agreement between theory and experiment; both show a clear decrease in bending stiffness with increasing bending angle. An important implication of these results is that the bending stiffness of FLG is not a single value for a given thickness, but instead depends on the geometry in which it is measured, a result that may partially explain the wide range of reported bending stiffness in the literature. Fig. 3: DFT calculations of bending stiffness in FLG and comparison with experiment. a , Plot of DFT-calculated bending stiffness versus bending angle for monolayer (1L) to five-layer (5L) graphene, bent along the zigzag direction. The bending stiffness decreases with bending angle for all N > 1 and plateaus at ~41°. b , Comparison of DFT and experimental bending stiffnesses versus thickness for monolayer to four-layer graphene. Open and filled symbols represent bending stiffness from DFT and experiment, respectively. Data are colour-coded by the bending angle \(\theta\) . Error bars for experimental bending stiffness values represent 95% confidence bounds, as determined by error propagation of geometric measurements. The DFT and experimental values exhibit a strong quantitative match, and both indicate a strong angle dependence for the bending stiffness. Power-law fits to the DFT simulation, \(B\propto {N}^{\gamma }\) , yield \(\gamma =2.2\pm 0.23\) for \(\theta =4.4^\circ\) (blue line) and \(\gamma =1.1\pm 0.022\) for \(\theta =81^\circ\) (red line). These fits show that the thickness scaling of bending stiffness changes with the curvature angle. Full size image We also find that the bending stiffness follows different scaling laws depending on the bending angle. We applied power-law fits \(B=c {N}^{\gamma }\) to the DFT simulations (Supplementary Fig . 11) as well as simulations using classical potentials, which allowed us to access larger systems up to \(N=10\) and lower angles down to 1° (see Methods and Supplementary Fig. 12 ). At the limits of low and high angles, we observe scaling laws that approach the predictionds from continuum mechanics. For a low angle of \(\theta =4.4^\circ\) , we obtain \(\gamma =2.2\pm 0.23\) (blue line, Fig. 3b ) from the DFT fits; using classical potentials, we find the scaling law continues to increase towards cubic at the lowest angles we simulated, yielding \(\gamma =3.1\pm 0.67\) for \(\theta =1^\circ\) . Conversely, for high bending angles, we obtain a nearly linear scaling through both simulation methods; for example, we obtain \(\gamma =1.1\pm 0.022\) from the DFT at \(\theta =81^\circ\) (red line, Fig. 3b ). Between these limits, the scaling power gradually decreases as the bending angle increases (Supplementary Fig. 11). Strikingly, our results show that above a threshold angle around 40°, FLG exhibits a nearly linear scaling law characteristic of a stack of frictionless plates, where each layer has the 1.5 eV bending stiffness of monolayer graphene. These results indicate the onset of superlubricity between the atomic layers of graphene at high bending angles. Figure 4a,b presents schematics of the two primary atomic deformations that can accommodate the differential stress induced by bending in 2D materials: in-plane strain within the layers (Fig. 4a ) or shear and slip between layers (Fig. 4b ). These models can be readily distinguished by comparing the number of atoms in each layer along the bend, which remains constant in the in-plane strain case but increases radially in the slip case. In Fig. 4c , we compare these models to a bright-field STEM image of curved 12-layer graphene. The number of atomic columns in each layer in the STEM image increases radially, confirming that the bending mechanism in FLG is dominated by interlayer shear and slip. Fig. 4: Atomic-scale bending mechanisms in FLG. a , Schematic for bending that is accommodated by in-plane strain in the graphene layers. b , Schematic for bending that is accommodated through interlayer shear and slip. c , Bright-field STEM image of 12-layer graphene bent to \(12^\circ\) . The number of atomic columns in the arc is higher for the outer graphene layers than for the inner layers, indicating bending consistent with the shear-slip model. Scale bar, 1 nm. d , Plot of the interfacial contribution to bending stiffness versus bending angle, derived from a simplified, two-chain Frenkel–Kontorova model. The interfacial contribution to bending stiffness decreases with bending angle and plateaus at around \(40^\circ\) , similarly to the DFT results in Fig. 3a . Inset: Cartoons of the interlayer registry of atoms in a 1D model for curved bilayer graphene, where the curvature is accommodated entirely by slip between layers. The atomic positions are plotted in cylindrical coordinates to show how atoms are aligned radially along the curve. The registry between layers decreases as the bending angle increases. Full size image Next, we model the impact of interlayer slip on the atomic structure of FLG. The inset of Fig. 4d shows a profile for curved bilayer graphene where the curvature is accommodated entirely by slip. This cartoon is drawn flat (that is, in cylindrical coordinates) to highlight the interlayer registry. Here, bending produces an effective lattice mismatch between adjacent layers that increases with bending angle. This behaviour is equivalent to the formation of extended dislocations or solitons 4 between layers. Adapting the concept of geometrically necessary dislocations 44 , the number of dislocations per layer is given by \(N={t}_{0}\theta /\left|\mathbf{b}\right|\) , where \({t}_{0}\) is the interplanar spacing and b is the Burgers vector as defined in Supplementary Fig . 9. This equation predicts the angle at which the outer layer contains exactly one more atomic column than the layer below, or equivalently the angle at which a full dislocation is present between adjacent layers. This value is dependent on the crystallographic orientation of the bend: 41.7° for bending along the zigzag direction and 24–36° for bending in the armchair direction. These angles correspond directly with the angular thresholds observed in the DFT simulations in Fig. 3a and Supplementary Fig. 9, demonstrating that a simple dislocation model can be used to predict the bending angles above which superlubricity dominates. In the shear-slip bending mechanism, the bending stiffness of 2D materials can be separated into two components: (1) the intrinsic bending stiffness of individual graphene layers and (2) the contribution from interfacial interactions between layers. Figure 4d plots the interfacial contribution to bending stiffness as a function of bending angle, given by a simplified Frenkel–Kontorova (F–K) model (for details see Supplementary Methods 1.5). The F–K model is commonly used to describe interfacial interactions in thin films, including the dynamics of friction 45 and the formation of solitons in bilayer graphene 4 . In Fig. 4d , we apply a simplified F–K model to describe curved bilayer graphene: each layer comprises a linear 1D chain of atoms connected by springs, and the atoms experience a sinusoidal atomic potential from the adjacent layer. By assuming an infinite in-plane spring constant—or equivalently, no in-plane strain—we force the system to follow the shear-slip bending mechanism to isolate and directly probe its effect on the bending stiffness. Here, the bending energy represents the change in interfacial energy resulting from changes in atomic registry between the layers. Notably, Fig. 4d qualitatively reproduces the drop-off in bending stiffness and the threshold angles seen in our experiment and DFT simulations. These results directly show that our experimental observation of FLG’s angle-dependent bending stiffness can be explained entirely by shear and slip. Put together, the analyses above unite continuum and atomic-scale models to predict, calculate and experimentally verify the phenomenon of slip-induced softening of FLG. Our results show that FLG relieves bending stress primarily through shear and slip between layers rather than in-plane strain. As FLG is gradually bent, its interlayer interactions transition between two limits: the strong coupling characteristic of Bernal-stacked graphite 4 and the weak, superlubric interactions of multiwalled carbon nanotubes 5 . This change in atomic registry and interlayer coupling directly results in a dramatic reduction of bending stiffness to \(B\propto {N}^{\gamma }\) , or equivalently \(B\propto {t}^{\gamma }\) , where \(1<\gamma <3\) , rather than the \(B\propto {t}^{3}\) behaviour of conventional thin films. Finally, we show that a simple dislocation model can predict the angular threshold for bending-induced superlubricity when a full dislocation is present between each layer. These behaviours occur in 2D materials because of their high anisotropy and low energy barrier for slip between atomic layers. For Bernal-stacked graphene, this energetic barrier is less than 2.1 meV per atom 4 , an order of magnitude lower than the 70–90 meV per atom barrier for slip in face-centred cubic nickel 46 . These results have significant implications for the mechanical properties of 2D materials and devices. Our findings indicate a new lower limit for the fabrication of ultrasoft, high-mobility electronic nanodevices. For ten-layer graphene, we show that the bending stiffness can be as low as 18 eV, three orders of magnitude lower than the bending stiffness predicted by conventional thin-film mechanics 18 . Although we have focused on the properties of graphene, our conclusions should generalize to other van der Waals-bonded materials. Finally, these results will be important for the design of new classes of highly curved nanosystems such as nanoelectromechanical systems, stretchable electronics and origami structures made from 2D materials. Methods Fabrication of graphene/hBN heterostructures To transfer FLG over hBN steps, we used established aligned transfer techniques 47 , 48 . First, we exfoliated graphite and hBN flakes separately onto a SiO 2 (285 nm)/Si substrate with the scotch tape method. Then, a few-layer graphene flake was transferred onto a PDMS block by attaching the graphene/SiO 2 /Si substrate onto PDMS and detaching the SiO 2 /Si substrate using a KOH solution. The PDMS block was fixed to a micromanipulator and, finally, the FLG was transferred onto an exfoliated hBN flake containing terraces or steps. After the final transfer, we annealed the sample under high vacuum at 350 °C for 14 h. TEM sample preparation First, we evaporated a protective layer of amorphous carbon (5–30 nm thick) on top of the heterostructure. We then fabricated cross-sectional TEM samples using standard focused ion beam lift-out procedures in an FEI Helios 600i Dual Beam FIB-SEM system. Final milling was performed at 2 kV to reduce sample damage, using a cryo-can to minimize redeposition. Aberration-corrected STEM imaging The samples were imaged in a Thermo Fisher Scientific Themis Z aberration-corrected STEM. The STEM was operated at 80 kV, below the knock-on damage thresholds of graphene and hBN. We used a convergence angle of 25.2 mrad. DFT calculations Atomistic simulations of FLG bending were conducted using DFT 42 , 43 , implemented by VASP 49 with projector augmented wave pseudopotentials 50 . A vdW-DF functional 51 was used to incorporate the van der Waals interaction between individual graphene layers. An energy cutoff of 400 eV was chosen for the plane wave basis, with a total energy convergence of 10 −4 eV. An 80-atom graphene sheet was used for each layer, with a GGA-PBE lattice constant of 2.46 Å. The dimensions of the flat graphene were \(49.2\ \mathring{\rm{A}} \times 4.27\ \mathring{\rm{A}}\) , and 30 Å of vacuum was included to avoid interaction between adjacent images in the z direction in each supercell. A \(2\times 6\times 1\) mesh was used to sample k-space points. To produce the deflection of graphene sheets, we reduced the size of the supercell in a given direction along the basal plane and we induced a geometric perturbation to produce out-of-plane deformation rather than in-plane strain. Through geometric optimization, we found the ground-state geometry of FLG for each supercell. Geometric relaxation was allowed until the forces on each atom were below 0.05 eV Å −1 . We used the fitting function \(f(x)={\sum }_{n=1}^{m}{a}_{n}\cos ({b}_{n}x)\) to describe the geometry of the deformed FLG, which enabled the evaluation of its curvature using \(\kappa =\frac{| f^{\prime\prime} (x)| }{{(1+f^{\prime} {(x)}^{2})}^{\frac{3}{2}}}\) . We found the bending energy of compressed multilayer graphene by subtracting the total energy of a flat, unstrained reference configuration from the total energy of the bent configuration with the same number of layers. Classical potential simulations Atomistic simulations of FLG are conducted using the Reactive Empirical Bond Order (REBO) 52 and Kolmogorov–Crespi (KC) 53 , 54 interatomic potentials, as implemented in the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) 55 package. We performed structural relaxation according to the damped dynamics minimization scheme ‘fire’ 56 . To ensure conformal bending of up to ten layers, each graphene sheet contains 50 unit cells in the zigzag direction and one in the armchair direction for a total of 200 carbon atoms. We analysed the structure (as detailed in the DFT section) to find the average curvature of compressed FLG and used the curvature and bending energy to find the average bending stiffness. Data availability The data and findings of this study are available from the corresponding authors on reasonable request. Change history 06 December 2019 A Correction to this paper has been published:

New research by engineers at the University of Illinois combines atomic-scale experimentation with computer modeling to determine how much energy it takes to bend multilayer graphene—a question that has eluded scientists since graphene was first isolated. The findings are reported in the journal Nature Materials. Graphene—a single layer of carbon atoms arranged in a lattice—is the strongest material in the world and so thin that it is flexible, the researchers said. It is considered one of the key ingredients of future technologies. Most of the current research on graphene targets the development of nanoscale electronic devices. Yet, researchers say that many technologies—from stretchable electronics to tiny robots so small that they cannot be seen with the naked eye—require an understanding of the mechanics of graphene, particularly how it flexes and bends, to unlock their potential. "The bending stiffness of a material is one of its most fundamental mechanical properties," said Edmund Han, a materials science and engineering graduate student and study co-author. "Even though we have been studying graphene for two decades, we have yet to resolve this very fundamental property. The reason is that different research groups have come up with different answers that span across orders of magnitude." The team discovered why previous research efforts disagreed. "They were either bending the material a little or bending it a lot," said Jaehyung Yu, a mechanical science and engineering graduate student and study co-author. "But we found that graphene behaves differently in these two situations. When you bend multilayer graphene a little, it acts more like a stiff plate or a piece of wood. When you bend it a lot, it acts like a stack of papers where the atomic layers can slide past each other." "What is exciting about this work is that it shows that even though everyone disagreed, they were actually all correct," said Arend van der Zande, a professor of mechanical science and engineering and study co-author. "Every group was measuring something different. What we have discovered is a model to explain all the disagreement by showing how they all relate together through different degrees of bending." To make the bent graphene, Yu fabricated individual atomic layers of hexagonal boron nitride, another 2-D material, into atomic-scale steps, then stamped the graphene over the top. Using a focused ion beam, Han cut a slice of material and imaged the atomic structure with an electron microscope to see where each graphene layer sat. The team then developed a set of equations and simulations to calculate the bending stiffness using the shape of the graphene bend. Graduate student Edmund Han, left, professor Elif Ertekin, graduate student Jaehyung Yu, professor Pinshane Y. Huang, front, and professor Arend M. van der Zande have determined how much energy it takes to bend multilayer graphene - a question that has long eluded scientists. Credit: Stephanie Adams By draping multiple layers of graphene over a step just one to five atoms high, the researchers created a controlled and precise way of measuring how the material would bend over the step in different configurations. "In this simple structure, there are two kinds of forces involved in bending the graphene," said Pinshane Huang, a materials science and engineering professor and study co-author. "Adhesion, or the attraction of atoms to the surface, tries to pull the material down. The stiffer the material, the more it will try to pop back up, resisting the pull of adhesion. The shape that the graphene takes over the atomic steps encodes all the information about the material's stiffness." The study systematically controlled exactly how much the material bent and how the properties of the graphene changed. "Because we studied graphene bent by different amounts, we were able to see the transition from one regime to another, from rigid to flexible and from plate to sheet behavior," said mechanical science and engineering professor Elif Ertekin, who led the computer modeling portion of the research. "We built atomic-scale models to show that the reason this could happen is that the individual layers can slip over each other. Once we had this idea, we were able use the electron microscope to confirm the slip between the individual layers." The new results have implications for the creation of machines that are small and flexible enough to interact with cells or biological material, the researchers said. "Cells can change shape and respond to their environment, and if we want to move in the direction of microrobots or systems that have the capabilities of biological systems, we need to have electronic systems that can change their shapes and be very soft as well," van der Zande said. "By taking advantage of interlayer slip, we have shown that the graphene can be orders of magnitude softer than conventional materials of the same thickness."

10.1038/s41563-019-0529-7

Medicine

Drug for pulmonary hypertension may become an option against cancer

Lucy Kappes et al, Ambrisentan, an endothelin receptor type A-selective antagonist, inhibits cancer cell migration, invasion, and metastasis, Scientific Reports (2020). DOI: 10.1038/s41598-020-72960-1 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-020-72960-1

https://medicalxpress.com/news/2020-12-drug-pulmonary-hypertension-option-cancer.html

Abstract Several studies reported a central role of the endothelin type A receptor (ETAR) in tumor progression leading to the formation of metastasis. Here, we investigated the in vitro and in vivo anti-tumor effects of the FDA-approved ETAR antagonist, Ambrisentan, which is currently used to treat patients with pulmonary arterial hypertension. In vitro, Ambrisentan inhibited both spontaneous and induced migration/invasion capacity of different tumor cells (COLO-357 metastatic pancreatic adenocarcinoma, OvCar3 ovarian carcinoma, MDA-MB-231 breast adenocarcinoma, and HL-60 promyelocytic leukemia). Whole transcriptome analysis using RNAseq indicated Ambrisentan’s inhibitory effects on the whole transcriptome of resting and PAR2-activated COLO-357 cells, which tended to normalize to an unstimulated profile. Finally, in a pre-clinical murine model of metastatic breast cancer, treatment with Ambrisentan was effective in decreasing metastasis into the lungs and liver. Importantly, this was associated with a significant enhancement in animal survival. Taken together, our work suggests a new therapeutic application for Ambrisentan in the treatment of cancer metastasis. Introduction The endothelin type A receptor (ETAR) is a G-protein coupled receptor (GPCRs) expressed on both non-immune (e.g., endothelial cells, vascular smooth muscle cells, fibroblasts) and immune cells such as neutrophils 1 , 2 , 3 , 4 , 5 . ETAR, present on vascular smooth muscle cells, has a well-known role in vasoconstriction, which upon agonist stimulation by Endothelin-1 (ET-1), induces cell contraction 6 , 7 . Beyond that, ETAR exerts pleiotropic effects on the progression of ovarian, prostate, colon, breast, bladder and lung primary tumors and tumor metastasis 8 , 9 , 10 , 11 , 12 , 13 . The expression of ETAR on tumor cells increases migration, proliferation, and survival 14 , 15 . ETAR has also been reported to be overexpressed on breast cancer tissue in comparison with non-neoplastic tissue 10 and tumor hypoxia induces breast carcinoma invasiveness in an ETAR-dependent manner 11 . Currently, there is a unique selective ETAR antagonist approved for clinical use, which is called Ambrisentan (Volibris). Ambrisentan has been successfully explored to treat pulmonary arterial hypertension (PAH), and its safety has been demonstrated by several clinical trials 16 , 17 , 18 . However, Ambrisentan remains to be explored as a potential therapy for cancer. Here, we aim to investigate the potential role of Ambrisentan for the treatment of cancer. Since Ambrisentan is a selective ETAR antagonist it does not interfere with the physiological vasodilator and clearance effects of the endothelin type B receptor (ETBR) 19 , 20 . This counterbalances the ETAR effects by transducing intracellular signals that result in the production of nitric oxide and vascular relaxation 21 . In the present study, we found that Ambrisentan inhibited tumor cell migration and invasion, while modulating the tumor transcriptome. Furthermore, in a pre-clinical breast cancer model, Ambrisentan reduced cancer cell metastasis to vital organs and decreased mortality. Taken together, our in vitro data using different tumor cell lines, and in vivo observations suggest a new application of Ambrisentan for the treatment of cancer. Results Ambrisentan demonstrates in vitro anti-tumor effects by impacting tumor cell migration and invasion In preliminary experiments, we tested the effect of exogenously added ET-1 on COLO-357 pancreatic carcinoma cells. However, the results showed no effect on migration (data not shown). It is well known that cancer cells secrete ET-1 constitutively. This could explain the lack of effect observed with exogenously added ET-1. We therefore assessed the effect of Ambrisentan on metastatic carcinoma cell lines in response to protease-activated receptor 2 (PAR2) activation, which is a GPCR that has been implicated in tumor progression 22 , 23 , 24 . For these experiments, we used the Oris Pro Cell assay, a technique employed to analyze the migration of adherent cells in real-time to a cell-free detection zone at the center of each well (Fig. 1 a). Ambrisentan reduced the migration capacity of COLO-357 pancreatic carcinoma cells in the presence or absence of PAR2 induction (Fig. 1 b). In concordance with these data, Ambrisentan also blocked the migration of other tumor cells such as OvCar3 ovarian carcinoma (Supplementary Fig. S1a ) and HL-60 promyelocytic leukemia cell line (Supplementary Fig. S1b ). We also analyzed the effect of Ambrisentan on the functional capacity of MDA‐MB231 breast adenocarcinoma, a cell line shown to be positive for ETAR expression (Fig. 2 a). The findings demonstrate that treatment with up to 100 µM Ambrisentan had no effect on viability of cancer cells (Fig. 2 b). However, the migration and invasion capacities of MDA-MB-231 cells, which are known to contribute to the tumorigenicity of these triple-negative breast cancer cells 25 were significantly inhibited after a 24 h-exposure to Ambrisentan in a dose-dependent manner (Fig. 2 c, d). Figure 1 The anti-tumor actions of Ambrisentan on metastatic pancreatic adenocarcinoma (COLO-357) cells. ( a ) Illustration of the Oris migration assay (left panel), which was used to assess cell migration. ( b ) Graphics (on the top) and representative images (on the bottom) of the inhibitory effect of Ambrisentan on migration of COLO 357 cells before and after stimulation by PAR2 agonist (PAR2 ag.). Assays were performed in quadruplicates. Error bars denote mean with SD; * p ≤ 0.05 (n = 3, Mann–Whitney test). Full size image Figure 2 Ambrisentan inhibits migration and invasion capacity of MDA-MB-231 cells without any impact on cell viability. (a) Flow cytometry histogram shows the extent of ETAR expression by MDA-MB-231 cells. (b) Impact of Ambrisentan on cellular vaibility of exponentially growing MDA-MB-231 cells treated with either vehicle (0.1% DMSO) or the indicated concentrations of Ambrisentan for 24 h. (c , d) Viable cells that were able to cross the 8-mm pores insert (migration assay; c ) and the matrigel matrix (invasion assay; d ) were quantified using the CellTiter-Glo luminescent cell viability assay. The data are expressed as means ± SD of 2–3 replicates per group and are pooled from 3 independent experiments. Asterisks denote statistically significant differences between Ambrisentan-treated cells compared to controls (*** p < 0.001; **** p < 0.0001). Full size image Since high toxicity towards immune cells could indicate a limiting feature 26 of ETAR blockade, we next assessed the effects of Ambrisentan on neutrophils from healthy donors. Neutrophils, which we recently showed to express ETAR 27 , are the most abundant peripheral blood-circulating leukocyte and have a very short lifespan 28 , 29 . Neutrophils exposed to Ambrisentan did not show apparent toxic effects as indicated by lack of cell apoptotic or necrotic signals while maintaining neutrophil nuclear integrity (Fig. 3 a). In addition, Ambrisentan did not affect the respiratory burst response (Fig. 3 b) or phagocytic capacity (Fig. 3 c) of neutrophils, which are both essential for the development of protective immune responses. Despite these results, Ambrisentan significantly decreased neutrophil motility in a concentration-dependent manner regardless of whether a chemoattractant agent (fMLP) was present in the bottom of the transwell plates or not (Fig. 3 d). Figure 3 ETAR blockade inhibits migration of neutrophils while showing no cytotoxic effect. ( a ) Left histogram displays the apoptotic cells stained by FITC-annexin V; middle histogram shows the necrotic cells stained by ethidium homodimer-III; right histogram demonstrates healthy donor cells stained by Hoechst. Heat-killed cells were used as experimental control. ( b ) Neutrophil activation was evaluated in response to phorbol-12-myristate-13-acetate (PMA) by measuring the respiratory burst using dihydrorhodamine 123 (DHR 123). ( c ) Fluorescence images (lower panels using 40 × and 100 × objectives) and graphic (upper panel) of neutrophil phagocytic capacity, showing no effect of Ambrisentan on phagocytosis. Error bars denote mean with SEM; (n = 3; Mann–Whitney test). ( d ) Dose response-inhibition of neutrophil (from healthy subjects) migration by Ambrisentan. Full size image Ambrisentan inhibits the activation of tumor transcriptome Next, we evaluated the effects of ETAR blockade using Ambrisentan on the transcriptome profile of COLO-357 cells by RNA sequencing (RNA-seq). Hierarchical clustering (Fig. 4 a) on the log 2 -transformed transcripts per million values (TPM) across all samples showed that Ambrisentan alone can change the transcriptome of COLO-357 cells. The transcriptome of COLO-357 cells treated with Ambrisentan plus the PAR2 agonist displayed a signature that was more similar with untreated (medium) cells when analyzed by hierarchical clustering compared to cells treated with PAR2 alone. In accordance, Gene Set Variation Analysis (GSVA) indicated that Ambrisentan alone is able to inhibit the activation of different reactome pathways that promote cancer initiation, progression, and metastasis (Fig. 4 b) such as caspase-mediated cleavage of cytoskeleton proteins and activation of the AP1 family of transcript factors. Likewise, Ambrisentan reduced PAR2-induced activation of NF-κB and MAPK pathways as well as the formation of tubulin folding intermediates and chemokine receptor activity. Collectively, these data suggest multiple inhibitory effects of ETAR blockade on tumor-promoting signaling pathways. To test this possibility, we performed an Ingenuity Pathway Analysis (IPA) analysis which predicted an inhibitory effect of Ambrisentan on PAR2 signaling (Fig. 5 ). In addition, IPA analysis suggested the potential of crosstalk between ETAR and growth factor receptors, such as epidermal growth factor receptor (EGFR), as well as transactivation signaling events with other GPCRs. This is consistent with previous studies 30 , 31 , 32 , 33 , 34 . Figure 4 Effects of ETAR inhibition on the transcriptome of COLO- 357 cells. ( a ) Heatmap of hierarchical clustering shows the transcriptome profile of COLO-357 cells in the absence or presence of ambrisentan and/or Par2 antagonist. The transcriptional levels are represented in a log2 scale. ( b) Gene Set Variation Analysis (GSVA) displays enriched (downregulated in blue and upregulated in orange) pathways in the absence or presence of ambrisentan and/or Par2 agonist. Full size image Figure 5 In silico analysis indicating an inhibitory effect of Ambrisentan on PAR2 signaling pathway. We applied QIAGEN’s Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City, ) tool to perform an in silico analysis of the effect of ETAR inhibition by Ambrisentan on PAR2 signaling. Molecules in red are predicted to be affected by Ambrisentan inhibition. Direct and indirect interactions are shown by solid and dashed lines, respectively. Full size image Ambrisentan inhibits tumor metastasis in vivo Based on the data above, we further explored the anti-tumor effect of Ambrisentan on tumor growth and metastasis by evaluating its effect in a pre-clinical syngeneic mouse 4T1 breast cancer model. 4T1 cells are triple negative mammary gland tumors and, similar to human breast cancer, readily metastasize to the lungs, liver, and bone 35 , 36 . We assessed the effect of Ambrisentan administration on the extent of 4T1 metastasis by determining the number of metastatic foci in liver tissue sections. The details of the treatment protocol are shown in Fig. 6 a. Flow cytometric analysis confirmed that 4T1 cells are positive for ETAR expression (Fig. 6 b). We found that Ambrisentan-treated mice had significantly fewer metastatic foci (~ 43% reduction) compared to vehicle controls (Fig. 6 c–h). These foci appeared clearly as small aggregates of tumor cells (Fig. 6 c, e) and stained strongly with anti-Ki67 antibody (Fig. 6 d, g), indicating a high proliferative state. The tumor foci had the characteristics of malignant cells including high nuclear cytoplasmic ratio, irregular contour, and abnormal mitotic figures (Fig. 6 e). Furthermore, we compared these foci with the primary tumor for confirmation of their cellular characteristics (data not shown). Figure 6 Assessment of liver metastasis in 4T1 breast tumor-bearing mice. ( a ) Schematic diagram of the treatment protocol for the orthotopic 4T1 breast cancer studies. Oral treatment with Ambrisentan was initiated 2 weeks pre-implantation of 4T1 cells and continued for another 2 weeks post implantation. Unless otherwise indicated, all animals were sacrificed on day 21 post implantation and organs/tissues were collected and processed for the indicated analysis. ( b ) Flow cytometry histogram shows ETAR expression on 4T1 cells. ( c-h ) Liver sections were prepared 3 weeks post implantation of 4T1 tumor cells and processed for H&E ( c, e, f ) and Ki67 ( d , g ) staining. Images taken at 40 × ( c, f ), 60 × ( e ) or 20 × ( d , g ) magnification are shown. Representative liver sections of control ( c-e ) or Ambrisentan-treated ( f , g ) mice are shown. Tumor metastatic foci consisting of a small cluster of tumor cells, strongly Ki67-positive, are circled in red. ( h ) The number of metastatic foci was determined for representative liver sections and calculated per mm 2 area. Asterisks denote statistically significant differences ( p < 0.05). The data is representative of 3 independent experiments ( n = 11–12 / group). ab antibody. Full size image By virtue of their high level of constitutive expression of monocyte chemoattractant protein-1 (MCP-1) and other chemokines 37 , 4T1 cells increased the accumulation of myeloid cells in the tumor tissue as well as in the blood and spleen of tumor-bearing mice 36 . Myeloid cell infiltration into the lung and liver of 4T1 tumor-bearing mice also occurred and correlated with cancer cell metastasis. Therefore, we used 6-color flow cytometry and sequential gating to analyze the extent of myeloid cell accumulation in the lungs of tumor-bearing mice with or without Ambrisentan pre-treatment. After the exclusion of doublets, debris, and non-viable cells, immune cells were identified using the pan-hematopoietic marker CD45 (Fig. 7 a, b). Based on their positive expression of CD11b, most of the gated CD45 + cells in the lungs are of myeloid origin. These myeloid cells can be broadly subdivided into Ly6G-positive (granulocytes) or -negative (monocytes) populations (Fig. 7 a, b). Pre-treatment with Ambrisentan reduced the accumulation of myeloid cells (66% and 54% for Ly6G + and Ly6G - populations, respectively) (Fig. 7 c–e), reflecting decreased metastasis of 4T1 cells in the lungs. Figure 7 Flow cytometric analysis of mouse lungs following Ambrisentan treatment. ( a , b ) Dot and contour plots of gating strategy used for the identification of major myeloid cell populations in 4T1 tumor-bearing mouse lungs at day 21 post- implantation. Panels a and b illustrate representative flow plots from lungs of control or Ambrisentan-treated mice, respectively. Analysis was done after gating on viable cells (not shown). ( c – e ) Quantification of myeloid cell infiltration in the lungs of tumor-bearing mice. The percentages of CD45 + hematopoietic cells ( c ), CD11b + /Ly6G + granulocytes ( d ), and CD11b + / Ly6G - monocytes ( e ) are shown. Asterisks denote statistically significant differences ( p < 0.05). The data is representative of 2 independent experiments using 3 mice per group ( n = 6). Full size image Ambrisentan improves host survival in orthotopically-implanted 4T1 model Next, we evaluated the effect of Ambrisentan administration (10 mg/kg/day) on tumor volume and host survival. While Ambrisentan demonstrated no reduction in local tumor growth within the mammary fat pad (Fig. 8 a), the overall survival of tumor-bearing animals improved significantly (Fig. 8 b). The median animal survival increased from 35 days in the control group to 45.5 days in Ambrisentan-treated mice (Fig. 8 b). It is important to note that Ambrisentan (at doses up to 100 µM) had no effect on 4T1 cell proliferation in vitro (data not shown). These findings suggest that the anti-tumor effect of Ambrisentan appears not to be via inhibition of cell proliferation but, rather, primarily through its effect on cancer cell metastasis. Figure 8 Increased mice survival after Ambrisentan treatment. Effect of Ambrisentan treatment on tumor growth ( a ) and host survival ( b ) after orthotopic implantation of 4T1 breast tumors. Ambrisentan was administered by daily oral gavage for 2 weeks pre-implantation and another 2 weeks post 4T1 tumor implantation. Animal survival was followed for up to 60 days. Numbers in parenthesis denote the number of mice per group. Asterisks denote statistically significant differences ( p < 0.05). The data is representative of 2 independent experiments. Full size image Discussion Drug repurposing represents an important pharmacological strategy for extending oncology therapies, and existing compounds originally developed for other purposes have the potential to be repurposed for cancer treatment 38 , 39 . The data reported herein suggests a potential therapeutic role of Ambrisentan in limiting the extent of cancer metastasis. Importantly, Ambrisentan has already been approved by the FDA and the European Medicines Agency (EMA) for the treatment of PAH 19 , 40 , 41 , 42 and connective tissue diseases such as Systemic Sclerosis based on extensive safety, toxicity and efficacy testing 43 . Our findings are in agreement with recent reports showing that the dual-specific ETAR and ETBR antagonist, Macitentan, has the ability to counteract chronic lymphocytic leukemia cells by inhibiting in vitro migration and proliferation 44 , as well as tumor growth and metastasis in resistant ovarian cancer xenografts in mice 45 . Likewise, a recent report from Eun-Ju Im and colleagues found serendipitously that sulfisoxazole, which inhibits exosome secretion, limits the growth and metastasis of MDA-MB-231 and 4T1 cells by targeting ETAR 46 . Cell migration, invasion and metastasis are complex mechanisms that involve the establishment and maintenance of cytoskeletal polarization and kinetics 47 triggered by homo- and heterodimeric receptor cross-talk and cross communication between second messengers 48 , 49 , 50 , 51 . In this context, ETAR participates in a complex network of events between GPCRs and several growth factor receptors to regulate cell migration 27 , 52 , 53 . Our preliminary transcriptome data suggests that it is not a single and linear cellular mechanism that is affected by ETAR blockade, but rather more intricate systemic and dynamic signaling events that might be influenced by Ambrisentan treatment through divergent and convergent pathways involving GPCR-mediated transactivation signaling mechanisms 9 , 33 , 34 , 54 . Our transcriptome analysis is limited by the fact that we only used PAR2 to stimulate the cells and pooled the RNA obtained from each condition in three independently performed experiments. This precluded a more detailed interpretation and analysis to understand the network of the signaling complex affected by Ambrisentan. ETAR mediates cellular processes by regulating several signaling pathways including NF-κB, HIF-1α, Akt, MAPK, Gαq/PKC, Src, EGFR, and GSK in an autocrine or paracrine fashion 33 , 55 , 56 , 57 , 58 , which are pathways involved in the development of cancer 59 , 60 , 61 , 62 , 63 , 64 , 65 , 66 . In fact, the results of our IPA analysis are consistent with these previous studies, which suggest the possibility that Ambrisentan’s inhibitory effects could also involve other tumor-promoting signalling pathways (e.g. EGFR) and transactivation signalling events 30 , 31 , 32 , 33 , 34 . Such possibility could explain the inhibitory effect of Ambrisentan in response to PAR2 agonist. In this context, additional experiments based on ETBR or ETAR silencing could provide insightful mechanisms on how the blockade of ETAR/ETBR might affect cancer metastasis. The effect of dual ET-1 receptor antagonists, such as Bosentan or Macitentan, in comparison to Ambrisentan, could also allow a better understanding of whether ETBR participates in the observed effects . Although Ambrisentan is a highly selective ETAR antagonist, its selectivity varies considerably according to the dose, cell model, species (animal or human), and receptor systems 67 , 68 . Therefore, future studies of the effect of Ambrisentan on different cancer cells will need to address its binding affinity to ETAR depending on the type of tumor investigated. Even though the concentration of Ambrisentan that we used here is in line with other studies assessing endothelin receptor antagonists (ERAs) in toxicity assays with glioblastoma and breast cancer cells, it is very likely that at high concentration off-target effects (ETBR and EGFR) 69 , 70 are also involved in the in vitro anti-tumor effects of Ambrisentan. It is important to emphasize, however, that our conclusions regarding the effect of Ambrisentan on cancer metastasis are primarily based on the findings of the in vivo breast cancer model. In our study, we assessed the effectiveness of Ambrisentan in the murine breast cancer model using drug concentrations (5–10 mg/kg) recommended for the treatment of patients with PAH 17 , 41 . Pharmacologically, taking into consideration the known species-specific differences in metabolic activity between mice and humans, the doses we used are at least one log below what the equivalent doses for mice would have been. In other words, a dose of 10 mg/kg in humans is equivalent to about 123 mg/kg in mice 71 . Consequently, the fact that we demonstrated anti-tumor effects of Ambrisentan at 10 mg/kg concentration in mice highlights its potential for in vivo treatment. If one were to extrapolate from the findings of the murine model, we would predict that Ambrisentan’s anti-tumor activity in humans could be achieved with a much lower drug dose than currently used to treat PAH. At this stage, it will also be important to investigate whether new therapeutic approaches such as local drug delivery systems could improve the anti-tumor potential of Ambrisentan. There is strong evidence of aberrant activation of ETAR in the development of cancer 9 , 15 , 46 , 72 , 73 , 74 , 75 . For instance, ETAR expression levels were found to be higher in breast cancer specimens compared with non-neoplastic breast tissue 10 . Additionally, tumor hypoxia has been shown to increase breast carcinoma invasiveness by releasing intracellularly stored ET-1 11 . ET-1 acting via ETAR is a growth factor for colorectal cancer cells 13 and ETAR activation of epithelial cancer cells, cancer-associated fibroblasts, and endothelial cells contributes to colorectal cancer growth and neovascularization 8 . However, earlier trials using endothelin receptor antagonist (ERAs) were not successful even though they showed an anti-tumor effect pre-clinically. The reasons for this failure might be that these drugs have been used as a monotherapeutic approach or in combination with standard chemotherapy for patients with established metastatic disease 9 , 72 , 76 , suggesting that ERAs have no effect on advanced tumors. Therefore, new therapeutic strategies need to explore the potential of Ambrisentan and other ERAs as adjuvant drug 77 for patients with non-metastatic disease to prevent metastasis 9 , 72 . Since Ambrisentan inhibited tumor metastasis in our orthotopic 4T1 breast cancer model, oral or local application of Ambrisentan for patients diagnosed with early stage tumor that has not yet spread to nearby tissues could improve the sensitivity of tumor cells to chemotherapeutic treatment. This would reduce the time course of therapy to which patients are currently submitted. A similar approach could also reduce the dosage of chemotherapeutic agents required for the treatment of aggressive types of tumors, including those that present with high metastasis and relapsing rates, such as breast cancer 77 , 78 , 79 . Finally, further work will be required to address the potential role of Ambrisentan in the treatment of cancer and to better understand the mechanisms how this drug reduces metastasis. For instance, it is necessary to investigate whether Ambrisentan promotes cancer apoptosis. Studies using starved cells to evaluate the effect of Ambrisentan on proliferation, migration, and invasion of synchronized cell cycle phase could also be insightful. Ultimately, experiments analysing cell migration and invasion using complementary approaches such as live-cell imaging will provide a more comprehensive analysis of how Ambrisentan affects biological processes and metastasis. Taken together, our data suggest a novel therapeutic role for Ambrisentan for the treatment of various metastasizing tumors. Methods Ambrisentan for in vitro assays Ambrisentan was obtained from Gilead Sciences (Foster City, CA). For all in vitro assays, except for the cancer cell viability assay with MDA-MB-231 cells in which Ambrisentan was dissolved in 0.1% DMSO, Ambrisentan was resuspended in RPMI (10 mM stock solutions, pH 7.0), stored at -20 °C until use, and Ambrisentan working solution (100 µM final concentration) obtained by dissolving the stock solution in RPMI medium. Cancer cell lines and culture conditions The inhibitory effect of Ambrisentan was analyzed using cancer cell lines: COLO-357 of metastatic pancreatic adenocarcinoma, OvCar3 ovarian carcinoma, HL-60, and MDA‐MB231 breast adenocarcinoma cell lines. The HL-60 promyelocytic leukemia cell line was cultured for 6 days in the presence of 1% dimethyl sulfoxide (DMSO) as previously described 80 . After 6 days of incubation, cells were harvested, counted, and their viability determined by Trypan blue exclusion before performing migration assays. All cell lines constitutively express ETAR and ETBR (Supplementary Fig. S2 ). They were cultured in 10% panexin NTA Serum (Pan-Biotech, Aidenbach, Germany) or 10% fetal bovine serum (FBS) + DMEM high glucose medium (Gibco-ThermoFisher Scientific, Waltham, MA, USA) as previously described 81 . The mouse mammary gland tumor cell line 4T1 36 was generously provided by Dr Jo Van Ginderachter (Vrije Universiteit Brussel, Belgium). Cells were grown in DMEM supplemented with 10% FBS (Gibco-ThermoFisher Scientific). To prepare cells for implantation, 4T1 cells were trypsinized, washed, resuspended in PBS to a single cell suspension, counted, and injected subcutaneously within 30 min. ETAR and ETBR expression ETAR and ETBR expression was evaluated by quantitative real-time PCR (qPCR) as previously described 82 . RNA extraction of tumor cell lines (COLO357, OvCar3, MDA-MB231 and HL-60) was performed using NucleoSpin RNA Kit (Macherey–Nagel GmbH & Co. KG, Düren, Germany) according to the manufacturer’s instruction. RNA integrity was checked on a 1% agarose/TRIS–acetate-EDTA gel supplemented with GelRed Nucleic Acid Gel Stain (Biotium, Fremont, CA, USA). 1 µg of total RNA was transcribed into cDNA using iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Inc., Hercules, CA, USA) according to the manufacturer’s instruction. qPCR was done using SsoAdvanced SYBR Green supermix (Bio-Rad Laboratories, Inc., Hercules, CA, USA) and run on the Bio-Rad CFX Connect Real-Time PCR System according to the manufacturer’s instruction. Specific primers and qPCR conditions are available upon request. ETAR expression by flow cytometry In addition to the analysis of ETAR and ETBR by real-time PCR (Supplementary Fig. 2 ), the expression of ETAR by MDA-MB-231 and 4T1 cells was confirmed at the protein level by flow cytometry using a modification of a previously published protocol 83 . Briefly, 3 × 10 5 cells suspended in staining buffer (PBS/1% FCS/ 0.1% NaN3) were pre-incubated with either Human Trustain Fcx or anti-mouse CD16/32 mAb (both from Biolegend, San Diego, CA, USA) to block FcγIII/II receptor sites on MDA-MB-231 or 4T1 cells, respectively. The cells were then stained with Zombie Aqua Viability dye (Biolegend). After washing, the cells were stained with ETAR-specific rabbit polyclonal antibody (Cat# MBS9203839; MyBiosource, San Diego, CA) for 30 min. After 3 rounds of washing in staining buffer, cells were incubated with FITC-conjugated goat anti-rabbit IgG secondary antibody (cat# Ab97050; Abcam, Cambridge, UK). Data were collected on 10,000 cells per sample using FACS Celesta (BD) and viable cells were analyzed using FlowJo software (BD Bioscience). Ambrisentan and cancer cell viability assay The effect of Ambrisentan on cancer cell viability was carried out using MDA-MB-231 cells, as detailed previously 25 . Briefly, MDA-MB-231 cells (5 × 10 3 cells/well) were seeded overnight in 96-well plate and then cultured in triplicate for 24 h with Ambrisentan dissolved in 0.1% DMSO. Control cultures were treated with 0.1% DMSO alone (the drug vehicle). Cell viability was determined after 24hrs incubation the CellTiter-Glo Luminescent Cell Viability Assay (Promega Corporation, Madison, WI), based on quantification of ATP, which signals the presence of metabolically active cells. Luminescence was measured using a Glomax Luminometer (Promega) and normalized to control. The data are presented as percent cell viability of experimental groups compared to that of control cells. If indicated, the inhibitory effect of Ambrisentan (100 µM) was tested after 1 h of incubation with different concentrations of drug. Cell migration and invasion assays The invasiveness of MDA-MB-231 cells was investigated using a Matrigel Invasion Chamber, as described previously 84 . In this assay, cancer cells must cross the matrigel matrix and the insert 8-µm pores (Costar, DC, USA). Cells (1 × 10 5 cells in 0.5 ml of media with or without Ambrisentan) were seeded into the upper chamber of the Matrigel system. The bottom chamber in the system was filled with DMEM supplemented with 10% FBS as a chemo-attractant and then incubated at 37 °C for 24 h. Cells that have passed through the Matrigel and the pores were counted using the above described viability assay. The migration assay was performed using the chamber system without the matrigel matrix. Neutrophil viability assays The effect of Ambrisentan on cell viability was analyzed by measuring neutrophil survival (apoptosis and necrosis) and function (respiratory burst and phagocytosis). After isolation of neutrophils as previously described 27 , the survival of neutrophils derived from healthy donors (all German subjects, age 22–45 years old) was analyzed using the Apoptotic/Necrotic/Healthy Cells Detection Kit (PromoCell, Heidelberg, Germany) according to the manufacturer´s instructions. The respiratory burst and phagocytosis capacities were assessed as previously described 80 , 85 . Briefly, 5 × 10 5 neutrophils were cultured for 1 h at 37 °C in the absence or presence of phorbol-12-myristate-13-acetate (PMA, 300 ng/ml, Sigma-Aldrich, St. Louis, MO), followed by dihydrorhodamine (DHR 123, Merck KGaA, Darmstadt, Germany) staining. Neutrophils were stained with anti-CD15, and respiratory burst analyzed on a BD FACS Canto II Cytometer (BD Biosciences), gated according to size (forward scatter, FSC) and granularity (side scatter, SSC). Neutrophil phagocytic capacity was analyzed in whole blood by flow cytometry using Staphylococcus aureus BioParticles fluorescein conjugate (Thermo Fisher Scientific, Waltham, MA). After 1 h incubation at 37 °C, red blood cells were lysed using a red blood cell (RBC) lysis solution (Qiagen, Santa Clarita, CA) and phagocytosis was analyzed by flow cytometry and fluorescence microscopy (EVOS FL Cell Imaging System, Oakwood, OH, USA) as previously described 80 . Transwell migration assay The inhibitory effects of Ambrisentan on neutrophil chemotaxis was analysed using the transwell migration assay (24-well plate, Corning, Inc., Corning, NY, USA) to analyse the response of non-adherent cells as previously described 27 . Samples of 5 × 10 5 neutrophils or HL-60 cells were incubated before chemotaxis assays for 1 h in the absence or presence of 100 µM Ambrisentan. The cells were transferred to the upper chamber, while RPMI containing 10 nM n-formylmethionyl-leucyl-phenylalanine (fMLP) was placed in the lower chamber. The plate was incubated at 37 °C for 1 h. Cells that migrated towards the lower chamber of the transwell plates were transferred to a 96-well plate and counted (cells/µl) by flow cytometry using a CytoFLEX Flow Cytometer (Beckman Coulter, Indianapolis, IN, USA), gating on neutrophils according to FSC and SSC to exclude cell debris. Images of the cells on the bottom surface of the transwell plates were obtained using a fluorescence microscope (EVOS FL Cell Imaging System, Oakwood, OH, USA). The neutrophil migration capacity was calculated in relation to the spontaneous migration (cells with medium) which we arbitrarily considered as 100%. Oris migration assay The migration of cancer adherent cell lines (COLO357 and OvCar3) was analyzed using the Oris migration assay 81 . COLO357 cells were cultured and allowed to migrate into the free space in the middle of the well for 48 h in the presence or absence of 100 µM PAR2 agonistic peptide (MoBiTec GmbH, Göttingen, Germany) and/or Ambrisentan. The cells were fixed and stained with a DiffQuick Cell Staining kit (Medion Diagnostics, Gräfelfing, Germany). After staining, an Oris detection mask was clipped to the bottom of the plate, and images were obtained with a blackfly camera on an Axioskop HBO 50 microscope (Zeiss, Oberkochen, Germany). The migration area was determined by analyzing the migration images with the Fiji module of the ImageJ software 86 . The pictures were analyzed with a pre-written macro 81 , analyzing each picture with the same thresholds for Grey values. RNA sequencing and bioinformatics analysis RNA-seq was performed as previously described with minor modifications 80 , 87 . After culturing for 24 h in the presence or absence of 100 µM PAR2 agonistic peptide and/or 100 µM Ambrisentan, the total ribonucleic acid (RNA) from COLO357 cells was obtained using TRIZOL (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. RNA integrity and concentration were assessed using the Agilent 2100 Bioanalyser RNA Nano chip and orthogonally validated by visualization of the integrity of the 28S and 18S band on an agarose gel. Three independent experiments were performed and cells of each experimental condition pooled. cDNA libraries were obtained using the Illumina CBot station, and HiScanSQ using the NEBNext Ultra Sample Preparation Kit (Illumina Inc., San Diego, USA) according to each manufacturer’s instructions. Sequencing was carried out using the Illumina HiSeq 4000 platform (150-nucleotide paired-end reads). The bioinformatics analysis of data obtained was performed as previously described 88 . After quality assessment using FastQC 89 , reads were pseudo-aligned with the human cDNA transcriptome from the Ensembl 82 using Kallisto 90 and differential gene expression was assessed using Sleuth 91 . The data obtained is available in Supplementary Table S1 . A hierarchical clustering heatmap of all expressed genes was created using perseus (MaxQuant, v1.11, Martinsried, Germany) 92 to show the Euclidian distance between the cell transcriptome under the different culture conditions. Mouse breast cancer model BALB/c mice were purchased from Jackson Laboratory (Bar Harbor, ME, USA), bred in the animal facility at the College of Medicine and Health Science (CMHS, UAE University), housed in plastic cages with a controlled light and dark cycle of 12 h each and received rodent chow and water ad libitum. For orthotopic implantation, 4T1 cells (2 × 10 4 cells in 100 µl volume, unless otherwise indicated) were injected subcutaneously into the fourth mammary fat pad of 8-week-old mice. Tumor growth was followed by quantitative determination of tumor volume twice weekly, as previously described 93 . Since Ambrisentan is poorly soluble in water, a working stock solution was prepared in dH 2 0 adjusted to a high pH (> 12). The stock solution was then re-adjusted to a physiological pH of ~ 7 with HCl to a final concentration of 20 mg/ml and kept at -20°c until use. Ambrisentan was diluted further to the desired concentration in ddH 2 0 before use. Before each experiment, the drug was further diluted to a working concentration of 1 mg/ml and was administered by oral gavage in a 200 µl volume per mouse (equivalent to ~ 10 mg/kg body weight). The experimental protocol involved random assignment of mice into two groups. Group I served as control and received a daily oral gavage of dH 2 0 for 5 days followed by a 2-day rest, for 2 consecutive weeks. Group II received Ambrisentan at 10 mg/kg body weight. At the beginning of week 3, all mice were implanted with 4T1 cells and treatment with Ambrisentan or dH 2 0 was continued for another 2 weeks post implantation. Mice were sacrificed at day 21 post-implantation and tumor, liver and lungs excised for further analysis. In separate experiments, the effect of Ambrisentan treatment on animal survival was assessed. Mice were implanted orthotopically with 4T1 cells (5 × 10 3 cells per mouse) and tumor volume and animal survival were monitored for up to 60 days. Mice exhibiting signs of morbidity (weight loss, lethargy, piloerection, hyperkyphosis) were humanely sacrificed by CO 2 asphyxiation. Lung tissue processing and flow cytometry After the mice were sacrificed, the lungs were removed, washed with PBS, cut into small pieces using a scalpel, transferred into GentleMACS C-tubes (Miltenyi Biotec, Germany), and processed using the lung dissociation kit and a GentleMACS dissociator (Miltenyi), according to the manufacturer’s instructions. Homogenized lungs were passed through a 70-mm nylon mesh to obtain a single-cell suspension. The resultant cells were counted using an automated cell counter (EVE, NanoEnTek, Seoul, South Korea) and processed for flow cytometry, as detailed previously 94 , 95 . Briefly, cells were pre-incubated with anti-mouse CD16/32 mAb (Biolegend) to block FcγIII/II receptor sites and then stained with Zombie Aqua Viability dye (Biolegend). After washing, the cells were stained with a mixture of fluorochrome-conjugated antibodies using the following panel of mAbs: CD45-Alexa Fluor 700, CD11b-Alexa Fluor 488, Ly6G-BV605, Ly6C-APC-Cy7, F4/80-PE, and MHC II-BV785 (all from Biolegend). Data were collected on 30,000 cells per sample using FACS Celesta (BD) and analyzed using BD FACS Diva software (BD Bioscience). Assessment of liver metastasis Liver tissues were fixed, embedded in paraffin and used to prepare thin sections (5 µm) using an established protocol 96 . Tissue sections were stained with hematoxylin and eosin and images captured with an Olympus BX51 microscope equipped with digital camera DP26 (Olympus Corporation, Tokyo, Japan). Quantification of micrometastases, defined as single-to small clusters of tumor cells, in the liver was done using H&E-stained sections and visualized in a stereo investigator system (Zeiss Imager M2 AX10, Germany). The total area of each section was scanned and measured. Digital images were used to quantify the number of tumor metastatic foci in each section and calculated as the number of foci per mm 2 area. Indirect immunostaining was also used to detect highly proliferative cells with a rabbit anti-mouse Ki67 antibody (Abcam, Cambridge, UK) followed by anti-rabbit-HRP (Cell signaling, Danvers, MA, USA). Peroxidase activity was determined using DAB chromogen (Dako, Carpinteria, CA, USA). Sections were then counter-stained with hematoxylin and visualized and photographed with an Olympus BX51 microscope. All slides were examined by a certified pathologist under blinded conditions. Statistical analysis Statistical significance was assessed by non-parametric Mann–Whitney tests. Data were expressed as median with range. For the animal studies, statistical significance between control and Ambrisentan-treated groups was analyzed by the unpaired two-tailed Student’s t -test. Survival analysis was performed by Kaplan–Meier survival curves and log-rank test (Mantel-Cox). The statistical analyses were performed using GraphPad PRISM 5.01 software (GraphPad Software, San Diego, CA), and differences with a p value ≤ 0.05 were considered significant. Ethics statement All participants provided written informed consent. All experimental procedures were performed in accordance with the Declaration of Helsinki and approved by the ethics committees of the University of Lubeck. All mouse assays were conducted in accordance with, and after approval of the Animal Research Ethics Committee of the United Arab Emirates University. Data availability The RNASeq data is provided as Supplementary Table S1. Additional data is provided upon reasonable request.

A drug used to treat pulmonary hypertension significantly reduced the capacity of tumor cells to migrate and invade other tissues in trials involving pancreatic, ovarian, breast cancer, and leukemia cell lines. Furthermore, in mice with an aggressive form of breast cancer, the drug reduced the incidence of metastasis in the liver and lungs by 47% and lengthened survival compared with untreated animals. The study is published in Scientific Reports. "The drug ambrisentan is an inhibitor of the endothelin type A receptor, which is known to play a role in vasoconstriction, so the drug is used to treat pulmonary hypertension [typically caused by autoimmune diseases such as lupus and systemic sclerosis]. In the laboratory, we found that the drug prevented migration of tumor cells to other tissues and had other effects we're still investigating," said Otávio Cabral Marques, a researcher at the University of São Paulo's Biomedical Sciences Institute (ICB-USP) in Brazil and principal investigator for the study, which was funded by FAPESP. Marques conducted the study while he was a postdoctoral fellow at the University of Freiburg in Germany, collaborating with researchers there and in the United Arab Emirates. He is currently principal investigator for a project supported by FAPESP via a Young Investigator Grant. Endothelin type A receptor is expressed in endothelium, the layer of cells that line the inner surface of blood vessels, and in the cells of the immune system. Other research has also shown its involvement in tumor growth and metastasis. "The effects of the drug appear not to be confined to preventing tumor cell migration, but also to include inhibition of neoangiogenesis, the formation of new blood vessels required to sustain tumor growth," Marques said. "We're currently doing experiments to confirm this. If so, the drug must have a systemic effect, preventing tumor migration to other tissues and inhibiting tumor growth by blocking the generation of new vessels." The drug's benefits in cancer treatment have yet to be proven. Its use without a physician's guidance can be harmful to health, especially in pregnancy. Experiments Using special techniques to measure cell migration, the researchers found that the drug significantly reduced both migration of tumor cells that received a stimulus and spontaneous migration. They tested ovarian, pancreatic, breast and leukemia cancer cell lines. Next, 4T1 cells derived from the mammary gland tissue of a mouse strain were injected into mice to mimic the initial stage of an aggressive form of breast cancer in humans. The mice were treated with the drug for two weeks before the injection and another two weeks afterward. In this experiment, the drug reduced metastasis by about 43% and enhanced median survival by about 30%. "Metastasis of 4T1 cells is very fast in mice, so we began treatment earlier in order to approximate what happens in humans," Marques explained. Marques is now preparing to perform clinical trials with other researchers at ICB-USP. The drug will be tested on a group of cancer patients undergoing chemotherapy to see if they recover better than the control group that will not be given the drug. Although the drug can be administered by mouth, which is an advantage, Marques wants to test applying it directly to the tumor in order to enhance its effect. The type of cancer on which it will be tested has not yet been decided.

10.1038/s41598-020-72960-1

Medicine

'Super-grafts' that could treat diabetes

Fanny Lebreton et al. Insulin-producing organoids engineered from islet and amniotic epithelial cells to treat diabetes, Nature Communications (2019). DOI: 10.1038/s41467-019-12472-3 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-019-12472-3

https://medicalxpress.com/news/2019-11-super-grafts-diabetes.html

Abstract Maintaining long-term euglycemia after intraportal islet transplantation is hampered by the considerable islet loss in the peri-transplant period attributed to inflammation, ischemia and poor angiogenesis. Here, we show that viable and functional islet organoids can be successfully generated from dissociated islet cells (ICs) and human amniotic epithelial cells (hAECs). Incorporation of hAECs into islet organoids markedly enhances engraftment, viability and graft function in a mouse type 1 diabetes model. Our results demonstrate that the integration of hAECs into islet cell organoids has great potential in the development of cell-based therapies for type 1 diabetes. Engineering of functional mini-organs using this strategy will allow the exploration of more favorable implantation sites, and can be expanded to unlimited (stem-cell-derived or xenogeneic) sources of insulin-producing cells. Introduction Although intraportal islet transplantation is an established therapy for patients with type 1 diabetes, maintaining long-term glucose control with this approach remains challenging, mainly due to considerable islet loss in the peri-transplant period 1 . Among the reasons for early graft loss, inflammation at the site of implantation and impaired revascularization appear as key factors 2 , 3 , 4 . Pancreatic islets have a dense blood supply which is inevitably disrupted by the isolation process 5 . In the first weeks after transplantation, oxygen and nutrients are delivered to avascular islets exclusively by diffusion until they become revascularized 6 . Therefore, new strategies aiming at protecting the islets from inflammatory insults and/or promoting graft revascularization may be effective for improving clinical islet transplantation outcomes. Recent studies have demonstrated the functionality of three-dimensionally assembled β-cell aggregates, or multicellular islet spheroids 7 , 8 . Modulating the cell composition by combining different cell types of islet spheroids leads to improvement of function and viability due to heterotypic cell–cell interactions and reproduction of the complex natural morphology of the islet 9 , 10 . This strategy could be brought further, by generating multicellular hybrid organoids consisting of several cell types, i.e., endocrine cells for regulated hormone release, and other cell types with cytoprotective and immunomodulatory properties with the aim to increase islet survival and function after transplantation. Over the last decades, human amniotic epithelial cells (hAECs) have gained interest in regenerative medicine due to their high proliferative capacity, multilineage differentiation, ease of access, and safety 11 . hAECs express surface markers found on human embryonic stem cells and secrete considerable amounts of proangiogenic and anti-inflammatory growth factors, including vascular endothelial growth factor (VEGF), basic fibroblast growth factor (bFGF), angiogenin (ANG), insulin-like growth factors (IGF), and their binding proteins (IGFBPs) 12 , 13 , 14 . In addition, hAECs secrete high levels of hyaluronic acid, which suppresses tumor growth factor β (TGFβ)—a potent profibrogenic cytokine 15 . Decreased levels of TGFβ expression were observed after AEC transplantation in mice with bleomycin- and CCl 4 -induced lung or liver injury 16 , 17 . This growth factor secretion profile and antifibrotic properties make hAECs attractive cells for a construct designed to enhance the engraftment and vascularization of islet cells. In this study we successfully generated viable and functional insulin-secreting organoids composed of hAECs and dissociated islet cells (ICs) and have shown that incorporation of hAECs into islet-cell constructs markedly enhances engraftment, viability and graft function in model of cell therapy for type 1 diabetes. Results Characterization of hAECs After initial seeding, hAECs rapidly formed proliferating clusters and grew within 5 days into a confluent cobblestone-shaped monolayer (Fig. 1a ). After culture, hAECs were characterized by flow cytometry and were positive for epithelial (CD326), mesenchymal (CD90, CD105), embryonic stem-cell (SSEA-4) and pluripotency (Oct-4) markers. Most importantly, the hAECs expressed non-classical class Ib histocompatibility antigens HLA-G and HLA-E (Fig. 1b, c , Supplementary Fig. 8 ). Human amniotic epithelial cells were negative for hematopoietic cell markers CD34, CD31, and CD45 (Fig. 1b ). Our results are consistent with previously reported findings 14 . Fig. 1 Characterization of hAECs. a Phase-contrast microscopic images of cultured hAECs at days 2 (left panel) and 5 (right panel) after seeding. Scale bar = 100 μm. b hAECs were characterized for various surface markers by FACS. Data are the means ± SEM, of cells obtained from sixteen different donors labeled with specific antibodies (% positive cells). c Immunohistochemical detection of cytokeratin (red) and OCT-4 (green) in cultured hAECs; nuclei are labeled by DAPI (blue). Scale bar = 50 μm Full size image Generation and in vitro assessment of organoids Figure 2a describes the process used herein to generate islet organoids by mixing ICs and hAECs. Both IC- and IC-hAEC aggregates formed round-shaped spheroids of uniform size with well-defined smooth borders within 5 days (Fig. 2b ). The average diameter of IC-hAEC organoids and IC spheroids were 139 ± 4 μm and 202 ± 2 μm (data are mean ± SEM, n = 12), respectively (Fig. 2c ). Although ICs and hAECs formed small monocellular islands within the organoid, most ICs were in contact with both cell types as shown by confocal laser scanning microscopy (Fig. 2d ). No evidence of cell loss was detected. There was no significant difference in cell viability between IC spheroids and IC-hAEC organoids (Supplementary Fig. 1 ). Fig. 2 In vitro characterization of islet organoids. a Schematic representation of islet organoid engineering. After labeling with Dio and DiI, ICs and hAECs were mixed, seeded and incubated several days on 3D agarose-patterned microwells to generate islet organoids. b Phase-contrast and corresponding fluorescence views of spheroids in one microwell on days 1 and 5. After 5 day culture (bottom), cells undergo compaction and spheroids appear to acquire a smooth border as compared with aggregated cells at day 1. Scale bar = 50 μm. c Diameters of IC spheroids and IC-hAEC organoids ( n = 12). **** p > 0.0001, unpaired Student’s t test. d Confocal views of islet-cell construct. Cell arrangement and composition of the islet organoid on day 14. Islet-derived cells stained for Insulin (green) and hAECs for human nuclear factor (red). Every ninth section of a z -stacked and the entire 3-D reconstructed islet heterospheroid (right panel) are shown. Scale bar = 50 μm. e Insulin mRNA expressed by IC spheroids and IC + hAEC organoids; insulin mRNA was analyzed by qPCR, arbitrary units (AU) after normalization to housekeeping genes. * p < 0.04, unpaired Student’s t test, n = 3. All data shown are mean ± SEM Full size image On day 5, harvested spheroids were assessed for insulin expression by qPCR analysis, which demonstrated that insulin mRNA expression was significantly upregulated in the IC-hAEC organoids as compared with IC spheroids (Fig. 2e ). The functionality of the spheroids was evaluated by glucose-stimulated insulin secretion (GSIS) assay. IC-hAEC organoids released considerably more insulin in response to high-glucose as compared with IC spheroids and showed significantly higher SI than controls (4.2 ± 0.4 vs 2.8 ± 0.3, data are mean ± SEM, n = 5). These data demonstrate that incorporation of hAECs into islet-cell constructs enhances β-cell function. Organoids maintain function after hypoxic stress in vitro Both IC-hAEC organoids and IC spheroids were cultured under hypoxic conditions (1% oxygen and 5% CO 2 at 37 °C) for 16 h to mimic the ischemic condition taking place in vivo in the early phase of engraftment 18 . This allowed us to examine whether hAECs were able to confer cytoprotection and help to maintain the functional capacity of ICs under ischemic stress. Incubation under hypoxia rapidly caused fragmentation of IC spheroids and increased cell death. By contrast, considerably fewer dead cells were observed within IC-hAEC organoids (Fig. 3a ). As anticipated, glucose-induced insulin secretion of monocellular spheroids was seriously impaired. By contrast, SI of the IC-hAEC organoids in GSIS assay was significantly higher (Fig. 3b ; Supplementary Table 1 ). These results were further strengthened by qPCR analysis, which showed about threefold higher insulin mRNA expression in IC-hAEC organoids as compared with controls (Fig. 3c ). Fig. 3 Organoid functionality after hypoxic stress. a Fluorescence views of IC spheroids and IC-hAEC organoids exposed to hypoxia and assessed for viability by a FDA/PI test; green (FDI) and red (PI) signals indicate live and dead cells respectively. Scale bars = 100 μm. b Insulin secretion, expressed as SI, of IC spheroids (magenta columns) and IC-hAEC organoids (blue columns) under normoxic and hypoxic conditions, * p < 0.03 and after hypoxic exposure, * p < 0.02, two-way ANOVA with Sidak’s multiple comparisons test, n = 5. c Insulin mRNA expressed by IC and IC + hAEC spheroids cultured under hypoxic conditions; insulin mRNA was analyzed by qPCR, and data presented as arbitrary units (AU) after normalization to housekeeping genes, *** p < 0.0003, unpaired Student’s t test, n = 3. d , e HIF-1α nuclear localization visualized by immunostaining and its upregulated expression in IC-hAEC spheroids. **** p < 0.0001, unpaired Student’s t test, n = 6. Scale bar = 50 μm. f Casp3, Casp8, Casp9, and Bcl2 mRNAs expressed by IC spheroids spheroids (magenta columns) and IC + hAEC organoids (blue columns) cultured under hypoxic conditions; data presented as arbitrary units (AU) after normalization to housekeeping genes. *** p > 0.0003, ** p < 0.0075, * p < 0.02, unpaired Student’s t test, n = 3. All data shown are means ± SEM Full size image To assess the possible molecular mechanisms behind the protective effect of hAECs on hypoxia-induced cell death and dysfunction, expression of HIF-1α, a key regulator of cell response to hypoxia was analyzed in spheroids. Immunofluorescence demonstrated a significant increase in the nuclear localization of HIF-1α in IC + hAEC organoids approaching 50% compared with IC spheroids after exposure to hypoxia (Fig. 3d, e ). The higher nuclear HIF-1α expression in IC-hAEC was correlated with a downregulation of the apoptotic genes Casp3, Casp8, and Casp9 and twofold upregulation of the antiapoptotic gene Bcl2 (Fig. 3f ) compared with IC spheroids alone. Taken together, these results suggest that the hAECs protect islet cells from ischemia-induced apoptotic injury and help to maintain glucose responsiveness through the upregulation of HIF-1α expression. Islet organoid transplantation improves diabetes reversal To assess whether incorporation of hAECs into the islet organoids could enhance engraftment and lead to better glycemic control, diabetic SCID mice were transplanted with a marginal mass of 150 IC-hAEC organoids (IC-hAEC group, n = 25), IC spheroids (IC group, n = 25), or hAECs spheroids (hAEC group, n = 5). Mice transplanted with IC-hAEC organoids exhibited enhanced glycemic control, compared with mice grafted with IC spheroids (Fig. 4a ). The average nonfasting blood glucose concentrations of mice in the IC-hAEC group were considerably lower than those in IC group at 1 month after transplantation (7.9 ± 1.1 mmol/l for IC-hAEC ( n = 13) vs 18.4 ± 2.1 mmol/l for IC ( n = 10), data are mean ± SEM, p < 0.0001, unpaired Student’s t test). The cumulative percentage of animals reaching normoglycemia was 96% in the IC-hAEC group vs 16% in the IC group at 1 month after the transplantation (Fig. 4b ). In cured animals, the median time to reach euglycemia was 5.1 ± 0.1 days in the IC-hAEC ( n = 24) group and 30 ± 9.2 days in the IC group ( n = 8) (data are mean ± SEM, p < 0.0001, unpaired Student’s t test). As expected, mice transplanted with hAEC spheroids remained diabetic. Removal of graft-bearing kidneys at different time points after transplantation led to recurrence of hyperglycemia in all mice within 24 h, indicating that the transplanted spheroids were responsible for normalized glucose levels in cured animals. Fig. 4 In vivo function of islet organoids. a Blood glucose measurements. **** p < 0.0001 IC-hAEC (blue squares, n = 25) vs IC (magenta circles, n = 25), * p IC vs, hAEC (gray triangles, n = 5), **** p < 0.0001 IC-hAEC vs hAEC, one-way ANOVA, with Tukey’s multiple comparison test. b Percentage of cured mice after islet spheroid transplantation. c . Intraperitoneal glucose tolerance tests. ** p < 0.01 IC-hAEC (blue squares, n = 10) vs IC (magenta circles, n = 10), one-way ANOVA, with Tukey’s multiple comparison test. Gray squares diabetic non-transplanted controls, gray circles non- diabetic, non-transplanted controls. d , e Insulin-positive area of each group visualized by the immunostaining and its percentage per given area 4 months after transplantation. **** p < 0.0003, unpaired Student’s t test, n = 3. Scale bars = 500 μm. f Rat insulin mRNA levels in retrieved grafts after marginal islet spheroid transplantation. Insulin mRNA was analyzed by qPCR, and data presented as arbitrary units (AU) after normalization to housekeeping genes. * p < 0.01 vs IC group, unpaired Student’s t test, n = 3. All data shown are means ± SEM Full size image To investigate the insulin secretory capacity of the graft in vivo, IPGTT was performed at 4 weeks post transplantation. As shown in Fig. 3c , glucose clearance of mice in the IC-hAEC group ( n = 10) was similar to that of a nondiabetic control at all time points after glucose loading. By contrast, the IC group ( n = 10) showed abnormal glucose tolerance. To further support the data obtained from the IPGTT, fasting serum insulin and C-peptide levels were measured in the same animals. Both insulin (242 ± 32 pmol/l in the IC-hAEC group vs 130 ± 29 pmol/l in the IC group ( n = 6), data are mean ± SEM, p = 0.02, unpaired Student’s t test) and C-peptide (1140 ± 43 pmol/l in IC-hAEC group vs 732 ± 124 pmol/l in the IC group ( n = 5), data are mean ± SEM, p = 0.01, unpaired Student’s t test) concentrations were significantly higher in the IC-hAEC group. These data demonstrate that incorporation of hAECs into the islet organoids enhances functional capacity of islet cells. Organoid transplantation enhances graft revascularization To evaluate engraftment and revascularization, graft-bearing kidneys were processed for histology. Immunohistochemical (IHC) staining showed larger β-cell mass, as assessed by the insulin-positive area per field in the IC-hAEC group compared with that of the IC group (Fig. 4d, e ) at 120 days posttransplant. This finding was further confirmed by qPCR analysis, which demonstrated that insulin mRNA expression levels were significantly higher (in the IC-hAEC group (Fig. 4f ). Similarly, more glucagon and somatostatin-positive cells were found by IHC in the removed grafts of IC-hAEC group compared with grafts of IC group (Fig. 5a–c ). Fig. 5 Immunohistochemical analysis of hormone production in the grafts. a – c Glucagon and somatostatin-positive cells quantified in each group in the field of view (magnification ×200), scale bars 50 μm. **** p < 0.0001 vs IC group, unpaired Student’s t test, n = 10, data are mean ± SEM Full size image To investigate whether incorporation of hAECs into the islet organoids promotes the process of revascularization, histological sections of the graft-bearing kidneys, harvested at different time points were processed for CD34 and CD31 immunostaining. Higher CD34 and CD31 staining on histological section was observed in IC-hAEC group compared with IC group (Fig. 6a , Supplementary Fig. 2 ). After quantification, CD34 staining was shown to be 2–4-fold higher in IC-hAEC group compared with IC group (Fig. 6b–d ). As expected, in both groups, CD34 staining was higher at day 28 compared with day 14. Fig. 6 Enhanced revascularization of the grafts in IC-hAEC group. a The blood vessels of the graft site were detected at day 14 and 28 using CD34 immunostaining. Scale bars = 20 μm. b – d The total number of vessels was calculated as the number of endothelial cells per mm 2 of the graft. The vessel/graft and vessel/β-cell ratios were calculated as a percent of the graft and insulin-positive area respectively. **** p < 0.0001 vs IC group, two-way ANOVA, with Tukey’s multiple comparison test, n = 10. All data shown are mean ± SEM Full size image Mechanisms of improved graft function and revascularization To elucidate the underlying mechanisms by which hAECs contribute to improved revascularization and function of the graft, we assessed possible differences in VEGF-A production between IC-hAEC and IC groups. Immunohistochemical staining showed higher expression of VEGF-A in the IC-hAEC group compared with in the IC group (Fig. 7a ). After quantification. VEGF-A staining in the graft was three times higher in IC-hAEC compared with IC group (Fig. 7b ). To assess whether the hAECs stimulated production of proangiogenic factors by the islet cells, rat-specific VEGF-A mRNA levels were measured in IC spheroids and IC-hAEC organoids. IC-hAEC organoids expressed considerably more VEGF-A mRNA than islet IC spheroids (Fig. 7c ). Moreover, we examined if endothelial cells in the graft tissue originated from hAECs. To this end, histological sections were exposed to an anti-human specific CD31 antibody instead of the anti-rodent CD34 or CD31 antibodies used above. No CD31 staining was found at any time point (Supplementary Fig. 3 ), suggesting that endothelial cells are not of human origin. These results indicate that human hAECs accelerate the revascularization process mainly by stimulating angiogenic factors in the islet cells, but not through differentiation into the endothelial cells. Fig. 7 The mechanisms of the enhanced vascularization and improved function of the IC-hAEC grafts. a The graft-bearing kidneys stained for VEGF-A and insulin at day 14 after transplantation. Scale bars = 100 μm. b The mean VEGF-A expression per insulin-positive area, **** p < 0.0001 vs IC group, unpaired Student’s t test, n = 10. c VEGF-A mRNA expressed by IC and IC + hAEC spheroids as analyzed by qPCR, data presented as arbitrary units (AU) after normalization to housekeeping genes, with the IC group set to, ** p < 0.001 vs IC group, unpaired Student’s t test, n = 3. d Representative images of the grafts labeled for collagen IV, laminin and insulin. Scale bars = 100 μm. e , f The Col IV-positive and the laminin-positive areas were calculated as a percentage of the graft area, **** p < 0.0001 vs IC group, unpaired Student’s t test, n = 10. g , h Expression of E-cadherin as analyzed by immunohistochemistry in the graft site was considerably upregulated in the IC-hAEC group. Scale bars = 100 μm. **** p < 0.0001 vs IC group, unpaired Student’s t test, n = 10. All data shown are mean ± SEM Full size image We finally examined whether hAEC incorporation into the islet organoids promoted the production of extracellular matrix proteins and adhesion molecules, which are essential to maintain islet morphology and promoting in turn β-cell survival and function 19 . IHC studies demonstrated that expression of collagen IV and laminin (two major basement membrane proteins) was higher in the IC-hAEC group (Fig. 7d–f ). Expression of E-cadherin, an adhesion molecule involved in the maintenance of β-cell viability and promoting insulin secretion 20 , was considerably upregulated in the IC-hAEC compared with IC group (Fig. 8a, b ). These results suggest that incorporation of hAECs into islet-cell constructs enhance basement membrane and production of E-cadherin, thus ensuring proper function of islet cells. Fig. 8 Function of a human-derived organoids transplanted in the epididymal fat pad. a Human C-peptide measurements of mice transplanted with 300 islet organoids (IC-hAEC group, blue squares, n = 10), or with 300 islet-cell spheroids (IC group; magneta circles, n = 8) *** p < 0.0007, unpaired Student’s t test. b Human C-peptide levels after intraperitoneal glucose challenge 4 weeks after transplantation. Magneta circles: IC, blue squares: IC-hAEC. ** p < 0.008, unpaired Student’s t test, n = 5. c Representative images of the graft stained for insulin (red) and glucagon (green). Scale bar, upper panel 500 μm, lower panel 50 μm Full size image hAECs remain within grafted organoids over 2 weeks To co-localize hAECs within transplanted IC-hAEC organoids, explanted graft-bearing kidneys were stained for anti-human nuclear antigen antibody. Our findings showed that while human-derived cells were abundantly present in the first 2 weeks after transplantation, their number gradually declined over time, and at the end of a 1-month period, only few HNA-positive cells were detectable (Supplementary Fig. 4 ). In vivo function and vascularization of human-derived grafts Islet material from two different human donors was used. Each mouse was transplanted with 300 human IC spheroids ( n = 8) or IC-hAEC organoids ( n = 10) into the epididymal fat pad. To monitor graft function after transplantation, human C-peptide levels in the blood were measured once a week. As shown in Fig. 9a , C-peptide levels gradually increased after transplantation and were significantly higher in the IC-hAEC group compared with IC spheroid controls. Glucose clearance was studied by IPGTT 1 month after transplantation. IC-hAEC organoids showed normal metabolic function, in contrast with the IC spheroid group (Fig. 9b ). Fig. 9 Revascularization of the grafts transplanted in the epididymal fat pad. a Fluorescent images of the grafts stained for blood vessels (CD31, red) and insulin (green). Scale bar = 20 μm. b Percentage of vessels per insulin-positive area. *** p < 0.0002 vs IC group, unpaired Student’s t test, n = 5. All data shown are mean ± SEM Full size image Grafts were removed one month after transplantation and processed for histological analysis to analyze engraftment and vascularization. Histological analysis of explanted grafts from the IC-hAEC group revealed healthy islet morphology. Grafts stained positive for insulin, glucagon and for the presence of endothelial cells in newly formed intra-islet micro vessels (Figs. 8c , 9a, b). In contrast, very little graft tissue was retrievable from mice transplanted with IC spheroids. Explants showed extensive loss of islet mass and poor vascularization. Consistent with these results, explanted grafts from the human IC-hAEC group exhibited considerably higher E-cadherin expression levels compared with IC grafts (Supplementary Fig. 5 ). These observations demonstrate that integration of hAECs into human islet-cell organoids significantly improves their functionality, viability and vascularization, and confirm the findings obtained with organoids derived from rodent islet cells. Discussion In this study, we have shown that incorporation of hAECs into islet-cell constructs markedly improved secretory function and viability in vitro, in conventional culture and in hypoxic conditions, and engraftment and graft function in vivo. Combined hAEC and islet-cell organoids hold great potential for cell-based therapies for type 1 diabetes. Recent studies have indicated that combining different cell types of cells into hybrid spheroids is a tissue engineering strategy able to provide “building blocks” for larger tissue constructs, with enhanced cell viability, physiologic function and proliferative ability 21 . Thus, the generation of islet-cell-based multicellular spheroids could be an interesting strategy towards the development of novel cell-based therapies in the treatment of diabetes. However, the generation of stable multicellular islet spheroids is quite challenging due to differences in the mode of cellular adhesion of pancreatic islet and other cell types used so far (mostly mesenchymal stem cells) 22 . In this study, we have successfully generated viable and functional islet organoids composed of hAECs and dissociated ICs. We did not observe any segregation of islet cells and hAECs into separate spheroidal units at any stage, as reported by other groups attempting to generate stable hybrid spheroids enriched with mesenchymal stem cells (MSCs) 23 , 24 . In contrast to what has been reported for MSCs, the even coaggregation of hAECs and islet-derived cells clearly demonstrated by confocal microscopy can be attributed to the epithelial origin of both cell types and the identical mode of cellular adhesion, mediated by the cadherin superfamily 25 , 26 . Significant islet loss in the early posttransplant period is one of the reasons for suboptimal outcomes of clinical islet transplantation. This occurs, at least in part, by anoikis (programmed cell death secondary to loss of cell-to-extracellular matrix contact) and necrosis caused by ischemia during the revascularization process 27 . Our findings demonstrate a clear protective effect of hAECs on islet cells in hypoxic conditions. While IC spheroids exposed to hypoxia display extensive necrosis and impaired function, IC-hAEC organoids preserve adequate glucose responsiveness and show considerable protection from cell death. HIF-1α is a transcription factor orchestrating compensatory responses to adapt to hypoxia through modulation of downstream genes involved in angiogenesis and cell survival 28 , 29 , 30 . In islets, HIF-1α upregulates genes involved in glucose metabolism such as glucose transporter 2 (GLUT2) and glucokinase (GCK) 31 . Moreover, recent studies have shown that upregulation of HIF-1α protects islets after transplantation, and thus improves islet transplantation outcomes 32 . In accordance with this, islet bicellular spheroids subjected to hypoxia showed a fivefold increase of HIF-1α expression, which was correlated with a downregulation of the apoptotic genes Casp3, Casp8, and Casp9, and a twofold upregulation of the antiapoptotic gene Bcl2. These findings strongly suggest that hAECs protect islet cells from hypoxic damage through HIF-1α. Moreover, we have observed that hypoxia led to significant loss of E-cadherin expression by IC spheroids (Supplementary Fig. 6 ), which was correlated with impaired insulin secretion in response to glucose stimulation. In contrast, IC-hAEC organoids showed preserved E-cadherin expression and adequate in vitro function. Our in vitro findings were corroborated with in vivo experiments, which demonstrated that transplantation of islet organoids enriched with hAECs resulted in larger β-cell mass engraftment and improved function. Moreover, transplantation of minimal mass of IC-hAEC organoids but not IC spheroids normalized blood glucose levels in STZ-induced diabetic SCID mice. Incorporation of hAECs into the islet-cell constructs accelerated the rate of graft revascularization, which in turn led to a superior engraftment. Several studies have shown that hAECs are able to promote endothelial cell proliferation and angiogenesis through the secretion of trophic factors 12 , 13 . Consistent with these data, our findings have demonstrated that hAEC enhance vascular density in the graft at 14 and 28 days after transplantation. Interestingly, we did not find human-derived endothelial cells in grafts. This finding clearly indicates that hAECs promote revascularization through the stimulation of angiogenic factors in the islet cells, but not through transdifferentiation of epithelial cells. The role of VEGF-A as a key regulator of islet vascularization and function is well known, and a substantial loss of islet vasculature and islet function has been observed in VEGF-A deficient islets 33 . On the other hand, manipulating islets to overproduce VEGF-A accelerates islet revascularization and function after transplantation 34 . Our findings showed a fourfold upregulation of VEGF-A mRNAs in the IC-hAEC organoids compared with IC spheroids suggesting that hAECs mediate the process of neovascularization by stimulating VEGF-A production from the β cells. Stimulation of VEGF-A production could also be attributed to upregulation of HIF-1α triggered by initial hypoxia after transplantation, as VEGF-A is known to be one of the primary target genes regulated by HIF-1α 35 . Enhanced revascularization mediated by hAECs contributed to superior engraftment of islet organoids as demonstrated by adequate blood glucose control, glucose tolerance and serum C-peptide concentrations. Preservation of E-cadherin expression, observed in IC-hAEC organoids but not in IC spheroids, could also contribute to the better engraftment and graft function in the presence of hAEC. Direct involvement of E-cadherin in the control of β-cell secretory function in response to glucose was recently demonstrated by Parnaud et al. 36 . Although data on the impact of hypoxic conditions on islet cells during the revascularization period, are lacking, hypoxia has been shown to downregulate E-cadherin expression in other cell types, mostly in the oncology field 37 , 38 . The mechanisms by which E-cadherin expression was preserved in the presence of hAECs have yet to be studied. Human amniotic epithelial cells abundantly produce extracellular matrix proteins 39 , which have been shown to be essential in promoting β-cell survival and function 19 , 40 , 41 . Consistent with reported data, superior engraftment in the IC-hAEC group was accompanied with a substantial increase of Col IV and laminin. Taken together, these results suggest that the integration of hAECs into islet-cell constructs significantly improves their functionality, viability and engraftment capacity through a variety of mechanisms, including better resistance to ischemia, accelerated revascularization and restoration of cell-to-matrix contacts. Other authors have demonstrated the immunomodulatory properties conferred by amniotic cells 42 , 43 , an observation that we have also made with our hAECs in preliminary experiments (Supplementary Fig. 7 ). We will next investigate whether these immunomodulatory properties could allow maintenance of the grafts with minimal or no immunosuppression. This strategy has great potential in the development of cell-based therapies for type 1 diabetes, since the engineering of spheroids into functional mini-organs would allow the exploration of more favorable implantation sites and could be expanded to unlimited sources of insulin-producing cells, such as stem cells or xenogeneic sources. Methods Antibodies and reagents

To save patients with a severe form of type 1 diabetes (characterized by the absence of functional insulin-producing cells), pancreatic cell transplantation is sometimes the last resort. The pancreas contains cell clusters—called islets of Langerhans—where cells that produce blood glucose regulating hormones are grouped together. However, the transplant process is long and complex: a significant part of the grafted cells die quickly without being able to engraft. By adding amniotic epithelial cells to these cell clusters, researchers at the University of Geneva (UNIGE) and the Geneva University Hospitals (HUG), Switzerland, have succeeded in creating much more robust "super-islets" of Langerhans. Once transplanted, more of them engraft; they then start producing insulin much more rapidly. These results, to be discovered in Nature Communications, would not only improve the success of cell transplants, but also offer new perspectives for other types of transplants, including stem cell transplantation. Today, islet transplantation is one of the last-chance options for patients with a particularly severe form of type 1 diabetes. The islets are removed from a donor's pancreas, isolated and then re-injected into the patient's liver. "The procedure is well controlled—about fifteen patients benefit from it every year in Switzerland—but nevertheless complex, says Ekaterine Berishvili, a researcher in the Department of Surgery at UNIGE Faculty of Medicine, who led this work. Many of the islets die along the way. It often takes several donors to treat one person, whereas we are in desperate need of donors." Placental cells to help grafts To improve the success of islet transplantation and the survival of transplanted cells, researchers in Geneva have sought to create new, more robust islets that would withstand the stress of transplantation better than natural islets. To do this, they came up with the idea of adding amniotic epithelial cells, taken from the wall of the inner placenta membrane, to the pancreatic cells. "These cells, very similar to stem cells, are already used in other therapies, such as corneal repair for example," says Thierry Berney, Professor in the Department of Surgery at UNIGE Faculty of Medicine and Head of HUG Transplant Division, who co-directed this work. "In our case, we found that they can promote the function of pancreatic cells, which is to produce hormones according to fluctuations in sugar levels." First step, in vitro: the addition of amniotic epithelial cells allowed the cell clusters to form regular spheres, indicating better intracellular communication and connectivity. Second step in vivo: the scientists transplanted their "super-islets" of Langerhans into diabetic mice, which quickly began to produce insulin. "Even with few cell clusters, our super islets adapted very well to their new environment and quickly became vascularized," says Fanny Lebreton, a researcher in the Department of Surgery at UNIGE Faculty of Medicine and the first author of this work. A good vascularization is indeed the key element of any transplantation: it allows to supply the new organ with oxygen and nutrients and guarantees their survival. In addition, the artificial islets quickly began to produce insulin. Improving oxygenation and protecting islets Amniotic epithelial cells are thus essential to islet survival and seem to act on two vital elements: the lack of oxygen, which usually kills a large number of transplanted islets, and the modulation of the host immune system to limit the risk of rejection. "In any transplant, the first step is to lower the recipient's immunity to limit the risk of rejection, says Ekaterine Berishvili. Amniotic epithelial cells have the unique characteristic of protecting the foetus, which is also a "non-self," from attacks by its mother's immune system. We believe that the same mechanism is at work to protect the grafts." The protective mechanism, observed here on cell transplants, could also take place in other types of transplants or even in xenotransplantation—where non-human cells or organs are transplanted into humans. These discoveries now need to be confirmed on human subjects. Since the use of amniotic epithelial cells is already common in other clinical settings without adverse side effects, this could be done relatively quickly. An important hope for all those awaiting a transplant.

10.1038/s41467-019-12472-3

Physics

Resetting the future of MRAM

F. Radu, R. Abrudan, I. Radu, D. Schmitz, H. Zabel: Perpendicular exchange bias in ferrimagnetic spin valves. Nature Communications, 2012. DOI: 10.1038/ncomms1728

http://dx.doi.org/10.1038/ncomms1728

https://phys.org/news/2012-03-resetting-future-mram.html

Abstract The exchange bias effect refers to the shift of the hysteresis loop of a ferromagnet in direct contact to an antiferromagnet. For applications in spintronics a robust and tunable exchange bias is required. Here we show experimental evidence for a perpendicular exchange bias in a prototypical ferrimagnetic spin valve consisting of DyCo 5 /Ta/Fe 76 Gd 24 , where the DyCo 5 alloy has the role of a hard ferrimagnet and Fe 76 Gd 24 is a soft ferrimagnet. Taking advantage of the tunability of the exchange coupling between the ferrimagnetic layers by means of thickness variation of an interlayer spacer, we demonstrate that perpendicular unidirectional anisotropy can be induced with desirable absolute values at room temperature, without making use of a field-cooling procedure. Moreover, the shift of the hysteresis loop can be reversed with relatively low magnetic fields of several hundred Oersteds. This flexibility in controlling a robust perpendicular exchange bias at room temperature may be of crucial importance for applications. Introduction After the discovery of the giant magnetoresistance 1 , 2 , the exchange bias (EB) effect has become an integral part of spintronics with implications for basic research and for numerous device applications like random access magnetic storage units and spin valves. Although its observation has been made only 60 years ago by Meiklejohn and Bean 3 when studying Co particles embedded in their natural oxide (CoO) matrix, the first EB system was engineered by nature long before, a few billion years ago, in grains of titanohaematite 4 . The origin of the EB effect is related to the magnetic coupling across the common interface shared by a ferromagnetic (FM) and an antiferromagnetic (AF) layer, when the system is cooled through the Néel temperature of the AF layer. The magnetic properties of the AF layer and the interface are crucial for understanding the magnetic properties of the EB systems 5 . Yet, these two magnetic components are difficult to access experimentally with laboratory tools because of a virtually vanishing magnetization of AF layers and a small volume of the magnetic interface. Important progress has been made by involving neutron and X-ray techniques to study the bulk part of the pinning layer and the interfacial magnetism 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 . For instance, in an archetypal CoO/FM bilayer, neutron scattering has revealed the origin of reduced EB. It was shown to be related to the anisotropic domain state of the AF system 12 . Using soft X-ray techniques, the frozen and rotatable spins 12 , 14 , 15 were observed in several systems, leading to a deeper understanding of the enhanced coercivity and the origin of the EB for systems with uncompensated AF interfaces. Larger efforts were dedicated to the understanding of the FM/AF EB systems with in-plane anisotropy of the ferromagnet 16 . Notably, Mangin and colleagues 17 , 18 , 19 , 20 have performed significant work on GdFe/TbFe ferrimagnetic soft/hard bilayers with in-plane anisotropy in which EB effects occur. EB was also studied in all coupled FM films with in-plane magnetized ferromagnets acting as pinning layers 21 , 22 . However, for applications, perpendicular uniaxial and unidirectional anisotropies are more desirable owing to a larger thermal stability of encoded information 23 . For instance, antiferromagnets in contact to ferromagnets with perpendicular anisotropy like Co/Pt multilayers and CoPt 3 layers have been shown to exhibit EB and all other macroscopic behaviour of EB systems, like training effect, enhanced coercivity and blocking temperature for the onset of the EB 23 , 24 , 25 , 26 . More recently, the perpendicular EB (PEB) was demonstrated in a novel exchange-coupled system Sm 1−0.028 Gd 0.028 Al 2 /SmAl 2 based on the zero magnetization ferromagnet Sm 1−0.028 Gd 0.028 Al 2 as the pinning layer 27 . The latter exhibits a compensation temperature where the antiparallel oriented spin and orbital moments cancel one another leading to a vanishing macroscopic magnetization of the ferromagnet 28 . In nearly all cases, the PEB is set through field-cooling or by in situ field growth procedures, similar to the AF/FM systems with in-plane anisotropy. In this article, we demonstrate that PEB can be set, controlled and reversed at room temperature in exchange-coupled ferrimagnetic systems, by making use of an interlayer exchange decoupling mechanism. As a prototype system, we have chosen artificial DyCo 5 /Ta(x)/Fe 76 Gd 24 spin-valve (SV) trilayers, where the hard ferrimagnet (HFi) DyCo 5 serves as a single magnetic domain pinning layer. In bulk form 29 and in thin films 30 , it has a magnetization compensation temperature and a spin reorientation temperature. As the orbital moment of Dy is nearly as large as the spin moment 31 , it is expected that DyCo 5 exhibits a large perpendicular anisotropy 32 , which qualifies it as a robust pinning layer. Owing to the vanishing orbital moment of Gd, the perpendicular magnetic anisotropy of Fe 76 Gd 24 (FeGd) is weak. Therefore, it acts as a soft ferrimagnetic layer (SFi). To induce a perpendicular unidirectional anisotropy, in a first step the soft and the hard ferrimagnets are partially decoupled by a Ta interlayer spacer, and in a second step hysteresis loops are measured, reversing the SFi magnetization while keeping the HFi magnetically saturated. In this way, the PEB can be set and even reversed without involving a field-cooling procedure. Results Perpendicular anisotropy with ferrimagnetic pinning layer The origin of EB in FM/AF and X/HFi (with X =FM or SFi) bilayers is related to the coupling across the interface shared by two adjacent magnetic layers. The magnetic state of the pinning layer interface has a crucial role in establishing a robust EB 5 . We exemplify below that, as a prerequisite for the occurrence of a strong PEB, the magnetic state of the interface of HFi pinning layer must be compensated. This contrasts the case of an AF pinning layer, where the magnetic state of the interface has to be uncompensated to establish a finite EB 5 . When the AF surface is fully uncompensated, one expects the largest EB to occur. In the opposite case, when the AF interface is fully compensated, the EB is expected to be zero 33 . The last situation is depicted in Fig. 1a . The magnetic moments in the AF layers are coupled antiferromagnetically. Across the interface, the coupling between neighbouring spins is assumed to be ferromagnetic. One can analyse the impact of this compensated magnetic state on the EB in terms of interfacial energy. The interfacial energy is calculated as E kl =− J i S k S l per pair of nearest neighbour spins at the interface 33 . Then, the EB field is proportional to 2* σ / A , where σ is the available energy and A is the unit area 33 . One observes in Fig. 1a that the first and the third (from the left) spin pair exhibits a parallel orientation; therefore they contribute negatively to the interface energy. However, the second and the fourth spin pairs contribute positively to the interface energy because the spins in the AF are oppositely oriented with respect to the FM spin. As a result, the interfacial energy vanishes for a compensated AF interface leading to a forbidden EB effect. Figure 1: Perpendicular unidirectional anisotropy with ferrimagnets. ( a ) Schematic view of a compensated AF interface. The interfacial energy is zero for this configuration, therefore no EB is allowed to occur. ( b – d ) A schematic view of a compensated hard ferrimagnetic interface in contact to a ferromagnet and a soft ferrimagnet, respectively. The larger circle designates the RE metal atom and the smaller one designates the TM atom. This interface does not exhibit frustrated bonds, leading to a maximum interfacial energy σ for EB to occur. The difference between ( c ) and ( d ) is the relative offset between the layers by one site. Full size image An opposite situation occurs for bilayers with ferrimagnetic pinning layers. This is depicted in Fig. 1b–d . For the sake of simplicity, we consider the HFi layer to be magnetically compensated. For instance, this can be achieved at the magnetization compensation temperature, where the magnetizations of the two sublattices are equal, leading to a vanishing total magnetization, just as in antiferromagnets. In Fig. 1b , the HFi layer is coupled to a FM layer. For this compensated HFi interface, the interfacial energy does not vanish. The first and third spin pairs contribute negatively to the total energy. The coupling between the transition metal and the rare-earth metal is AF across the interface (and within the HFi layer). Therefore, the second and the fourth spin pairs contribute also negatively to the interfacial energy. As a result, a compensated interface of a HFi layer leads to a maximum EB. When the pinned layer is a SFi, a similar situation occurs as shown in Fig. 1c,d . All the spin pairs contribute negatively to the interface; therefore the compensated nature of HFi leads to maximum interfacial energy for both FM and SFi pinned layers. The occurrence of a highest available interfacial energy for a compensated interface of a HFi layer does not necessarily lead to a shift of the hysteresis loop for the pinned layer. A critical condition has to be obeyed for the EB to occur 5 , namely Kt / J >1, where K is the HFi anisotropy, t is the thickness of the HFi layer and J is the interfacial coupling constant. Therefore, when the coupling is large, the above condition prevents the EB to occur. Our approach here is to make use of an interlayer spacer for tuning down the coupling strength to values that fulfills the critical condition for EB. Magnetic properties of the ferrimagnetic films The magnetic properties of the individual HFi and SFi ferrimagnetic layers prepared as separate samples were studied by soft X-ray magnetic circular dichroism (XMCD). In Fig. 2a , we show the temperature dependence of the coercive field and the normalized remanent magnetization for the single DyCo 5 (250 Å) layer. These data have been extracted from the hysteresis loop measurements at the L 3 resonant energy of Co. The DyCo 5 layer exhibits two magnetic transition temperatures, a magnetization compensation temperature and a spin reorientation temperature. The magnetization compensation temperature occurs at about T comp =120 K where the sum of the orbital and spin moments of the Co layer equals the magnetic moment of Dy, leading to a vanishing net magnetization of the film. For systems with uniaxial magnetic anisotropy, the coercive field is proportional to the magnetic anisotropy of the film and inversely proportional to the total magnetization 5 . Therefore, at the magnetization compensation temperature, the coercive field diverges as the net magnetization approaches towards zero, as shown in Fig. 2a . The compensation temperature can also be identified by the inverted hysteresis loop below T comp as compared with the one measured above T comp as shown in the insets of Fig. 2a . Below T comp , the Dy magnetic moment becomes dominant and therefore the Co moment is oriented antiparallel with respect to the applied magnetic field. Above T comp , the magnetic moment of Co is larger than the one of Dy and therefore, it is oriented parallel to the external field. At higher temperatures, the DyCo 5 film exhibits a spin reorientation temperature ( T R =350 K). We observe that the uniaxial magnetic anisotropy rotates from out-of-plane, below T R to in-plane, above T R as clearly verified by the different shapes of the hysteresis loops in both regions shown in the insets of Fig. 2a . Below T R , the hysteresis loops of the DyCo 5 remain squared as exemplified in the inset of Fig. 2a , and exhibit high remanent magnetization equal to the saturation magnetization. These square loops with their high remanence at room temperature are nearly ideal properties of films with uniaxial perpendicular anisotropy. The hysteresis loop above T R has a very low remanent magnetization, and is characteristic for a hysteresis loop measured along the hard axis of a thin magnetic film with uniaxial anisotropy. Figure 2: Ferrimagnetic properties of the individual films. ( a ) The temperature dependence of the coercive field (filled black circles) and the remanent magnetization (open red circles) extracted from hysteresis loops measured for a 250-Å-thick DyCo 5 sample. The magnetization compensation temperature ( T comml_mp DyCo 5 =120 K) is clearly visible at the temperature where the coercive field diverges, whereas the reorientation temperature ( T R DyCo5 =350 K) is distinguishable as a change of the remanent magnetization. The insets show hysteresis loops below T comp , between T comp and T R , and above T R . ( b ) The temperature dependence of the coercive field (filled black circles) and the normalized remanent magnetization (open blue circles) for the FeGd film extracted from the hysteresis loops. In the inset are two representative hysteresis loops measured below and well above the compensation temperature ( T comp FeGd =14 K). The external magnetic field was applied perpendicular to the sample surface. Full size image It is essential that the SFi exhibits a low coercive field and a high magnetic remanence characteristic for a well-defined perpendicular uniaxial anisotropy. These magnetic properties of the SFi layer were tuned by means of stoichiometry variation. The characterization of a single Fe 76 Gd 24 (500 Å) layer that fulfils the required magnetic behaviour is shown in Fig. 2b . The hysteresis loops were measured at the L 3 resonant energy of Fe. The compensation temperature of the FeGd film was tuned 34 to lower temperatures in order to obtain a square hysteresis loop with a high remanent magnetization at room temperature. This ideal squareness with a low coercive field of about 60 Oe as well as the compensation nature of the FeGd alloy are clearly demonstrated in the insets of Fig. 2b . Decoupling the ferrimagnetic layers When the SFi (FeGd) and the HFi (DyCo 5 ) layers are set in direct contact, the exchange coupling between them is so strong that the magnetization of the SFi layer follows the one of the HFi layer. Therefore, in order to observe PEB, it is essential to partially decouple the ferrimagnetic layers as presented further on. In Fig. 3 , we show the hysteresis loops of the Si 3 N 4 /Ta(50 Å)/DyCo 5 (250 Å)/Ta(x Å)/FeGd(500 Å) trilayers for several Ta thicknesses. The 50 Å Ta layer next to the Si 3 N 4 substrate serves as an adhesion layer and has no role in our further discussion. When the FeGd and DyCo 5 films are in direct contact, the coercive field is the same for both layers, H c =350 Oe as observed in Fig. 3a . Fe and Co are aligned parallel to the field and exhibit a parallel interfacial exchange coupling. Dy and Gd magnetic moments are also parallel exchange coupled but antiparallel oriented with respect to the external field. This is the most favourable configuration leading to a strong coupling between these layers. The other situation, when measuring in between the compensation temperatures of the two ferrimagnets, would lead to frustrated magnetic moments at the interface because of competing interactions between a rare earth dominated ferrimagnet and a transition metal dominated one. In order to allow EB to occur, one needs to partially decouple these two layers. This is achieved by introducing a Ta interlayer of suitable thickness. We observe that for 1, 2 and 3 Å thick Ta interlayer, the DyCo 5 and FeGd films are still reversing together as shown in Fig. 3b . A sizeable decoupling appears only when Ta is about 5 Å thick ( Fig. 3c ). When the Ta thickness is further increased to 10 Å, the interlayer exchange coupling strength diminishes and therefore the coercive field of the FeGd layer further decreases ( Fig. 3d ). For fully decoupled layers, the coercive field of the SFi layer is H c =60 Oe ( Fig. 3e ). Note that the coercive field of the DyCo 5 layer does not vary considerably as a function of the spacer layer thickness, which is in strong contrast with the large variation of the coercive field of the FeGd layer. This further confirms the hard and soft nature of the magnetic properties of these two films, in agreement with their much different values for the orbital moments. Given that the total magnetizations of the SFi and HFi layers do have finite values at room temperature, a SV configuration can be induced for the partially decoupled trilayers. Reversing the net magnetization of FeGd in a saturated state of DyCo 5 layer leads to an antiparallel orientation of the net magnetizations of both ferrimagnetic films, just like in classical FM SV systems 35 . One conceptual difference between our perpendicular ferrimagnetic SV and the FM SV systems is that the interlayer is used here to decouple only partially the ferrimagnetic layers, while for the classical case (AF/FM/spacer/FM) an interlayer spacer is used to fully decouple the FM-free layer from the FM pinned layer. Figure 3: Element specific XMCD hysteresis loops measured for samples with different interlayer Ta thicknesses. For a Ta thickness of 0 Å ( a ) and 2 Å ( b ), both ferrimagnetic layers reverse together. For 5 Å ( c ) and 10 Å ( d ) Ta thicknesses, the magnetizations of FeGd and DyCo 5 are partially decoupled. The hysteresis loops of individual layers are shown in ( e ). The measurements were performed at room temperature. The statistical errors are smaller than the symbol size. Full size image Room temperature control of PEB Taking advantage of the partial decoupling of the FeGd and DyCo 5 , we demonstrate in Fig. 4 that the hysteresis loops of FeGd can be shifted and the shift's sign even reversed without involving a field-cooling procedure. In the left column, we show the measured data corresponding to the sample with a Ta interlayer thickness of 5 Å. When the hysteresis loop is measured between −900 Oe to +900 Oe and backwards, both element-specific hysteresis loops of the DyCo 5 and FeGd layers are centred with respect to the external field as shown in Fig. 4a . However, after saturating the system in a positive field higher than the coercive field of HFi and measuring the hysteresis loop of the FeGd layer from 300 to −300 Oe and backwards, a shift of the hysteresis loop occurs ( Fig. 4b ), revealing the presence of a unidirectional anisotropy. The EB and the coercive fields are H EB =−80 Oe and H c =112 Oe, respectively. Note that the coercive field of the biased FeGd is higher than the coercive field when unbiased, as expected (compare Fig. 4b with Fig. 3e ). Please also note that the hysteresis loops in Fig. 4a , although centred with respect to the field axis, are not unbiased. They are centred because we reverse the unidirectional anisotropy by reversing the DyCo 5 layer during the field cycle. As a result, the EB field adds to the coercivity in the Fig. 4a . For the scenario in Fig. 4b , the DyCo 5 film remains saturated in a single magnetic domain state, and through interlayer exchange coupling a unidirectional anisotropy is mediated in the FeGd layer, according to the coupling mechanism described at the beginning of this section. When the system is saturated in a negative field smaller than the coercive field of the HFi layer and a subsequent hysteresis loop is measured from −300 to +300 Oe and backwards, the magnetization curve of FeGd is shifted to the opposite side, as shown in Fig. 4c . This demonstrates that the unidirectional anisotropy can be reversed in relatively low fields and without involving any field-cooling procedure. The single magnetic domain state of the pinning layer is of crucial importance also for the stability of the EB field. In Supplementary Fig. S1 , we show that the coercive and EB fields remain stable as a function of the loop index, demonstrating that a training effect is essentially absent. A magnetic multi-domain state in the pinning layer not only diminishes drastically the absolute value of the EB field but it also mediates a strong EB reduction upon consecutive cycling (training effect) 12 , 21 . Figure 4: Exchange-biased hysteresis loops for two representative samples. Left column: ( a–c ) the hysteresis loops measured for a sample with a 5 Å Ta interlayer spacer. Panel ( a ): the hysteresis loops of the DyCo 5 and FeGd films were measured from 900 to −900 Oe. Panel ( b ): the system was saturated in a positive field of +3,000 Oe and consecutively a hysteresis loop was measured between +300 Oe and −300 Oe. Panel ( c ): the system was saturated in a negative field of −3,000 Oe and consecutively a hysteresis loop was measured between −300 Oe and +300 Oe. Right column: ( d–f ) the same scenario was followed for a sample with a Ta interlayer thickness of 10 Å. The dotted vertical lines mark the +300 Oe and −300 Oe fields. The statistical errors are smaller than the symbol size. Full size image To further demonstrate that the absolute value of the PEB field can also be controlled, we show in the right column of Fig. 4 the magnetization for a sample with 10 Å Ta interlayer thickness. The same procedure (as for the sample with 5 Å Ta) was applied to reveal the occurrence of the unidirectional anisotropy in FeGd. The characteristic hysteresis loops shown here are measured by probing the rare earth elements (at the M 5 resonant energy), which demonstrates also their intrinsic AF orientation with respect to the transition metal. The coercive field and the EB for the SFi layer are clearly smaller as compared with the sample with thinner Ta thickness. Therefore, by varying the interlayer thickness one can tune in a controlled manner the absolute value of the EB and coercive fields. Discussion The coupling mechanism with HFi pinning layers is fundamentally different as compared with the established mechanism of EB with AF pinning layers. While for the latter one it is required that the AF exhibits a magnetically uncompensated interface for EB to be established, in the former case a compensated HFi interface is required for the largest EB effect to occur. Moreover, for FM/AF EB systems with in-plane anisotropy, the thickness of the interlayer does not exhibit a cutoff behaviour. Submonolayer dusting of the interface is sufficient to dramatically reduce the absolute values of the EB and coercive fields 36 . This is different for our system, where a much larger thickness of Ta, equal to 5 Å, was needed to allow EB to occur. The critical thickness for the interlayer can be understood based on models for EB 5 . For instance, in AF/FM bilayers, the onset of EB occurs for a critical AF thickness, which is given by the condition R =( K AF t AF )/ J EB ≥1, where the K AF and t AF are the anisotropy and the thickness of the AF, respectively, and J EB is the interfacial coupling constant. This implies that the AF thickness needs to be increased until the AF layer can resist the interfacial torque that acts during the rotation of the ferromagnet 15 . In our case, the thickness and the anisotropy of the HFi layer are held constant; therefore the interfacial coupling has to be decreased by increasing the thickness of the interlayer until the condition above is satisfied. The dipolar coupling between the macroscopic magnetization of the two ferrimagnetic layers also has a role, contributing, however, minimally to the absolute critical thickness of the interlayer. In fact, the enhanced coercive fields of the SFi hysteresis loops in Fig. 4 can be partially attributed to dipolar coupling. For instance, when the FeGd magnetization reverses first, the net magnetizations of the HFi and SFi layers are in an unfavourable antiparallel situation, therefore an enhanced coercive field is observed. This is different with respect to the EB with in-plane anisotropy where the stray fields are not present leading to a vanishing dipolar coupling. In conclusion, we have demonstrated that PEB in ferrimagnetic DyCo 5 /Ta(x)/Fe 76 Gd 24 alloys, where the DyCo 5 layer has the role of a single magnetic domain hard ferrimagnet and the FeGd alloy acts as a soft ferrimagnet, can be controlled in a flexible and robust manner. Taking advantage of the tunability of the interlayer exchange coupling, we have shown that PEB can be induced without making use of a field-cooling procedure. This was achieved by carefully engineering the magnetic properties of the soft ferrimagnet and the thickness of the partially decoupling interlayer. Owing to the strong coupling between the DyCo 5 and FeGd, the EB occurs above a critical thickness of Ta interlayer, at 5 Å. The absolute value of the EB can be tuned through the thickness of the interlayer spacer and even reversed with relatively low magnetic fields of several hundred Oersteds. This flexibility in controlling a robust PEB at room temperature may be of outmost importance for applications in modern ultrafast storage media 37 . Methods Sample preparation The samples were grown by magnetron sputtering (MAGSSY chamber at HZB) in an ultraclean Argon atmosphere of 1.5×10 −3 mbar with a base pressure of <5×10 −9 mbar at the deposition temperature of 300 K. The stoichiometry of the ferrimagnetic alloys was controlled by varying the deposition rate of separate chemical elements in a co-evaporation scheme. As substrates, we used Si 3 N 4 membranes. XMCD from ferrimagnetic films XMCD was measured in a transmission geometry 38 , 39 at BESSY II using the ALICE diffractometer 40 installed at the PM3 bending magnet beamline. The study of the DyCo 5 layer in high fields was performed with the high-field endstation at the beamline UE46-PGM1 installed at the electron storage ring BESSY II. The measurements were performed by detecting the external field-dependent transmitted intensity of the circularly polarized soft X-rays with the energy set to the experimental L 3 resonant edges of Co(780.4 eV) and Fe(709.4 eV), and M 5 resonant edges of Gd(1,187 eV) and Dy(1,297 eV). Circularly polarized soft X-rays are impinging at normal incidence on the sample placed in an external magnetic field set parallel to the beam direction. The transmitted intensity is measured by a Si photodiode, which is insensitive to applied magnetic fields. Typical experimental spectra for a DyCo 5 /Ta/FeGd sample are shown in Fig. 5b–e . By varying the energy of the circularly polarized light and measuring the normalized transmitted intensity for two opposite saturating fields, one observes a strong variation of intensity at the absorption edges. The experimental XMCD curves are then extracted as the logarithm of the ratio of the transmitted intensities for positive and negative field, respectively. Inspecting the polarity of the XMCD signal at the absorption edges, one obtains information about the relative orientation of the sublattice magnetization components. For instance, the XMCD at the Co L 3 edge is negative whereas the XMCD at the M 5 edge of Dy is positive. This shows that the magnetic moments of Co and Dy are oriented antiparallel with respect to each other. Similarly, the XMCD signals at the L 3 edge of Fe and at the M 5 edge of Gd have opposite signs, demonstrating the ferrimagnetic nature of the layers. Most important, for the case of interfacial (or interlayer) coupled films, the coupling sign between the magnetic moments of adjacent magnetic layers can be directly extracted owing to the element-specific sensitivity of the XMCD technique. In our case, the transition metal (Fe and Co) are oriented parallel with respect to each other and the rare earth metals (Gd and Dy) are also parallel oriented. Figure 5: XMCD from ferrimagnets. ( a ) Sketch of the transmission geometry with circularly polarized soft X-rays. The normalized transmitted spectra are shown in ( b , c ), measured for opposite orientations of the magnetic field (±3 kOe). In ( d ) and ( e ) the experimental XMCD spectra are shown. By fixing the photon energy at specific absorption edges and measuring the transmitted intensity while sweeping the external magnetic field, one obtains the element-specific hysteresis loops. Typical hysteresis loops measured at the L 3 absorption edges of Fe and Co, and M 5 absorption edges of Dy and Gd are shown in ( f ) and ( g ), and ( h ) and ( i ), respectively. The measurements have been performed for a DyCo 5 /Ta(5 Å)/FeGd sample at a temperature of 250 K, which is well above the compensation temperatures for both DyCo 5 and FeGd layers. Full size image The element-specific hysteresis loops are recorded by setting the photon energy to the experimental L 3 absorption edges of the transition metal or to the M 5 absorption edges of the rare earth and measuring the transmitted intensity while sweeping the magnetic field. Typical hysteresis loops obtained for Co, Dy, Fe and Gd are shown in Fig. 5f–i . The polarity of the hysteresis loops reflects also the relative orientation of the elemental magnetic moments. Additionally, it provides intrinsic magnetic parameters, that is, the coercive and EB fields for each probed element. In this way, element-specific hysteresis loops can be used to probe the complex magnetic behaviour of exchange-coupled films, demonstrating also the coupling ground state of these layers. Additional information How to cite this article: Radu, F. et al . Perpendicular exchange bias in ferrimagnetic spin valves. Nat. Commun. 3:715 doi: 10.1038/ncomms1728 (2012).

In close collaboration with colleagues from Bochum, Germany, and the Netherlands, researchers from the Helmholtz-Zentrum Berlin, Germany, have developed a novel, extremely thin structure made of various magnetic materials. It is suitable as a kind of magnetic valve for data-storage units of the most recent generation and makes use of effects in the context of so-called spintronics, with which, in addition to the (re-)charging process, magnetic characteristics of the electrons can also be used for information-processing and -storage. The advantage of the new structure: data remain intact even after the electric current has been switched off and the memory can be re-written more or less indefinitely. The scientists published their results in Nature Communications. Everything began with basic academic curiosity. "First of all we just wanted to create a defined anisotropy with two thin, stacked ferrimagnetic layers", says Florin Radu, physicist at the Institute for Complex Magnetic Materials of the Helmholtz-Zentrum Berlin (HZB) and principal author of the research paper. In other words, the researchers just wanted to create a structure in which a magnetic characteristic within the material changes in a well defined way. Experts in this field define this as magnetic hysteresis. It describes the behaviour of magnetic substances vis-à-vis an externally-applied magnetic field. However, the task proved to be much more difficult; the magnetic energies at the interfaces turned out to be so powerful that the magnetization of the films reverses together. It was necessary to place an additional, non-magnetic layer made of tantalum between the ferrimagnetic layers in order to diminish this effect. What the scientists saw next was truly astounding; the system behaved fundamentally differently as compared to the conventional systems made of ferromagnetic and anti-ferromagnetic layers. The ferrimagnet described as magnetically "soft", which consists of the chemical elements iron and gadolinium, unexpectedly indicated an alteration in the hysteresis, while the existing magnetism remained unaltered for the "hard" ferrimagnetic film that consists of the chemical elements dysprosium and cobalt. This discovery paves the way for an even more vigorous research in the field of spintronics.. "Know how, Show how!!", thus proclaims the research maxim of Radu. "I would not be surprised to see this discovery implemented into PC's, smart phones and tablets in the future", he predicts. For his invention the so-called spin-valve the HZB filed a patent application this week. Nowadays, the data storage units are either volatile or non-volatile. For the former, the information is lost as soon as the device is switched off, and for the latter the information remain intact for many years. Due to thermal effects, they are also practically unusable after about ten years. In particular, when the bits are only a few nanometres in size, they lose stability. Once lost, the magnetization direction of the hard magnetic layer cannot easily be set again in the original direction.. This leads irretrievably loss of data. This stability issue can now be addressed with the new spin-valve concept. By tunning the magnetic properties of the hard ferrimagnetic layer, the so-called RAM memory building-blocks (RAM stands for random access memory) can be manufactured with controlled life-time of the stored information of weeks, months or years.. Thereafter, the magnetic orientation can be resettedin the original state, which increases considerably the overall life expectancy of the information as compared to the existing non-volatile MRAM (Magnetoresistive Random Access Memory). These memory building-blocks are now certainly highly sought-after in the field of micro-electronics, but have not been able, to date, to be established in the markets due to high costs and technical problems. With the spin-valve concept by Radu and his colleagues, electronic devices can now be developed that, similar to the MRAM technology, are operable immediately after being switched on and allow their data storage units to be re-written more or less indefinitely.

10.1038/ncomms1728

Earth

People prefer 'natural' strategies to reduce atmospheric carbon

Shannan K. Sweet et al, Perceptions of naturalness predict US public support for Soil Carbon Storage as a climate solution, Climatic Change (2021). DOI: 10.1007/s10584-021-03121-0 Journal information: Climatic Change

http://dx.doi.org/10.1007/s10584-021-03121-0

https://phys.org/news/2021-05-people-natural-strategies-atmospheric-carbon.html

Abstract Soil Carbon Storage has emerged as a feasible strategy for removing carbon dioxide from the atmosphere, raising important questions regarding whether the general public supports the strategy as a means to address climate change. We analyzed data from a national probability survey of 1222 US adults who reported believing in climate change at least “somewhat” to estimate public support for Soil Carbon Storage and how it compares to other leading Carbon Dioxide Removal (CDR) strategies. Overall, a majority of the sample expressed support for Soil Carbon Storage—regardless of whether the strategy involved the use of biochar (a form of charcoal made from organic matter) or not (55% and 62%, respectively)—placing Soil Carbon Storage ahead of Bioenergy plus Carbon Capture and Storage (32%) and Direct Air Capture (25%), and behind only Afforestation and Reforestation (73%), in terms of public support. In addition, perceiving Soil Carbon Storage as “natural” strongly predicted individual-level support, a pattern that held for every CDR strategy featured on the survey. Results demonstrate broad US public support for Soil Carbon Storage as a climate change mitigation strategy at a time when scientists and policymakers are actively considering the political, not just technical, feasibility of different climate solutions. Access provided by MPDL Services gGmbH c/o Max Planck Digital Library Working on a manuscript? Avoid the common mistakes 1 Introduction 1.1 Carbon Dioxide Removal (CDR) strategies According to the Intergovernmental Panel on Climate Change (IPCC 2018 ), anthropogenic climate change—due largely to carbon dioxide emissions (Cox et al. 2000 ; Solomon et al. 2009 )—increased the global average temperature to 1 °C above pre-industrial levels in 2017. However, the same IPCC report concluded that maintaining a global temperature increase of less than 1.5 °C would substantially reduce risks to ecosystems and humans relative to 2 °C of warming (IPCC 2018 ). There is a growing consensus that meeting this target will be difficult to achieve via reductions in greenhouse gas emissions alone. In addition to reducing emissions, strategies that create “negative emissions” scenarios—for example, Carbon Dioxide Removal (CDR) strategies—will likely also be needed (Van Vuuren et al. 2018 ). CDR strategies remove carbon dioxide from the atmosphere using chemical and/or biological solutions to capture and store carbon dioxide (Corner and Pidgeon 2014 ; Field and Mach 2017 ; Lawrence et al. 2018 ; Rogelj et al. 2018 , Smith et al. 2016 ). As several CDR strategies have emerged in recent years, scholars have recognized that gaining policy traction in this area will require not only demonstrating the scientific feasibility of these techniques (Caldeira et al. 2013 ; Fuss et al. 2018 ) but also understanding public opinion on the issue, including the psychological factors that predict support for different strategies (Campbell and Kay 2014 ; Minx et al. 2018 ; Shrum et al. 2020 ). This may be particularly important in the USA, where the issue of climate change and proposed policy solutions have been politically polarized for decades (Dunlap et al. 2016 ; McCright and Dunlap 2011 ). Although recent public opinion data reveal that a majority of the US public is concerned about climate change and its implications for society (Borick and Rabe 2012 ; Brulle et al. 2012 ; Leiserowitz et al. 2019 ; Schuldt et al. 2020 ; Van Boven and Sherman 2018 ), there is limited consensus on which strategies should be used to address the issue (Corner et al. 2013 ; Faran and Olsson 2018 ; Kahan et al. 2015 ; Preston 2013 ). This is partly because public opinion is often shaped by subjective risk perceptions, whether it regards a global pandemic (Dryhurst et al. 2020 ; Walter et al. 2012 ; van der Weerd et al. 2011 ) or new technological breakthroughs (Bassarak et al. 2017 ; Sjöberg 2000 ; Slovic 2000 ). When it comes to CDR strategies in particular, members of the public and key stakeholder groups may not yet have strongly formed opinions. As a consequence, they may rely on pre-existing mental schemas and cognitive associations in deciding whether or not to support a given strategy for removing carbon dioxide from the atmosphere (Shrum et al. 2020 ). 1.2 Soil Carbon Storage As CDR strategies have gained attention in scientific circles, a growing number of studies suggest that sequestering carbon in soils may be a worthwhile approach to addressing climate change (Christoff 2016 ; García-Tejero et al. 2020 ; Lal 2004; Minasny et al. 2017 ; Vermeulen et al. 2019 ). One such approach involves managing land, such as farm and grazing lands, forests, and wetlands, in ways that store increased amounts of carbon in the soil, thus keeping it out of the atmosphere (i.e., Soil Carbon Storage) (Bossio et al. 2020 ; Doetterl et al. 2015 ). Another approach receiving recent attention (Minx et al. 2017 ) involves the same land management approaches mentioned above plus the use of heat to convert unused plant material or manure on farms into a form of charcoal called biochar (Demirbas and Arin 2002 ; Gurwick et al. 2012 ; Lehmann and Joseph 2015 ), which is then mixed into the soil in order to store carbon for long periods of time (i.e., Soil Carbon Storage with Biochar) (Winsley 2007 ; Laird 2008 ; Woolf et al. 2010 ). Although sequestering carbon in soil to mitigate climate change is scientifically feasible (Lal et al. 2003 ; Sykes et al. 2020 ), some have argued there may be significant political barriers to implementing Soil Carbon Storage in the USA, given that key stakeholders, such as farmers, tend to be lean politically conservative and may be resistant to government regulations aimed at addressing climate change (Amundson and Biardeau 2018 ; Amelung et al. 2020 ). Also, while a small number of studies have investigated public perceptions of Soil Carbon Storage as a climate change mitigation strategy (Glenk and Colombo 2011 ; Kragt et al. 2016 ; Shrum et al. 2020 ), few have examined how the public reacts to the use of biochar as a component of the process (Wright et al. 2014 ). Moreover, past research on this topic has been conducted almost entirely outside of the USA (e.g., Jobin and Siegrist 2020 ), where attitudes toward climate change mitigation may be less politicized or controversial. To address this gap, the present study examines public support for Soil Carbon Storage—with and without biochar—using a national probability survey of US adults, allowing us to estimate the overall level of US public support as well as the factors that best predict support for soil carbon strategies. In doing so, we build on recent research into public perceptions of CDR strategies besides Soil Carbon Storage, which suggests that public support may be driven to a substantial degree by whether a given strategy is perceived as “natural” (Wolske et al. 2019 ; Raimi et al. 2020 ). 1.3 The role of perceived naturalness in public support for CDR strategies Although limited research has examined public support for Soil Carbon Storage specifically, recent research suggests that US public support for CDR strategies may depend on the extent to which the strategy is perceived to “tamper with nature.” Specifically, Wolske et al. ( 2019 ) randomly assigned survey respondents to read about one of the following CDR strategies: Afforestation and Reforestation (AR), which involves planting trees in previously unforested and previously forested areas; Bioenergy plus Carbon Capture and Storage (BECCS), which involves growing and harvesting plants as a fuel source and which removes CO 2 from the air and stores it deep underground after the fuel is burned; and Direct Air Capture (DAC), which involves the passing of air over or through chemicals that absorb CO 2 , typically using large fans, and depositing the recovered CO 2 into long-term geologic storage. To the extent that the public perceives some strategies as more natural than others, we might expect positive receptivity and support to follow suit, given that a preference for naturalness has been documented in various domains (Rozin et al. 2004 , 2012 ). For instance, Wolske et al. ( 2019 ) found lower support for strategies that were perceived to tamper more with nature (e.g., BECCS and DAC) as compared to strategies that were perceived to tamper less with nature (e.g., AR). 2 Current research We pursued two primary objectives in the present study. First, we sought to build on recent research on the role that perceptions of naturalness and attitudes toward tampering with nature play in US public support for CDR strategies (Wolske et al. 2019 ) with an expanded set of carbon removal options that included Soil Carbon Storage (SCS) and Soil Carbon Storage with Biochar (SCSB), which limited research has examined in the US context. Second, we examined these questions using a national level, probability-based sample of the US public. Prior research on this topic has recruited respondents from online opt-in samples with quota sampling to match US demographic categories. While research finds that such non-probability methods sometimes yield estimates that cohere with those from probability surveys (see Motta et al. 2019 for a discussion of measurement effects in climate change surveys), others have noted that these different sampling approaches can yield different results (Goldberg et al. 2019 ). Given the policy implications of public opinion surveys on this topic, it is important to examine whether associations between perceptions of naturalness and public support for CDR strategies emerge in probability-based surveys, which may better enable researchers to generalize survey results to the overall US public on this timely environmental issue. We pursued two primary hypotheses based on previous research. First, we expected that perceiving SCS as more “natural” would predict increased public support, given prior evidence that perceptions about tampering with nature predict public support for CDR strategies (i.e., AR, BECCS, and DAC) (Wolske et al. 2019 ). Second, we expected that respondents scoring higher on a scale measure of discomfort with altering the natural world—Aversion to Tampering with Nature (ATN) (Raimi et al. 2020 )—would report more support for strategies that may be seen as more “natural” on average (e.g., SCS), and less support for strategies seemingly less “natural” on average (e.g., DAC). Finally, as an exploratory research question, we sought to test whether respondents reacted differently to soil carbon strategies depending on whether they did or did not include the use of biochar, a more technical component that may result in the strategy being perceived as less “natural.” We addressed this question by embedding a between-subjects experiment in which respondents were randomly presented with either SCS or SCSB. 3 Methods 3.1 Setting and participants We analyzed survey data from a probability-based sample of 1222 US adults recruited by the National Opinion Research Center (NORC) at the University of Chicago using the AmeriSpeak® Panel ( ) from September 19 to October 4, 2019. AmeriSpeak® randomly samples households using area probability and address-based sampling with a known, non-zero probability of selection from the NORC National Sample Frame. The panel provides sample coverage of approximately 97% of the US household population. This study was offered in English-only and was administered exclusively as a Web survey, due to the use of images as well as text in our research materials. Because our main survey questions pre-supposed that climate change is a real phenomenon, we screened for belief in climate change using the question “Do you believe climate change is really happening?” with response options being “Yes, definitely,” “Yes, somewhat,” and “No.” Of the original sample of 1393 respondents who agreed to take the survey, eligibility was limited to those indicating they “definitely” or “somewhat” believe climate change is happening (1284, or 92% of the sample); of these, 64 respondents did not qualify as completes according to NORC, leaving N = 1222 for the analytic sample. A summary of sample demographics appears in Table 1 . Table 1 Summary of unweighted sample demographics for the analytic sample ( N = 1222), including the number ( N ) and proportion (%) of valid respondents Full size table 3.2 Measures 3.2.1 Perceptions of naturalness and support for Carbon Dioxide Removal strategies Building on the study by Wolske et al. ( 2019 ), we solicited respondents’ perceptions of naturalness and policy support for five CDR strategies: Afforestation and Reforestation (AR), Bioenergy plus Carbon Capture and Storage (BECCS), Direct Air Capture (DAC), Soil Carbon Storage (SCS), and Soil Carbon Storage with Biochar (SCSB). We polled respondents immediately after providing them with brief and scientifically accurate descriptions of each strategy that were created or adapted from multiple sources (e.g., Campbell-Arvai et al. 2017 ; Meko 2016 ; Wolske et al. 2019 ; see Supplementary Material ). We provided these descriptions because we expected, based on prior work, that many members of the public would be unfamiliar with CDR strategies (Campbell-Arvai et al. 2017 ; see also Corner et al. 2012 ). To minimize respondent burden, respondents viewed just three out of these five CDR strategies, which were presented in random order to account for possible order effects. Because of our focus on the Soil Carbon Storage strategies, all respondents viewed one of the two versions of this strategy (i.e., SCS or SCSB) as part of a between-subjects experimental design; in addition to this, respondents evaluated two of three strategies from the remaining set (i.e., AR, BECCS, and DAC), selected at random. This design allowed us to examine how public support for Soil Carbon Storage compared to other, previously polled CDR strategies, while testing the replicability of previous findings (Wolske et al. 2019 ) in a probability-based sample. To measure perceived naturalness, respondents were asked to rate how much a given CDR strategy “is natural,” “tampers with nature,” and “disturbs the natural order” on a scale ranging from 1 = Strongly disagree to 7 = Strongly agree (the latter two being reverse-coded for analysis). We combined these three items to form the composite variable, perceived naturalness, for each of the five CDR strategies (αs .68 to .77). 3.2.2 Support for CDR strategies To assess our main dependent variable, support, respondents were asked how likely they would be to support each CDR strategy using the question: “How likely are you to support [...] as a Carbon Dioxide Removal strategy?” where 1 = Very unlikely , 2 = Somewhat unlikely , 3 = Neutral , 4 = Somewhat likely , and 5 = Very likely . 3.2.3 Aversion to tampering with nature To assess individual differences in attitudes toward human intervention in natural systems, near the end of the survey, respondents completed the Aversion to Tampering with Nature (ATN) scale from Raimi et al. ( 2020 ). Specifically, respondents were asked to rate their level of agreement with each of the following five statements: “People who push for technological fixes to environmental problems are underestimating the risks”; “People who say we shouldn’t tamper with nature are just being naïve” (reverse-coded); “Human beings have no right to meddle with the natural environment”; “I would prefer to live in a world where humans leave nature alone”; and “Altering nature will be our downfall as a species” (1 = Strongly disagree to 5 = Strongly agree ) (Wolske et al. 2019 ). A numerical average was computed to yield an ATN score for each respondent ( M = 3.96, SD = 1.02, α = .72). 3.3 Analytic strategy Our analysis begins by examining how public opinion toward Soil Carbon Storage approaches compares to other Carbon Dioxide Removal strategies. In doing so, we report analysis of variance (ANOVA) models testing for the experimental effect of including biochar on perceptions of naturalness and support. We then turn to a set of regression models examining the extent to which perceptions of naturalness and tampering with nature attitudes predict public support for the different Carbon Dioxide Removal strategies. In each of the main regression models, we regressed the support variable onto respondents’ perceived naturalness rating for that CDR strategy and their Aversion to Tampering with Nature (ATN) score. In addition, as covariates, we include political ideology (1 = Very liberal to 7 = Very conservative ), Footnote 1 education (four-category; dummy-coded with the lowest category, Less than high school , as the reference group), sex (coded as 1 = Male and 2 = Female ), and age (as a continuous variable in years) given their examination in prior research on climate change public opinion. Finally, for the purposes of generalizing results to the segment of the US adult public that reports believing in climate change, we present weighted analyses unless otherwise noted. 4 Results 4.1 Reactions to Soil Carbon Storage in comparison to other CDR strategies To examine how the public’s reactions varied across each of the five CDR strategies, we first computed mean-level perceived naturalness scores and mean-level support for each strategy: Afforestation and Reforestation (AR), Bioenergy plus Carbon Capture and Storage (BECCS), Direct Air Capture (DAC), Soil Carbon Storage (SCS), and Soil Carbon Storage with Biochar (SCSB) (Table 2 ). Perceived naturalness scores varied substantially across strategies, with AR ( M = 5.24, SD = 1.27), SCS ( M = 4.54, SD = 1.21), and SCSB ( M = 4.36, SD = 1.13) being perceived as more natural in comparison to the scale midpoint (i.e., 4 = Neither agree nor disagree ), and with BECCS ( M = 3.62, SD = 1.26) and DAC ( M = 3.60, SD = 1.18) being perceived as less natural in comparison to the scale midpoint (| t |s ≥ 7.9, p s < .001 for the midpoint comparisons). Support varied similarly across strategies, with AR ( M = 3.97, SD = 1.06), SCS ( M = 3.68, SD = 1.09), and SCSB ( M = 3.49, SD = 1.04) receiving greater support in comparison to the scale midpoint (i.e., 3 = Neutral ), and with BECCS ( M = 2.83, SD = 1.21) and DAC ( M = 2.66, SD = 1.17) receiving less support in comparison to the scale midpoint (| t |s ≥ 4.0, p s < .001 for the midpoint comparisons). Expressed in percentage terms, whereas a majority of respondents reported being “somewhat” or “very” likely to support AR, SCS, and SCSB (73%, 62%, and 55%, respectively), only a minority indicated the same level of support for BECCS and DAC (32% and 25%, respectively). Table 2 Summary of weighted statistics for perceived naturalness and support for each Carbon Dioxide Removal (CDR) strategy: Afforestation and Reforestation (AR); Bioenergy plus Carbon Capture and Storage (BECCS); Direct Air Capture (DAC); Soil Carbon Storage (SCS); and Soil Carbon Storage with Biochar (SCSB). N s vary between perceived naturalness and support due to missing data. Percent support combines the two highest categories “Somewhat likely” and “Very likely” Full size table 4.2 Testing reactions to different Soil Carbon Storage strategies While the above results suggest that both soil carbon strategies (SCS and SCSB) were perceived as highly natural and supported by a majority of respondents, did these reactions nevertheless differ significantly across the two versions (i.e., with vs. without biochar)? Following Miratrix et al. ( 2018 ), we report unweighted sample average treatment effects to examine this question. Footnote 2 Indeed, ANOVA testing for the experimental effect revealed that SCS was both perceived as more natural ( F (1,1211) = 7.24, p < .01) and received greater support ( F (1,1212) = 5.82, p < .05) than SCSB. 4.3 Perceived naturalness and aversion to tampering with nature as predictors of support for CDR strategies To better understand the predictors of public support for the different CDR strategies, we turn to the regression results. Recall that we conducted a set of five regression models, and in each, support for a given CDR strategy was regressed onto perceived naturalness, ATN score, and the aforementioned covariates. The results for each of the five models are displayed in Table 3 . As expected, perceived naturalness was a significant positive predictor in all models ( B s > .299, p s < .001). ATN, in contrast, was a significant negative predictor in three of the five models, namely, for DAC (−.109, p < .01), BECCS ( B = −.078, p < .05), and SCSB ( B = −.149, p < .001). Table 3 Summary of weighted regression coefficients (beta ( B ) and standard errors ( SE )) depicting significant predictors (* p ≤ .05; ** p ≤ .01; *** p ≤ .001) for support for Carbon Dioxide Removal (CDR) strategies: Afforestation and Reforestation (AR); Bioenergy plus Carbon Capture and Storage (BECCS); Direct Air Capture (DAC); Soil Carbon Storage (SCS); and Soil Carbon Storage with Biochar (SCSB). Predictors include perceived naturalness ratings for each CDR strategy, Aversion to Tampering with Nature (ATN) score, and demographic variables (i.e., political ideology, education, sex, and age) Full size table Furthermore, we conducted an additional set of exploratory regression models that incorporated an additional predictor—namely, belief in anthropogenic climate change (to account for belief in the human vs. natural causes of climate change; Funk and Kennedy 2016 )—as well as select interaction terms (e.g., the education by political ideology interaction; see Hamilton 2011 ; Schuldt et al. 2020 ) (see Supplementary Table S1 ). Footnote 3 Notably, perceived naturalness remained a significant predictor in all five models ( B s > .168, p s < .05). In contrast, the pattern of effects for ATN was more variable. Whereas ATN scores negatively predicted support for BECCS but not DAC in the simplified models, the opposite was observed in the expanded models ( B = −.105, p ≤ .01 for DAC; B = −.064, p > .05 for BECCS) (see Supplemental Table S2 for details). We return to this point about the consistency of perceived naturalness versus ATN scores as predictors of policy support in the discussion below. 4.4 Demographic predictors of support for CDR strategies Finally, demographic variables emerged as significant predictors in all five CDR strategies (Table 3 ), although the patterns of association differed across models. Political ideology was a significant predictor in four out of five models—namely, for BECCS ( B = −.099, p < .001), DAC ( B = −.101, p < .001), SCS ( B = −.062, p < .05), and SCSB ( B = −.058, p < .05)—such that conservatism was associated with lower levels of support for these strategies, echoing numerous prior findings on the relationship between political ideology and climate policy support (e.g., Funk and Hefferon 2019 ; Gillis et al. 2021 ). Education was a significant positive predictor of support for AR, SCS, and SCSB ( B s ~ .3 to .4, relative to Less than high school ), but somewhat expectedly, a negative predictor in the case of BECCS ( B = −.362, p < .01 for Some college , and B = −.432, p < .001 for Bachelor degree or above , relative to Less than high school ). Sex (female) was a significant predictor for SCSB only ( B = .108, p < .05), while age negatively predicted support for BECCS ( B = −.011, p < .001), DAC ( B = −.007, p < .001), SCS ( B = −.005, p < .05), and SCSB ( B = −.005, p < .05). No other significant relationships were observed. Footnote 4 5 Discussion 5.1 Perceptions of naturalness and support for Carbon Dioxide Removal strategies Amid rising attention to Carbon Dioxide Removal strategies to help mitigate the effects of anthropogenic climate change, there is a need to better understand how the general public reacts to these approaches, as well as the psychological factors that predict public support. Soil Carbon Storage, which involves managing land in ways that increase the amount of carbon stored in soils, thus keeping it out of the atmosphere (Minasny et al. 2017 ), is one such strategy that is receiving increased attention from scientists and policymakers (Minx et al. 2017 , 2018 ; Vermeulen et al. 2019 ). Yet, little is known about the extent to which the general public supports Soil Carbon Storage as a climate change mitigation strategy. In a probability-based survey of the US public, we find that a majority of respondents expressed support for Soil Carbon Storage as a climate change mitigation strategy, whether or not it involved biochar—a process that converts unused plant material or manure into a form of charcoal that is then mixed into the soil—although Soil Carbon Storage received more support than Soil Carbon Storage with Biochar (62% and 55%, respectively). Moreover, the Soil Carbon Storage strategies trailed only Afforestation and Reforestation (AR) (73%) in terms of overall public support, and garnered significantly more support than either Bioenergy plus Carbon Capture and Storage (BECCS) (32%) or Direct Air Capture (DAC) (25%). Notably, this ordering of strategies in terms of policy support exactly matched their ordering in terms of perceived naturalness. Stated differently, the three strategies that enjoyed majority support (i.e., Afforestation and Reforestation, Soil Carbon Storage, and Soil Carbon Storage with Biochar) were rated significantly above the scale midpoint in terms of perceived naturalness, whereas the two strategies that garnered minority support (Bioenergy plus Carbon Capture and Storage and Direct Air Capture) were rated significantly below the scale midpoint in terms of perceived naturalness. When the association between perceived naturalness and policy support was examined more closely in a set of regression models that included covariates, perceptions of naturalness emerged as a significant predictor of support for every CDR strategy on the survey. These associations remained robust in expanded regression models that incorporated additional main effect and interaction terms (see Supplemental Table S1 ), further complementing previous work suggesting that such perceptions are a critical factor in public support for techniques that remove carbon dioxide from the atmosphere (Wolske et al. 2019 ). 5.2 Aversion to tampering with nature and support for Carbon Dioxide Removal strategies In addition to perceptions of Carbon Dioxide Removal strategies as “natural,” we included an individual-difference measure of one’s discomfort with altering the natural world—the Aversion to Tampering with Nature (ATN) scale (Raimi et al. 2020 )—as a predictor of Carbon Dioxide Removal strategy support in all regression models. The ATN findings were more mixed, as this variable was a negative predictor of support for three out of the five Carbon Dioxide Removal strategies, namely, DAC, BECCS, and SCSB. Although we cannot be certain why aversion to tampering with nature predicted support for some strategies but not others, the pattern of results observed in our main regression models suggests that this attitudinal disposition may matter more when evaluating strategies that are generally perceived as less “natural,” given that this variable did not predict support for Afforestation and Reforestation or Soil Carbon Storage—the top-two strategies in terms of perceived naturalness. At the same time, we note that this pattern of effects changed somewhat in the supplemental regression models, which controlled for belief in anthropogenic climate change and select interaction effects, such that aversion to tampering with nature no longer predicted support for Direct Air Capture but did predict support for Soil Carbon Storage. Future research may wish to further explore the relationship between aversion to tampering with nature and support for climate dioxide removal strategies, including variability in this relationship across strategies. 5.3 Implications for building public support for Soil Carbon Storage The present results contribute in significant ways to our understanding of US public support for Soil Carbon Storage as a climate change mitigation strategy (Shrum et al. 2020 ; Wright et al. 2014 ) and offer insights for communicators and policymakers. First, results of our embedded experiment revealed higher public support for Soil Carbon Storage than for Soil Carbon Storage with Biochar—yet, both versions of the strategy enjoyed majority support, suggesting that the role of biochar in public perceptions may be of minor practical importance. At the same time, we observed differences between the two versions that may matter for building public support for soil carbon solutions, such as the biochar version being perceived as less natural, and aversion to tampering with nature negatively predicting support for that version only. Together, these results suggest that Soil Carbon Storage that does not involve biochar may be especially politically feasible, something that policymakers—including President Biden, who highlighted Soil Carbon Storage on the presidential campaign trail (Gustin 2019 )—may wish to note. These results also carry implications for messengers seeking to build support for soil carbon and other CDR strategies among the public. Like with Afforestation and Reforestation, our findings suggest that the high naturalness perceptions enjoyed by both Soil Carbon Storage strategies may be a critical factor in their broad support, consistent with the documented preference for the “natural” over the “unnatural” (e.g., Rozin et al. 2004 ). Therefore, those seeking to bolster support for soil carbon sequestration as a climate mitigation strategy may wish to frame messaging in ways that highlight these naturalness associations—for example, by emphasizing elements of the strategy that may be perceived as most natural (i.e., the soil itself) or its co-benefits for agriculture and ecosystem functioning, and by de-emphasizing the strategy’s more technical elements. At the same time, we note that the correlational nature of our regression analyses prevents us from speaking directly to any causal association that may exist between perceptions of naturalness and support for Soil Carbon Storage, or the other Carbon Dioxide Removal strategies we examined. 6 Limitations of the study Although our work complements and extends on previous findings (Jobin and Siegrist 2020 ; Visschers et al. 2017 ; Wolske et al. 2019 ) by using a probability-based sampling approach to enable greater generalizability to the US public, and by examining public support for Soil Carbon Storage as well as Soil Carbon Storage with Biochar, we note some study limitations. We anticipated that the Carbon Dioxide Removal strategies we asked about would be unfamiliar to many of our respondents and, accordingly, we provided brief, scientifically accurate descriptions of each strategy to respondents before measuring their attitudes and policy preferences. Nevertheless, it is possible that the survey responses we analyze here are less crystalized, or perhaps less enduring, than those on other topics that the US public may consider more often (e.g., presidential approval or belief in climate change). Furthermore, although the descriptions of the Carbon Dioxide Removal strategies we provided to respondents were intended to inform and not persuade, it is possible that the presence of the descriptions, or particular elements therein (e.g., the images used to visually represent each strategy), resulted in higher levels of policy support than would be observed in everyday contexts, or otherwise affected the results. As these Carbon Dioxide Removal strategies rise on the public agenda and receive increased media coverage, it will be important to more regularly survey the public to track trends in public support and to examine whether perceptions of naturalness remain a predictor moving forward. In addition, we reiterate that while our respondents were drawn from a probability-based survey panel that is constructed to be representative of US households, respondents who indicated they did not believe climate change is really happening ( N = 109, or 8% of the sample) were not eligible to complete the questionnaire. While this methodological choice was motivated by a desire to bolster the interpretability of our main measures, which pre-suppose that climate change is happening, we acknowledge that some who deny the reality of climate change may nevertheless feel neutral to positive about Soil Carbon Storage or other Carbon Dioxide Removal strategies—a possibility that our design cannot address. As such, the findings reported here are more accurately described as being generalizable to the portion of the US public that accepts the reality of climate change—a large and growing share of the public (Leiserowitz 2007 ; Leiserowitz et al. 2019 ). 7 Conclusions Overall, the present study suggests that Carbon Dioxide Removal strategies centered on Soil Carbon Storage enjoy widespread support among the US public as a climate change mitigation strategy, while underscoring the role of naturalness perceptions for this support as well as public support for various Carbon Dioxide Removal strategies. The importance of perceptions of naturalness suggests that public support may be garnered for soil carbon sequestration, including biochar, based on its alignment with natural carbon cycles. International science policy and integrated assessment models must consider public perception in projections of climate mitigation, and policymakers should bear in mind the greater public support for Soil Carbon Storage compared to Direct Air Capture and BECCS in the coming years. As climate solutions continue to rise on the public agenda, future research should track how public opinion on Soil Carbon Storage evolves over time, including the extent to which naturalness perceptions remain a key predictor of support over-and-above well-established predictors of climate policy support, such as political ideology. Data availability Data and syntax are available at: Notes We analyze political ideology rather than party affiliation because of our conceptual interest in political worldview rather than political identity and because of ideology’s stronger association with climate change opinions that has been documented in prior research (e.g., Fielding et al. 2012 ; Cruz 2017 ). Nevertheless, when party affiliation (Democrat, Republican, Independent/Other; dummy-coded with Republican as the referent group) is substituted for political ideology in our main regression models, the findings involving perceived naturalness and aversion to tampering with nature remain substantively unchanged. Weighted analyses revealed substantively unchanged treatment effects on both perceived naturalness ( F (1,1217) = 6.98, p < .01) and support ( F (1,1220) = 9.97, p < .01). Belief in anthropogenic climate change was measured immediately after respondents qualified for the survey, with the item “Do you think climate change is caused more by human activities, more by natural changes in the environment, or by both equally?” The full set of interaction terms included perceived naturalness by ATN; perceived naturalness by political ideology; perceived naturalness by belief in anthropogenic climate change; and education by political ideology. For the unadjusted bivariate relationships between key study variables, see the correlation matrix in Supplemental Table S2.

Soil carbon storage, carbon capture and storage, biochar—mention these terms to most people, and a blank stare might be the response. But frame these climate change mitigation strategies as being clean and green approaches to reversing the dangerous warming of our planet, and people might be more inclined to at least listen—and even to back these efforts. A cross-disciplinary collaboration led by Jonathon Schuldt, associate professor of communication at Cornell University, found that a majority of the U.S. public is supportive of soil carbon storage as a climate change mitigation strategy, particularly when that and similar approaches are seen as "natural" strategies. "To me, that psychology part—that's really interesting," Schuldt said. "What would lead people, especially if they're unfamiliar with these different strategies, to support one more than the other? Our study and others suggest that a big part of it is whether people see it as natural." The group's paper, "Perceptions of Naturalness Predict U.S. Public Support for Soil Carbon Storage as a Climate Solution," published May 26 in the journal Climatic Change. Co-authors include Johannes Lehmann, the Liberty Hyde Bailey Professor in the School of Integrative Plant Science (SIPS), Soil and Crop Sciences Section (CALS); Dominic Woolf, senior research associate in SIPS; Shannan Sweet, postdoctoral associate in the Lehmann Lab; and Deborah Bossio of the Nature Conservancy. Schuldt's team analyzed results from a survey of 1,222 U.S. adults who reported believing in climate change at least "somewhat," to estimate public support for soil carbon storage and how it compares to other leading carbon dioxide removal strategies. Mitigation strategies—solar and wind power, electric vehicles and sustainable land use and biodiversity, to name a few—are already capturing much attention as the world grapples with rising temperatures, melting ice caps and increasingly violent weather events. Survey data came from an online poll conducted Sept. 19 to Oct. 4, 2019, by NORC at the University of Chicago, a leading survey research firm. The team solicited respondents' perceptions of naturalness and policy support for five CO2 removal strategies: afforestation and reforestation; bioenergy plus carbon capture and storage; direct air capture; soil carbon storage; and soil carbon storage with biochar. Each respondent viewed a randomized group of three options and was asked to estimate the likelihood that they'd support that strategy. They were also asked to rate their level of agreement with each of five statements related to humans' tampering with nature. In the final analysis, perceived naturalness was a strong indicator of support for soil carbon storage as a climate change mitigation strategy. Of the five CO2 removal strategies, support was highest (73%) for afforestation and reforestation; soil carbon storage ranked second, supported by 62% of those polled. And in this politically divided time, Schuldt said, support for soil carbon storage crossed the aisle. A total of 72% who identified as Democrats supported the strategy; among Republicans, 52% were in support. "We expected, and found, that Democrats support all kinds of climate strategies more than Republicans do," Schuldt said. "But the error I think we sometimes make is that we categorize all Democrats as being for it, and all Republicans as being against it. That's not true." Ultimately, Schuldt said, the goal is to allow policymakers to present the public with palatable options for addressing climate change. "There is a whole range of solutions out there," he said. "Then the question politically becomes, where do you start? Which one has the most buy-in? I think our data help speak to that."

10.1007/s10584-021-03121-0

Medicine

The rat's whiskers: Multidisciplinary research reveals how we sense texture

Maysam Oladazimi et al, Conveyance of texture signals along a rat whisker, Scientific Reports (2021). DOI: 10.1038/s41598-021-92770-3 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-021-92770-3

https://medicalxpress.com/news/2021-07-rat-whiskers-multidisciplinary-reveals-texture.html

Abstract Neuronal activities underlying a percept are constrained by the physics of sensory signals. In the tactile sense such constraints are frictional stick–slip events, occurring, amongst other vibrotactile features, when tactile sensors are in contact with objects. We reveal new biomechanical phenomena about the transmission of these microNewton forces at the tip of a rat’s whisker, where they occur, to the base where they engage primary afferents. Using high resolution videography and accurate measurement of axial and normal forces at the follicle, we show that the conical and curved rat whisker acts as a sign-converting amplification filter for moment to robustly engage primary afferents. Furthermore, we present a model based on geometrically nonlinear Cosserat rod theory and a friction model that recreates the observed whole-beam whisker dynamics. The model quantifies the relation between kinematics (positions and velocities) and dynamic variables (forces and moments). Thus, only videographic assessment of acceleration is required to estimate forces and moments measured by the primary afferents. Our study highlights how sensory systems deal with complex physical constraints of perceptual targets and sensors. Introduction The study of perception needs to consider the special physical (and chemical) properties of the sensory objects and their interaction with the sensors. In the tactile system, and most prominently in texture discrimination, such physical constraints include friction, arising with moving mechanical contact of integument (hair and skin) and touched object. Frictional phenomena with presumptive high impact on tactile coding are the so-called ‘stick–slip events’ (slips), short jerky movements that originate from storing and releasing energy into elastic object deformations on the microscopic level 1 . Here, we focus on rats’ vibrissae, or whiskers, tactile hairs that the animals actively move across textures. Rodent vibrissae are tapered giving them extraordinary elasticity 2 , 3 , and pliability 4 , 5 , 6 . Vibrissae movements vary in characteristic ways 7 , 8 , 9 , such that, in principle, the animal could choose to vary properties of the whisking motion, in order to optimize its performance in a challenging perceptual context. Tracking a point on the vibrissae shaft when rats touched textured surfaces 10 , 11 has demonstrated that sequences of frictional stick–slip events can carry a substantial amount of texture information, as well as information about the context, e.g. speed and distance to the texture surface 12 . System identification methods have identified that microslip-like features in the vibrotactile domain are well encoded on the tactile neural pathway 13 , 14 , 15 , 16 , 17 , 18 , 19 . Finally, behavioural studies suggest that such stick–slip signals determine perception 20 , 21 . From the study of biomechanical properties of whiskers, it has been suggested that dynamical variables, such as force and moment, are likely to be the definitive factors for stimulation of the mechanoreceptors in the follicle 2 , 4 . Unfortunately, observations of motion of intact whiskers used by a behaving animal are largely limited to whisker kinematics. It is therefore crucial to find out how kinematic variables associated with the whisker’s tip relate to dynamical variables. Moreover, it remains to establish the principles of mechanical transmission of signals along the whisker beam, from the highly pliable tip towards the much stiffer follicle, where neural signals are generated. It would seem that mathematical modelling is necessary to faithfully translate measured whisker kinematics into dynamics. At present, whisker mechanics has mainly been studied using either linear beam theory 4 , 22 or using a quasi-static approximation of a Euler elastica with non-uniform axial properties 23 , 24 , 25 . These approaches are useful for establishing the linear vibration frequencies in contactless whisking, and the quasi-static nonlinear deformations, like buckling, that likely occur when the pliable tip 6 is in continuous contact with a surface. For example, Goss and Chaouki 25 , using elastic beam theory, worked out criteria for the establishment of tip and line contacts, depending on the distance of the beam to the surface (see also 26 , 27 ). Also, the quasi-static Elastica2D code introduced in 24 has been successfully used by 28 to measure the axial follicle force. But none of these techniques can be used to measure the rapid dynamic forces in the transmission of stick–slip waves from the tip to the follicle. In the spirit of this earlier work, we therefore adapt the elastica formulation to include rapid dynamical deformation while the whisker is in contact with a surface. To do this we employ Cosserat rod-mechanics formulation (e.g. 29 ), to model the dynamics of velocity, force and moment at every position along the central axis of the whisker, with complex boundary conditions that can capture the impulsive forces caused by transitions between stick and slip contact. The earlier models likely capture the DC component of the forces at the follicle (which is likely to sense proximity and hardness) whereas our model is able to capture the rapid AC component of these forces, which we argue is how texture is likely to be coded. In the present work we include the case of tip contact only, but following the approach outlined in 25 it is straightforward in principle to extend our mathematical formulation to include line contact. In particular, a simple mechanical argument shows that the forces transmitted in the presence of line contact are dynamically equivalent to those of a shorter rod undergoing tip contact. We combine modelling with a limited number of biomechanical measurements to identify new biomechanical phenomena about how slip-stick transitions at the whisker tip are transmitted to force and moment information at its base. We will show that deflections at the tip, characterized by low moments and large excursions, are rapidly transmitted and result in large moment / short excursion movements at the base, and we will demonstrate that the non-linear modelling can recreate these observed characteristics. Our results strongly suggest that slips generated at the tip are presented in robust fashion to the base, the location of the neurites. Results For videography, the follicle end of a plucked whisker (Fig. 1 A,B) was glued to a vertical shaft that was made to rotate at constant angular velocity, using a stepper motor (Fig. 1 C). The gluing was such as to provide an effective clamped boundary condition, with the whisker free to bend in an approximately horizontal plane. The tip was allowed to contact with the interior of a vertically mounted semi-cylindrical arena, the midpoint of which coincided with the rotation centre of the whisker. We measured the fine-scale, spatio-temporal kinematics of the whisker shaft while the tip was in moving contact with the arena surface clad with sandpapers of different roughness 12 . A rat whisker C4 was used for the experiments. Its length ( \(l\) ) was 28.36 mm, and its radius at base was 69 μm. (Core findings of the study were confirmed by measurement of two more whiskers, one C3 and the other D3. Whiskers within one animal or across animals vary in their geometric outline, a variance that we did not attempt to systematically explore with the present experiments). The whisker’s shape was conical with the exception that at the very tip it was truncated at the point where the radius assumed 3 μm. The shaft rotation speed was 420°/s (in some runs also 840°/s and 1260°/s). Two different surface distances were used, \(x=l-1\) mm or \(x=l-7\) mm. Care was taken to align the intrinsic curvature of the whisker with the horizontal plane and to measure only the movement toward the concave side of the whisker. In the rat’s face such movement would largely correspond to whisker protraction. The movement of the free whisker shaft (i.e. from the parts of the tip not in contact with the sandpaper to the parts close to the follicle that were not fixed and visually obstructed by the glue and rod) was monitored by a camera mounted above operating at ultra-high frame rate (4 kHz if not stated otherwise) (Fig. 1 C). Figure 1 Whisker and biomechanical measurement. ( A ) Rat whisker C4. The rat ‘s head and location of whisker field is shown. The position of C4 is the green dot in the magnified whisker field. Conical shape, length \(l\) and distance \(x\) (cf. panel C) are shown. (Note that we repeated core measurements using also a C3 and a D3 whisker, the locations of which are indicated by grey dots) ( B ) Microscopic images of whisker tip (left) and base (right). ( C ) Experimental set up (view from the recording camera). The rotating rod is seen on top. At the bottom, the half-cylinder holding the sandpaper is shown. Experimental variables position on whisker (length) \(l\) , driving speed \(v\) , and distance \(x\) are indicated. ( D ) Stick–slip events in a whisker moving across a texture. Videographic analysis of whisker shape and location in \(x\) and \(y\) during one protraction (all frames of the video are shown, the sandpaper was located at distance 1 mm less than whisker length: i.e. at \(x=27.36\) mm). Red: all frames below the driving velocity—pointing to sticking periods. Green: instances with local maximum of acceleration—pointing to slips. The first frame captured when the whisker was moving free in air (no texture) is shown for comparison (violet). Inset: Cross-correlogram of accelerations at the tip ( \(x=24\) mm, top white line) vs. that at the base ( \(x=3\) mm, bottom white line). The grey lines indicate time lag \(t=0\) (abscissa) and correlation coefficient \(r=0\) (ordinate). ( E ) Method to identify stick and slip events. Sticks (red) were found by thresholding the velocity trace (at driving speed 420°/s) and minimizing the trace below that limit. Slips (green) were found by maximizing the acceleration strips above threshold (2 standard deviations found with movement in air/no contact). On top all events are shown again aligned on the time axis. Further, a short sequence of 4 slips and 7 sticks is shown in the blow-up. Full size image Image processing techniques were used to extract the position of the centreline of the whisker at every frame (resolution in space: 14 µm; in time: 0.25 ms, see “ Methods” section). From these data, the normal velocity (in the plane of rotation swept out by the whisker) and acceleration at each point on the shaft were assessed. Kinematic traces of the point 3 mm from the base were used to define ‘sticking phases’ as intervals when the normal velocity reached below the rotational speed of the stepper motor, as well as ‘slipping phases’ as intervals in which whisker acceleration exceeded a value of two standard deviations of the distribution of acceleration measurements during contactless movement (‘movement in air’, as done before in 12 ). Colouring the whisker positions according to these sticking and slipping phases reveals prominent and alternating frictional stick–slip events (Fig. 1 D), as has been described before 10 , 11 , 30 . Minimizing velocity in sticking phases and maximizing acceleration in slipping phases yielded the operationally defined ‘stick and slip events’ (or ‘sticks’ and ‘slips’ for short; green and red dots in Fig. 1 E, cf. 12 ). Slip events can be clearly discerned in the acceleration traces of both whisker tip and base, and, confirming an earlier study 12 , they change their appearance when in moving contact with either the P80 or P1200 sandpapers, or when the whisker is engaged with different distances to the texture (Fig. S1 ). As a note of caution, we wish to emphasize that while the definition of sticks and slips is useful to demonstrate the general relationship of biomechanical variables with stick and slip phases (see below), the precise distinction between slips and sticks in individual cases cannot always be done in straightforward ways, e.g. because in reality there can be complex frictional phenomena like micro-slips and creep-like movements. High transmission speed along the whisker To elucidate the speed of biomechanical transmission from tip to base of the whisker, we plotted the acceleration of tip and base in the plane of whisker movement (tip: \(a_{x} (s = 24~\;{\text{mm}})\) ; base: \(a_{x} (s = 3\;{\text{~mm}})\) ), and calculated their cross-correlation. We found that the acceleration at these two points is negatively correlated, i.e. when the tip speeds up, the base slows down (Fig. 1 D, inset). The negative correlation was precise in time: only accelerations at time lag 0 were negative, indicating that conveyance of vibrations are ultra-fast, i.e. non-discriminable using our camera frame rate of 4000 Hz (time bin: 0.25 ms). This finding generalized to all contact conditions studied (i.e. textures, distances, Fig. S1 ). Whiskers vibrate in the second bending mode Next, we wanted to find out how the observed opposing movement directions near the tip and the base of the whisker plays out along the entire whisker beam. To that end, we estimated the local curvature (at all pixels on the whisker centreline) by determining the angle \(\theta\) spanned by the vector orthogonal to whisker base and the tangential force acting on the whisker beam (Fig. 2 A inset). Curvature \(\kappa\) is defined by the spatial derivative of \(\theta\) ( \(\kappa (s) = ~d\theta\) / \(ds\) ). We plot curvature \(\Delta \kappa ={\kappa }_{i}-\kappa\) (i.e. curvature relative to the intrinsic curvature \({\kappa }_{i}\) of the whisker), at all points along the shaft at each sampled time point during a single sweep (Fig. 2 A–C). When moving in air (with the sandpaper-coated screen removed) the whisker oscillated between roughly its intrinsic shape ( \(\Delta \kappa~ = ~0\) ; arrow heads in Fig. 2 A), and a deflection with the tip curving backward ( \(\Delta \kappa (tip){{~ > ~}}0\) ) and the base curving forward \(( \Delta \kappa (base)~ < ~0\) ; between arrow heads in Fig. 2 A), corresponding to an oscillation in the second bending mode. The result of this vibration mode is that the change in curvature relative to the intrinsic shape changes along the beam, with a node located at about 8 mm distance from the follicle. The oscillation frequency was close to 200 Hz (Fig. 2 A). Figure 2 Curvature of the whisker in moving contact. ( A ) Movement in air. Curvature of each point along the beam and across time is colour coded. Inset: Calculation of curvature \(\kappa\) . The angle \(\theta\) at each point of the beam is measured and the curvature calculated as \(\kappa \left(s\right)=\theta \text{'}(s)=d\theta /ds.\) We plot \(\Delta \kappa\) , in which the intrinsic curvature of the whisker (at rest in air) is subtracted. ( B ) Curvature as in A when in contact with a smooth sandpaper (P1200) and ( C ) a rough sandpaper (P80). For both the distance was (distance \(x=l-1\) mm; speed \(v=420\) °/s). Arrows point to curvature changes evoked by stick–slip events being transmitted rapidly along the beam and therefore appearing as vertical stripes. ( D ) Spectra of base movements shown in ABC. Inset shows the same data rescaled to dB. Note the prominent peak in the spectrum at ~ 200 Hz in the ‘free in air’ condition, indicating the 2nd bending mode. Full size image When the whisker engaged with the smoother sandpaper (grain size P1200; at a distance 1 mm less than its length), \(\Delta \kappa\) at the tip changes to forward direction ( \(\Delta \kappa (tip)~ < ~0\) ) as expected (Fig. 2 B,C). Thus, the free vibrations of the whisker were damped, but they were nevertheless still visible at about the same frequency (arrow heads, Fig. 2 B). In addition, rapid waves corresponding to the onset of a slip, became visible. These are seen as the irregular vertical stripes, two of which are marked by arrows. The vertical nature of these stripes again suggests their ultra-fast transmission along the beam. Similar results were observed with P80, the rougher sandpaper (Fig. 2 C). The second bending mode was robust against variations of texture engagement. With stronger engagement, either by increasing roughness (Fig. 2 C) or decreasing distance (Fig. S1 ) the node shifted only slightly toward the base (to around 5 mm from the base). In summary, we note that second order bending with opposite curvature at the tip and base with a node at a distance of 5–8 mm from the base are a characteristic feature of whisker vibrations in varying contact situations. The spectra of whisker vibration when engaged at distance \(l-1\) mm are shown in Fig. 2 D. The stable second harmonic of the oscillation stands out with movement in air (green) and against the smooth P1200 sandpaper (blue). The fundamental frequency was nearly absent. This oscillatory pattern breaks down when the whisker is engaged with the rough P80 sandpaper (red) because the signal from the slip and stick events begins to dominate the signal. Weak and wide tip excursions are transformed into strong and short ones at the base The curvature measurements provide a basis to infer the moment \({M}_{na}(s,t)\) (spanned by the normal force F n and the axial force F a see inset in Fig. 3 B) about the vertical axis at every point along the whisker along its centreline \(s\) and time \(t\) . The effect of \({M}_{na}\) is to locally bend the whisker beam at the point in question within its plane of movement. At each point position \(s,\) \({M}_{na}\) can be calculated, from the beam’s Young’s modulus ( \(E\) ), its measured curvature \(\kappa (s\) ), intrinsic curvature of the whisker \({\kappa }_{i}(s\) ) (Fig. 3 A), and the second moment of area of its circular cross-section Figure 3 Moment amplification during transmission from whisker tip to base. ( A ) Curvature of whisker in contact with P80 (speed 420°/s). Each line represents one time bin (0.25 ms). Two frames toward the extremes of the curvature are coloured to demonstrate a node of vibration at around 10 mm from the tip (2nd bending mode). ( B ) Moment \({M}_{na}\) as calculated from Eq. ( 2 ). Left Inset: Schematic of normal and axial forces ( \(F_{n} ,\) F a ) and moment ( \({M}_{na}\) ) acting in the plane of whisker movement. Moments are negligible at the tip and small negative moment builds up a short distance from the tip. (cf. right inset). The left bundle of lines indicate moment calculated using \(\kappa\) (instead \(\kappa -{\kappa }_{i}\) , cf. Equation 2). Two arbitrary frames are highlighted in green to demonstrate the node of vibration. ( C ) Moment \({M}_{na}\) at tip (s = 24 mm) and base (s = 3 mm) of the whisker (note the three orders of magnitude difference in scale). The red areas indicate periods of sticking ( \(v<420\) °/s). ( D , E ) Average moment \({M}_{na}\) with respect to stick (red) and slip (green) events (as identified using the method in Fig. 1 B). Nine correlograms taken from traces measured with three different sandpapers and using three driving speeds are shown. Full size image $$I\left( s \right) = \frac{\pi }{4}r(s)^{4}$$ (1) Assuming a linearly elastic constitutive model for a conical rod with circular cross-section, we then write $$M_{{na}} \left( {s,t} \right) = EI\left( s \right)\left[ {\kappa \left( {s,t} \right) - \kappa _{i} \left( s \right)} \right],$$ (2) (see also Eq. 6 and related text in materials and methods). Combining Eqs. ( 1 ) and ( 2 ), given the highly tapered nature of the whisker, we note there will be a continuous build up in strength of moment along the whisker as we approach the base end. Thus, tiny forces deflecting the distal pliable part of the whisker lead to a build-up of moment along the whisker beam. This guarantees that the base of the whisker signals a robust moment signal in response to slight tip deflections (Fig. 3 B). To quantify the amplification of moment we calculated the ratio of variances of moments measured at the site 1 mm distance from the tip and same distance from the base of the whisker ( \(\frac{{var}_{tip}}{{var}_{base}}\) ). This ratio was 1.4e−4 for the C4 whisker shown in Fig. 3 B. We studied two more whiskers (one C3 and one D3). These were measured with the drum rotating (cf. Fig. 4 ) instead of rotating whisker, and analysed in exactly the same way. They yielded still smaller ratios, i.e. higher amplification (C3, length 44.56 mm, ratio: 2.6e−4; D3, length 45.22 mm, ratio: 2.57e−4). Interestingly, the intrinsic curvature \({\kappa }_{i}\) acts to shift the range of moments experienced by the follicle and attached neurites such that a sign change results, a fact which likely contributes to the phenomenon that the whisker vibrates mainly in its second bending mode (cf. Fig. 2 and S1 , S2 ). The time series of M na at tip and base and their inverse relationship can be appreciated in Fig. 3 C. Triggering moment at the base by stick and slip events revealed a systematic relationship between the two variables: on average, moment builds up during stick (Fig. 3 D) and is released during slip (Fig. 3 E). Figure 4 Micro-force measurement at the whisker base. ( A ) Left inset: Photograph of the measurement set-up. The sensor was mounted on an immobile ledge directly above the rotation axis of a rotating drum. The drum was perforated to save weight and contained a wall to fix the sandpaper. Centre: Schematic of measurement set-up. The whisker (green) was mounted on the sensor (dark grey and pink), and brought in contact with the sandpaper (violet), which in turn was rotated by the drum (grey arrow). Two measurements were performed. First, the forces in the plane of whisker movement ( \(F_{n} ,\) F a ) were directly measured by the piezoresistive sensor (right inset). Second, the lateral acceleration at the base (s = 3 mm) ( \({a}_{x}\) ) was assessed by videography (cf. Fig. S1 ). ( B ) The top graph plots \({a}_{x}\) (red) on top of \({F}_{n}\) , the bottom one plots \({F}_{n}\) and \({F}_{a}\) . Acceleration is in excellent correspondence to the forces acting on the base. ( C ) Average normal force triggered on stick events. \({F}_{n}\) builds up during sticks (as well as \({F}_{a}\) , not shown). Compare with the parallel increment of moment \({M}_{na}\) as calculated from curvature measurements (Fig. 3 D). Full size image In summary these results suggest that the taper of the whisker together with its intrinsic curvature play an important role in determining the range of moments experienced in the follicle, the site of neurites of primary afferents, and that stick–slip events feature prominently in the robust moment signal that is received there. Acceleration is a proxy for whisker-beam forces and moments In order to verify the calculations of moment from kinematic data we chose to directly measure the force acting at the whisker base when in contact with the sandpaper. To this end we used a piezoresistive force sensor probe that is capable to measure multiaxial forces in the nano- to microNewton range (Fig. 4 A; Fig. S3 31 ). The whisker was glued to the probe’s cantilever and the sandpaper was brought into contact with the whisker tip and moved past it using a rotating drum. In this configuration, the force sensor was able to measure the normal (F n ) and axial force (F a ) acting at the whisker base (light and dark blue arrows in Fig. 4 A). At the same time, we used ultra-fast videography (at a 9.6 kHz frame rate) as described above to measure the acceleration close to the base ( \({a}_{x}\) , red in Fig. 4 ). We find that F n, F a , and a x , are strongly correlated ( \(\left| {r_{{F_{n} }} ,_{{a_{x} }} } \right| = 0.9349;\;\left| {r_{{F_{a} }} ,_{{a_{x} }} } \right| = 0.9302;\left| {r_{{F_{n} }} ,_{{F_{a} }} } \right| = 0.9657\) ; Fig. 4 B). In consequence, the average force observed during sticks shows a clear peak, as exemplified by event-triggered average of F n plotted in Fig. 4 C. In summary, the time series of acceleration measured close to the base provides a reasonable approximation to the time series of forces acting on the base. However, we wish to emphasize that the usage of acceleration as a proxy for forces at the base suffers some limitation because moment (and possibly also forces) are subject to a large augmentation from tip to base while acceleration is not (Fig. S1 ). Cosserat mechanics with friction model recreates1 whisker-beam dynamics It has been shown that whisker bending in air can be estimated quite well by models based on linear beam theory 4 , 23 . However, the conical whisker’s highly pliable tip together with frictional forces acting at it renders it highly unlikely that linear beam theory will be sufficient. Rather, it is clear that a formulation is required that allows arbitrarily large, geometrically exact deformations. One way to see this is to calculate the critical Euler load \(P_{E}\) defined by $$P_{{E~}} = \frac{{\pi ^{2} EI}}{{4l^{2} }},$$ (3) with \(E\) being the Young Modulus, \(I\) the second moment of area (Eq. 1 ), and \(l\) the rod length. Applying reasonable assumptions about material constants of a homogenous, isotropic conical rod, this equation yields critical Euler loads in the order of 1 nN. Buckling of the beam is thus expected with minimal forces—widely exceeding those we have measured to be acting at the whisker’s tip. We therefore turned to Cosserat mechanics, which is able to account for gross geometrical non-linearities (see e.g. 29 ). Cosserat models capture the dynamics of the space-centroid line (the line intersecting the centre of all circular cross-section areas of the conical rod; Fig. S4A ) using a set of coupled partial differential equations in space and time. We deem such spatio-temporal coupling essential to recreate rapidly propagating and highly fluctuating frictional movements at the whisker tip. Such motion could not be captured by a quasi-static Euler–Bernoulli formulation, in which the temporal dimension is neglected. It is noteworthy that the whisker used here has an aspect ratio (length \(l\) to cross-sectional radius \(r\) ) of more than 400. For such slender objects it is widely accepted that the effect of shear along the rod’s longitudinal axis is negligible and can be ignored. In this case, a simpler mathematical formulation can be used, which assumes that cross-section areas are normal to the tangent of the space-centroid line (see materials and methods for details). The whisker is modelled as a truncated cone with length \(l\) (from base to truncation point, the latter is called ‘tip’ throughout the paper), and homogeneous linearly elastic material properties (i.e. a constant Young Modulus \(E\) ). The elasticity of the rod for bending movement perpendicular to its longitudinal axis will increase significantly from base to tip, because the circular cross section decreases (see eqns. (1) and (2)). The central variable to describe the whisker mechanics is the local curvature \(\kappa \left(s,t\right)\) defined as the first derivative of the local curvature angle \(\partial \theta (s,t)/\partial s\) . We suppose that the rod has an intrinsic curvature profile \({\kappa }_{i}\left(s\right)\) , which we obtained from measurements of the real whisker. As in the experiments, we assume that the rod moves in the horizontal plane laid out by the intrinsic curvature (with tip bent in the direction of sandpaper movement, Fig. S4A ). Accurate modelling the effects of high-speed dynamical friction between two dry surfaces is a current topic of active research within the field of tribology, with no consensus emerging on which, if any, closed-form low-order model is able to capture the dynamical consequences of surface roughness (see e.g. 32 ). We have chosen to use a so-called rate-and-state frictional law (see 33 , 34 , 35 ). This approach models the dynamical friction coefficient \(\mu\) and resulting friction force \({F}_{T}=\mu p\) (Fig. S4 ) based on parameters of past sliding history (the interface ‘state’), and the relative velocity between the contacts (the ‘rate’ of deformation, see 34 ). This formulation is richer than classical Coulomb friction as it additionally allows setting a parameter \(L\) called ‘slip length’, sometimes considered a proxy for roughness. Many geometric and material property parameters were taken as measured. A full list of model parameters is given in Table 1 . Table 1 List of all input parameters of the Cosserat and the friction model. Full size table Modelling details of frictional movements of whiskers in contact with specific surfaces will need future versions of the frictional model including also material properties and asperity size and distributions of surface materials. Here we content ourselves with varying \(L\) to adjust model behaviour qualitatively to match our principal experimental findings. In particular we asked whether stick–slip can be captured with our approach, and how they are represented by kinematic and dynamical model outputs. We further aimed at capturing the experimentally observed generation of increasing moments along the whisker beam, second bending mode and the rapid transmission of slip events along the beam. Model dynamics Analysing model kinematic output in steady state conditions (using \(L =10\) μm, \(v =0.2\) m/s, \(\alpha ~ = ~1\) ) showed that stick and slip phases and events were readily generated by the model. The choice of parameter values is explained in materials and methods (see Table 1 for a full list). Note that \(v =0.2\) m/s corresponds to a shaft rotation of 420°/s. Base angular velocity (Fig. 5 A) undershot driving velocity regularly and was followed by transients surpassing accelerations observed without contact. Searching for local extremes in these cases identified a stick–slip pattern similar to that seen in experiments (Fig. 5 A,B) (cf. Fig. 1 B). The model also captured the experimental fact that the whisker tended to vibrate at the second bending mode: tip and base accelerations were negatively correlated (Fig. 5 B,C; correlation coefficient \(r=-0.36\) ; the experimental data of comparable conditions [distance \(l-1\) mm and smooth sandpaper surface, cf. Fig. 2 B] yielded \(r=-0.47\) ). Further, the experiment showed a tight correlation between kinematic and dynamical variables, a fact which the model captured: acceleration and normal force were correlated at a coefficient \(r=0.44\) . This correlation coefficient is lower than that seen in the experiment, perhaps reflecting that the model contact is with a smooth surface while the experiment used sandpaper surfaces containing asperities. Tip and base moment occur in opposite directions reflecting the second mode of bending (Fig. 5 D,E,F, cf. Fig. 3 C) (please note that the model used \({\kappa }_{i}=0\) for ease of computing [see materials and methods] explaining the shift of moment toward negative values, and therefore, a more moderate negative correlation [ \(r=-0.34\) ] as compared to the experiment). Plotting the average moment around stick events (Fig. 5 G) we note that, as observed with the real whisker (cf. Figures 3 E, 4 C), stick events are followed by an increase in moment along the whisker. Thus, it appears that during a stick event, the moment builds up, and is subsequently released during the slip. Figure 5 Kinematic and dynamical output variables of the model (see full set of output variables in Table 2 ). Stick and slip events were identified as in the videographic measurements (cf. Fig. 1 E, dots in base kinematic traces in which the identification was done, and lines throughout, red: sticks; green: slips) ( A – C ) Kinematic variables ( D – F ) Dynamical variables. ( G ) The average normal force to the whisker base builds up to a maximum around a stick event. ( H ) Spectra of moments for model runs using different values of \(L\) [5,50,500]. Top and bottom are the same data plotted as dB or PSD. Medium values of \(L\) match the experimental data (cf. Fig. 2 D) best, as this model best recreates the dominance of the 2nd bending mode (2nd harmonic is marked by an asterisk). ( I ) Curvature output ( \(\Delta \kappa\) ) of the model using \(L=50\) μm. (experimental data cf. Fig. 2 B). ( J ) Moments \({M}_{na}(s,t)\) (each line represents moments at one time point along \(s\) ). A few lines have been coloured green to demonstrate the node of vibration at 2nd bending mode. Comparable experimental data are in Fig. 3 B. Full size image Table 2 List of all output parameters of the Cosserat and the friction model. Full size table To adjust \(L\) we ran the model varying \(L\) (in the range from 5 to 500 μm). The resulting position spectra (Fig. 5 H) revealed that the predominance of the second bending mode as observed in the experiment were best represented by middling values of \(L\) (here 50 μm). Using the model with \(L = 50\;\mu {\text{m}}\) , to plot curvature \(\Delta \kappa ={\kappa }_{i}-\kappa\) as done for experimental data in Fig. 2 , yielded qualitatively similar oscillations to the ones observed with real whisker movement in contact with a smooth surface (Fig. 5 I; cf. Fig. 2 B). Calculating moments along the whisker beam revealed a strong increment from tip to base with the node of bending located in the lower half of whisker length (Fig. 5 J), as seen in the experiment (Fig. 3 B). As reported above, nodes (the signature of the second bending mode) were observed between 5 and 15 mm. The range of node locations are demonstrated by the four moment excursions highlighted in Fig. 5 J. Discussion This study provides fundamental insights about how vibrations are conveyed from tip to base of a rat conical whisker in moving contact with a texture. Biomechanical measurements reveal that conveyance of vibrations from tip to base is ultrafast, and utilizes the second mode of bending. Tip movements are transmitted to the base as moment, a dynamical variable. Due to tapering, the moment, which is vanishingly small at the tip, builds up to significant amounts at the follicle. A relative measure of forces acting on the follicle—their change in time—is available from kinematic data in the form of acceleration of the whisker beam close to the follicle. Stick–slip frictional movements feature prominently in dynamical signals conveyed down the beam. We demonstrate the novel mechanistic principles in a limited number of whiskers, such that the detailed intra and inter-individual variation of these phenomena has to be worked out in the future. We further established a realistic model of moving-contact whisker biomechanics based on Cosserat geometric theory linked to a state-rate friction law, which is the first to capture the rapid spatio-temporal transmission of frictional stick–slip movements along the whisker. This model qualitatively recreated the fast conduction, second order bending, and the conversion of frictional stick–slip movements at the whisker tip into robust bending moment at its base. Ultrafast transmission and second-order bending mode The measurements of conveyance speed exceeded the temporal resolution of our ultrafast camera system (conveyance time from tip to base ( \(< ~0.25\;{\text{ms}}\) ). The conical shape of whiskers and the keratinous material suggest high pliability of the whisker, as has been previously observed 6 . This is reflected by our observation of a strong correlation between lateral deflection ( \({a}_{x}\) ) and normal force ( \({F}_{n}\) ). The inverse relationship of bending at tip and base was a surprising finding, and is explained by the robust observation that the static deformation of the whisker is in its second bending mode. Our mathematical model captured the phenomenon in relative terms (antagonistic relative movement) and partially in absolute terms (absolute direction of movement). We therefore conclude that the second bending mode is mechanistically brought about by the elastic and geometrical properties of the beam incorporated in the Cosserat model. A further role in specifying details of second bending mode vibration is played by the intrinsic curvature of the whisker. The typical movement direction in strongly whisking species (like the rat studied here) has been associated with distinct innervation patterns most conspicuously at the inner conical body of the follicle 35 . The observed inverse relationship of tip vs. base deflection (and amplified moment; see next paragraph) will likely be instrumental to decipher the mechano-electrical transduction in those specifically formed and distributed end organs. Transmission of frictional stick–slip movements to the follicle Combining measurement of acceleration and force acting at the whisker base we found that the acceleration time series is a good proxy for the forces acting on the base: It is highly correlated with normal force \({F}_{n}\) and axial force \({F}_{a}\) . This observation is likely to be of benefit for investigators, who work with behaving animals and who can thus only measure kinematic variables (positions, velocity and acceleration) but wish to infer dynamical variables (forces and moments) acting on the follicle. By direct measurement of acceleration, one could hope to capture at least the relative amplitude of dynamical variables. Absolute values of acceleration fluctuations at the whisker base, however, seem to vary significantly with the overall position of the node point of the second bending mode. A more robust measurement seems to be curvature fluctuations, and hence fluctuations of moment at the follicle. Indeed, we showed that the whisker appears to be an effective device for turning large distance, weak excursions of the pliable tip into short excursion, high moment fluctuations at the follicle. The above-mentioned correlation of normal and axial forces, \({F}_{n}\) and \({F}_{a}\) , acting on the whisker can be intuitively understood: Whenever normal force builds up—i.e. during stick phases—axial force (directed toward the tip) is increased because during a stick, the tip of the whisker stays behind and thus tends to pull out the base from its fixture. During slip the situation is roughly the inverse. In this study we were able to verify these intuitive relationships directly by assessing stick and slip events from videographic velocity and acceleration data 12 , with simultaneous assessment of moment (calculated from curvature, Fig. 3 ), and direct piezoresistive force measurements (Fig. 4 ). We found that moment \({M}_{na}\) and normal and axial forces \(F_{n} ,\) build up during whisker stick events. In reverse, slip events are associated with the relaxation of these variables. Thus, we demonstrate that slips are robustly represented by dynamical variables along axial and normal dimensions. Experiments that found vivid responses of primary afferents after ramp-and-hold movements of the whisker along the two dimensions 36 support the notion that the existent highly specialized classes of end organs 37 partly have evolved to pick up follicle forces and moments along the two dimensions. The prominence of representations of stick–slip sequences in dynamic variables, shown here, suggests that it is those stick–slip events that dominate transduction into primary afferent action potentials. Prominent coding of temporally local deflection waveforms in primary afferents 14 , and the reported sparse but reliable response of S1 neurons to slips 16 provide supporting evidence for this conjecture. As previous studies have concordantly shown that the kinematic profile of slips contains substantial texture information 10 , 11 , 12 , we hypothesize that the biomechanical transmission of stick–slip sequences, revealed here, sets the stage for neuronal coding of frictional movements and their hypothesized exploitation for purposes of texture discrimination 1 . Variation of biomechanical phenomena across whiskers and individuals Mystacial whiskers are all long conical structures, the principle feature giving rise to the experimental data and modelling results reported here. In this initial study we focussed on the identification of new biomechanistic principles and did not attempt to give a detailed view about the variability of these principles. Slip transmission across whisker types on the snout, across animals, across the growth cycle of hair, and to different types of whiskers found in other species, are all topics left to be studied in future work. This said, we expect the core features reported here to generalize in some way across cone-shaped whiskers, as they are uniquely related to the tapering and pliability of the whisker tip. The fact that different whiskers (e.g. the small and immobile microvibrissae vs the mobile macrovibrissae) are used in different behavioural contexts 38 , imply that variations of slip coding might be found there. Specific behavioural functions of the well-described systematic changes in whisker morphology across the mystacial pad 5 are unknown. The fact that encoding textures with slips varies systematically across the rat’s mystacial pad 12 , and whiskers of different locations show different frequencies of vibration when moving in free air 2 , 3 , point to a possible gradient in biomechanical mechanisms, also when in touch with a surface, that need to be aligned to the yet unknown functional aspects of whiskers ordered along arcs and rows. Mathematical model of whisker bending when in moving contact We have established a mathematical model that, to our knowledge, is the first that exceeds quasi-static approaches as implemented before 6 , 23 , 24 , and uses an analytic mathematical formulation to capture critical dynamical features of whisker vibration under conditions of moving contact with a surface. Amongst them are the generation of stick–slip sequences and their transmission and translation into dynamical variables, the increment of moment along the beam, the second bending mode of vibration, and the correlation of kinematic variables with forces acting on the whisker base. The output of the model thus provides access to dynamical mechanical quantities (moments and forces), supposedly critical for tactile perception. We hold that the model will be well suited for future parametric studies towards understanding texture identification, as it implements two novel features that set it apart from previous approaches. Firstly, the model for the first time incorporates a frictional model, which we deem critical to study the sweeping touch across surfaces (which necessarily involves frictional contact), typically executed with the aim of texture identification. Secondly, the model is the first to use the Cosserat formulation of the whisker beam, which allows for arbitrarily large, geometrically exact deformations. With these two novel features in place, our model prepares the ground to model texture identification—without doubt an important function of whisker movements. There are several aspects of the whisker vibrations described here that will benefit from non-linear geometry and friction model. The first is the ultra-rapid transmission of texture information from tip to follicle. Another particular aspect of tapered whiskers will surely require both novel aspects of the model: compared to hair of cylindrical shape (e.g. vellum hair), tapered whiskers rather quickly buckle when in moving contact to a surface. Such buckling will happen at the tapered parts close to the whisker’s tip, and will switch ‘point contact’ (only tip in contact with surface) to ‘line contact’ (contact made by a longitudinal stretch of the tapering whisker close to the tip). If, as appears highly likely, buckling is an issue in whisker related touch, both novel elements of our model, the frictional as well as the Cosserat sub-models, will be indispensable tools. In the present work, we simply followed the mathematical prediction that force is only transmitted from the lift-off point of the line contact. Thus, we assumed that a whisker with line contact over a length \({\boldsymbol{l}}_{0}\) is dynamically equivalent to whisker of length \(\boldsymbol{l}-{\boldsymbol{l}}_{0}\) in point contact. While this may be correct at a first level of approximation, details in changing frictional forces when going from line to point contacts and vice versa may still turn out to be an important aspect in texture discrimination. Methods Kinematic measurement using videography Moving contact in a constant context (i.e. fixed whisker speed and object distance) was established using a published protocol (Fig. 1 A–C) 12 . In short, the whisker C4 was plucked from a dead animal (Sprague–Dawley, male, age 4 months), sacrificed for an experiment unrelated to the present report. Three whiskers (C4, C3, D4) were used in this study. Care was taken to secure the whisker in its entire length, that the follicle was included, and the hair shaft was devoid of kinks. Length, diameter and intrinsic curvature were measured using microscopic pictures before and after imaging the whisker. These measurements did not yield any measurable difference before and after the experiment. Confirming previous reports, the whisker was approximately of a conical shape 5 , 22 . The most salient divergence from a pure cone shape were shape irregularities on the last hundreds of microns of the tip. The cone’s tip was cut at the point where the beam measured 3 μm in diameter. The exact measurements of the whisker cone were: follicle length: 1056 μm; diameter at the base: 138.6 μm; tip diameter: 3.08 μm; length (s) from follicle to tip; 28.48 mm. Sandpapers of two grades were used: P80 (rough) and P1200 (smooth), both part of the standard series issued by the Federation of European Producers of Abrasives with mean grain diameters of 201 and 15.3 μm respectively. Free movement in air (contactless) of the whisker was also recorded for reference. Rotation speed was set to 420°∕s which is representative of a lower speed of whisking observed in vivo 39 . In some cases velocities of 840 and 1260°/s were used in addition (Fig. 3 D,E). The whisker was clamped at its follicular site so that the whisker’s axis was perpendicular to the rotational axis of a shaft rotated by a stepper motor (Orientalmotor, Tokyo, Japan). A high-speed camera (GMCLTR1.3CL-SSF LTR Mikrotron, Unterschleissheim, Germany) with a Tokina objective (Tokina 100 mm f/2.8, AT-X PRO—Macro, 16 × 16 μm 2 / pixel size, Kenko Tokina Co., Ltd., Japan) was positioned above the rotational plane of the whisker in order to record its planar motion. The acquired videos had a resolution of 480 × 270 pixels at 4 kHz (data in Fig. 3 ). Sandpapers were mounted on a cylindrical rigid plastic shield that was positioned in such a manner that its central axis coincided with the axis of the rotational shaft. Two different plastic shields were used, each precision-made using a 3D printer, having the geometry of cylindrical segments with radius 1 mm and 7 mm less than the whisker length (Fig. 1 A). Hence the distance of the follicle to the textured surface can be varied, although it was held constant for the duration of a particular rotational movement. One measurement cycle consisted in forward movement of 60° across the sandpaper (convex surface of the whisker leading, as with protraction in the intact animal). The first 15 ms of trial time containing movement transients from rest to steady state were discarded. Force measurements The force measurements (Fig. 4 ) were done using piezoresistive force sensors 31 , specially designed and fabricated for this purpose in the laboratory of authors I.S. and K.N. Design and the dimensions of the piezo-resistive force sensor is shown in figure S3 . The sensor lever thickness and width was 20 μm, and 195 μm respectively. This type of sensor is sensitive to forces as small as 10 nN (see details in 31 ). The whisker was mounted using micromanipulation on the lever using UV activated glue. The force sensor plus attached whisker were then carefully mounted on a central ledge, such that the whisker base was located in the centre of a rotational drum holding a sandpaper-clad cylindrical arena, identical to the one used for videography. The difference to the videography was that force sensor and vibrissa were held in place and the arena moved around them (sensor/vibrissa movement proved incompatible with the integrity of the sensor). Rotation velocity was 420°/s and distance of the arena to the whisker was 1 mm less than the whisker length. Thus, the relative speed and distance of movement of whisker across sandpaper was identical in experiments depicted in Figs. 1 , 2 , 3 and 4 . For videography during force measurement a camera (Fastcom Mini WX100, Photron, Tokyo, Japan) at a resolution of 1600 × 360 pixels at 9.6 kHz was used (data in Fig. 4 ). Data analysis We used Matlab to write a bespoke algorithm based on image contrast to track the whisker’s coordinates in the plane of its motion from each frame, leading to a time-series for a set of discrete points \(\left\{ {x_{i} \left( t \right),~y_{i} \left( t \right)} \right\}\) along the whisker’s axis with the spatial resolution of 42 µm. Briefly, lighting was adjusted such that the acquired images were close to pure black and white with the white whisker on a black background. Starting from a manual starting point at the base the algorithm iteratively traced out the outline of the whisker by detecting the two steps in contrast (black to white and white to black) on a circle with radius of 154 µm around the last point that indicated the contours of the whisker. The centreline of the whisker was constructed as the centre between the contour steps and the starting point for the next iteration was chosen as end of the so-far constructed centreline (Fig. 1 D shows a time sequence of centrelines). The resolution of the image was 14 × 14 µm. The data at each time-step was used to compute an arclength coordinate \(s\) and the signed curvature at each material point along the whisker, respectively given by $$s~ = \int_{{x_{0} }}^{x} {\sqrt {1~ + ~\left( {\frac{{dy}}{{dx}}} \right)^{2} } } ~dx,\quad ~\kappa ~ = ~x^{\prime } \left( s \right)y^{\prime\prime} \left( s \right) - ~y^{\prime } \left( s \right)x^{\prime\prime } \left( s \right),$$ (4) where the prime refers to the derivative with respect to the arclength \(s\) , the derivatives being numerically evaluated by finite differences and the integral using the cumulative trapezoidal method. In keeping with simple estimation of axial stiffness, we make the assumption that the whisker is inextensible. Note that, as the spatial sampling frequency varied slightly from one frame to another (mostly due to loss of data points near the whisker’s tip), we used spline interpolation (via the Matlab function interp1 to form a regular spaced grid of the arclength from the raw data, imposing an arclength increment of ds = 1 μm. The trajectory of any material point labelled by its arclength coordinate is then stored in an array of points \((x\left(s,t\right),y\left(s,t\right))\) . We also monitored the time series of the position angle \(\psi \left( {s,t} \right): = \tan ^{{ - 1}} [y(s,t)/x(s,t)]\) . The angular velocity is then \(\dot{\psi }=\partial \psi (s,t)/\partial t\) , as obtained from the numerical time derivative of \(\psi (t)\) . Subtracting the solid rotation of the whisker, from the imposed motion, we finally derive the rotation fluctuations kinematic variables \(\stackrel{\sim }{\psi }=\psi -{\Omega }_{shaft}t\) and \(\partial \stackrel{\sim }{\psi }/\partial t=\dot{\psi }-{\Omega }_{shaft}\) . Mathematical model of a clamped conical rod The whisker is modelled as an inextensible elastic rod of length \(l\) that is constrained to move in a plane, as depicted in Fig. S4 . In figure S5 and related mathematical formulations, we present precise details of parameter, step sizes, boundary conditions needed to implement the model. The rod is assumed to have uniform material properties, but to have non-uniform cross-section representing the tapered nature of the whisker. The shape of the non-deflected beam corresponds to a truncated cone of cross-sectional area $$A\left( s \right) = \pi r(s)^{2} ~,$$ and second moment of area given respectively by Eq. 1 . Here \(r\) is the whisker’s radius, which is assumed to vary linearly with the arclength coordinate \(s\in \left[0,l\right],\) via $$r\left( s \right) = r_{0} \left( {1 - \frac{s}{{l_{c} + l}}} \right)$$ (5) where \({r}_{0}\) is the base radius and \({l}_{c}\) is the truncation length (i.e. the length of the tiny tapered whisker that would need to be added to make a perfect cone). This means the ratio of tip to base radius is \(\frac{{r}_{1}}{{r}_{0}}=\frac{{l}_{c}}{{l}_{c}+l}\) . As is common in rod mechanics, we assume the rod to be unshearable and inextensible, and, in the absence of any evidence to the contrary, we suppose the rod to be linearly elastic (in an Euler–Bernoulli sense) with material properties that are uniform along its length. That is, the density \(\rho\) and Young’s modulus \(E\) are taken to be constant, with the only longitudinal variation of the corresponding linear density of mass \(\rho A(s)\) and flexural rigidity \(EI(s)\) , due to the rod’s tapered geometry. Similarly, inspired by 40 and to tame high frequency vibrations, we also include a geometrically spatially varying Kelvin-Voigt damping with coefficient \(\delta\) . With these assumptions, the mechanics of the whisker is described in terms of the local angle \(\theta (s)\) between its centerline and the \(x\) axis. It follows that the strain variable of the rod is measured in terms of its local curvature \(\kappa \left( {s,t} \right) = ~\partial \theta \left( {s,t} \right)/\partial s.\) For whiskers, it may be assumed furthermore that the rod has intrinsic curvature, such that \(\kappa ={\kappa }_{i}\left(s\right)\) in a load free configuration. As a result, only the moment (bending couple) constitutive equation is needed, which, in terms of the rod’s curvature \(\kappa\) is written following 40 as $$M\left( {s,\kappa ,\dot{\kappa }} \right) = EI\left( s \right)\left[ {\kappa - \kappa _{i} \left( s \right)} \right] + \delta A\left( s \right)~\dot{\kappa }.~$$ (6) For simplicity, all numerical results presented here were computed in the case \({\kappa }_{i}\left(s\right)\) =0, which was found to make negligible difference to the transmission of dynamic information. This simplification can also be argued mathematically because rotary inertia is several orders of magnitude smaller than bending forces. The rod was modelled to be in contact with a plate fixed at a finite distance \(\stackrel{-}{y}<l\) from the base of the rod, and moving tangentially at a constant speed \(v\) in the direction normal to the undeflected rod’s axis. Speed \(v\) can be related to the angular velocity used in the experiments via the simple relation \(v = ~\overline{y} \Omega _{{shaft}} .\) In all computations we assume tip contact, between the whisker and the plane. The normal pressure \(p(t)\) at the tip is an unknown dynamical variable that is solved for as part of the problem. The base (follicle end) of the rod is assumed to be fixed within some rigid substrate representing the experimental shaft, leading to ideal clamped boundary conditions. Under the above assumptions, the mathematical model is developed as follows. Material points of the rod are labelled by the Lagrangian coordinate \(s\in [0,l]\) measured in the undeformed rectilinear configuration. The goal of the model is to describe the motion of each material point \(r\left(s,t\right)=[x(s,t),y\left(s,t\right)]\) along the centreline of the rod. Let \(f\) and \(g\) be the Cartesian components of the force at each material point and let the corresponding moment be \(m\) . Under these hypotheses, conservation of linear momentum in each Cartesian direction and of angular momentum lead to the equations of motion $$\rho A\left( s \right)\ddot{x} = f^{\prime},~\quad ~\rho A\left( s \right)\ddot{y} = g^{\prime},\quad \rho I\left( s \right)\ddot{\theta } = M^{\prime} + g~\cos \theta - f~\sin \theta,$$ (7) in which partial derivatives with respect to time and arclength are denoted with a dot and a prime respectively. Inextensibility means that the forces \(f\) and \(g\) at each position \(s\) are Lagrange multipliers, i.e. determined by kinematics not determined by constitutive laws. The inextensibility constraint requires that $$x^{{\prime2}} + y^{{\prime2}} = 1~,$$ (8) which is automatically satisfied by the differential equations $$x^{\prime} = \cos ~\theta {\text{ }}~,~\quad y^{\prime} = \sin ~\theta ~.$$ (9) Hence, with appropriate initial and boundary conditions, the Eq. 9 can be integrated to obtain the position of the rod \([x\left(s,t\right),y\left(s,t\right)]\) at each point along the rod and each instance in time. The base-end boundary conditions are straightforward. We suppose that this end is the origin of the Cartesian co-ordinates and that the rod is clamped, which gives $$x\left( {0,t} \right) = y\left( {0,t} \right) = 0,~\quad ~~x^{\prime}\left( {0,t} \right) = 0,\quad ~~~\theta \left( {0,t} \right) = 0.$$ (10) The boundary conditions at the tip end are less straightforward. The plate is assumed to occupy the half-space \(y=\stackrel{-}{y}\) and be rigid. The tip is assumed to experience tangential and normal pressures \(f_{T}\) and \(p(t)\) , which leads to boundary conditions of the form $$y\left( {l,t} \right) = \overline{y} ,~\quad ~~~~f\left( {l,t} \right) = f_{T} ,~\quad ~~~~g\left( {l,t} \right) = - p,~~\quad ~~~~M\left( {l,t} \right) = 0.$$ (11) Condition (Eq. 11 , 1) when combined with (Eq. 9 ) leads to the integral constraint $$\overline{y} = \smallint _{0}^{l} \sin ~\theta {\text{ }}~ds,~$$ (12) which allows for determination of the unknown pressure \(p(t)\) . Friction model A friction law was required to determine the tangential force \({f}_{T}(t)\) in terms of the pressure \(p\) and the velocity \(v\) of the plate’s motion. There is no single accurate model for frictional contact, especially in high-frequency dynamic environments, (see 32 ). It is common to assume that the normal and tangential interfacial forces are related by a ‘coefficient’ of friction \(\mu\) with a Coulomb or generalised Coulomb law between the slip velocity if the ratio of tangential to normal forces exceeds \(\mu\) . For solid interfaces, experimental evidence (see references in 33 , 41 ) suggests that a match with experimental data may be obtained if \(\mu\) is considered as a function of the past sliding history often modelled as an internal state variable that measures the state evolution of the frictional surface. Such so-called rate-and-state friction laws also have the advantage that they avoid the singularities associated with non-smooth Coulomb-like friction laws. Here we use a formulation 34 in which we suppose \(\mu\) to depend both on the velocity \(v\) and a dimensionless, internal relaxation variable \(\varphi \left(t\right)\) , which measures the state of the interface and quantifies the interfacial resistance to slip. Specifically $$\left\{ {\begin{array}{*{20}l} {f_{T} = \mu \left( {\upsilon ,\varphi } \right)p~~~~~} \hfill \\ {\mu = F\left( {\upsilon ,\varphi } \right) = ~a~\sinh ^{{ - 1}} \left[ {\gamma _{*} \left( {\frac{v}{{V_{*} }}} \right)\varphi ^{{\frac{b}{a}}} } \right]~ \approx \mu _{*} + a\ln ~\left( {\frac{v}{{V_{*} }}} \right)~ + ~b\ln \left( \varphi \right);~\gamma _{*} \equiv \frac{{\exp \left( {\frac{{\mu _{*} }}{a}} \right)~}}{2}} \hfill \\ {\dot{\varphi } = - G\left( {\upsilon ,\varphi } \right) \approx - \frac{{\varphi - \varphi _{{ss}} \left( \upsilon \right)}}{{t_{{ \star }} (v)}};~~~~~\varphi _{{ss}} \left( \upsilon \right) \equiv \frac{{V_{*} }}{\upsilon };~~~~~t_{{ \star }} \left( \upsilon \right) \equiv \frac{L}{\upsilon }.} \hfill \\ \end{array} } \right.$$ (13) The downside of the rate-and-state approach is that there are extra parameters that must be chosen (note that 33 , 41 discuss ways for the experimental determination of such rate-and-state parameters). Specifically, \({\mu }_{*}\equiv \mu [{V}_{*},{\varphi }_{ss}({V}_{*})]\) is a kinetic coefficient of friction of reference and the parameters \(a\) and \(b\) respectively characterise the strength of the instantaneous velocity and state dependence. Note that, in the steady-state sliding situation, it is customary to distinguish between “velocity-strengthening” \((a-b>0)\) and “velocity-weakening” \(\left(a-b<0\right)\) friction laws, the latter case allowing the possibility of stick–slip oscillations to exist. These parameters, as well as the parameter \({V}_{*}\) , which is a characteristic reference velocity, can be associated to the microphysics of creep (see e.g. 34 and references therein). The phenomenological sliding memory parameter \(L\) represents a “slip-length” whose microscopic origin is still a matter of debate (see e.g. 32 and references therein). The slip-length \(L\) in the state evolution law (Eq. 13 ) models the characteristic length (or equivalently the timescale of order \(L/v\) ) over which the frictional response to velocity jumps relaxes to a new sliding equilibrium 42 , 43 and therefore we use L as a proxy for surface roughness. We stress that the rate-and-state framework of friction is now well established for multi-contact interfaces over a wide range of scales. Nevertheless, dry friction modelling at the macro-scale based on microscopic measurements remains a topic of active research in the field of tribology, and further microscopic experimental work would however be needed to experimentally determine parameter regions of applicability of the rate-and-state formalism to whisker tribology. Numerical implementation and parameter fitting The set of equations (Eq. 7 ) with constitutive equation (Eq. 6 ) and boundary conditions (Eqs. 10 and 11 ), associated with constraints (Eq. 9 and 12 ) constitute a well-posed system of algebraic partial differential system to be solved for the unknowns \(\theta \left( {s,t} \right),~f\left( {s,t} \right),~g\left( {s,~t} \right)\) as well as the unknown scalars \(p(t)\) and \(\varphi (t)\) . The equations of motion of the whisker are numerically solved using the method of lines 48 . The discretization in space is achieved from the first order finite difference scheme proposed by 40 . The resulting large system of ordinary differential–algebraic equations of index 2 is then integrated in time using the implicit Runge–Kutta method of order 5 Radau IIA 44 (see 45 for a Matlab implementation ). With respect to parameter fitting, the whisker truncation length \({l}_{c}\) and density \(\rho\) are readily calculated from the rod conical geometry that we assume. The value of Young’s modulus is based on the natural frequency of the second mode of vibration of the whisker that is exhibited from the power spectral density computed from the free-on-air data (see Fig. 2 D). We tuned the value of \(E\) from computing the frequency response of the whisker with a set of ringdown numerical experiments, i.e. the oscillatory response of the whisker whose tip has been displaced by applying a force. Our value is consistent with previous published estimates (e.g. 2 , 3 , 46 , 47 ). The speed of sound \({c}_{s}=\sqrt{E/\rho }\) then follows directly. The driving plate location \(\stackrel{-}{y}/l=0.98\) is chosen for the present geometry so that regional contact of the whisker tip is avoided, the tip contact remaining close to tangency. In absence of any tribological data for the system whisker/sandpaper, our choice of parameter values for the friction model is ad hoc and based on orders of magnitude commonly used in the relevant literature (see also Fig. S5 and related mathematical formulations). Change history 14 December 2021 A Correction to this paper has been published:

How we sense texture has long been a mystery. It is known that nerves attached to the fingertip skin are responsible for sensing different surfaces, but how they do it is not well understood. Rodents perform texture sensing through their whiskers. Like human fingertips, whiskers perform multiple tasks, sensing proximity and shape of objects, as well as surface textures. Mathematicians from the University of Bristol's Department of Engineering Mathematics, worked with neuroscientists from the University of Tuebingen in Germany, to understand how the motion of a whisker across a surface translates texture information into neural signals that can be perceived by the brain. By carrying out high precision laboratory tests on a real rat whisker, combined with computation models, the researchers found that whiskers act like antennae, tuned to sense the tiny stick-slip motions caused by friction between the surface and the tip of the whisker. "One of the most striking things we found both in the experiments and the theory was the thousand-fold amplification of tiny force signals perceived by the tip of the whisker to that received by the neurons at the whiskers base. Suddenly we realized that the whisker is acting like an amplifier, taking micro-scale stick-slip events and rapidly turning them into clean pulses that can be picked up and processed by the brain," said Professor Alan Champneys from the University of Bristol, co-lead of the modeling work with colleague, Dr. Robert Szalai. Dr. Thibaut Putelat carried out the detailed numerical modeling. The research Conveyance of texture signals along a rat whisker, published in the journal Scientific Reports from the publisher Nature, reveals the tapering of the whisker has the effect of amplifying tiny high-frequency motions into appreciable pulse-like changes in forces and movement at the whisker follicle. In turn, the nerve cells in the follicle sense these changes and transmit them to the brain. "It is almost as if the morphology of the whisker is designed to convey these friction-induced signals as "AC" waves on top of the "DC" motion of the whisker that conveys the information on surface proximity and hardness. "These AC waves are too small and too rapid to be perceived by the human eye. However, in approaching this problem in a multidisciplinary fashion, we have been able to reveal these waves with clarity for the first time," said Professor Champneys. "The findings have implications for human touch too, where the morphology of finger-print ridges is more complex, but might similarly distinguish between AC and DC signals as our brain tries to disentangle multiple information streams about what we are feeling," said Dr. Maysam Oladazimi, who carried out the experiments as part of his Ph.D. The findings could have far-reaching benefits including how textures could be designed to provide optimal cues for the visually impaired, for human safety operation in low light environments, or for immersive artistic installations. "This research opens several avenues for future work. As neuroscientists, we are interested in developing a more detailed understanding of neural signaling pathways in texture discrimination, while our colleagues in Bristol are keen to explore implications for the design of future robotic sensing systems," said Professor Cornelius Schwarz, who led the experiments at the University of Tuebingen. Professor Champneys said the research was of particular value to haptic-sensing in the field of robotics, where robots literally feel their environment and is the focus of much current research, especially for robots that need to act autonomously in the dark, such as in search and rescue missions. Professor Nathan Lepora and colleagues at the Bristol Robotics Laboratory are pioneers in this field. "This transnational interdisciplinary collaboration between experimentalists and mathematical modelers was exciting. The results from the computer models and from the laboratory experiments went hand in hand—it was only through a combination of the two that we were able to make our breakthrough," said Professor Champneys.

10.1038/s41598-021-92770-3

Physics

Scientists create a quantum computer memory cell of a higher dimension than a qubit

Alexey A. Melnikov et al, Quantum walks of interacting fermions on a cycle graph, Scientific Reports (2016). DOI: 10.1038/srep34226 Journal information: Scientific Reports

http://dx.doi.org/10.1038/srep34226

https://phys.org/news/2016-12-scientists-quantum-memory-cell-higher.html

Abstract Quantum walks have been employed widely to develop new tools for quantum information processing recently. A natural quantum walk dynamics of interacting particles can be used to implement efficiently the universal quantum computation. In this work quantum walks of electrons on a graph are studied. The graph is composed of semiconductor quantum dots arranged in a circle. Electrons can tunnel between adjacent dots and interact via Coulomb repulsion, which leads to entanglement. Fermionic entanglement dynamics is obtained and evaluated. Introduction Quantum walks are quantum counterparts of classical random walks 1 , 2 . Unlike the state of classical walker, quantum walker’s state can be a coherent superposition of several positions. Quantum walks found applications to various fields, for example, to the development of a new family of quantum algorithms 3 , 4 , 5 , 6 or to the efficient energy transfer in proteins 7 . And recently, quantum walk dynamics is used as an underlying mechanism for quantum-enhanced decision-making process in reinforcement learning 8 , 9 , for which schemes of experimental realization in systems of trapped ions and superconducting transmon qubits were proposed 10 , 11 . There are plenty of theoretical and experimental results in the field of single-particle quantum walks 12 , 13 , but walks with multiple identical walkers, in both non-interacting and interacting cases, are less explored. In this paper we study quantum walks of identical particles for quantum information processing purposes. It is known that entanglement creation plays a pivotal role in most of the branches of quantum information. Here we introduce a method for generating a two-qudit (two d-level systems) entangled state by implementing continuous-time quantum walks on a cycle graph. This technique allows us to observe diverse structures of entangled subsystems of high dimensions, a preparation of which is of importance 14 , 15 . To relate theoretical study with feasible experimental implementations we consider realistic models of quantum walks 16 . The physical system we choose as a suitable candidate for quantum walks implementation is an array of tunnel-coupled semiconductor quantum dots. Quantum dots in semiconductors can be used as building blocks for a construction of a quantum computer, where quantum dots positions provide a spatial degree of freedom of a quantum particle 17 , 18 , 19 , 20 . It was shown that a spatial location of an electron in one of two semiconductor quantum dots can serve for encoding a qubit 17 , 18 and errors that occur mostly because of the interaction with acoustic phonons can be corrected 21 , 22 . In this paper we study quantum dots arranged in a circle, where each quantum dot can be populated by no more than one electron. By placing two identical particles in this system, one can define higher dimensional quantum states, qudits. If electrons are close enough they can also influence each other via Coulomb interaction. First, we analyze dynamics of non-interacting particles, and then we proceed to the case of interacting electrons. Results The remainder of the paper has the following structure: first, we introduce the model of symmetrical two-electron quantum walk on a cycle graph of arbitrary size. After the introduction of the model we study the dynamics of electrons in the cases with and without an interaction between them. The case of interacting electrons is studied in details and the scheme for entangling gate between two qudits, represented by two electrons, is proposed. Then we summarize the results and discuss possible applications of the proposed scheme. Framework The system under consideration contains two electrons. Each electron can sit in one of N quantum dots arranged in a circle 23 . Dots themselves can be formed from the two-dimensional electron gas by field of gates and the population of electrons in these dots can be controlled by potentials on gates. Each position in the circle can be occupied by at most one electron. The position of an electron can be measured by quantum point contact detectors, which are placed near quantum dots such that an electron in a certain quantum dot decreases an electric current in the detector by increasing a potential barrier. Therefore a lower current detects an electron and a higher current indicates an absence of an electron (an empty quantum dot), correspondingly. Experimentally, lateral structures of this geometry with different number of quantum dots were realized. Among them, a double quantum dot, which can be viewed as a circle with N = 2 sites, is the most studied configuration and is used to create a solid state qubit 24 , 25 , 26 , 27 . Beyond this, triple quantum dots with circular and linear geometries were studied in detail both theoretically and experimentally 28 . A concept of a scalable architecture was demonstrated by fabricating quadruple 29 , 30 , 31 and quintuple 32 quantum dots. In all experiments, a high degree of control over the precise number of electrons in each quantum dot was demonstrated by measuring stability diagrams. Moreover, it was shown that it is possible to tune the tunnel coupling between neighbouring quantum dots by changing the voltage on the gate that spatially separates these dots, see e.g. ref. 26 , where the tunnel coupling was shown to be an exponential function of the gate voltage. Similar techniques and technologies could be used for a fabrication of circles of larger sizes. The described circle of semiconductor quantum dots is mathematically represented as a cycle graph with quantum dots being vertices of this graph. Edges of the cycle graph connect only nearest neighbours and represent possible tunnel transitions of electrons. We enumerate the vertices within the graph, from 0 to N − 1. The localization of an electron in the 0-th, 1-st, … or ( N − 1)-th quantum dot is described by corresponding quantum states |0〉, |1〉, … or | N − 1〉, as shown in Fig. 1 for N = 2 K , . As a straightforward result, the states |0〉, |1〉, … and | N − 1〉 can be viewed as the basis states of a qudit, whose amplitudes squared correspond to the probabilities of detecting an electron. Note that because electrons cannot occupy the same energy level, i.e. the same vertex on a cycle graph, | ii 〉 two-qudit basis states are impossible for all i ∈ [0, 2 K − 1]. Figure 1: A cycle graph with N = 2 K vertices, where each vertex is viewed as a level in the N -level quantum system. Two electrons are initially placed in the 0-th and K -th positions, i.e. to an initial state. This initial state is chosen in order to achieve a high symmetry in this system. Full size image Electrons are initially placed in opposed vertices of the graph, as depicted in Fig. 1 , but can later change their positions by hopping between neighbouring vertices. This process is a continuous-time quantum walk governed by the Hamiltonian, which we introduce below. Electrons walk and spread due to tunneling through the barrier of controlled height between the quantum dots. For the sake of simplicity we assume that electrons spins are always up (1/2), which can be the case, for example, in a strong magnetic field. Wave function of two indistinguishable fermions in form of |Ψ( t )〉 = | ψ ( t )〉|↑↑〉, an antisymmetric coordinate part of which is where | m , k 〉 is the state with the first and second electron being in the m -th and k -th vertex, respectively; | ψ ( m , k ) 〉 is the state of electrons occupying vertices m and k , which corresponds to a product state of two uncorrelated systems. These electrons have to be treated as indistinguishable because their wave functions overlap spatially in the quantum dots. The state in Eq. (1) is a superposition of electrons being in different vertices with time-dependent amplitudes ω mk ( t ), which form a matrix ω ( t ) 33 . The matrix ω ( t ) is antisymmetric, i.e. ω T ( t ) = − ω ( t ), and takes into account the antisymmetric nature of the fermionic wave function. The normalization of the | ψ ( t )〉 state gives an additional condition on ω ( t ): The wave function specified by ω matrix fully characterizes a fermionic state, however a correct definition of its subsystems is required for studying properties of the system. The problem of reduced fermionic density operators was addressed recently in refs 34 and 35 , where it is shown that parity superselection rule should be applied to the fermionic state and the unique definitio of the reduced density operator is provided. In our case, the total number of fermions is constant, which leads to the standard procedure of obtaining the reduced state ρ 1 ( t ): This definition is used later to study the entanglement properties of the system. Non-interacting indistinguishable electrons The dynamics of the two-electron fermionic state depends on an arrangement of quantum dots: if quantum dots are close enough – electrons will interact through Coulomb repulsion, otherwise electrons do not interact. First, we consider the case without interaction and move to the case with interaction afterwards. The evolution of an electron in an array of tunnel-coupled semiconductor quantum dots can be modelled by a continuous-time quantum walk, which is defined by a Hamiltonian with nearest-neighbour interactions 23 . By analogy, the evolution of two electrons can be modelled by a continuous-time quantum walk of two particles that is governed by the Hamiltonian where Ω is the tunneling frequency, which corresponds to the potential barrier height between neighbouring quantum dots. This Hamiltonian is defined for N > 2 ( K > 1), for the smallest graph ( K = 1) the Hamiltonian is equal to the half of the one in Eq. 4, i.e. H O /2, since in the circle of two dots clockwise and counterclockwise jumps correspond to the same transition. One can see, that the Hamiltonian H O only changes the spatial part of the total fermionic wave function |Ψ( t )〉, leaving the spin part unchanged. In other words, the walk is performed in the space of coordinates of quantum dots, and the spins remain parallel as they were initially prepared. Therefore, the spin part of the wave function factors out from the evolution and will not be taken into account below. The remaining part of the total wave function, the antisymmetric spatial part, evolves according to the Schrödinger equation , where the unitary operator can be shown to map any antisymmetric fermionic wave function to antisymmetric one. An exact matrix representation of this unitary operator can be obtained analytically for small K , but in general can only be computed numerically. In Methods we provide exact solutions of the Schrödinger equation for K = 2, 3, 4. Exact solutions let us observe the periodic dynamics for K = 2 and 3 with periods T = π /2Ω and 2 π /3Ω, respectively, and aperiodic dynamics for K = 4 (see Methods for details). From these results we conclude that in general the dynamics is aperiodic, as it was also shown in the case of discrete-time quantum walks on cycles 36 , 37 . Although the dynamics is aperiodic, it is known that by waiting enough time, an arbitrary precision of returning to the initial state can be achieved, as shown in Methods for K = 4. The possibility to achieve an arbitrary precision of the state revival holds for all K and is known from the Poincaré recurrence theorem 38 , 39 , although in general for different K it might take different time to achieve the same level of precision. In experiment, the wave function | ψ ( t )〉 cannot be directly observed, the measured data corresponds to a population in each quantum dot, i.e an average number of electrons in each dot. For this reason our function of interest is the population λ i in the vertex i of the cycle graph. The population λ i is equal to the probability to detect an electron in the vertex i and is related to the amplitudes ω mk of the wave function | ψ ( t )〉: Figure 2 shows population dynamics λ i ( t ) for the smallest K = 2, 3, 4 and 5. The solution in the case of K = 1 is trivial | ψ ( t )〉 = | ψ (0)〉 and is not shown. From Fig. 2(a,b) one can immediately deduce that the charge dynamics is periodic, confirming the analytical results for the periods of quantum walks T = π /2Ω and 2 π /3Ω in the case of K = 2 (a) and K = 3 (b), respectively. The dynamics in the case of K = 4 (c) is, however, aperiodic, which is proven in Methods. But as we discussed before, a nearly full state revival can be observed in this system, in particular in the case of K = 4 (c) and K = 5 (d). Figure 2: The average number of electrons λ i in a vertex vs . time. The initial state is . ( a ) Quantum walk dynamics for K = 2, initial state is fully recovered after the time π /2Ω. ( b ) Quantum walk dynamics for K = 3, initial state is fully recovered after the time 2 π /3Ω. ( c ) Quantum walk dynamics for K = 4, initial state is partially recovered after the time , and . ( d ) Quantum walk dynamics for K = 5, initial state is partially recovered after the time Ω t ≈ 2.8 π , 4.4 π and 7.2 π . Full size image A population distribution dynamics , similar to the one shown in Fig. 2 , can be obtained by having only one electron initially prepared in a superposition of |0〉 and | K 〉 coordinate states, where the scaling factor of 1/2 comes from the reduction of the total charge in the system. This can be seen from the right part of Eq. 5 – the position of the second particle k is irrelevant, the distribution λ i only depends on the position of the first particle. The probability to find this single electron in a certain node is half of the probability of finding one of two non-interacting electrons, which is also a consequence of Eq. 5. Hence a quantum walk of two non-interacting particles can be simulated by a one-particle walk, whose dynamics was studied in refs 23 , 40 and 41 . But because it is not straightforward to initialize an electron in a superposition of being in different nodes, two-particle walk can be used for studying one-particle walks with arbitrary initial conditions. Interacting indistinguishable electrons Here we consider the case of two identical electrons that interact through Coulomb repulsion. The mutual repulsion between electrons becomes apparent when the distance between the quantum dots is such that the emerged Coulomb energy induced by one of the electrons prevents the second electron to tunnel to the adjacent dot. In order to model a fermionic quantum walk we approximate the Coulomb interaction by restricting the positions of electrons: electrons cannot be in the same or neighbouring vertices of the graph, and the effect of repulsion is negligible in all other situations, i.e. an electron does not “feel” an electric field of the distant electrons, if the distance between them is more than one empty quantum dot. This approximation is reasonable because neigbour dots are generally closer to each other than to metallic gates forming them so interaction can be strong, while interaction of electrons at distant dots is substantially suppressed not only by larger distance of interaction but also by screening due to presence of metallic gates between and nearby them. The Hamiltonian with the restriction of not being in the same and neighbouring vertices of the cycle graph is Similar to the case of non-interacting electrons, we obtain analytical solutions of the Schrödinger equation for small dimensions of the cycle graph K = 2, 3 and 4 (the case of K = 1 is unfeasible, because it is impossible to place two strongly repelling electrons in two quantum dots). The results, provided in Methods, demonstrate that there exists a period of quantum walks for K = 3, but not for K = 4. Hence, in general, the quantum walk of interacting particle on a cycle aperiodic. This fact can be seen in the population dynamics λ i ( t ) plotted in Fig. 3 for K = 3, 4, 5 and 6. Figure 3: The average number of electrons λ i in a vertex vs . time. Mutual repulsion between electrons is taken into account. The initial state is . ( a ) Quantum walk dynamics for K = 3, initial state is fully recovered after the time . ( b ) Quantum walk dynamics for K = 4, initial state is partially recovered after the time and . ( c ) Quantum walk dynamics for K = 5, initial state is partially recovered after the time Ω t ≈ 4.6 π , 8.5 π . ( d ) Quantum walk dynamics for K = 6, initial state is partially recovered after the time Ω t ≈ 3.9 π , 6.0 π . Full size image Fermionic entanglement by means of a quantum walk It is known that interactions between particles create quantum entanglement between these particles 42 , 43 . The qualification and quantification of an entanglement between several subsystems is one of the most important issues in quantum information theory. However, by describing an entanglement of two fermions we cannot use the standard definition of entanglement of distinguishable particles, because for identical particles the Hilbert space has no longer a tensor product structure. More specifically, the Hilbert space of two electrons is an antisymmetric product, not a direct product 44 , 45 . To define entanglement of indistinguishable fermions one can use the Slater rank 33 , 46 , 47 . The Slater rank is the minimum number of Slater determinants, and this number is an analogue of the Schmidt rank for the distinguishable case. Fermions are called separable iff the Slater rank is equal to one. That is quantum entanglement arise in a pure state if there is no single-particle basis such that a given state of electrons can be represented as a single Slater determinant Fermionic quantum correlations defined above are the analogue of quantum entanglement between distinguishable systems and are essential for quantum information processing with indistinguishable systems. However these correlations should be quantified differently from the case of distinguishable systems by taking into account the definition of fermionic entanglement. Defining good measures of fermionic entanglement remains a field of active research 48 , 49 . In this paper we use three fermionic entanglement measures: von Neumann entropy 50 , 51 , linear entropy 50 , 51 and fermionic concurrence 49 . Von Neumann entropy of the pure state where ρ 1 is the single-particle reduced density matrix defined in Eq. 3, and ξ j are the nonzero eigenvalues of the ρ 1 matrix. It was shown, that a pure state ρ has the Slater rank equal to one iff S vN ( ρ ) = 0 33 , 52 , 53 . An entanglement criterion for states of two fermions can also be formulated in terms of the linear entropy which is the approximation of the von Neumann entropy. A pure state ρ has the Slater rank equal to one iff S L ( ρ ) = 0. We also use the fermionic concurrence 49 which by analogy with the linear entropy gives 0 for separable states and nonzero values for entangled fermionic states. In addition, the fermionic concurrence in Eq. 10 is normalized between 0 and 1. We calculate S vN ( ρ ( t )), S L ( ρ ( t )) and C f ( ρ ( t )) functions using Eqs 8–10, respectively, for K = 3, 4, 5 and 6. These entanglement measures are shown in Fig. 4 . It can be seen that two electrons are initially separable, but after a time, which increases with K , they become entangled. In the case of K = 3, shown in Fig. 4(a) , the entanglement dynamics is periodic, as expected due to the periodicity of the wave function. The maximum entanglement is achieved at times , , for the state Figure 4: Entanglement measures: fermionic concurrence C f (solid black), von Neumann (dotted red) and linear (dashed blue) entropy. ( a ) K = 3. Entanglement exhibits periodic dynamics with the period . ( b ) K = 4. Entanglement dynamics is aperiodic. At time Ω t ≈ 7 π there is a sudden drop of entanglement with the local minimum of entanglement at time Ω t ≈ 7.5 π . ( c ) K = 5. At time Ω t ≈ 8.5 π there is a drop of entanglement because of the partial revival of the initial state. ( d ) K = 6. At time Ω t ≈ 6.0 π there is a drop of entanglement because of the partial revival of the initial state. Full size image The minimum entanglement corresponds to the separable state | ψ (0,3) 〉, which is present at times , . The evolution of entanglement for K = 4 ( N = 8 vertices) is shown in Fig. 4(b) . One can see that the particles entangle slower (initial slope in Fig. 4 ) than in case of K = 3, because electrons are initially further away from each other and it takes more time for particles to meet each other. The evolution is aperiodic and the entanglement never disappears, but because of a partial revival of the initial separable state, the entanglement of electrons drops suddenly at times of the largest overlap with the initial state Ω t ≈ 3.1 π and Ω t ≈ 7.5 π (see also Fig. 3 ). The maximum entanglement is achieved for multiple states. For instance, at times , and the following fermionic state is generated Figure 4(c,d) show the entanglement dynamics for higher cycle graph dimensions K = 5 and 6, respectively. Similar to the case of K = 4, the dynamics is aperiodic and maximal entanglement is achieved for many states. There are also occasional drops of entanglement caused by a partial return to the initial state | ψ (0, K ) 〉. The fermionic entanglement initiation described here is due to the Coulomb interaction. This repulsive interaction can be interpreted as a condition that restricts the positions of electrons – a quantum walk of one electron is conditioned on the state of the second electron and vice versa. In contrast to the interacting case, non-interacting electrons do not have this conditioned dynamics; dynamics of electrons is independent from each other. It can easily be shown that in absence of this Coulomb repulsion condition, entanglement is not initiated and all mentioned fermionic entanglement measures are equal to zero throughout the entire quantum walk evolution. Indeed, the Hamiltonian in Eq. 4 that governs the evolution of non-interacting electrons, leads to an independent unitary dynamics of two electrons where by U O ( t ) we denote a time-dependent unitary matrix that acts locally on a subspace of one particle. Because the initial state of two electrons is separable, local operations clearly cannot create an entangled state. To verify this we compute the reduced density state from Eq. 3: Using the explicit form of the reduced density matrix ρ 1 ( t ) that we obtained, we compute and , which leads us to the observation that all entanglement measures, S vN ( ρ , t ), S L ( ρ , t ) and C f ( ρ , t ) are equal to zero for all times t . Because these measures are zero iff the fermionic state ρ is separable, we conclude that, as expected, only by allowing electrons to interact, one can introduce fermionic entanglement in the system of coupled quantum dots that we consider. Electrons for quantum information processing We showed that a variety of entangled fermionic states can be created by means of quantum walks. However, it is not apparent how useful are these fermionic states for quantum information processing, because of inconsistency between fermions and qubits 34 . Here we show how one can define qudits by using the freedom of dividing the graph into two subgraphs. We divide the cycle graph into two equal parts: the first subgraph contains the vertices , the second subgraph contains the vertices . Experimentally, this division can be realized by raising a potential barrier between the two pairs of quantum dots, and dots, and between and dots. As we demonstrate below, in our framework, due to the symmetry of the initial state, it is possible to see that electrons are confined in different subgraphs with the unit probability. In this case, we say that an electron in the upper subgraph (vertices ) and an electron in the lower subgraph (vertices ) represent two distinguishable qudits. Below we show that this definition of two qudits in terms of the upper and the lower subspaces allows obtaining highly entangled states of two qudits. The described “cuts” of the circle are depicted in Fig. 5(a–d) for K = 3 (a), K = 4 (b), K = 5 (c) and K = 6 (d). Figure 5(a) schematically shows two qutrits (three-level systems with basis states |0〉, |1〉 and |2〉) defined on a cycle graph. At time , as we have shown before, the quantum dynamics on a cycle graph with 6 vertices leads to the state in Eq. 11 with two-particle correlation matrix shown in the right part of Fig. 5(a) . One can see, that if one raises a potential barrier between quantum dots 1 and 2, 4 and 5, as shown in the left part of Fig. 5(a) , one traps electrons in separate subgraphs, because at this time the particles can only be detected in the strictly opposite sites at the circle. After defining qudits at this time step we obtain an entangled state Figure 5 The left part of each figure ( a – d ) shows a scheme of the cycle graph with 2 K vertices divided into two subgraphs, each of which represents a state space of a qudit. The subspace of the first qudit is shown in blue (vertices 0, …, , , …, 2 K − 1), the subspace of the second qubit is shown in violet (vertices , …, ). Dashed black arrows show the possible transitions between the vertices in the redefined graph. Brown curves represent the type of entanglement (see text for details). The right part of each figure ( a – d ) shows a matrix of two-particle correlations of quantum walkers in position space. The element ( j , i ) of the correlations matrix corresponds to a probability of detecting two electrons in quantum dots j ∈ {0, …, N − 1} and i ∈ {0, …, N − 1}. ( a ) K = 3. The entangled two-qutrit state corresponds to a fermionic state obtained at times , . ( b ) K = 4. The entangled two-ququart state corresponds to a fermionic state obtained at time . ( c ) K = 5. The entangled two-qudit state corresponds to a fermionic state obtained at time t = 24.3/Ω. ( d ) K = 6. The entangled two-qudit state corresponds to a fermionic state obtained at time t = 25.7/Ω. Full size image The brown curves in Fig. 5(a) represent the type of a superposition in Eq. 15, which corresponds to the Bell-type entanglement. Figure 5(b) depicts a larger cycle graph with 6 vertices. Similar to the case of K = 3, two qudits are defined on the circle at a certain time . At this time one can separate two halves of the circle and obtain the following state of two ququarts Although this state is similar to the one in Eq. 15 up to a local phase, the type of entanglement in space of the graph is different and shown in the left part of Fig. 5(b) . From the right part of the Fig. 5(b) one can see that electrons are distributed in different subgraphs. The cases of K = 5 and K = 6 are shown in Fig. 5(c,d) , respectively. As one can see from correlation matrices, the same types of quantum correlations can be achieved with high probability. The fermionic entanglement dynamics studied in this paper is obtained for a pure state of a quantum system. However, in experiment the quantum state is subjected to decoherence. In particular, a change in the state of the qudits can be caused by the white noise from the quantum point contact detectors, which was shown to be one of the major concerns facing experimental realization of quantum walks in quantum dots structures 23 . This noise can be modelled by a depolarizing channel that acts on a density matrix of two fermions ρ as follows which is the solution of the differential equation dρ ( t )/ dt = −Γ( ρ ( t ) − ρ M ) with the initial state ρ (0) = | ψ (0, K ) 〉 〈 ψ (0, K ) |, where Γ is the relaxation rate that corresponds to the coupling between the quantum dot and the quantum point contact. The density matrix ρ M is the maximally mixed state of the coordinate part of two electrons. Because of the antisymmetric fermionic state the maximally mixed state is not the normalized identity matrix, but is defined as follows Combining the dissipative dynamics from Eq. 17 with the coherent evolution with Hamiltonian H C from Eq. 6 we write the general expression for the fermionic density matrix We are able to analyze the combination of the two different processes and to do the simplification of the expression due to the relation [ H C , ρ M ] = 0, which implies that the operator commutes with ρ M . The decoherence described by Eq. (17) leads to errors in quantum information stored in the system of electrons. In order to quantify this error we use the measure of decoherence 22 , 54 to quantify the amount of errors, where the operator norm of the matrix X is given by with spec( X ) being the spectrum of the operator X . The measure of decoherence D can be thought of as a probability of obtaining an error. In our scenario getting an error would correspond to getting a completely classical state of two particles, which are uniformly distributed over the circle, instead of getting entangled states shown in Fig. 5 . As shown in Fig. 6 , where we plot an error for each state in the set from Fig. 5 (entangled states with K = 3, 4, 5 and 6), this error can become large for large circles and strong couplings Γ between the quantum dot and the quantum point contact. Figure 6: The dependance of an error per state preparation on the relaxation rate is depicted. Four curves correspond to the states shown in Fig. 5(a–d) . Full size image In addition to the described noise, the system of electrons is subjected to the inevitable phase noise caused by the deformation interaction of electrons with acoustic phonons 18 . As a result, the energy levels in quantum dots where electrons reside become not fully determined, which effectively distorts the nondiagonal elements of the density matrix as follows where I is the identity matrix, for i < K and E K = e − γ /2 I with γ = Ξ 2 /2 ħπ 2 ρs 3 a 2 . Note that the | ψ K 〉 state is written in the basis of separated electrons, which corresponds to the states in Eqs 15 and 16. The following parameters are taken for electrons in silicon: effective deformation potential Ξ = 3.3 eV, speed of sound s = 9.0 × 10 3 m/s, density ρ = 2.33 g/cm 3 , and quantum dot size a = 10 nm 18 . For this set of parameters and K = 4 the obtained phase error is D ≈ 1.4 × 10 −5 , which suggests that this additional phase error is negligible and the error is mostly determined by the depolarizing noise in the range of relaxation rates we consider in Fig. 6 . Discussion In the paper we considered the dynamics of two-particle fermionic system. We analyzed quantum walks in two possible setups, which lead to a walk with and without interaction between electrons. We preferred quantum walks approach to quantum information processing for a number of reasons. Quantum walks dynamics is a natural process for many quantum systems compared to more artificial gate implementation. It is therefore easier to build and implement in experiment. One may also hope that relatively complex gates sequences could be replaced by simpler quantum walks processes. It was shown before that one can do arbitrary quantum operations using only particles free propagation 12 . One way to realize quantum walks algorithms is to use silicon quantum dots that form a cycle graph. We showed the electrons entanglement dynamics in this structure. The value of fermionic entanglement was calculated using measures in Eqs (8)–(10), which were proven to correctly quantify entanglement 49 , 50 , 51 . We showed that fermionic entanglement can be used to prepare quantum states for quantum information processing. These highly entangled states of qudits can be obtained by only using the free quantum evolution of identical particles, without relying on any additional manipulations with electrons. In addition, we supplemented our protocol of obtaining entangled states with analytical solutions for certain sizes of a graph and proved a general aperiodic nature of the continuous-time quantum walk of identical particles on a cycle graph. Methods In this section we explain the fermionic quantum walks dynamics in details by obtaining explicit analytical solutions of the Schrödinger equation. Throughout the section we use the series expansion of the quantum walk unitary operator where H is H O in case of non-interacting electrons and H C in case of interacting electrons. This expansion is useful in the case of a continuous-time quantum walk on a circle, because due to the cyclic conditions the number of fermionic states that can be observed is bounded. Period of quantum walks of non-interacting particles We first start our analysis with the case of quantum walks of non-interacting indistinguishable electrons, whose dynamics is described by the Hamiltonian H O in Eq. 4. We first consider the smallest sizes of the cycle graph with K = 1 (2 vertices), K = 2 (4 vertices), K = 3 (6 vertices) and show the periodicity of the underlying dynamics. Next we show that, in general, the dynamics is aperiodic, i.e. there is no time T ≠ 0 s.t. | ψ ( T )〉 = | ψ (0)〉, by obtaining the solution for K = 4 (8 vertices). A cycle graph with 2 vertices For K = 1 the evolution of the state is trivial: , because the initial state is the eigenstate of the Hamiltonian H O . This is expected due to the fact that the system of two quantum dots has only two energy levels both occupied by electrons, and because of the Pauli exclusion principle these electrons cannot change their positions. A cycle graph with 4 vertices By computing the lower powers of H O for K = 2 we observe that . Hence, using the observation we reduce Eq. 22 for this size of the graph to An overlap of the state | ψ ( t )〉 with the initial state is equal to , therefore the dynamics of the wave function (as well as the population λ i and the fermionic entanglement functions) is periodic. The period of the dynamics is T = π /2Ω, after this time the initial state is fully revived (we neglect a global phase). It is worth noting, that at time t = π /Ω the unitary matrix is equal to the identity matrix I , so any initial state is recovered after this time. The specific choice of the symmetric initial state we use recovers twice more frequent. A cycle graph with 6 vertices Similarly to the case of K = 2, we compute the lower powers of H O and obtain the relation , which leads us to the state An overlap of the state | ψ ( t )〉 with the initial state is equal to , therefore the dynamics is periodic with the period T = 2 π /3Ω. It is worth noting, that at time t = 2 π /Ω unitary matrix is equal to identity matrix I , so any initial state is recovered after this time. The specific choice of the symmetric initial state we use recovers 3 times more frequent. A cycle graph with 8 vertices The case of K = 4 is already more involved. We first divide the sum from Eq. 22 in two sums with even and odd powers of H O , respectively: where and An application of the to the unnormalized states in Eqs 26 and 27 preserves their structure, and only changes the coefficients and , respectively: By using the relations from Eqs 26–28 we compute the sum in Eq. 25 and obtain the solution of the Schrödinger equation An overlap of the state | ψ ( t )〉 with the initial state is . This overlap is unit only when t = 0, therefore there is no period of quantum walks for K = 4. However, by choosing the time t s.t. and , i.e. , with , the overlap |〈 ψ (0)| ψ ( t )〉| is close to unity. By waiting enough, an arbitrary precision can be achieved. Period of quantum walks of interacting particles We next move to the case of interacting particles, which quantum dynamics is described by Eq. 22 with the Hamiltonian H C . The minimal graph size in the case of repulsive electrons is K = 2, for which the dynamics is trivial with stationary solution . A cycle graph with 6 vertices By computing the lower powers of the H C for K = 3, we see that , hence An overlap of the obtained solution with the initial state is . Therefore the dynamics of the wave function is periodic with the period , which means that after the time T the initial state is fully revived. A cycle graph with 8 vertices Similar to the case of non-interacting electrons, the dynamics for the size K = 4 is more involved. We first decompose the sum from Eq. 22 in the following way: where and An application of the to the unnormalized states in Eq. 32 and 33 preserves their structure, and only changes the coefficients and , respectively: By using the relations from Eqs 32–34 we compute the sum in Eq. 31 and obtain the solution of the Schrödinger equation where and . An overlap of the obtained solution with the initial state is . This overlap is unit only when t = 0, therefore there is no period of quantum walks for K = 4. However an arbitrary precision of state revival can be achieved by choosing the time t s.t. , i.e. t = 2 πn / ω + , with . This happens approximately, e.g. for Ω t = 10 π Ω/ ω + ≈ 3.1 π ( n = 5, k ≈ 2) and Ω t = 24 π Ω/ ω + ≈ 7.5 π ( n = 12, k ≈ 5). Additional Information How to cite this article : Melnikov, A. A. and Fedichkin, L. E. Quantum walks of interacting fermions on a cycle graph. Sci. Rep. 6 , 34226; doi: 10.1038/srep34226 (2016).

Scientists from the Institute of Physics and Technology of the Russian Academy of Sciences and MIPT have let two electrons loose in a system of quantum dots to create a quantum computer memory cell of a higher dimension than a qubit (a quantum bit). In their study published in Scientific Reports, the researchers demonstrate for the first time how quantum walks of several electrons can help to implement quantum computation. "By studying the system with two electrons, we solved the problems faced in the general case of two identical interacting particles. This paves the way toward compact high-level quantum structures," says Leonid Fedichkin, associate professor at MIPT's Department of Theoretical Physics. In a matter of hours, a quantum computer would be able to hack through the most popular cryptosystem used by web browsers. As far as more benevolent applications are concerned, a quantum computer would be capable of molecular modeling that accounts for all interactions between the particles involved. This in turn would enable the development of highly efficient solar cells and new drugs. To have practical applications, a quantum computer needs to incorporate hundreds or even thousands of qubits. And that is where it gets tricky. As it turns out, the unstable nature of the connection between qubits remains the major obstacle preventing the use of quantum walks of particles for quantum computation. Unlike their classical analogs, quantum structures are extremely sensitive to external noise. To prevent a system of several qubits from losing the information stored in it, liquid nitrogen (or helium) needs to be used for cooling. Plenty of schemes have been proposed for the experimental realization of a separate qubit. In an earlier study, a research team led by Prof. Fedichkin demonstrated that a qubit could be physically implemented as a particle "taking a quantum walk" between two extremely small semiconductors known as quantum dots, which are connected by a "quantum tunnel." From the perspective of an electron, the quantum dots represent potential wells. Thus, the position of the electron can be used to encode the two basis states of the qubit—|0? and |1?—depending on whether the particle is in one well or the other. Rather than sit in one of the two wells, the electron is smeared out between the two different states, taking up a definite position only when its coordinates are measured. In other words, it is in a superposition of two states. The blue and purple dots in the diagrams are the states of the two connected qudits (qutrits and ququarts are shown in (a) and (b) respectively). Each cell in the square diagrams on the right side of each figure (a-d) represents the position of one electron (i = 0, 1, 2, ... along the horizontal axis) versus the position of the other electron (j = 0, 1, 2, ... along the vertical axis). The cells color-code the probability of finding the two electrons in the corresponding dots with numbers i and j when a measurement of the system is made. Warmer colors denote higher probabilities. Credit: MIPT If an entangled state is created between several qubits, their individual states can no longer be described separately from one another, and any valid description must refer to the state of the whole system. This means that a system of three qubits has a total of eight basis states and is in a superposition of them: A|000⟩+B|001⟩+C|010⟩+D|100⟩+E|011⟩+F|101⟩+G|110⟩+H|111⟩. By influencing the system, one inevitably affects all of the eight coefficients, whereas influencing a system of regular bits only affects their individual states. By implication, n bits can store n variables, while n qubits can store 2n variables. Qudits offer an even greater advantage, since n four-level qudits (aka ququarts) can encode 4n, or 2n×2n variables. To put this into perspective, 10 ququarts store approximately 100,000 times more information than 10 bits. With greater values of n, the zeros in this number start to pile up very quickly. In this study, Alexey Melnikov and Leonid Fedichkin obtain a system of two qudits implemented as two entangled electrons quantum-walking around the so-called cycle graph. To make one, the scientists had to "connect the dots," forming a circle (once again, these are quantum dots, and they are connected by quantum tunneling). The entanglement of the two electrons is caused by the mutual electrostatic repulsion experienced by like charges. It is possible to create a system of even more qudits in the same volume of semiconductor material. To do this, it is necessary to connect quantum dots in a pattern of winding paths and have more wandering electrons. The quantum walks approach to quantum computation is convenient because it is based on a natural process. Nevertheless, the presence of two identical electrons in the same structure was a source of additional difficulties that had remained unsolved. The phenomenon of particle entanglement plays a pivotal role in quantum information processing. However, in experiments with identical particles, false entanglement can arise between electrons that are not interacting, which must be distinguished from genuine entanglement. To do this, the scientists performed mathematical calculations for both cases, viz., with and without entanglement. They observed the changing distribution of probabilities for the cases with six, eight, 10, and 12 dots, i.e., for a system of two qudits with three, four, five, and six levels each. The scientists demonstrated that their proposed system is characterized by a relatively high degree of stability. The qubit is the basic element of a quantum computer. It has two basis states, viz., |0⟩ and |1⟩. The distinction between the classical bit and its quantum counterpart lies in more than just fancy brackets (these are the standard quantum mechanical notation for a state). The essential property of a qubit is its ability to be in a superposition of the two basis states: A|0⟩+B|1⟩. A classical bit, on the other hand, can only have one of the two values (0 or 1). The term “qudit” can be used to refer to higher-level quantum systems with more than two states. Credit: MIPT So far, scientists have been unable to connect a sufficient number of qubits for the development of a quantum computer. The work of the Russian researchers brings computer science one step closer to a future when quantum computations are commonplace. And although there are algorithms that quantum computers could never accelerate, others would still benefit enormously from devices able to exploit the potential of large numbers of qubits (or qudits). These alone would be enough to save us a couple of thousand years.

10.1038/srep34226

Physics

Symmetric graphene quantum dots for future qubits

L. Banszerus et al, Particle–hole symmetry protects spin-valley blockade in graphene quantum dots, Nature (2023). DOI: 10.1038/s41586-023-05953-5 Journal information: Nature

https://dx.doi.org/10.1038/s41586-023-05953-5

https://phys.org/news/2023-05-symmetric-graphene-quantum-dots-future.html

Abstract Particle–hole symmetry plays an important role in the characterization of topological phases in solid-state systems 1 . It is found, for example, in free-fermion systems at half filling and it is closely related to the notion of antiparticles in relativistic field theories 2 . In the low-energy limit, graphene is a prime example of a gapless particle–hole symmetric system described by an effective Dirac equation 3 , 4 in which topological phases can be understood by studying ways to open a gap by preserving (or breaking) symmetries 5 , 6 . An important example is the intrinsic Kane–Mele spin-orbit gap of graphene, which leads to a lifting of the spin-valley degeneracy and renders graphene a topological insulator in a quantum spin Hall phase 7 while preserving particle–hole symmetry. Here we show that bilayer graphene allows the realization of electron–hole double quantum dots that exhibit near-perfect particle–hole symmetry, in which transport occurs via the creation and annihilation of single electron–hole pairs with opposite quantum numbers. Moreover, we show that particle–hole symmetric spin and valley textures lead to a protected single-particle spin-valley blockade. The latter will allow robust spin-to-charge and valley-to-charge conversion, which are essential for the operation of spin and valley qubits. Main Carbon-based materials, such as monolayer and bilayer graphene, are interesting hosts for spin and spin-valley qubits thanks to their weak spin-orbit (SO) coupling 7 , 8 , 9 , 10 and weak hyperfine interaction 11 , 12 . Bilayer graphene (BLG) is attracting particular attention because it presents a gate-tuneable band gap 13 , E g , that can be used to electrostatically confine charge carriers into quantum point contacts and quantum dots (QDs) 9 , 10 , 14 , 15 . The small size of this gap (up to around 120 meV) allows the formation of ambipolar QDs, which is difficult to achieve with standard semiconductors 16 , 17 , 18 . Another attractive feature is that, at low energy, charge carriers have an orbital magnetic moment caused by a finite Berry curvature 3 , 19 . These orbital magnetic moments are aligned perpendicular to the BLG plane and allow control over the valley degree of freedom because they have opposite signs for the two valleys ( K and K ′) and for electrons and holes, as illustrated in Fig. 1a . This property is a consequence of the particle–hole symmetry that is imprinted on the low-energy Hamiltonian of bilayer graphene, $${H}_{{\rm{BLG}}}=-\,\frac{1}{2m}{\Psi }^{\dagger }\left[\left({p}_{x}^{2}-{p}_{y}^{2}\right){\sigma }_{x}+2{p}_{x}{p}_{y}{\sigma }_{y}{\tau }_{z}\right]\Psi +\frac{{E}_{{\rm{g}}}}{2}{\Psi }^{\dagger }{\sigma }_{z}\Psi ,$$ as well as on the intrinsic Kane–Mele SO coupling term \({H}_{{\rm{SO}}}\,=\) \(\frac{1}{2}{\Delta }_{{\rm{SO}}}{\Psi }^{\dagger }{\sigma }_{z}{\tau }_{z}{s}_{z}\Psi \) 7 , 8 . Here, m = 0.033 m e is the effective mass of the charge carriers in BLG with free electron mass m e , momentum operators p i and s i , τ i and σ i being Pauli matrices ( i = x , y , z ) acting on the spin, valley and sublattice space, respectively. Both H BLG and H SO are invariant under particle–hole transformation \({\rm{K}}\) , which effectively flips the sublattice, valley and spin indices, \({\rm{K}}{\Psi }^{\dagger }{{\rm{K}}}^{-1}={\sigma }_{y}{\tau }_{x}{s}_{y}\Psi \) . As a consequence, the hole spectrum in BLG mirrors the electron spectrum around the K and K ′ points. Fig. 1: Particle–hole symmetry in BLG and the formation of electron–hole DQDs. a , Low-energy dispersion relation of gapped BLG at the K and K ′ points. Arrows indicate the orientation of the valley-dependent orbital magnetic moment of electrons (blue) and holes (red). b , Each orbital state in the electron and hole QDs holds four single-particle states due to the spin and valley degree of freedom. The SO gap, Δ SO , splits the fourfold degeneracy of each orbital into two Kramers’ pairs. Black (coloured) arrows indicate the orientation of the spin (valley) magnetic moment. The first electron shell in the conduction band is separated from the first hole shell in the valence band by E g and the confinement energy of the QD. c , Schematic cross-section of the device. The van der Waals heterostructure consists of a hBN–BLG–hBN–graphite stack. The gate pattern comprises a layer of SGs, forming a narrow channel, and two layers of FGs to define and control the QDs. d , Schematic of the valence and conduction band-edge profiles along the p-type channel. Finger gates (L, C and R) form an e–h DQD. e , Charge stability diagram of the device at V SD = 1 mV. Red dashed lines indicate charging lines of a hole QD and black dashed lines indicate charging lines of electron QDs, with the electron (hole) occupation number labelled in black (white). Dashed circles mark the formation of single e–h DQDs. The position of the QDs along the channel is interchanged for the red and black circles (see Extended Data Fig. 1 ). Full size image This symmetry remains preserved in BLG QDs, where the orbital states are quantized and form shells of four states grouped by the SO coupling into Kramers’ doublets, \(\left|K\uparrow \right\rangle ,\left|{K}^{{\prime} }\downarrow \right\rangle \) and \(\left|{K}^{{\prime} }\uparrow \right\rangle ,\left|K\downarrow \right\rangle \) . Every electron state in the energetically lower (higher) Kramers’ pair in the conduction band has a corresponding hole state in the energetically higher (lower) Kramers’ pair in the valence band, as illustrated in Fig. 1b . This is in stark contrast to the single-particle spectrum of QDs in carbon nanotubes, where the SO coupling caused by the curvature breaks particle–hole symmetry 20 . Here we show that this symmetry is almost perfectly preserved also when electrons and holes are physically separated into different QDs, and that the resulting ordering of electron and hole states leads to a very strong single-particle blockade in the transport through electron–hole (e–h) double quantum dots (DQDs). Bilayer graphene electron–hole DQD devices The DQD devices are fabricated as shown schematically in Fig. 1c . They consist of BLG encapsulated between two crystals of hexagonal boron nitride (hBN) resting on a graphite crystal that acts as a back gate (BG). A pair of metallic split gates (SGs) on top is used to create a narrow conducting channel. Two layers of interdigitated finger gates (FGs) across the channel are used to modulate the band-edge profile along the channel and to confine electrons and holes in neighbouring QDs 10 , 21 , as illustrated in Fig. 1d . The charge stability diagram in Fig. 1e gives an overview of the different charge configurations of the investigated device. At FG voltages V L and V R ≲ 4.1 V, a hole QD is formed underneath the central gate (C) whereas at V L and V R ≳ 4.1 V an electron–electron DQD is formed under the left (L) and right (R) gates. The dashed lines indicate the charge transitions of the electron (black) and hole (red) QDs. The intersections of the electron and hole charging lines correspond to the formation of an ambipolar electron–hole DQD (Extended Data Fig. 1 ). We now focus on the charge transition (0h,0e) ↔ (1h,1e), highlighted by the dashed black circle in Fig. 1e . For positive bias voltages, a steady tunnel current through the DQD involves the transition (0h,0e) → (1h,1e), which is only possible if electron–hole pairs with opposite quantum numbers (for example, \({\left|K\downarrow \right\rangle }_{{\rm{h}}}\) and \({\left|{K}^{{\prime} }\uparrow \right\rangle }_{{\rm{e}}}\) ) can be continuously created. Because the SO coupling has opposite sign for electrons and holes, this is possible only in two configurations, α and β , illustrated in the two panels of Fig. 2a . These two configurations, which are energetically offset by 2Δ SO , result in two sharp current peaks in the bias triangle, as shown in Fig. 2b . The value of Δ SO extracted from the separation of the peaks along the detuning axis, Δ ε = 2Δ SO = 140 ± 10 μeV, is in excellent agreement with that measured in an electron–electron DQD realized in the same device (Δ SO = 68 ± 7 μeV; see Extended Data Fig. 2 ) and in other experiments 9 , 10 , 22 . The separation of the two resonances remains constant when applying a perpendicular magnetic field; only their position changes with respect to the baseline of the bias triangle, as can be seen in Fig. 2c . By contrast, the separation between α and β increases slightly when applying a parallel magnetic field and a third resonance, γ , appears between, as shown in Fig. 2d . This behaviour can be readily understood, as explained below. Before turning to this, we consider the case of negative bias voltage applied to the electron–hole DQD device. Fig. 2: Finite bias spectroscopy and spin-valley blockade. a , Schematics of the alignment of energy levels in the first shells of an e–h DQD for positive bias. Creation of e–h pairs is possible only when states of opposite spin and valley-quantum numbers are aligned, and hence transport occurs only via the ground state ( α ) and excited state transition ( β ). b , Charge stability diagram of the (0h,0e) → (1h,1e) transition at V SD = 1 mV. Dashed lines represent the outline of the bias triangle and co-tunnelling lines, and the white arrow indicates the direction of increasing detuning, ε , between the two QDs. c , Charge stability diagram as in b , but for B ⊥ = 0.4 T. The transitions α and β shift with respect to the outline of the bias triangle but their separation is not changed. d , Charge stability diagram as in b , but for B ∥ = 0.75 T. Separation of peaks α and β is slightly increased and an additional transition, γ , appears between. e , Schematic as in a , but for negative bias. Electrons and holes tunnelling into the DQD from the leads must recombine to facilitate current flow. As soon as charge carriers with incompatible spin- or valley-quantum number occupy the QDs, transport is blocked. This blockade cannot be lifted with detuning energies lower than the band gap (see lower panel). f – h , Charge stability diagrams measured at B = 0 T, B ⊥ = 0.4 T and B ∥ = 0.75 T as in b – d , respectively, but recorded at negative bias, V SD = −1 mV, probing the (1h,1e) → (0h,0e) transition. Transport is blocked, except for faint co-tunnelling effects and small currents at the corners of the bias triangles (dashed blue circles). Insets show simulations of the current at the corners of the bias triangles. Complementary data obtained from another DQD are shown in Extended Data Fig. 3 . Full size image Single-particle electron–hole blockade In the case of negative bias voltage, tunnel transport through the DQD requires the continuous annihilation of electron–hole pairs—(1h,1e) → (0h,0e)—which is possible only if electron and hole have opposite quantum numbers. However, because electrons and holes tunnel in from the leads with random quantum numbers, transport is blocked as soon as the QDs are occupied by charge carriers with incompatible quantum numbers, as shown in Fig. 2e . This blockade is robust up to very high detuning energies because it can be overcome only by involving the next unoccupied orbital state, which is separated in energy from the low-energy spectrum by the band gap, E g ≈ 20 − 120 meV (ref. 13 ). By contrast, the singlet–triplet Pauli blockade observed in conventional semiconductors 23 , 24 (including BLG 25 ), is observed only below the singlet–triplet splitting, and additional orbital states are separated only by the confinement energy (typically 0.05−1.70 meV). The single particle e–h blockade is shown in Fig. 2f , in which the tunnel current inside the (1h,1e) → (0h,0e) bias triangle is entirely suppressed except for a faint contribution that can be attributed to co-tunnelling. The current is slightly enhanced only at the corners of the bias triangle (dashed circles), which correspond to configurations where the electron or hole states are aligned with the Fermi level of source or drain, respectively. Under this condition, incompatible charge carriers can tunnel back into the leads and new ones, possibly with matching quantum numbers, can enter the QD, lifting the blockade and allowing a small current through the DQD. This is confirmed by measurements on another DQD Extended Data Fig. 3 and by simulations based on the single-particle spectrum of BLG QDs and Pauli’s master equation 26 Extended Data. Fig. 5 , which shows lifting of the blockade only at the corners of the bias triangle as shown in the insets of Fig. 2f–h . The application of perpendicular magnetic fields leaves the blockade intact and reduces the co-tunnelling current below the noise floor 10 , 27 (Fig. 2g ). The difference between the average current inside the bias triangle and the co-tunnelling current outside is below 10 fA and is not measurable. For parallel magnetic fields the blockade also remains intact but with a stronger enhancement of the tunnel current at the corners of the triple point, which is in agreement with the simulation (Fig. 2h ). Protected electron–hole blockade at finite B -field The concept of the symmetry-protected spin-valley blockade presented above is based on careful analysis of all possible transitions between single-hole and single-electron states in the DQD, and on magnetotransport measurements that support the assignment of the states involved in the various transport processes. The magnetic field-dependent energy dispersion of the first hole and electron states is depicted in Fig. 3a . A perpendicular magnetic field, B ⊥ , lifts the degeneracy of the Kramers’ pairs and each state shifts due to the spin and valley-Zeeman effect \(\Delta E({B}_{\perp })=\frac{1}{2}(\pm {g}_{{\rm{s}}}\pm {g}_{{\rm{v}}}){\mu }_{{\rm{B}}}{B}_{\perp }\) . Here, μ B is the Bohr magneton, g s = 2 the spin g -factor 28 , 29 and g v the valley g -factor, which quantifies the strength of the valley-dependent magnetic moment 19 . From Fig. 2c and Extended Data Fig. 4b , we extract g v ≈ 15 for our DQD system. Electrons and holes with opposite quantum numbers undergo the same Zeeman shift and, therefore, the splitting between the α and β transitions remains constant with B ⊥ . This is clearly reflected in the magnetotransport measurements presented in Fig. 3b,c , which show the current measured along the detuning axis ε of the (0h,0e) → (1h,1e) triple point (arrow in Fig. 2b ) as a function of B ⊥ . Fig. 3: Probing the single-particle particle–hole symmetric spectrum. a , Energy dispersion of the first four single-particle electron (blue) and hole (red) states in parallel and perpendicular magnetic fields. Kane–Mele SO splitting, which has opposite signs for the valence and conduction band, polarizes the spins out of plane for zero magnetic field. Coloured arrows represent the required detuning of permitted transitions (creation or annihilation) between hole states in the left QD and electron states in the right QD. b , Current through the DQD for positive bias, V SD = 1 mV, as a function of B ⊥ and of detuning. The direction of the detuning axis is indicated by the white arrow in Fig. 2b . Because the position of the resonances α and β shifts with respect to the baseline of the bias triangle as a function of B ⊥ (Fig. 2b,c ), we plot here the current as a function of an effective detuning \(\widetilde{\varepsilon }\) , defined such that for each value of B ⊥ the position of resonances α and β is symmetric with respect to \(\widetilde{\varepsilon }=0\) . c , Same measurement as in b , but for a second DQD (Extended Data Figs. 3 and 4 ). d , Current through the DQD as a function of \(\widetilde{\varepsilon }\) and B ∥ for positive bias voltage. Coloured circles indicate peak positions of the current. Coloured lines indicate the expected positions of local maxima, as given by the required detuning for each transition (compare with length of arrows in a ). e , Simulation of current as a function of \(\widetilde{\varepsilon }\) and B ∥ based on Pauli’s master equation. f , Current through the DQD as a function of ε and B ⊥ for negative bias, V SD = −1 mV. g , Same as in f , but for B ∥ . Full size image The situation is different for in-plane magnetic fields, B ∥ , where the spin-Zeeman effect competes with SO coupling, which polarizes the spins out of plane for zero B -field 8 . With increasing B ∥ , spins are canted into the BLG plane, aiming for the same spin direction within a Kramers’ pair, and for opposite spin directions between Kramers’ pairs (Fig. 3a , left). The energy difference between Kramers’ pairs increases according to \(\Delta E({B}_{\parallel })=\pm \,\frac{1}{2}\sqrt{{\Delta }_{{\rm{SO}}}^{2}+{({g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{B}_{\parallel })}^{2}}\) . This means that the required detuning for the α transition decreases whereas that for the β transition increases, as can be seen in Fig. 3d . Both α and β are suppressed for B ∥ > 0.4 T, because the overlap between states in different Kramers’ pairs vanishes with increasing tilt of the spins into the BLG plane. Instead, transitions involving states from the same Kramers’ pair of electron and hole states become possible—that is, \({| K\leftarrow \rangle }_{{\rm{e}}}\leftrightarrow {| {K}^{{\prime} }\to \rangle }_{{\rm{h}}}\) or \({\left|{K}^{{\prime} }\to \right\rangle }_{{\rm{e}}}\leftrightarrow {\left|K\leftarrow \right\rangle }_{{\rm{h}}}\) . They appear as a third resonance, γ , which is visible in Fig. 3d . This behaviour is reproduced by our simulation, assuming the energy spectrum in Fig. 3a with Δ SO = 70 μeV, g v = 15 and g s = 2. We confirmed the robustness of the blockade for both perpendicular (Fig. 3f ) and parallel (Fig. 3g ) magnetic fields by repeating the measurements in Fig. 3b,d for negative bias. The robustness is a direct consequence of (1) the separation of higher orbital states by the band gap (Fig. 2e , lower panel) and (2) the spin and valley texture of our system (that is, the ordering of the states) imposed by particle–hole symmetry, which forbids ground-state–ground-state transitions by both spin and valley selection rules independently of the applied magnetic field. This is in stark contrast to the conventional singlet–triplet Pauli blockade 23 , 24 , which exists only in a limited range of detuning and magnetic field and is protected only by spin selection rules that can be lifted by relaxation mechanisms 23 Extended Data. Table 1 and to Extended Data. Fig. 6 . It should be noted that perfect particle–hole symmetry is not necessary to achieve a robust blockade, but only the ordering of electron and hole states needs to be symmetric. Symmetric state ordering is strongly protected because it is intrinsic to the BLG band structure and no conditions have been observed that would allow lifting of the blockade Extended Data Table 2 . The robustness of the single-particle blockade also indicates the absence of spin- or valley-flipping processes and that mixing between states within a Kramers’ pair is negligible. Mechanisms breaking electron–hole symmetry In the following, we demonstrate that not only do we have a symmetric state texture but observe indeed near-perfect symmetry of the particle–hole states in our system. This is not a priori granted, because electrons and holes are physically separated into two different QDs on different layers of the BLG (Fig. 4a ), giving rise to two mechanisms that can potentially break the perfect symmetry of the system. The first of these is a difference in the valley g -factors for electrons and holes, which might be possible because the valley g -factor sensibly depends on the geometry of the QD 18 , 19 and electrons and holes sit in two different QDs. Different valley g -factors—that is, \({g}_{{\rm{v}}}^{{\rm{e}}}\ne {g}_{{\rm{v}}}^{{\rm{h}}}\) —would break the particle–hole symmetry of the system at finite B ⊥ (Fig. 4b ) and lead to a splitting of the α and β resonances with increasing B ⊥ Extended Data. Fig. 7 . The second mechanism that could break particle–hole symmetry is a difference in SO coupling for electrons and holes, \({\Delta }_{{\rm{SO}}}^{{\rm{t}}}\ne {\Delta }_{{\rm{SO}}}^{{\rm{b}}}\) . This could originate from the fact that, at low k -values, electrons and holes are located on different layers of BLG 3 , 22 . Therefore, they can experience different proximity-enhanced SO couplings caused by varying interatomic distances or crystallographic orientations between BLG and the top and bottom hBN crystal. Assuming different SO coupling energies in the top (t) and bottom (b) layer which breaks the layer symmetry, the Kane–Mele SO Hamiltonian of BLG takes the form 22 $${H}_{{\rm{SO}}}=\frac{1}{4}{\Psi }^{\dagger }\left[\left({\Delta }_{{\rm{SO}}}^{{\rm{t}}}+{\Delta }_{{\rm{SO}}}^{{\rm{b}}}\right){\sigma }_{z}-\left({\Delta }_{{\rm{SO}}}^{{\rm{t}}}-{\Delta }_{{\rm{SO}}}^{{\rm{b}}}\right){\sigma }_{0}\right]{\tau }_{z}{s}_{z}\Psi ,$$ which breaks the layer symmetry Fig. 4a,b . Such a layer-dependent SO coupling would cause splitting of the γ transition with a separation proportional to the asymmetry of the SO coupling between the two layers, \(| {\Delta }_{{\rm{SO}}}^{{\rm{t}}}-{\Delta }_{{\rm{SO}}}^{{\rm{b}}}| \) . Fig. 4: Quantitative assessment of particle–hole symmetry. a , Schematic of electron and hole QDs located on the two different layers of the BLG sheet, potentially experiencing different proximity-enhanced SO couplings on each layer. Black arrow indicates the direction of the applied displacement ( D )-field. b , Energy dispersion of the first four single-particle electron (blue) and hole (red) states as a function of B ⊥ assuming \({g}_{{\rm{v}}}^{{\rm{e}}}\ne {g}_{{\rm{v}}}^{{\rm{h}}}\) and \({\Delta }_{{\rm{SO}}}^{{\rm{t}}}\ne {\Delta }_{{\rm{SO}}}^{{\rm{b}}}\) . c , Line cuts at B ⊥ = B ∥ = 0 along the black and grey arrows in Fig. 3d,e . d , Line cuts at B ∥ = 0.8 T along the black and grey arrows in Fig. 3d,e . Peaks α , β and γ are fitted to Gaussian line shapes. e , Average width ⟨ Γ α β ⟩ of resonances α and β for different ranges of B ⊥ . Error bars indicate the 80% percentile of Γ α β . f , As in e , but for ⟨ Γ α β ⟩ and ⟨ Γ γ ⟩ at B ∥ = 0.66−0.80 T, where all three resonances are readily distinguishable. Full size image To quantify an upper bound for these effects, we extract the full width at half-maximum ( Γ ) of resonances α , β and γ by fitting Gaussian line shapes and assuming a constant background and an equal width of peaks α and β (Fig. 4c,d ). For increasing B ⊥ we observe a slight broadening of the α and β resonance, as shown in Fig. 4e . Attributing this broadening entirely to a difference of valley g -factors between the electron and hole QD (rather than, for example, to variations in tunnel coupling) we get an upper limit for the valley g -factor mismatch below 1% of g v (Extended Data. Fig. 8 ), which is consistent with the high symmetry of our gate design. For parallel magnetic fields, Fig. 4f shows that the average width ⟨ Γ α β ⟩ is comparable to ⟨ Γ γ ⟩ , indicating almost identical SO couplings in the two layers. From the uncertainty of the line widths, we estimate the layer asymmetry of the SO coupling to be \(| {\Delta }_{{\rm{SO}}}^{{\rm{t}}}-{\Delta }_{{\rm{SO}}}^{{\rm{b}}}| < 5\,{\rm{\mu }}{\rm{eV}}\) —that is, below 10% of Δ SO . This careful analysis confirms that the system is very close to perfect particle–hole symmetric, with a rich and well-understood single-particle spectrum, in full agreement with the Kane–Mele spin-orbit Hamiltonian 7 . Impact of electron–hole symmetric-level texture The electron–hole symmetric-level texture leads to a strong single-particle spin and valley blockade that is robust up to very high detuning energies and independent of magnetic field, in strong contrast to the conventional two-particle singlet–triplet Pauli blockade 23 , 24 , 25 . This robust single-particle blockade will allow for high-fidelity spin-to-charge and valley-to-charge conversion, providing a reliable readout scheme for spin and valley states 30 . Combined with microwave control, this will enable the study of spin and valley coherence times by electron spin-resonance 31 or electron dipole spin-resonance techniques 32 , 33 , and open the door for full spin and valley qubit operation. BLG electron–hole QDs are also well suited for implementation of long-range qubit coupling in extended chains of ambipolar quantum dots 30 , a goal that is also under exploration in other material systems such as silicon 34 . Furthermore, the single-particle blockade presented in this work makes electron–hole symmetric DQDs a highly interesting system for implementation of gate-tuneable, single-photon THz detectors 35 , entangled electron–hole pair pumps or highly efficient Cooper pair splitters 36 , 37 . When coupling a BLG electron–hole DQD to a superconductor, conventional elastic co-tunnelling is forbidden by spin and valley selection rules, enabling pure Cooper pair splitting and crossed interband Andreev reflection 38 . This also makes such electron–hole QDs coupled to a superconductor an interesting building block for a topological Kitaev chain 39 , 40 . Methods Sample fabrication The devices are fabricated from mechanically exfoliated BLG flakes encapsulated between two hBN crystals, of approximately 25 nm thickness, using conventional van der Waals stacking techniques. A graphite flake is used as a BG. Cr/Au SGs with a lateral separation of 150 nm are deposited on top of the heterostructure. Isolated from the SGs by 15-nm-thick atomic layer-deposited Al 2 O 3 , we fabricate two layers of 70-nm-wide FGs with a pitch of 150 nm. Details of the fabrication process can be found in ref. 15 . Measurement technique All measurements are performed in a dilution refrigerator at a base temperature of 10 mK using standard DC measurement techniques. QDs are following previous studies of gate-defined BLG QDs 14 , 18 , 21 . Throughout the experiment, a constant BG voltage of V BG = −1.73 V and a SG voltage of V SG = 1.56 V are applied to define a p-type channel between source and drain. The estimated band gap is around 20 meV. For better comparability, the data in Fig. 3b,c,f,g are shown symmetrized around zero magnetic field. Charge stability diagrams for opposite bias voltages Extended Data Fig. 1 compares charge stability diagrams measured at positive and negative bias voltage in DQD #1 (compare with Figs. 1 – 3) . Dashed lines indicate charge transitions of the electron (black) and hole (red) QDs. Electron–hole DQDs are formed at the intersections of these charging lines. For the left electron–hole DQD (DQD #3, (0h,0e) ↔ (1h,1e) transition; red circle), transport is blocked at positive bias whereas for the right electron–hole DQD (DQD #1, (0h,0e) ↔ (1h,1e) transition; black circle), transport is blocked at negative bias. The data in the main manuscript are obtained from DQD #1. Extraction of Δ SO from measurements of a single-electron DQD in the same device To compare the measured value for Δ SO in the electron–hole DQD and to demonstrate that the magnitude of the SO gap is symmetric for electrons and holes, we present measurements of Δ SO in an electron–electron DQD. Extended Data Fig. 2 shows a close-up of the first triple point of an electron–electron DQD formed in the same device (compare with Extended Data Fig. 1 ). Transport via both a ground state and an excited state can be observed. We extracted their energy splitting by fitting two Lorentzian peaks to a line cut through the triple point (Extended Data Fig. 2b ). The determined value of Δ SO = 68 ± 7 μeV is in good agreement with those observed in the electron–hole DQD regime. A detailed discussion of Δ SO and the single-particle spectrum in the electron DQD in this device is given in ref. 10 . Additional dataset for another electron–hole DQD #2 in the same device A second electron–hole DQD is studied, formed with a different set of gate fingers on the same gated bilayer graphene device presented above (DQD #2; Extended Data Fig. 3 b). The single-electron–single-hole transition, (0h,0e) → (1h,1e), is highlighted by the dashed circle in the charge stability diagram (Extended Data Fig. 3c ). Measurements of that bias triangle are shown in Extended Data Fig. d–k for different V SD and magnetic fields, showing good agreement with the data presented for DQD #1 in Fig. 2 . In contrast to the data presented above, co-tunnelling was more pronounced due to strong coupling of the hole QD to the reservoir. The magnetic field-dependent spectrum of the first electron and first hole states is shown in Extended Data Fig. 4a (compare with Fig. 3a ). Extended Data Fig. 4b–e shows measurements complementary to that presented in Fig. 3 , recorded for DQD #2 shown in Extended Data Fig. 3b . The measurement shown in Extended Data Fig. 4b shows that the difference in detuning energy between α and β is independent of B ⊥ , the energy splitting measures Δ ε = 150 ± 10 μeV, which corresponds to 2Δ SO . The background current originates from co-tunnelling in the bias transport window (its onset is highlighted by the white dashed line), which shifts in energy with increasing ∣ B ⊥ ∣ . This is due to the fact that the bias window is defined by the (forbidden) ground-state transition, \({\left|{K}^{{\prime} }\uparrow \right\rangle }_{e}\leftrightarrow {\left|{K}^{{\prime} }\uparrow \right\rangle }_{h}\) , which requires less detuning for increasing ∣ B ⊥ ∣ . The same measurement for parallel magnetic fields (Extended Data Fig. 4c ) shows the effect of spins continuously canted into the BLG plane. The difference in detuning of transitions α and β increases while a third resonance, γ , emerges. Extended Data Fig. 4d,e show magnetotransport data in the single-particle blockade regime. The spin-valley blockade is not lifted under the influence of both in-plane and out-of-plane magnetic fields. Transport via co-tunnelling is suppressed at increasing B ⊥ as (also in this case) the tunnelling barriers became more opaque due to magnetic confinement. The robustness of the single-particle blockade indicates the absence of spin- or valley-flipping processes and that mixing between the states within a Kramers’ pair is negligible. Overall, the data are in good qualitative and quantitative agreement with those presented in Fig. 3 . Simulation of magnetotransport through an electron–hole DQD We simulate transport within the DQD bias triangles and along the detuning cuts by solving the rate equations for the electron and hole QD states presented in Fig. 3a following the approach used in ref. 41 . The energy of the respective electron ( σ z = +1) and hole ( σ z = −1) states is given by $${H}_{{\rm{e}}}=\frac{1}{2}{\Delta }_{{\rm{SO}}}{\tau }_{z}{s}_{z}+\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{\bf{B}}\cdot {\bf{s}}+\frac{1}{2}{g}_{{\rm{v}}}{\mu }_{{\rm{B}}}{{\rm{B}}}_{z}{\tau }_{z}$$ (1) $${H}_{{\rm{h}}}=-\,{H}_{{\rm{e}}},$$ (2) with the spin and valley g -factors g s = 2 and g v = 15, the Bohr magneton μ B , the proximity-enhanced (intrinsic) Kane–Mele spin-orbit coupling, Δ SO = 70 μeV, and the Pauli matrices s i and τ i , which act on spin and valley, respectively. We approximate the effect of the right (R) and left (L) finger gate on the charging energy of the system by $${E}_{{\rm{c}}}({N}_{{\rm{R}}},{N}_{{\rm{L}}})=e{N}_{{\rm{R}}}{V}_{{\rm{R}}}+e{N}_{{\rm{L}}}{V}_{{\rm{L}}},$$ (3) with the absolute value of the elementary charge, e , QD occupation numbers N L = −1 (1h) and N R = 1 (1e) and gate voltages V R and V L . To describe transport through the electron–hole DQD we focus on the (0,0) → (−1,1) → (−1,0) → (0,0) charge cycle and consider only sequential tunnelling. There are in total 25 possible states of the system χ = (hole QD state, electron QD state) with $$\chi =(\overline{{\varphi }_{{\rm{h}}}}\,,\overline{{\psi }_{{\rm{e}}}})$$ (4) $$\overline{{\varphi }_{{\rm{h}}}},\overline{{\psi }_{{\rm{e}}}}\in \{0,K\uparrow \,,K\downarrow \,,{K}^{{\prime} }\uparrow \,,{K}^{{\prime} }\downarrow \}.$$ (5) Here, \(\overline{{\varphi }_{{\rm{h}}}}\,,\overline{{\psi }_{{\rm{e}}}}\) describes the state of the left and right QD, which includes the four single-particle states, as well as the QD being empty. We assume no mixing between lead and QD states and equal tunnel probabilities to and from the leads for all states, γ L,R = 1.7 GHz. Thus we obtain the transition rates between QD states involving tunnelling processes from the leads (L,R) by computing $${W}_{\chi \leftarrow {\chi }^{{\prime} }}^{{\rm{L,R}}}={\gamma }^{{\rm{L,R}}}\,f({E}_{\chi }-{E}_{{\chi }^{{\prime} }}-{\mu }^{{\rm{L,R}}}),$$ (6) with the Fermi function, f , at T = 0.1 K, and electron and hole QD states ϕ h , ψ e . Note that hole states tunnel only to the left lead and electron states tunnel only to the right lead. For interdot transitions, we assume no mixing of electron and hole states due to the small interdot tunnel coupling. For simplicity, relaxation is neglected. We obtain the rates of interdot transition by computing $${W}_{(0,0)\leftarrow ({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})}^{{\rm{inter}}}={W}_{({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})\leftarrow (0,0)}^{{\rm{inter}}}={G}^{{\rm{inter}}}\,\langle {\varphi }_{{\rm{h}}}| {\psi }_{{\rm{e}}}\rangle \frac{1}{\sqrt{2\pi \sigma }}\exp \left(-\frac{{({E}_{(0,0)}-{E}_{({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})})}^{2}}{4{\sigma }^{2}}\right),$$ (7) with the interdot tunnel rate γ inter = 6 kHz and \({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}}\in \{K\uparrow \,,K\downarrow \,,{K}^{{\prime} }\uparrow \,,{K}^{{\prime} }\downarrow \}\) . Gaussian energy smearing models the experimentally observed peaks with an estimated width of the resonances Γ = 40 μeV. We expect that this smearing originates from voltage fluctuations of the FGs. The overlap between electron and hole states is given by \(\langle {\varphi }_{{\rm{h}}}| {\psi }_{{\rm{e}}}\rangle ={({\sigma }_{y}{\tau }_{x}{s}_{y})}_{{\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}}}\) to ensure that only electrons and holes with opposite quantum numbers are created (or annihilated). With equation ( 7 ) we implicitly assume that the states in the left and right QD have no coherent phase relation. We solve the master equation of the probabilities, P χ , of the system being in state χ , $${\dot{P}}_{\chi }=\sum _{{\chi }^{{\prime} }}({W}_{\chi \leftarrow {\chi }^{{\prime} }}\,{P}_{{\chi }^{{\prime} }}-{W}_{{\chi }^{{\prime} }\leftarrow \chi }{P}_{\chi }),$$ (8) in the stationary limit, \({\dot{P}}_{\chi }=0\) , normalizing the probabilities to ∑ χ P χ = 1. In the stationary limit, we can compute the current through the DQD by computing the current flow from the right QD to lead R: $${I}^{{\rm{R}}}=e\sum _{{\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}}}({W}_{({\varphi }_{{\rm{h}}},0)\leftarrow ({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})}^{{\rm{R}}}{P}_{({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})}-{W}_{({\varphi }_{{\rm{h}}},{\psi }_{{\rm{e}}})\leftarrow ({\varphi }_{{\rm{h}}},0)}^{{\rm{R}}}{P}_{({\varphi }_{{\rm{h}}},0)}).$$ (9) We followe this procedure for different magnetic fields and different gate voltage combinations, V L , V R . The results are shown in Extended Data Fig. 5 , in which we were able to reproduce the experimental data from Fig. 2b–d,f–h . Additionally we simulate the current along the detuning axis of the (0h, 0e) → (1h, 1e) triple point as a function of parallel magnetic field, which is presented in Fig. 3e . Comparison of singlet–triplet Pauli blockade and single-particle electron–hole blockade In this section we compare the robustness of the single-particle electron–hole blockade with conventional singlet–triplet spin Pauli blockade in different material systems, which is summarized in Extended Data Table 1 . For the conventional singlet–triplet spin Pauli blockade in unipolar DQDs (that is, electron–electron or hole–hole DQDs), a finite tunnelling probability into higher orbital states (or valley states in silicon) lifts the blockade and sets a limit to the detuning window (typically in the order of 0.05−1.70 meV depending on the material system and QD size) in which the DQD remains Pauli blocked, as illustrated by Extended Data Fig. 6c,d . In a unipolar BLG DQD, spin blockade has been observed with a detuning range limited by both B -field-dependent excited-state splitting and orbital splitting to a maximum of 1.7 meV, but only for magnetic fields larger than 0.6 T 25 . In the single-particle electron–hole regime, however, the blockade is robust over the entire experimentally available detuning window rather than over merely a small region, which is illustrated in Extended Data Fig. 6a . This is due to the fact that the next-higher accessible states for increasing detuning are separated not only by the confinement energy of the QDs but also by the sum of the confinement energy and size of the band gap, which is typically in the order of 20−120 meV 13 . Additionally, due to particle–hole symmetry, electron and hole states have a symmetric spin and valley texture. This means that ground state \(\left|{K}^{{\prime} }\uparrow \right\rangle \) for the electron QD infers that the ground state in the hole QD is \(\left|{K}^{{\prime} }\uparrow \right\rangle \) . Because of this, ground-state–ground-state transitions are twofold blocked, by both spin and valley selection rules. Crucially, the ground-state–ground-state transition remains blocked independent of the applied perpendicular or parallel magnetic fields, making it possible to tune the system without losing the single-particle blockade. This is in contrast to the conventional singlet–triplet spin Pauli blockade, which is blocked only by spin selection rules and where the blockade is lifted as soon as the Zeeman splitting of the triplet state equals the singlet–triplet splitting, Δ S–T . For qubit operations in most systems, this is not a major obstacle because the involved states tune only weakly with magnetic field and one usually remains at low B -fields to have the longest spin (and/or valley) lifetimes. What makes our system stand out, however, is the fact that it is strongly tuneable by magnetic fields (due to high valley g -factors (10–70)) while still maintaining the blockade across the whole detuning range (making it robust for electron or hole shuttling or other long-range coupling mechanisms 30 ). Moreover, it is noteworthy that whereas relaxation processes that bring the QD to its ground state do not lift the blockade in our electron–hole system, they do lift the blockade in singlet–triplet Pauli blockade systems 23 . Finally we note that the electron–hole blockade is robust against charge noise up to a magnitude in the order of the charging energy, E C , in the case where the electron–hole DQD is tuned to the most noise-resilient configuration, the centre of the (1h,1e) charge occupation. Typical charging energies in BLG QDs measure 2–10 meV 18 , 27 , 42 , 43 , 44 . These values are typically around three orders of magnitude higher than the effective charge noise, which can be estimated from the effective electron temperature broadening of the Coulomb peaks of a QD. In our setup, the effective electron temperature is around 60 mK 17 , corresponding to an energy scale of around 5 μeV, which is indeed three order of magnitudes lower than the charging energy of DQD #1, E C = 6 meV. When considering material systems that posses a valley degree of freedom, using an electron–hole single-particle blockade has additional advantages. Together, spin and valley degree of freedom give rise to many possible states for QD systems with few particles. For example, a two-electron single QD with two spin and two valley degrees of freedom already has 16 possible states 27 , which results in a total of 256 transitions for two electrons ((1,1) → (0,2)) in a unipolar DQD. Even though this number is reduced by spin and (much weaker) valley selection rules 25 , the large number of transitions (which depend on the device geometry and applied magnetic field) strongly reduces the parameter window and requires the fine-tuning of a device into spin or valley blockade 25 , which becomes a disadvantage when scaling the system to many DQDs—that is, for quantum computing applications. Hence, being able to operate the blockade in the single-particle regime is very convenient because the number of possible transitions is drastically reduced. Also, in the case of unipolar BLG DQDs, spin blockade is absent even for magnetic fields smaller than 0.6 T. These high magnetic fields result in large spin splittings ( \(\Delta E=\sqrt{{\Delta }_{{\rm{SO}}}^{2}+{({\mu }_{{\rm{B}}}{g}_{{\rm{s}}}{B}_{\perp })}^{2}}\) ), which correspond to experimentally challenging qubit frequencies in excess of 25 GHz. It is very important to note that perfect particle–hole symmetry is not necessary for the presence of single-particle blockade: only the texture of the states is required to have the same ordering for particles and holes. All mechanisms changing this texture are strongly suppressed, as discussed in the next section. Breaking the symmetry of the spin and valley texture in BLG The invariance of the Hamiltonian of BLG, H BLG , under particle–hole transformation K (see above) rsults in a mirror symmetric spectrum for electrons and holes in which the electron and hole ground states have the same spin and valley quantum number. Thus, the ground-state–ground-state transition is blocked by both spin and valley selection rules. This transition remains blocked, independent of the applied magnetic field. In this section we discuss potential mechanisms that break the symmetry of the spin and valley texture—that is, the ordering of electron and hole states. Such mechanisms therefore have the potential to lift the single-particle blockade. Extended Data Table 2 summarizes the potential mechanisms and requirements in regard to breaking the symmetry of the spin or valley texture. The single-particle Hamiltonian describing the states in the electron ( H e with σ z = +1) and hole ( H h with σ z = −1) QD under the influence of an external magnetic field is given by $$\begin{array}{r}{H}_{{\rm{e}}}=+\,\frac{1}{2}{\Delta }_{{\rm{SO}}}{\tau }_{z}{s}_{z}+\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{\bf{B}}\cdot {\bf{s}}+\frac{1}{2}{g}_{{\rm{v}}}^{{\rm{e}}}{\mu }_{{\rm{B}}}{{\rm{B}}}_{z}{\tau }_{z},\\ {H}_{{\rm{h}}}=-\,\frac{1}{2}{\Delta }_{{\rm{SO}}}{\tau }_{z}{s}_{z}-\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{\bf{B}}\cdot {\bf{s}}-\frac{1}{2}{g}_{{\rm{v}}}^{{\rm{h}}}{\mu }_{{\rm{B}}}{{\rm{B}}}_{z}{\tau }_{z}.\end{array}$$ Breaking of the symmetric spin and valley texture in our electron–hole DQD system could have arisen from either (1) strong asymmetries—in the end sign flips—of the individual terms (Δ SO , g s or g v ) in the single-particle Hamiltonian, H e and H h , describing the states in the two QDs or (2) the introduction of extrinsic Bychkov–Rashba-type SO coupling with varying strength to both QDs. Spin and valley g -factors To allow the spin-Zeeman and valley-Zeeman terms to break the symmetric ordering of the electron and hole states in the two QDs, variations in the spin or valley g -factor are insufficient: a sign change of the g -factor in the two QDs would be required. Because the spin g -factor of electrons and holes in BLG has been found—in agreement with graphene and carbon nanotubes 29 , 45 —to be in good agreement with the Landé g -factor of free electrons, g s ≈ 2 (refs. 14 , 46 ), there is no reason to expect any asymmetry or even a sign change here. Therefore, the spin-Zeeman term cannot break the symmetry of the level texture and this term will not be discussed further. Although the precise value of the valley g -factor depends on both the QD confinement potential and the applied electric displacement field 18 , 41 , 47 , the sign of the valley g -factor for a confined electron in BLG depends only on the sign of the Berry curvature. This is a robust property of the BLG band structure, dependent only on the valley and band indices. Hence, a sign reversal of the spin- and valley-Zeeman terms between the two QDs can be excluded. Layer-dependent Kane–Mele spin-orbit term The intrinsic Kane–Mele spin-orbit Hamiltonian, which acts on the states of the four-band Hamiltonian ( H BLG above) is given by ref. 7 , \({H}_{{\rm{SO}}}=\frac{1}{2}{\Delta }_{{\rm{SO}}}{\sigma }_{z}{\tau }_{z}{s}_{z}\) . It is symmetric under particle–hole transformation but may be subject to (small) corrections due to a substrate-dependent, proximity-induced enhancement or potential partial cancellations (not yet observed) of Δ SO . Such a correction to Δ SO only affects particle–hole symmetry if both electron and hole QDs experience a different effective SO gap, due either to layer-dependent variations of Δ SO or a strong spatial variation of Δ SO . In the more general setting, the spin-orbit Hamiltonian can be described by the form $${H}_{{\rm{SO}}}=\frac{1}{4}{\Psi }^{\dagger }\left[\left({\Delta }_{{\rm{SO}}}^{{\rm{t}}}+{\Delta }_{{\rm{SO}}}^{{\rm{b}}}\right){\sigma }_{z}-\left({\Delta }_{{\rm{SO}}}^{{\rm{t}}}-{\Delta }_{{\rm{SO}}}^{{\rm{b}}}\right){\sigma }_{0}\right]{\tau }_{z}{s}_{z}\Psi ,$$ with different spin-orbit coupling energy gaps \({\Delta }_{{\rm{SO}}}^{{\rm{t}}}\) and \({\Delta }_{{\rm{SO}}}^{{\rm{b}}}\) for the top (t) and bottom (b) layers of the BLG 8 . \({\Delta }_{{\rm{SO}}}^{{\rm{t}}},{\Delta }_{{\rm{SO}}}^{b}\) correspond to λ I 1 and \({\lambda }_{I1}^{{\prime} }\) in ref. 8 . To allow these mechanisms to alter the spin and valley texture, they must be sufficiently large to flip the sign of Δ SO in one of the QDs: \({\rm{sgn}}({\Delta }_{{\rm{SO}}}^{{\rm{t}}})\ne {\rm{sgn}}({\Delta }_{{\rm{SO}}}^{{\rm{b}}})\) . In Figs. 1b and 3a and Extended Data Fig. 4a it can be seen that a sign change in Δ SO for either electrons or holes is required to change the ordering of states. Note that a decrease of, or even a sign change of the SO gap has not been observed experimentally to date. Measurable differences between \({\Delta }_{{\rm{SO}}}^{{\rm{t}}}\) and \({\Delta }_{{\rm{SO}}}^{{\rm{b}}}\) are likely to occur only if the BLG is encapsulated between two different materials with very different SO coupling 48 , 49 , 50 . In previous publications it has been shown that the SO gap in similar devices is always very close to 70 μeV (ref. 9 , 10 ), with deviations from this value in the order of only 10%, which is close to measurement uncertainty. An interesting side note: a strong asymmetry between \({\Delta }_{{\rm{SO}}}^{{\rm{t}}}\) and \({\Delta }_{{\rm{SO}}}^{{\rm{b}}}\) leads not only to different spin-orbit gaps at zero magnetic field but also to a finite noncollinearity between spin states in the electron and hole QD for intermediate in-plane magnetic fields. This is because tilting of the spin states into the plane with increasing B ∥ occurs at higher magnetic field for increasing Δ SO . This effect, however, is suppressed when the spin-quantization axis is along the in-plane or out-of-plane direction—that is, either \(\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{B}_{\parallel }\ll {\Delta }_{{\rm{SO}}}+\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{B}_{\perp }\) or \(\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{B}_{\parallel }\gg {\Delta }_{{\rm{SO}}}+\frac{1}{2}{g}_{{\rm{s}}}{\mu }_{{\rm{B}}}{B}_{\perp }\) . Breaking of particle–hole symmetry due to Bychkov–Rashba spin-orbit coupling Since the inversion symmetry of BLG is explicitly broken in our case by applying a perpendicular electric displacement field, extrinsic (Bychkov–Rashba) spin-orbit coupling poses an additional mechanism to break the particle–hole symmetry in our electron-hole DQD system. The corresponding term in the Hamiltonian is given by 8 $${H}_{{\rm{B}}{\rm{R}}}=\frac{1}{2}{\lambda }_{{\rm{e}}{\rm{x}}}{\Psi }^{\dagger }({{\sigma }}_{y}{s}_{x}+{\tau }_{z}{{\sigma }}_{x}{s}_{y})\Psi .$$ The extrinsic (Bychkov–Rashba) SO coupling, λ ex , scales linearly with the applied electric displacement field ( λ ex corresponds to λ 3 in ref. 8 ). To understand the influence of the extrinsic (Bychkov–Rashba) term we note that, for Fermi energies close to the band edge, the sublattice space is equivalent to the layer space and therefore to the conduction and valence band. This is caused by the fact that excess charge is strongly layer polarized, leading only to a small admixture of sublattices 3 , 22 . The extrinsic SO term H BR couples the two sublattices via σ x , y and therefore to the two layers, which are subject to a potential difference due to the electric displacement field. As a consequence, the extrinsic spin-orbit term is suppressed proportional to \( \sim {\lambda }_{{\rm{ex}}}^{2}/{E}_{{\rm{g}}}^{2}\) . Theoretical predictions of λ ex are at least three orders of magnitude smaller than the band gap ( E g ), rendering extrinsic spin-orbit coupling irrelevant for our system 8 , 10 . Investigation of transport properties for asymmetric valley g -factors in electron and hole QDs We argue above that the spin and valley texture of single-particle spectra, along with the single-particle blockade, is preserved even when the valley g -factor in the two QDs is not the same (but has the same sign). In this section we discuss how asymmetric valley g -factors could quantitatively affect the transition spectrum of the electron–hole DQD for both forward and backward bias direction. Extended Data Fig. 7a–d shows how we simulate current through the device as a function of detuning energy \(\widetilde{\varepsilon }\) and perpendicular magnetic field, B ⊥ , for different combinations of valley g -factors in the hole and electron QD, respectively. As clearly visible in Extended Data Fig. 7a,b , both the α and β transition split due to a difference in valley g -factors (coloured lines in Extended Data Fig. 7a ) by \(\Delta E=\frac{1}{2}{\mu }_{{\rm{B}}}| {g}_{{\rm{v}}}^{{\rm{e}}}-{g}_{{\rm{v}}}^{{\rm{h}}}| {B}_{\perp }\) . For equal valley g -factors, α and β do not show any B ⊥ dependence, as shown in Extended Data Fig. 7c . A tiny asymmetry in valley g -factors is allowed withoutnotably changing the observed features for magnetic fields below 1 T, as shown in Extended Data Fig. 7d , where a g -factor asymmetry of 0.1 is assumed. We also investigate the effect of asymmetric valley g -factors on the robustness of the observed electron–hole single-particle blockade for the reverse bias direction. As discussed above, the Pauli blockade does not require an unbroken particle–hole symmetry but relies on the specific spin and valley texture—that is, the ordering of states, in particular of the ground state. At finite perpendicular magnetic field the electron and hole ground states have the same spin and valley-quantum numbers, preventing the annihilation of electron–hole pairs independently of the valley g -factors of the two QDs. Therefore the blockade is robust to valley g -factor asymmetries. This can be seen in Extended Data Fig. 7e–h , in which we repeat the simulations of Extended Data Fig. 7a–d but for reversed bias. Indeed the single-particle blockade remains robust and independent of the detuning and strength and direction of the applied magnetic field for asymmetries in the valley g -factors. To quantitatively estimate valley g -factor asymmetry in our experiment, we fit Gaussian peaks of width Γ to the detuning cuts shown in Fig. 3b , allowing for a constant background and assuming equal width for both peaks, α and β . Such a fit is exemplarily shown in Extended Data Fig. 8a . The fitted width of the two peaks increased slightly for increasing B ⊥ , as shown in Extended Data Fig. 8b . In the worst-case scanario, when attributing this effect entirely to a difference in electron and hole g -factors rather than, for example, changes in tunnel broadening, we obtaine a maximum g -factor difference of g v ≈ 0.1 (compare with Extended Data Fig. 7d ). Data availability The data supporting the findings are available in a Zenodo repository under accession code . Code availability The simulation code is available in a Zenodo repository under accession code

Quantum dots in semiconductors such as silicon or gallium arsenide have long been considered hot candidates for hosting quantum bits in future quantum processors. Scientists at Forschungszentrum Jülich and RWTH Aachen University have now shown that bilayer graphene has even more to offer here than other materials. The double quantum dots they have created are characterized by a nearly perfect electron-hole-symmetry that allows a robust read-out mechanism—one of the necessary criteria for quantum computing. The results were published in Nature. The development of robust semiconductor spin qubits could help the realization of large-scale quantum computers in the future. However, current quantum dot based qubit systems are still in their infancy. In 2022, researchers at QuTech in the Netherlands were able to create 6 silicon-based spin qubits for the first time. With graphene, there is still a long way to go. The material, which was first isolated in 2004, is highly attractive to many scientists. But the realization of the first quantum bit has yet to come. "Bilayer graphene is a unique semiconductor," explains Prof. Christoph Stampfer of Forschungszentrum Jülich and RWTH Aachen University. "It shares several properties with single-layer graphene and also has some other special features. This makes it very interesting for quantum technologies." One of these features is that it has a bandgap that can be tuned by an external electric field from zero to about 120 milli-electronvolt. The band gap can be used to confine charge carriers in individual areas, so-called quantum dots. Depending on the applied voltage, these can trap a single electron or its counterpart, a hole—basically a missing electron in the solid-state structure. The possibility of using the same gate structure to trap both electrons and holes is a feature that has no counter part in conventional semiconductors. "Bilayer graphene is still a fairly new material. So far, mainly experiments that have already been realized with other semiconductors have been carried out with it. Our current experiment now goes really beyond this for the first time," Christoph Stampfer says. He and his colleagues have created a so-called double quantum dot: two opposing quantum dots, each housing an electron and a hole whose spin properties mirror each other almost perfectly. The double quantum dots were produced at the Helmholtz Nano Facility, the central technology platform for the production of nanostructures and circuits in the Helmholtz Association. Credit: Forschungszentrum Jülich / Sascha Kreklau Wide range of applications "This symmetry has two remarkable consequences: it is almost perfectly preserved even when electrons and holes are spatially separated in different quantum dots," Stampfer said. This mechanism can be used to couple qubits to other qubits over a longer distance. And what's more, "the symmetry results in a very robust blockade mechanism which could be used to read out the spin state of the dot with high fidelity." "This goes beyond what can be done in conventional semiconductors or any other two-dimensional electron systems," says Prof. Fabian Hassler of the JARA Institute for Quantum Information at Forschungszentrum Jülich and RWTH Aachen University, co-author of the study. "The near-perfect symmetry and strong selection rules are very attractive not only for operating qubits, but also for realizing single-particle terahertz detectors. In addition, it lends itself to coupling quantum dots of bilayer graphene with superconductors, two systems in which electron-hole symmetry plays an important role. These hybrid systems could be used to create efficient sources of entangled particle pairs or artificial topological systems, bringing us one step closer to realizing topological quantum computers." The research results were published in the journal Nature. The data supporting the results and the codes used for the analysis are available in a Zenodo repository.

10.1038/s41586-023-05953-5

Biology

Scientists confirm dorado catfish as all-time distance champion of freshwater migrations

Ronaldo B. Barthem et al, Goliath catfish spawning in the far western Amazon confirmed by the distribution of mature adults, drifting larvae and migrating juveniles, Scientific Reports (2017). DOI: 10.1038/srep41784 Journal information: Scientific Reports

http://dx.doi.org/10.1038/srep41784

https://phys.org/news/2017-02-scientists-dorado-catfish-all-time-distance.html

Abstract We mapped the inferred long-distance migrations of four species of Amazonian goliath catfishes ( Brachyplatystoma rousseauxii, B. platynemum, B. juruense and B. vaillantii ) based on the presence of individuals with mature gonads and conducted statistical analysis of the expected long-distance downstream migrations of their larvae and juveniles. By linking the distribution of larval, juvenile and mature adult size classes across the Amazon, the results showed: (i) that the main spawning regions of these goliath catfish species are in the western Amazon; (ii) at least three species— B. rousseauxii, B. platynemum, and B. juruense —spawn partially or mainly as far upstream as the Andes; (iii) the main spawning area of B. rousseauxii is in or near the Andes; and (iv) the life history migration distances of B. rousseauxii are the longest strictly freshwater fish migrations in the world. These results provide an empirical baseline for tagging experiments, life histories extrapolated from otolith microchemistry interpretations and other methods to establish goliath catfish migratory routes, their seasonal timing and possible return (homing) to western headwater tributaries where they were born. Introduction The Amazon has two main groups of migratory fish species, and they belong to the orders Siluriformes (catfishes) and Characiformes (characins) 1 , 2 , 3 . Major fisheries in the Amazon are based on knowledge of the seasonal upstream and downstream movements of fish, although the life cycles are poorly known by fishermen because the species enter and leave local fishing areas 4 , 5 , 6 . Long-distance fish migrations (>1,000 km) that exclusively or partially involve freshwater are known for salmon 7 and eels 8 but have also been inferred for Amazon goliath catfishes of the family Pimelodidae 9 , 10 , 11 , 12 . Although the migrations of most catfish species are poorly known, the general pattern reported is upstream movement to spawn, downstream passive drifting and even active migration of young juveniles to enter nursery habitats in river channels, floodplains or estuaries 7 , 13 , 14 , 15 . In the Amazon, goliath catfishes are major river channel and estuarine predators that are represented by a paraphyletic group of six extant and one fossil species of the genus Brachyplatystoma 16 , 17 , 18 with a maximum known adult fork length (FL) of 60–280 cm 9 ( Fig. 1 ). Figure 1: Migratory goliath catfishes ( Brachyplatystoma , Pimelodidae). ( A ) B. vaillantii (piramutaba in Portuguese, pirabutón in Spanish); ( B ) B. rousseauxii (dourada in Portuguese, dorado in Spanish); ( C ) B. platynemum (babão in Portuguese, mota flemosa in Spanish); ( D ) B. juruense (zebra in Portuguese, zebra in Spanish); ( E ) Dorado migrations exploited by fishermen. The Santo Antônio Dam on the Madeira River now drowns the Teotônio Rapids (shown here) where B. rousseauxii (species in photo) and B. platynemum were previously exploited and easily detected when migrating. Photos by M. Goulding. Full size image The first hypotheses of goliath catfish migration focused on B. rousseauxii and B. vaillantii , the most important commercial species, and their dependence on the Amazon River estuary as their nursery and inland rivers as feeding and spawning areas 9 . The spawning areas, however, were only identified as being located in the western Amazon, a vast region of at least 2 million km 2 that includes parts of Colombia, Ecuador, Peru, Bolivia and Brazil. More recent biological studies based on commercial fisheries in all major western Andes-Amazon tributary basins 19 , 20 , 21 , 22 , 23 established the seasonal, and in some cases year-round, presence of mature goliath catfishes in the western Amazon to at least a few hundred km downstream of the Andes. However, these investigations could not verify the spawning areas, leaving the possibility that they are located farther upriver and closer to the Andes. We present data on the distribution of goliath catfish ( Brachyplatystoma ) adults, larvae and juveniles across the Amazon Basin, including areas in or near the Andes. Even without tagging experiments, the general differential distribution of sub-adult (downstream) and adult (upstream) goliath catfish size-classes 9 , as well as otolith microchemistry data 10 , strongly indicates that long-distance upstream goliath catfish migration occurs. Seasonal upstream goliath catfish movements are also visually obvious at some cataracts, such as those of the Madeira River ( Fig. 1 ) in Brazil, before dams were built 4 , and those of the Caquetá River in Colombia 6 . Therefore, it is reasonable to hypothesize that long-distance downstream migrations of young fish occurs, otherwise there would be no recruitment to nurseries, in some cases as far downstream as the Amazon River estuary. To test the goliath catfish migratory hypothesis related to western spawning regions, we mapped the presence of mature adults complemented by the seasonal and geographical variation in abundance and length of larvae and juveniles in river channels across the Amazon, including in the estuary. Two complementary geographical and temporal perspectives were used. The first focused on the Madeira Basin, the Amazon’s largest sub-basin with headwaters in the Andes, and where year-round collections were performed. The second included most of the Amazon and available data from all years. Our study shows that spawning for at least B. rousseauxii, B. juruense and B. platynemum occurs in or near the Andes and demonstrates conclusively that long-distance downstream migration of their larvae and juveniles occurs. The great variation in B. platynemum larvae size across the Amazon indicates a much wider nursery area than just the Andean region. Size-class data for B. vaillantii indicate long-distance migration and spawning in the western Amazon but do not confirm it near the Andes. Our results are also discussed in light of published goliath life history hypotheses derived from genetic and otolith isotope signatures. Results Adult Distribution and Gonadal Stage Fish length and gonadal data derived from specimens captured in fisheries in major Andes-Amazon sub-basins in Brazil, Bolivia and Peru, complemented by published data from studies based in Colombia 19 , 23 and Ecuador 24 , show the wide distribution of mature B. rousseauxii, B. platynemum and B. juruense in all major turbid rivers with headwaters in the Andes, including the Amazon River main channel ( Fig. 2 and Table 1 ). Mature fish are defined as ripe individuals, that is, individuals with fully developed ova or testes 25 . Brachyplatystoma rousseauxii has the widest distribution. In addition to whitewater (turbid) rivers arising in the Andes, it is found in many clearwater and blackwater tributaries that arise on the Brazilian and Guiana Shields. With the exception of the lower Tocantins River 26 , which is part of the Amazon River estuary, B. rousseauxii is rarely registered in the fisheries of blackwater and clearwater rivers, an indication of its rarity in these drainages. The only Brazilian or Guiana shield rivers where mature long-distance migratory goliath catfishes were found was the Branco, a semi-turbid tributary of the Negro River. Of the goliath species considered, only B. rousseauxii was present in the Branco River, but it is of minimal importance in fisheries there and is reported by fishermen to be relatively rare. Figure 2: Distribution of sites investigated and locations of mature goliath catfishes. ( a ) Sites investigated by the authors and others from published data. ( b – e ) Locations of mature goliath catfishes by species. Figure was created by authors with ArcGIS for Desktop Advanced 10.2, MapPublisher 9.6 tool inside Adobe Illustrator CC, and Adobe Illustrator CC, 2.0. Full size image Table 1 Summary of goliath catfish data locations where mature goliath catfishes were captured in major rivers in Brazil, Bolivia and Peru by the authors, complemented by documented data from studies in Colombia and Ecuador: L1 19 , L2 23 , L3 22 , and L4 24 . Full size table In contrast to the other three widely distributed species discussed in this paper, B. vaillantii is rare or naturally absent above the Madeira Rapids in the southwestern Amazon of Bolivia and Peru. The new fishways around the Madeira dams, however, could allow B. vaillantii to migrate more easily to Bolivian and Peruvian waters in the southwestern Amazon, which would alter the population dynamics of this species and perhaps other species on which it preys. Mature B. rousseauxii, B. platynemum and B. juruense are only abundant, as indicated by fisheries catches, between approximately 55–250 m of elevation and only in turbid rivers at least 3,500 km upstream from the Amazon River mouth ( Fig. 2 and Table 1 ). Considering only adult stages, B. vaillantii has the widest distribution of the long-distance migratory goliath catfishes because of its presence in the far eastern Amazon, including the estuary 9 , 27 . In the estuary, however, B. vaillantii adults are not mature as defined above. Mature B. vaillantii adults are known as far upstream as the Pongo de Manseriche, an Andean gorge of the Marañón River at 246 m elevation in northern Peru and 4,847 km upstream of the Amazon River mouth. Year-round gonadal stage data reported for B. rousseauxii, B. platynemum and B. juruense captured over a 12-month period by local fishermen in the upper Madeira (2003–2004) and Ucayali (2004–2005) basins of Peru provided unequivocal proof of the presence of only mature goliath catfishes in or near the Andes to at least 198 m elevation and 5,788 km from the Amazon River mouth ( Figs 2 and 3 and Table 2 ). Mature B. rousseauxii and B. platynemum were abundant in commercial fisheries in both areas during our sampling periods, whereas B. juruense was only captured in the Ucayali River. Other goliath catfishes ( B. filamentosum, B. capapretum , and B. tigrinum ) are also present in these rivers but in smaller quantities, as indicated by commercial fisheries 28 , 29 . Figure 3: Mature goliath catfish from commercial catches near the Andes indicate the presence of spawning populations. Monthly precipitation and monthly capture of mature B. rousseauxii, B. platynemum , and B. juruense between April 2002 and April 2004 in Puerto Maldonado (Madre de Dios River, Upper Madeira Basin) and between July 2004 and July 2005 at Atalaya (near the confluence of the Ucayali and Urubamba Rivers). B. juruense was not captured in the Madre de Dios River during our study period. Precipitation was used as a proxy for river level since there were few data for the latter in the Andean region. Full size image Table 2 Ready-to-spawn or recently spawned goliath catfish ( Brachyplatystoma rousseauxii ) near the Andes. Full size table All 5,348 B. rousseauxii , 2,985 B. platynemum and 528 B. juruense specimens we examined from commercial fisheries near the Andean Piedmont or in the immediate pre-Andean area were sexually mature adults with fully developed gonads or recently spawned fish ( Table 2 ). In contrast to the downstream regions, all goliath catfish captured near the Andean Piedmont had empty stomachs, suggesting that their presence in these areas was not related to trophic migration but to spawning. Although goliath catfishes were present in every month in the far western Amazon, the commercial captures from which our data were derived indicated strong seasonal variation ( Fig. 3 ). The seasonal differences between commercial captures in the Madre de Dios (upper Madeira Basin) and Ucayali and Urubamba Rivers (Ucayali Basin) ( Fig. 3 ) may reflect distinct headwater migration patterns or differences in fishing, considering that fishing was virtually halted in the Madre de Dios during the high water discharge period (November-February) due to the large quantities of downstream-moving wood that are a danger to nets. In the Ucayali and Urubamba Rivers, ready-to-spawn B. rousseauxii, B. juruense and B. platynemum were most abundant in commercial fisheries during the rainy and warmer months corresponding to the higher river discharge period from October to March ( Fig. 3 ). Since monthly fishing effort was approximately equivalent, landings can be considered a proxy for seasonal migratory fish abundance. Furthermore, goliath catfishes were captured by commercial fisheries in river channels with downriver drifting gill nets during all months. Since drifting gill nets only capture fish moving upstream, they reveal the direction of movement. Seasonal Larval Densities as Spawning Indicators Only B. rousseauxii larvae were captured throughout most of the year in standardized density sampling, including during the low water period, but in much lower quantities than during the December floods ( Fig. 4 ). The other two species present in the sampling area, B. platynemum and B. juruense , were poorly or not represented during 6–8 months in density samples that included parts of the high, low and rising water periods, and only the latter species was captured during the low water period. The highest larval densities of all species combined occurred between November and January, months when river discharge, depth and current velocity were rapidly increasing 30 . Mature B. platynemum were also fairly common in Madre de Dios commercial fisheries from May to August, which included the falling and low water period, verifying their presence in the region ( Fig. 3 ). Although B. juruense larvae were captured between November 2004 and August 2005, no adults were reported in the commercial fisheries of the Madre de Dios for this period, although the species is known in the study area 31 ( Fig. 3 ). Figure 4: Goliath catfish larvae densities based on standardized sampling as indicators of spawning periods. The monthly drifting larvae densities of goliath catfishes in the Madre de Dios River of the upper Madeira Basin in relation to total monthly precipitation (mm) levels. Larvae densities were estimated based on ichthyoplankton samples from the Madre de Dios River between November 2004 and August 2005. Precipitation data were used as a proxy for river level since data for the latter were not available. Precipitation data are for the city of Puerto Maldonado were supplied by the Servicio Nacional de Meteorología e Hidrología del Peru (SENAMHI). Full size image Growth of Larvae and Juveniles If the goliath catfish spawning areas and nurseries are separated by thousands of kilometers as we hypothesize, it should be reflected in the larvae and juvenile growth in the downstream direction. We tested this hypothesis for the Madeira Basin, for which seasonally sequential data were available, and for the Amazon as whole, including data from various years. The growth of fish in their early phase may be described by Brody’s equation, which relates fish size (L t ) to time (t) in the exponential equation L t = a × e K × t 32 , where a and K are constants. We used Brody’s equation but substituted the time variable with distance to fit the relationship between larvae/juvenile length and their distance from the mouth of the Amazon River. The size distributions of B. rousseauxii and B. juruense in the Madeira Basin ( Fig. 5 ) and the Amazon as a whole ( Fig. 6 ) fit exponential curves (p < 0.01 for both species). By the time B. rousseauxii enters the Amazon River from the Madeira River, it has reached juvenile size (approximately 20 mm). For B. juruense , the available data indicate an increase in median length from the headwaters to the upper Madeira in Brazil, after which the median length decreases, indicating a wide spawning region for the species ( Fig. 6 ). Juvenile stages of B. juruense are most common in the central Amazon, approximately 2,000–2,500 km from the Amazon River mouth, though much less is known about this species than the other goliath species considered and more collections are needed in the lower Amazon River, where juveniles are also known. Figure 5: Downriver migration of goliath larvae/juveniles in the Madeira River system. Box-plots of the lengths of goliath catfish drifting larvae from near the Andes in the Madre de Dios River to the Madeira River in Brazil, approximately 1,600 km downriver. Collection sites are indicated by the numbers above the box-plots: 1- Madre de Dios River at Los Amigos River confluence, 2- Madre de Dios River at Puerto Maldonado, 3- Madeira River at Porto Velho, and 4- Madeira River at Humaitá. Data were fit to an exponential curve, where Lmm is the size in millimeters and km is the distance in kilometers: B. rousseauxii: Lmm = 269.1 −0.00101× km , r 2 = 0.67, F (1, 327) = 670, p < 0.001; B. juruense: Lmm = 18.5 −0.00034× km , r 2 = 0.22, F (1, 76) = 23.1, p < 0.001; B. platynemum : r 2 = 0.17 F (1, 108) = 1.93 p > 0.05; B. vaillantii : r 2 = 0.08, F (1, 34) = 2.9, p > 0.05. Full size image Figure 6: Downriver migration of goliath larvae/juveniles in the Amazon Basin. Fish length scatterplots of goliath catfish drifting larvae from near the Andes to the Amazon River estuary in Brazil. Data were fit to an exponential curve, where Lmm is the size in millimeters and km is the distance in kilometers: B. rousseauxii: Lmm = 81.8 −0.00065× km , r 2 = 0.84, F (1, 491) = 670, p < 0.001; B. juruense: Lmm = 33.9 −0.00048× km , r 2 = 0.57, F (1, 130) = 172.8, p < 0.001; B. platynemum : r 2 = 0.04, F (1, 151) = 6.3, p > 0.05; B. vaillantii: Lmm = 50.9 −0.00085× km , r 2 = 0.64, F (1, 1035) = 1818, p < 0.001. Maps in figure was created by authors with ArcGIS for Desktop Advanced 10.2, MapPublisher 9.6 tool inside Adobe Illustrator CC, and Adobe Illustrator CC, 2.0. Full size image A seasonal factor for all species is that varying current speeds during sampling periods could affect the larval size distribution along the downstream movement. If current speed greatly influenced the distribution of juvenile size, then a much less precise pattern would be expected for B. rousseauxii since it spawns during varying periods of the year with different current speeds. However, B. rousseauxii shows the best fit of all species to the downstream migratory growth model. The wide distribution of small larvae of B. platynemum in the Madeira Basin and the Amazon as a whole and their highly mixed length-class distribution do not fit exponential curves, strongly suggesting that spawning for this species is widespread, that more than one population exists 33 and that the nurseries include a large area in the western and central, and perhaps even eastern, Amazon ( Figs 5 and 6 ). Larvae and juveniles of B. vaillantii have been captured in the lower stretches of the Madeira River and upriver near the Madeira Rapids, at least 3,129 km upstream of the estuary. The presence of small larvae (<5 mm) in this region suggests the proximity of a spawning area, but we have too few sample points to detect the growth of B. vaillantii larvae during their downriver migration ( Fig. 5 ). Nevertheless, the data for B. vaillantii larvae and juveniles for the entire Amazon fit an exponential curve (p < 0.01), corroborating the long-distance downstream larval and juvenile migration hypothesis ( Fig. 6 ). Given the presence of ripe adults of this species in the Marañón River in Peru at its outlet from the Andes, it is possible that some spawning occurs much farther upstream in the western Amazon than our data indicate. No small B. vaillantii larvae (<5 mm) were found in the Amazon River within 1,500 km of the estuary. Discussion We argue that the distribution of mature size classes and general downriver movement, concurrent growth and regional size-class differences of goliath catfish larvae and juveniles of B. rousseauxii, B. juruense and B. vaillantii indicate that the western spawning areas and downstream nurseries of these species are widely separated in the Amazon ( Figs 2 , 3 and 6 ). The greatest distances measured between spawning and nursery areas, as determined by larvae and juvenile presence, were 5,786 km for B. rousseauxii , 4,238 km for B. juruense and 3,129 km for B. vaillantii . The most extreme migration is undertaken by B. rousseauxii , which spawns in the far western Amazon but uses the estuary as its nursery, for a maximum known life history migratory cycle of all size classes of approximately 11,600 km ( Figs 5 and 6 ). The size-class distribution of older juveniles and adults indicates that it takes 1–2 years to reach the Andes during upstream migration from the estuary 9 , 22 , 34 . Based on B. rousseauxii size classes recorded from fisheries catches in the western Amazon 19 , 22 and farther downstream 9 , adults do not return to the estuary but remain in a large area of the western Amazon. The annual migration of adults subsequent to reaching the western Amazon for the first time, however, is probably much shorter; however, based on the distribution of adult size classes, it could still be 1,000–2,000 km or more. Because no other strictly freshwater long-distance fish migrations 7 close to those discussed in this paper have been reported, B. rousseauxii undertakes the longest migration in the world, considerably longer than previously hypothesized 9 . Brachyplatystoma rousseauxii migration also surpasses the maximum life cycle migration (6,000 km) reported for anadromous salmon ( Oncorhynchus ) 35 and is nearly as long as that of the European eel ( Anguilla anguilla Linnaeus 1758), including the freshwater and marine phases 8 . It is also possible that B. platynemum undertake migration from the estuary to the Andean region similar to B. rousseauxii , as juveniles of the former are relatively common in artisanal fisheries in the estuary 36 , but much less is known about them 37 . The general life history patterns of B. platynemum and B. juruense are similar to those of B. rousseauxii , with a more restricted, but still wide, separation of spawning and nursery areas. The more mixed larvae and juvenile size classes of B. platynemum across the Amazon indicate that their spawning areas are neither exclusively in the far western Amazon nor are their nurseries restricted to or mostly in the eastern Amazon. In contrast to B. rousseauxii and B. platynemum , neither adult nor larval B. juruense are known in the estuary, and the nursery of the latter species appears to be in the central Amazon. Of the long-distance migratory goliath catfishes, B. vaillantii is the only species whose adults and young share the estuary. To date, the maximum distance from the estuary that B. vaillantii larvae have been captured is 3,129 km. However, the presence of mature B. vaillantii adults in the Marañón River at the Pongo de Manseriche Gorge (4,847 km upstream) and in the Napo of Ecuador (4,754 km upstream) suggests the possibility of spawning in the far western Amazon. Based on commercial catches, B. vaillantii is most abundant in the Amazon River mainstem and was rare, if at all present, in Bolivia and Peru above the Madeira Rapids in Brazil. As mentioned above, the new dam bypasses might allow the species to pass the Madeira Rapids, which were previously a barrier 4 . The presence and abundance of mature goliath catfish in commercial fisheries in the Andean region is a reliable indicator of upstream movement to spawn. The absence of these fish in commercial fisheries during some or all months near the Andes, however, should not be interpreted as direct evidence that they are not present in the region, as fishing activity must also be considered. The striking differences between the monthly relative captures in the Madre de Dios and Ucayali-Urubamba Rivers is most likely due to the absence of fishing in the former rather than the absence of mature fish. The high catches in the Madre de Dios River during the falling river level period during all three years for which data are available should not be interpreted as a greater abundance at this time of year but rather that fishing is possible during these months ( Fig. 3 ). The presence of spawning B. rousseauxii and B. platynemum in the Madre de Dios River channel during the high water period is corroborated by the relative abundance of their larvae ( Fig. 4 ). The higher density of fish larvae during the rising water period is thus a better indicator of the main reproductive period of goliath catfishes than monthly commercial captures. The best indicator, however, is larval flux, which is the absolute density value of larvae per unit of time in a river section 38 . Drift densities of young fish generally decrease with increased river depth and flow velocity due to the dilution effect caused by higher turbulence in a much greater volume of water 39 . The larval flux index was not used due to the difficulty of obtaining accurate river discharge data near the Andes because of the few hydrological stations that exist. Considering that goliath catfish larvae densities were calculated without discharge calculations and were highest at the beginning of the rainy season and decreased during the January and March peak discharge periods 30 , it is possible that the highest larval flux would be in the latter period if river discharge was considered in the larvae density algorithms. Fisheries data from the middle Caquetá River in Colombia also indicate that upstream-moving mature goliath catfish are most abundant during the high water period 40 . While upstream areas near the Andes have yet to be investigated as spawning sites in the Caquetá Basin, large B. rousseauxii have been reported farther upstream to at least 170 m elevation in the neighboring Putumayo River 23 . In contrast to our data and that of the Caquetá, a gonadal review of 15,000 B. rousseauxii specimens captured by commercial fisheries during a 5-year period near Iquitos, Peru at 76 m elevation and approximately 4,000 km upstream of the Amazon River mouth hypothesized that breeding occurs mostly during the low water period and ends at the beginning of rising water 22 . The fisheries in both areas are intensive throughout the year, and fishing effort bias is minimal 22 , 40 . However, those studies analyzed the monthly proportions of mature females by combining maturation stages 3 (advanced maturation) and 4 (ripe) 25 . The uncertainty of the time between stages 3 and 4 raises doubts about combining the two to identify exactly when spawning occurs. It is unknown how far downstream of the Andes or immediate pre-Andean area that long-distance migratory goliath catfish spawn, and the Caquetá and Iquitos data alone do not confirm this. Nevertheless, the presence of maturing populations throughout the year in the Iquitos Region, an area of large rivers and floodplains where prey are more abundant 9 ( Fig. 2 , Table 1 ), suggests that migratory adult goliath fish, especially B. rousseauxii , remain in this area to feed and develop their gonads for subsequent reproductive cycles after their first arrival. A second study in the Iquitos area that included only larval abundance concluded that breeding occurs mostly during receding or low water 41 . Considering that spawning fish are present during all months in the Ucayali headwaters upriver of this site ( Fig. 3 ), it is reasonable to expect the presence of fish larvae downstream near Iquitos during all months. The larval results are based only on abundance without reference to water volume, the latter of which is needed to calculate the absolute abundance index to compare high and low water periods. A similar but slightly later peak in drifting larval abundance than that of the Madre de Dios River was found for B. rousseauxii in the Madeira River, 1,613 km downstream from the former sampling site, with maximum fluxes (larvae or juveniles/second) in January and February and minimal downstream migration in September 38 . Another study using bottom-trawl sampling in the same region of the Madeira River reported different results for B. rousseauxii , with relatively high larvae and juvenile abundance (number of fish/haul) during the low water period 42 . However, the study did not consider the bias introduced by the inverse effect of decreased water volume on larval density; thus, seasonal comparisons may not be accurate because the absolute number of larvae may be greater during the high water period but in a much greater volume of water. As indicated by the upper watershed areas in the Madeira and Ucayali Basins, the goliath catfish spawning zone is located in a mountainous to lowland transition area above about 170 m with relatively high channel slope (declivity = 0.16 m/km) and a lowland downstream area with a much lower declivity (generally less than 0.01 m/km) ( Table 1 ). The river channels in the spawning zone, and in the Andes in general below 300 m, are characterized by gravelly bottoms, as opposed to muddy and soft substrates farther downstream, greater turbidity than downstream, shallower river channel depth (7–20 m versus 40 + m downstream), higher pH (up to 7.9 versus up to 7.1 downstream), higher conductivity (up to 287 microsiemens/cm versus 70 microsiemens/cm downstream), highly saturated O 2 levels (up to 8.2 mg/l versus up to 6.4 mg/l downstream), and lower average water temperature (26.8 °C versus 28 °C downstream) 20 , 43 . Despite these striking physical differences, there are too few data to support a hypothesis of spawning site selection by long-distance migratory goliath catfishes. Furthermore, limnological and geomorphological data alone cannot be used as explanatory variables because biological factors, such as avoidance of egg and larvae predators, could also play a critical role 9 . The evolution of long-distance migratory goliath catfish life histories in connection with the western Amazon and western Orinoco, where they also occur, could reflect an ancient evolutionary spawning association with the Andes, as Andean fossils of their genus are known from at least the Miocene 12–11 Ma 16 , although the elevations at that time were probably not much higher than 200 m. The paleo-Amazon-Orinoco in which the ancient catfish species lived flowed north in the low foreland basin of the Andes 44 . Brachyplatystoma vaillantii, B. rousseauxii , and B. platynemum range widely beyond the Amazon, with all three found throughout the Orinoco Basin and the first two also in the large, short rivers of the Guianas. Juveniles have also been reported in river channels of the Orinoco Basin 45 . The geographical genetics of three goliath catfish species have been studied, but only B. platynemum presented clear population segregation for the Amazon River mainstem and Madeira River 33 , whereas B. vaillantii showed no genetic segregation 46 , 47 . Genetic studies of B. rousseauxii presented mixed results but indicated a relatively homogenous population in the Amazon Basin, although homing behavior could not be totally excluded 48 , 49 . Regional genetic differentiation of B. rousseauxii in the large headwater region of the upper Madera Basin in Bolivia could also indicate homing behavior 49 . The present genetic evidence indicates that long-distance migratory goliath catfish species have few distinct populations in the Amazon Basin, but it does not convincingly eliminate the possibility of homing. Recent studies have reconstructed the theoretical movement and migration of individuals of the Brachyplatystoma species by comparing strontium isotope signatures ( 87 Sr/ 86 Sr) along transverse sections of their otoliths with the isotope signatures of major river water types 10 , 11 . The main results show that the juvenile phases of B. rousseauxii and B. vaillantii have strontium isotope signatures of the western Andean tributaries and the Amazon River mainstem, and few present signatures for other water types. The estimated mean size (37 ± 16 mm (mean ± S.D.) of B. rousseauxii when it reaches the Amazon River during its downriver drift migration in the Madeira Basin has also been calculated using strontium isotope signatures and the relationship between otolith radius and body length 10 . The length range of B. rousseauxii larvae and juveniles captured in the Madeira River in our study varied from 5 to 42 mm ( Fig. 5 ), in agreement with that extrapolated from otoliths. The otolith data corroborate that the nursery habitat of these two species is mostly in turbid rivers, as directly indicated by our study and indirectly indicated by fisheries data 9 . Although considering the eastern Amazon as an obligate region of early goliath catfish life cycles, otolith isotope interpretations also point to the possibility that some of the large eastern clearwater tributaries might be used as nurseries for B. rousseauxii and B. vaillantii . The eastern estuary (Marajó Bay, Pará River and the lower Tocantins) is heavily influenced by the clearwater Tocantins River, especially during the Amazon River’s low water period 50 ; thus, it should be expected that at least part of the estuary goliath catfish nursery population would have a clearwater river otolith strontium signature. Our field studies do not indicate the presence of young (<20 cm) B. rousseauxii and B. vaillantii in the Xingu or Tapajós Rivers, and they are absent or extremely rare in the fisheries of those areas. A multi-element otolith microchemistry study also indicated that B. rousseauxii resided in the Amazon estuary for the first 1.5–2.0 years of life based on Sr:Ca and Ba:Ca ratios 12 . By presenting the convergence of three lines of evidence—the distribution of mature size classes, downstream migration of larvae/juveniles and otolith signatures—that strongly suggest long-distance goliath catfish migration in the Amazon, this study presents a significant step towards achieving a holistic understanding of the longest freshwater fish migrations in the world and at a mega-basin scale. Furthermore, it sets the hypothetical stage for eventual tagging experiments to understand the exact environmental cues that the fishes use during various life history movements and to test empirically recent homing hypotheses 10 and at what sub-basin level they occur. The methods used in this paper also raise uncertainties that the present data alone cannot address, such as why the goliath catfishes that use the eastern or central Amazon as nurseries migrate thousands of kilometers upstream to spawn in the far western Amazon associated with the Andes or nearby uplands rather than reproduce much closer in the middle or even lower reaches of the Amazon River mainstem or in the eastern tributaries arising on the Brazilian and Guiana Shields. Finally, the previous limitations imposed by inferring migration based on the wide separation of nurseries and spawning areas are largely eliminated in this paper by proof of long-distance downstream larvae/juvenile movements in the river channels. Of special relevance is the expected infrastructure development in the Andes, especially the combination of dams, headwater deforestation and mining activity 51 , 52 , which could present major threats to important spawning areas ranging from Colombia in the north to Bolivia in the south. Andean large dams will most likely be much different than those already constructed elsewhere in the Amazon, specifically their high walls. Even if high-wall dams are located upstream of spawning sites, they would greatly alter sediment and nutrient cycles downriver where spawning occurs. The long-distance migratory goliath catfishes provide a profound biological indicator of ecosystem health from the Andes to the freshwater Amazon River plume in the Atlantic, and the impacts on them should be considered in all major infrastructure development. Methods Definition of Long-Distance Migratory Goliath Catfishes A modern cladistic classification of the goliath catfish genus Brachyplatystoma includes seven species 17 , of which five of the extant six species are easily recognized by commercial fishermen. We consider five ( rousseauxii, vaillantii, platynemum, juruense and tigrinum ) to be long-distance (>1,000 km) migratory species. We did not include B. tigrinum in this study because it is relatively rare in fisheries, and our data for it were minimal, although small numbers of individuals are commonly present in association with the upstream movement of B. rousseauxii . The largest goliath catfishes are B. filamentosum and B. capapretum , both commonly reaching more than 2.5 m in length and 100 kg in weight. We found no field evidence that they perform long-distance migration (>1,000 km), and their mixed size classes across the Amazon indicate that they spawn widely and in various river types, although shorter migrations are probably involved 12 . They are also the only species that commonly enter floodplain waters 53 , whereas the long-distance migratory species are largely confined to river channels and the estuary. Adult Goliath Catfish Distribution and Size Classes Data for the presence of migratory goliath catfishes, their size classes and maturity stages were obtained by extensive field surveys of local fisheries conducted by the authors since the late 1970 s in six major regions: (1) the Amazon River mainstem from its estuary in Brazil to the confluence of the Ucayali and Marañón Rivers in Peru; (2) major tributary basins (Madeira, Ucayali and Marañón) with headwaters in the Andes; (3) the Purus and Juruá Basins, whose headwaters are associated with the Fitzcarrald Arch, which has a post-Andean uplift origin that produced their low hilly areas, such as the Serra do Divisor 54 ; (4) Guiana Shield basins (Negro, Branco and Trombetas); (5) Brazilian Shield basins (Trombetas, Xingu and Tapajós); and (6) the Amazon estuary, including Marajó Bay and offshore fresh waters in the Atlantic. The data from the Andean tributaries in Colombia and Ecuador were obtained from the literature. Published goliath catfish data, including fisheries catches, maturity stages and size classes, were especially important for the Caquetá and Putumayo basins, which have been investigated by Colombian scientists 19 , 23 . Photographic proof and coordinates of goliath catfishes for the Napo Basin near the Andes was provided in a recent thesis 24 . The presence of the migratory goliath catfishes and their maturity stages were verified by the authors from specimens in urban markets and complemented by local interviews. The maturation stages of females and males were based on macroscopic gonadal characteristics: ripe, ready-to-spawn individuals, spent or very recently spent 25 . Wherever present, goliath catfishes are known to the local people, and nearly all oral reports we received were eventually verified by actual specimens either from fishers or our own observations. Mature fish were considered abundant in a fishing area when they were among the 20 most important food species, frequent when they regularly appeared in markets in small quantities, and rare when observed by the authors but their presence was not well known by local fishers. Andean surveys based on interviews to establish the maximum elevations that goliath catfishes reach were conducted in the Amazon River’s two largest Andes-Amazon sub-basins, the Madeira and Ucayali. For the Madeira Basin, the Madre de Dios sub-basin in Peru was surveyed from the city of Puerto Maldonado in the lowlands (<200 m) to tributaries 4,000 m upstream, including the Manu, Alto Madre de Dios and Inambari Rivers, whose headwaters rise in the high Andes. Surveys in the Ucayali Basin ranged from near Cusco at 3,300 m to Atalaya at 200 m and included the Vilcanota, Urubabamba, Tambo and Ucayali Rivers. No evidence of long-distance migratory goliath catfishes was found above 250 m elevation in the Andes ( Table 1 ). Two large datasets of unpublished goliath catfish fork length and gonadal maturity stages are reported here for the first time: Madre de Dios River commercial fisheries based in the city of Puerto Maldonado between April 2002 and April 2004 in the Upper Madeira Basin of Peru and Urubamba and Ucayali commercial fisheries between July 2004 and August 2005 based in the city of Atalaya ( Table 2 ). The Madre de Dios fisheries are within 150 km of the Andes, and those of Atalaya are at the edge of the Andes. Ichthyoplankton Data Given the lack of tagging experiments to detect upstream movements in the days or weeks before spawning, a more substantial understanding of goliath catfish reproductive periods in the western Amazon is provided by the seasonal downstream drift of their larvae, which indicates when spawning occurs. We define larvae as the fish phase between hatching to complete loss of embryonic and larval organs 55 , with the juvenile stage beginning at approximately 20 mm 56 . Fish larvae larger than 2.7 mm were measured, and they were identified by the number of caudal and pre-caudal myomeres 56 . Year-round larvae and juvenile collections were conducted in the Madeira, the Amazon’s largest sub-basin, and these results were compared to all the data available for the Amazon, including all specimens captured in various years, seasons and locations. For the Madeira Basin, the Madre de Dios River in southeastern Peru was selected as the site near the Andes to test for the presence of headwater larvae because commercial fisheries based in the city of Puerto Maldonado revealed the at least seasonal presence of mature goliath catfishes near the Andes 20 . Total length and abundance data of goliath catfish larvae and juveniles were investigated near the Andes in the Madre de Dios, Ucayali and Urubamba Rivers; in the lowlands in the large channels of the Madeira and Amazon rivers; and in the freshwater areas of the Amazon River estuary. Samples were collected during daylight hours with ichthyoplankton nets with a mesh size range of 300–500 μm and trawl nets with a mesh size of 5 mm 30 , 56 , 57 . The fish larvae collection method was approved by the Brazilian Institute of the Environment - IBAMA (SISBIO- permission number 4419) in Brazil and the Dirección General de Extracción y Producción Pesquera para Consumo Humano Directo from the Ministry of Production in Peru, with in-country legal deposit protocols of the Museo de Historia Natural, Lima. Standardized larval density data 30 were collected weekly using ichthyoplankton nets in the Madre de Dios River between November 2004 and August 2005 ( Fig. 4 ). The sampling was conducted at five points across five cross-channel transects along 12 km of the river. Two samples were taken at each site: near the surface (1 m depth) and near the bottom (70% of the maximum depth). A mechanical flowmeter (General Oceanics 2030 R) was installed at the mouth of the net to estimate the volume of water sampled. The monthly drifting larvae densities were estimated based on the average of all samples. River Distance Calculations The distances from the Amazon River mouth to the studied areas were determined using the Barrier Analysis Tool (BAT) as an extension of ArcMap 10.2, which was developed for The Nature Conservancy (Software Developer: Duncan Hornby of the University of Southampton’s GeoData Institute). The tool uses point data to divide a routed river network (polylines with node to node coding) into connected networks from which a direct path distance calculation can be made. To test the long-distance downstream larvae and juvenile migration hypothesis, the size of larvae and juveniles in the ichthyoplankton samples from the Madre de Dios, Madeira and Amazon Rivers and the estuary were related to the distance of the sampling points to the Amazon River mouth by the exponential equation L = a × e b× D , where L is the size in millimeters and D is the distance in kilometers from the Amazon River mouth. An exponential model was chosen because it had biological significance and provided the best data fit. Climate Data Hydrological data were not available for most of the Andean areas we surveyed; thus, we used local precipitation data as a proxy for river level seasonality. Precipitation data were supplied by the Servicio Nacional de Meteorología e Hidrología del Peru (SENAMHI). Because there were no precipitation records for Puerto Maldonado (Madre de Dios) in November 2004, we used the average November precipitation registered in a historical series of 43 years. Note Herewith the authors declare that the study (manuscript no: SREP1622389T) was completed in accordance with the laws of Brazil and Peru and the fish larvae collection method was approved by the Brazilian Institute of the Environment - IBAMA (SISBIO- permission number 4419) in Brazil and the Dirección General de Extracción y Producción Pesquera para Consumo Humano Directo from the Ministry of Production in Peru, with in-country legal deposit protocols of the Museo de Historia Natural, Lima. Additional Information How to cite this article : Barthem, R. B. et al . Goliath catfish spawning in the far western Amazon confirmed by the distribution of mature adults, drifting larvae and migrating juveniles. Sci. Rep. 7 , 41784; doi: 10.1038/ srep41784 (2017). Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

An international team of scientists has confirmed that the dorado catfish (Brachyplatystoma rousseauxii) of the Amazon River basin holds the record for the world's longest exclusively freshwater fish migration, an epic life-cycle journey stretching nearly the entire width of the South America continent. The finding, published today in the journal Scientific Reports-Nature, is an output of WCS's Amazon Waters Initiative, sponsored by the Science for Nature and People Partnership (SNAPP) led by WCS (Wildlife Conservation Society), The Nature Conservancy (TNC) and the National Center for Ecological Analysis and Synthesis (NCEAS). The study reveals the amazing life histories of the dorado and three other "goliath" catfish species that spawn in the western headwaters of the Amazon. The journey of the dorado catfish specifically begins with adults and pre-adults traveling thousands of kilometers upriver from the Amazon River estuary to spawning areas in or near the Andes Mountains. The breeding fish never return to their nursery areas, but the newborn catfish do, migrating thousands of kilometers in the opposite direction to complete the cycle. The dorado catfish was found to have a life-cycle migration of approximately 11,600 kilometers (more than 7,200 miles). Growing up to nearly 6 feet in length, the dorado catfish is sometimes called the gilded catfish due to its silver and gold skin with no scales. The dorado and the other three closely related species are widely distributed in the Amazon River basin and are among the most important commercialized species. The paper's authors warn that looming development plans for proposed dams, mining operations, and deforestation—especially in the Amazon's headwaters where these fish spawn—could imperil these long-distance migrants and the fishing industries that rely on them. Growing up to six feet in length, the dorado catfish is sometimes called the gilded catfish due to its silver and gold skin. Credit: Michael Goulding/WCS. "This is the first time that scientific research has linked the full range of these fish species, some of which stretch from the Andes to the Amazon River estuary abutting the Atlantic Ocean," said lead author Ronaldo Barthem of Museu Paraense Emílio Goeldi of Brazil. "These findings can now inform effective management strategies for these fish, some of which are important for fishing industries in the region." "One of the biggest threats to the dorado catfish and other fish species is headwater infrastructure development in the Andes that could heavily impact the spawning grounds of the world's longest freshwater migrants," said WCS aquatic scientist Michael Goulding, a co-author on the study. Whereas the journeys of fish such as salmon and eels are well known, the migratory movements of goliath catfish have been less understood and poorly documented, although it was suspected that goliath catfish make the longest freshwater fish migrations in the world. A map of dorado's life-cycle migration range. Credit: WCS. To confirm this hypothesis, the research team mapped the long-distance movements of four goliath catfish species (Brachyplatystoma rousseauxii, B. platynemum, B. juruense, and B. vaillantii) using information on the presence of adult fish, juveniles, and larvae across the Amazon. The scientists also statistically analyzed the downstream migrations of juvenile fish and larvae from the headwaters of the Amazon's tributaries to the nurseries. The study results revealed that adult dorado catfish traveling upstream from the Amazon estuary may take as long as 1-2 years to reach their spawning grounds in or near the Andes. The scientists also confirmed that at least two other goliath catfish species (B. platynemum, and B. juruense) spawn near or in the Andean foothills, and suspect that there are other species as well. The researchers added that the newly published goliath catfish study will serve as a foundation for later research. Tagging experiments and analyses of isotopes in calcium deposits in the inner ear of fish can both be used to study the migratory pathways of goliath catfish and the seasonal timing of those movements. "Many questions remain about these incredible fish, such as why they travel so far to reproduce and do they return to place of birth to spawn," said Goulding. "Now we have a baseline that will help direct the trajectory of future research and conservation efforts."

10.1038/srep41784

Earth

Wildfire smoke may have amplified Arctic phytoplankton bloom

Mathieu Ardyna et al, Wildfire aerosol deposition likely amplified a summertime Arctic phytoplankton bloom, Communications Earth & Environment (2022). DOI: 10.1038/s43247-022-00511-9 Journal information: Communications Earth & Environment

https://dx.doi.org/10.1038/s43247-022-00511-9

https://phys.org/news/2022-09-wildfire-amplified-arctic-phytoplankton-bloom.html

Abstract Summertime wildfire activity is increasing in boreal forest and tundra ecosystems in the Northern Hemisphere. However, the impact of long range transport and deposition of wildfire aerosols on biogeochemical cycles in the Arctic Ocean is unknown. Here, we use satellite-based ocean color data, atmospheric modeling and back trajectory analysis to investigate the transport and fate of aerosols emitted from Siberian wildfires in summer 2014 and their potential impact on phytoplankton dynamics in the Arctic Ocean. We detect large phytoplankton blooms near the North Pole (up to 82°N in the eastern Eurasian Basin). Our analysis indicates that these blooms were induced by the northward plume transport and deposition of nutrient-bearing wildfire aerosols. We estimate that these highly stratified surface waters received large amounts of wildfire-derived nitrogen, which alleviated nutrient stress in the phytoplankton community and triggered an unusually large bloom event. Our findings suggest that changes in wildfire activity may strongly influence summertime productivity in the Arctic Ocean. Introduction The intensity, frequency, and duration of fires is rapidly changing globally 1 , altering the global carbon cycle and climate 1 , 2 , 3 . High latitude regions of the Northern Hemisphere (>50°N) have dense boreal forests and peatlands subject to major wildfire activity, emissions from which have approximately doubled (north of 60°) over the last decade 4 . The Arctic Oscillation-induced temperature increase appears to be critical for driving earlier snowmelt and fire activity, particularly in southeastern Siberia 5 . Aerosols and gases emitted from wildfires are predominantly carbonaceous in composition, but smoke plumes also carry significant amounts of bio-essential nutrients such as phosphorus 6 , 7 , nitrogen (N) 8 , and iron 9 , 10 . Although Russian observation stations do not routinely record information about N species, N deposition in other northern high latitude regions (e.g., North American High Arctic) is enhanced by wildfire smoke 11 , 12 . Consequently, wildfires can impact the Earth’s biosphere by altering plant productivity, biodiversity, and ultimately ecosystem carbon storage 13 , 14 . Over the past two decades, wildfires have released substantial amounts of carbon in North America (60 Tg C year −1 ) and Asia (124 Tg C year −1 ) 15 . Major Arctic wildfire source regions include Canada, Scandinavia, and Russia 4 , but also Greenland 16 . Russia accounts for approximately two-thirds of the total burnt area within these countries 4 , highlighting the importance of understanding how changing fire activity in Russia under a warming climate could impact marine biogeochemical processes 17 . Depending on the type of vegetation burned, the return interval for boreal forest fires ranges from a few decades to many centuries 18 . Return intervals are projected to decrease in the future, leading to more frequent and severe fires. More severe fires have the potential to release more aerosol to the atmosphere per unit time burned, and thus nutrients deposited within their plumes can be expected to increase. The coupling of wildfires and marine biogeochemical cycles is a recent development in our understanding of the Earth System 19 , 20 , and the impact of increasing boreal wildfires is yet to be assessed for Arctic Ocean marine primary production. Here, we suggest that boreal wildfires directly affect Arctic Ocean primary production by providing a new source of N, the macronutrient primarily limiting biological productivity in these waters 21 , 22 , and thus stimulating phytoplankton growth 23 . Results In summer 2014, ocean color satellites captured one prolonged, or several short, phytoplankton blooms (reaching 1–2 mg chlorophyll a m −3 , Fig. 1 ) up to 850 km south of the North Pole in the eastern Eurasian Basin. The pervasive cloud cover did not allow for continuous characterization of bloom dynamics, but snapshots of the ocean surface clearly revealed anomalously high chlorophyll a concentrations in the eastern Eurasian Basin, north of the Laptev Sea interior shelf. Given the spatial extent and short periods between missing retrievals, it was likely one single prolonged summer bloom. Also captured by satellite remote sensing were the exceptional sea ice conditions; although the onset of melt was two weeks later than the climatological mean, the consolidated ice pack disappeared in July and August at the most rapid rate ever recorded for this region and much further north (Fig. S1 ). Fig. 1: Large summer phytoplankton bloom near the North Pole (eastern Eurasian Basin) in summer 2014. Satellite-derived mean chlorophyll a concentration within the region of the bloom (28–155°E, 80–85°N) during the summer of 2014 ( a ). Dot color represents which satellite sensor (MODIS Aqua, Terra, or VIIRS) is used. Dot size is relative to the number of observations obtained (i.e., pixels). The blue line is the climatological daily average of surface chlorophyll a concentration over the period 2003–2019 (except 2014) with the shading envelope corresponding to the interval between the first and third quartiles. Sea-ice concentration and sea surface temperature, for the full period July 28–August 31 ( b ), and for the three time periods July 27–28, August 13 –15, and August 29–31 ( c – e , respectively). Sea-ice concentration and chlorophyll a concentration, for the same dates as b – e , shown in panels f – i . For b – i : location of the bloom is within the dotted box (28–155°E, 80–85°N) and the continental shelf (bottom depth <50 m) is shown by cross-hatching. Full size image To ascertain the potential source of new nutrients fueling this high latitude bloom we assessed several plausible mechanisms (see the supplementary results for a comprehensive evaluation of all potential mechanisms) of new N supply to the N-depleted surface ocean in summer (Fig. S2 ). Summarizing, storm events can mix N-rich deeper waters to the surface; however, winds remained weaker than the 10 m s −1 threshold generally required to induce rigorous vertical mixing 24 , 25 (Fig. S3 ). Likewise, upwelling in the Arctic can transport deep nutrient-rich waters to the surface that support intense marine production 26 , 27 , 28 . The relative importance of this mechanism depends on regional factors such as sea ice cover, shelf depth, and wind direction with respect to the shelf break 29 . While upwelling favorable southeast winds over the Laptev Sea slope were observed in July–August 2014 (Fig. S3 ) they were moderate (max ~9 m s −1 ). Furthermore, the strong topographically-controlled eastward boundary current (positive zonal component, Fig. S4 ) clearly inhibited shelf break upwelling in this region, as confirmed by temperature and salinity mooring data (Fig. S4 ). On the contrary, the temperature and salinity sections show a downwelling event along the shelf slope. Thus, we argue that neither storm-induced mixing nor shelf break upwelling provides the N that stimulated the observed bloom north of the Laptev Sea in summer 2014. Ocean dynamics may trigger changes to phytoplankton growth rates. Polyakov et al. (2017) 30 argued that in recent decades (including 2014) increased vertical mixing in winter months (peaking in April) was driven by weakened halocline stratification and enhanced sea-ice production in the eastern Eurasian Basin. However, surface winter nutrient inventories are typically exhausted within two weeks by the spring bloom at the time of sea-ice retreat (which in 2014 began in early July; Fig. S1 ). Thus, any vertically mixed winter nutrients were likely exhausted prior to the onset of this late summer bloom. Further analyzing available records, we found a massive sub-surface anticyclonic eddy capable of introducing nutrients from deeper depths into surface waters (see section 2.3 in the Supplementary Notes ; Figs. S5 and S6 ). However, its period of activity, low intensity in the upper water column, and position downstream of the current that influenced the bloom, suggest it had no effect on the bloom. The lateral advection of nutrients, particularly from river inflows, can be an additional source of new N to the open ocean. However, in situ studies clearly show that the Laptev Sea slope serves as a strong barrier preventing continental shelf waters from escaping into the deeper basin (especially in summer 2014 in the Laptev Sea 26 , 27 , 28 ). Excluding these physical mechanisms of N supply that increase phytoplankton growth promotes the hypothesis that an increased nutrient supply from the atmosphere supported the observed phytoplankton bloom. During the summer intense wildfires extend over large areas of boreal forests and peatland (Fig. S7 ), producing extensive smoke plumes that include N compounds (i.e., nitrous oxide, nitrite, and ammonia). Analysis shows Arctic wildfire activity and pollution enhancements are most pronounced in July and August 31 , 32 , when the 2014 bloom was observed. The region 115–125°E and 60–70°N, within the Sakha (Yakutia) Republic in Russia, is directly upwind of the observed bloom and MODIS recorded a burnt area (1486 kha) in July and August (Fig. 4a ) that was approximately three-fold higher than the decadal average (Fig. S8 ). Furthermore, this examined Sakha Republic region is 7.5× smaller than Canada above 60°N, but the area burnt there in 2014 was two-thirds of the total area burnt in Canada above 60°N (Fig. S8 and Supplemental Text). The 2014 fires in Sakha, Russia were exceptionally large, and thus may provide an indication of future fire behavior. Atmospheric circulation can transport continental wildfire smoke from Russia northwards to the Arctic Ocean. The Copernicus Atmosphere Monitoring Service (CAMS) reanalysis dataset (Fig. 2 ) and satellite images combined with back-trajectory analysis (Fig. 3b ) both show how wildfire smoke emitted from Russia (Fig. 3a ) traveled over the Laptev Sea and the eastern Eurasian Basin to reach the anomalous Arctic bloom of August 2014. Tiksi (128.9°E, 71.6°N; 1 m above sea level) is a remote coastal Siberian measurement station situated along this transport path. During the summer of 2014, four clear aerosol peaks (aerosol optical density; AOD) were recorded at Tiksi. All AOD peaks are dominated by fine mode aerosols indicative of a wildfire source, as opposed to coarse mode aerosol which would indicate other local natural aerosol sources (e.g., sea spray or dust). High elemental carbon (EC) concentrations measured at Tiksi (Fig. S9 ) further support the domination of AOD by fire aerosols. CAMS AOD over the bloom, and over the Laptev Sea en route to the bloom, shows three August peaks (Fig. 2 ). A time-lag between the original Tiksi AOD peak and two subsequent AOD peaks en route suggests that the smoke was transported from the continent to the bloom over a couple of days (see also: Fig. S10 ). However, the mid-July AOD peak at Tiksi is not observed over either Arctic Ocean region of interest, suggesting that smoke from fires in early July may have taken a different trajectory than later in July and August. Fig. 2: Aerosol content over ground reference site and region of interest. Locations of the different regions of interest ( a ). Comparison of the aerosol optical depth at 550 nm based on AERONET measurements (level 2.0), CAMS archive, and CESM simulations with FINN fire emissions over the Tiksi-AERONET site ( b ). The color scale indicates the fraction of fine mode aerosol contributing to AOD as retrieved from the AERONET-Solar processing. Spatially averaged AOD for the regions of interest ( c ). Full size image Fig. 3: Wildfires in Siberia, from nitrogen emissions to deposition in the eastern Eurasian Basin. Modeled FINN fire emissions of N (July 15–August 14 2014 anomaly versus the 2002–2019 mean) ( a ) with a box showing the location of panels ( c , d ). MODIS visualization of the aerosol plume and corresponding HYSPLIT back-trajectories highlighting atmospheric transport pathways (in red) to the location of the bloom ( b ). Simulated N deposition flux in 2014 ( c ) and the related deposition anomaly versus the 2011–2016 mean N deposition flux ( d ). Full size image The Community Earth System Model (CESM; see “Methods”) 33 includes N deposition from wildfires and is used here to estimate N deposition fluxes to the Artic Ocean. Simulations suggest that in late July–August 2014, N deposition fluxes were anomalously high (Fig. 3c ), reaching over double that of the preceding and following years (Fig. 3d ). Fires and plume transport directions are episodic 34 (Fig. S10 ), and results show five consecutive deposition pulses, including one strong pulse in late July, three smaller pulses in early August, and one final large pulse on 21 August (Fig. 4b ). Since Arctic Ocean waters are relatively stratified compared to other ocean basins, the estimated residence time of these atmospherically deposited aerosols could be sufficiently long to allow any associated N to accumulate, over a period of days 35 , to a sufficiently high enough concentration for phytoplankton uptake and subsequent growth. Fig. 4: Nitrogen stock in boreal peat reserves and deposition in the region of the bloom (longitude: 128–155°E; latitude: 75–85°N). Stocks and spatial distribution of nitrogen contained in boreal peat reserves ( a ) and 2014 wildfire locations (a; teal dots). Time series of the estimated magnitude of nitrogen wet and dry deposition integrated over the region (128–155°E, 75–85°N) from CESM simulations ( b ). Uncertainty in deposition fluxes is estimated from missing peat burning emissions and the fire inventory uncertainties (factor of 2–5). Values given in Table 1 are the integrated deposition fluxes as shown here. Nitrogen in peat from Hugelius et al. (2020) 46 . Wildfire locations from MODIS where burnt area was > 0 in July and August 2014. Full size image To assess whether the fires provided sufficient exogenous N to enhance production, we established a nutrient budget for the region based on current models of atmospheric N deposition (Table 1 ). Several lines of evidence suggest that the Fire INventory from NCAR (FINN) emission dataset 36 , used here in CESM simulations, likely underestimates boreal fire emissions. First, McCarty et al. (2021) 4 showed that Arctic FINN fire emissions are between a factor of 2–5 lower than all other major fire emission inventories. Next, while the CAMS reanalysis dataset reproduces AOD observations at Tiksi (Fig. 2 ), both in magnitude and timing, atmospheric modeling of aerosol transport using FINN emissions in CESM captures only the timing of the AOD peaks; the AOD magnitude is lower than both CAMS and observations by a factor of 2–5, with the best-captured peak being the first one which may not have reached the bloom. In addition, Eckhardt et al. (2015) 37 showed that global modeling, on average, underpredicts measured aerosol observations of EC taken at Tiksi by a factor of 3. Finally, CESM CO concentrations (a tracer of combustion) are lower than pan-Arctic observations by the same factor of ~2–5 (Fig. S11 ). Applying adjustments of 2–5× suggests that wildfires provided between ~12 and 30% (Table 1 : standard model ×2 and ×5) of the N required to fuel the enhanced biomass associated with the bloom. However, the atmospheric deposition model we used is missing a significant source of N in sub-polar environments from N-rich peat fires (Fig. 4a ), which are not represented in the FINN emission dataset. Assuming a ×3.5 factor is required to account for additional N emissions from peat fires (see “Methods”), the successive N deposition events we observed in the summer of 2014 could then support between ~40 and 100% of the total N content of the bloom (Fig. 4b and Table 1 ). Despite uncertainties in our knowledge of N emissions, and subsequent atmospheric deposition to aquatic ecosystems, the budget reconciles N transport when including peat fires in this region. Thus, in very N-depleted and highly stratified conditions in the Arctic Ocean, especially in the Central Arctic, significant deposition of N from peat fires will likely impact Arctic Ocean phytoplankton growth and productivity. Table 1 Nitrogen budget for wildfire aerosol fueled Eastern Eurasian Basin phytoplankton bloom. Full size table Discussion and conclusions Here, our objective is not to demonstrate that this unusual summer bloom is entirely explained, or triggered, by the wildfire aerosol N deposition, but rather that it is most likely a significant contributor to its development and/or duration. Other mechanisms have been investigated and excluded as potential significant N sources, but it is plausible that other unidentified N sources may contribute to this unusually long bloom in the highly stratified and oligotrophic eastern Eurasian Basin. Nevertheless, the potential importance of wildfire aerosol deposition as an emerging mechanism in modulating Arctic Ocean biogeochemical cycles needs to be highlighted. Under the effect of climate change, these deposition events can supply surface waters with significant levels of nutrients, as in other ocean basins 38 , 39 . It is very likely that future fire-related deposition events will become more frequent and intense, with increasingly severe wildfires in peatlands and boreal forests. However, episodes of high fire activity remain unpredictable and are thus difficult to measure; from the deposition of aerosols to their impact on biogeochemical cycles, which may have a time-lag of several days 35 . The impact of wildfires (and more specifically aerosol deposition) on sea ice thermodynamic and dynamic properties is not investigated here, although they are likely to be interconnected 40 , 41 . Soot and associated nutrients from wildfires deposited on ice can (1) lead to increased melting by reducing its albedo, (2) be transported over large distances with sea ice (even in the Central Arctic) and, (3) when the ice melts, increase nutrient and freshwater fluxes that stratify the water column and make the upper ocean more suitable for phytoplankton growth. Given the degree to which shifts in sea ice surface properties can drive large-scale changes in under-ice ecology 42 , 43 , understanding the effects of wildfire-driven changes to sea ice, and the related effect on Arctic ecosystems, ought to be explored in future work. Knowledge about atmospheric aerosol nutrient sources and deposition patterns is scarce at the pan-Arctic scale, with nearly all marine aerosol nutrient observations being collected below the Arctic circle 44 , 45 . Significant reserves of all peatland N (~80%) are currently stored within northern high latitude peatlands 46 (Fig. 3a ). Global warming is projected to result in reductions to the permafrost of peatlands by 50–100% with a warming of 2–6 °C relative to preindustrial times 46 . If increases in precipitation do not offset soil moisture losses through warming, then vegetation water stress increases and peatlands dry further and faster, making fuel more combustible and leading to an overall elevated risk in sustained major fire outbreaks. In this study, global climate model simulations, including N species contained in wildfire smoke, are used to test the hypothesis that Arctic biogeochemical cycles are sensitive to changes in boreal wildfire activity. However, the 2014 wildfires only occurred within regions of moderate peatland N stocks (Fig. 4a ). Wildfires have been detected within regions containing peatland with a higher N store (Fig. S7 ), and thus future increased boreal wildfire activity combined with a more readily mobilized N stock from thawed peatland may rapidly amplify impacts of human perturbations to Arctic ecosystems. Russian burnt area accounts for ~2/3 of all high latitude regions combined in the present day 4 , yet it is understudied when compared to North America. Different N species have different atmospheric lifetimes and thus transport potential. The N species contained in smoke plumes is thus likely to be an additional important consideration in determining the changing contribution of wildfires to aerosol burdens and marine fluxes in different Arctic regions 47 . Increased observational efforts will thus improve model simulations and provide a better understanding of the impact of boreal wildfires on Arctic Ocean productivity. Considering the N-depleted nature of Arctic Ocean surface waters, N-bearing aerosol deposits originating from wildfires will undoubtedly have repercussions on the nutrient and carbon cycles. Especially during the summer, when phytoplankton growth is severely N-limited 4 , these new N inputs could stimulate phytoplankton productivity and may partly explain the ongoing increase in annual primary production in the Arctic Ocean 48 , 49 . Phytoplankton growth is controlled by many factors, both physical and biogeochemical. Aerosol deposition, including from wildfires, is a source of new nutrients in many remote ecosystems 50 . In this study, global aerosol transport modeling suggests that Siberian wildfires supplied between 12 and 100% of the required N to support a large Arctic bloom in the summer of 2014, with the mass of N emitted from peat fires identified as a main uncertainty. Yet, nutrient aerosol addition may not always result in increased primary productivity 17 . Addressing the question of what initial conditions support aerosol-mediated phytoplankton growth should be explored further and will aid in understanding the evolution of the biogeochemical couplings between the land, ocean, and atmosphere under human-mediated climate change. These amplifying climate-driven changes, in addition to late summer/fall storms, will certainly promote secondary/fall blooms and thus contribute to the potential borealization of Arctic marine ecosystems 25 , 51 . The cascading effects of wildfire aerosols on different components of the Arctic ecosystem (land, atmosphere, sea ice, and ocean) create multiple questions that need to be assessed, quantified, and integrated into Arctic studies, in order to understand their implications on marine biogeochemistry in a changing global climate. Methods Ocean color and surface parameters Ocean color Chlorophyll a concentration, was inferred from retrievals of the three main ocean color satellite missions in orbit over the time period of this study (2014): MODIS-Aqua, MODIS-Terra, and VIIRS-SNPP; note that similar processing was used to generate the daily chlorophyll a climatology using satellite data available between 2003 and 2019. The composite products of the three missions were generated based on the recommendations of the Ocean Color - Climate Change Initiative 52 , but for a finer spatial resolution corresponding to a pixel edge of ~1 km. The remote sensing reflectance R rs and chlorophyll a were obtained after atmospheric correction performed through the SeaDAS software 53 and the combination of the spectral ratio and color index algorithms for chlorophyll a 54 , 55 . Note that the latter parameter was also estimated through regionally-tuned algorithms showing similar results (see Fig. S12 ). Image reprojection, binning, and aggregation were performed using the Sentinel Application Platform (SNAP) software, developed by Brockmann Consult . Thanks to the high number of data acquisitions in this period (up to 8 for a given satellite per day), bad quality pixels were filtered out, and basic statistics were performed to provide quality controlled daily values. First, for each individual image, pixels were filtered out if one of the following quality flags provided by the SeaDAS algorithm is true: ATMWARN, ATMFAIL, HIGLINT, HILT, HISATZEN, STRAYLIGHT, CLDICE, CHLFAIL, and MAXAERITER. Due to particularly complex environmental conditions in Arctic seas for optical remote sensing, another pixel filtering procedure was performed when at least one the spectral bands exhibited R rs values smaller than 0.0003 sr −1 or if the ratio between R rs at 412 and 443 nm was >2.5. Note that pixels with aerosol optical thickness >0.2 were also filtered out to avoid misinterpretation of pixel information in the presence of aerosol plumes from the wildfires. The second step consists of calculating the daily median and standard deviation values for each pixel location of a given satellite mission. Once the daily product is obtained, the composite image was constructed by averaging the pixel values of the three distinct satellite missions. Sea surface temperature The SST was obtained from the MODIS-Aqua level-2 NASA products derived from long-wave (11–12 µm) thermal radiation. Daily SST images were generated following the similar approach to that applied for the ocean color products but using the following flags to filter out bad quality pixels: BTDIF, SSTRANGE, BTNONUNIF, CLOUD. Atmospheric transport modeling CESM version 1.5 33 was used with the interactive chemistry version of the Community Atmospheric Model (CAM-chem) as the atmospheric component 56 following the set-up described in Bernstein et al. (2021) 57 . Aerosols in CAM-chem are represented by four modes (Aitken, accumulation, coarse, and a primary carbon) 58 . All simulations were performed on a horizontal resolution of 0.9 × 1.25° and 56 vertical levels with offline meteorology nudged to GEOS5 59 meteorological analysis. Dust and sea salt are prognostically calculated following the MAM4 default configuration 58 . Anthropogenic emissions are taken from HTAP-2 60 . Daily fire emissions (from wildfires, agricultural fires, and prescribed burning) are taken from the Fire INventory from NCAR (FINN) dataset 36 version 1.6 and prescribed vertically following AeroCom recommendations 61 up to a maximum plume height of 5 km. The CAMS reanalysis dataset incorporates satellite-derived AOD reanalysis using the ECMWF Integrated Forecasting System 62 . In this way, CAMS reduces bias with observations, and in the context of this study, provides an AOD reference with which to compare CESM results. The Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) 63 , developed by NOAA’s Air Resources Laboratory, was used to simulate back trajectories from the location of the bloom (82°N, 130°E, August 12, 2014). Current fire emission inventories combine data from different sources, on fire fuel loading, biomass consumed, and species-specific emission factors to estimate emissions over the area burned. This results in substantial uncertainties between commonly used datasets which differ by a factor of 4 globally and by factors of 3–15 regionally 64 , 65 . A second uncertainty is that when fire emissions are prescribed in global model simulations, there is a well-known low bias compared to aerosol optical depth measurements 2 , 66 , leading to regional adjustments of fire emissions commonly between a factor of 2–3. The fire emission database FINN used here does not include peatland fire emissions, and thus model simulations do not transport any N from peat fires, despite fires occurring within regions containing peatlands (Fig. 4a ). For our study region, FINN currently contains the lowest estimates of fire carbon emissions between all inventories: the highest estimates being a factor of ~5 larger 14 . In addition to peat fires being a large missing carbon source they are also a significant missing source of nitrogen owing to some species (e.g., HCN and NH 3 ) being emitted from smoldering peat fires at a factor of ~10 higher, per unit biomass consumed, than for flaming savanna fires 8 or a factor of 3.5 higher than boreal fires 8 . Therefore, to estimate the missing source contribution of peat fires to the N deposition flux, an emission ratio of 3.5 (peat fires:boreal fires) is used as a bias correction, based on emission factor differences between boreal forests and peatlands for some N species 8 and the missing PM 2.5 contribution from tropical peatland fires when using the FINN dataset 67 . Assuming linearity in transport of peat fire emissions with those from forest fires, we applied this factor 3.5 bias correction factor to the ‘inclusion of peat biomass’ nitrogen deposition flux in Table 1 . Burned area was estimated using the collection 6 MODIS Global Burned Area Product MCD64CMQ (climate model grid) 68 . Data availability The NABOS oceanographic dataset in the Laptev Sea was obtained from . AERONET data for the Tiksi site (PI: Brent Holben) were downloaded from . Ocean color data from MODIS-Aqua, MODIS-Terra, and VIIRS-SNPP and sea surface temperature from MODIS-Aqua were downloaded from and . The sea ice concentration and melt onset dates were obtained from the NSIDC ( ). The daily CESM aerosol optical depth and nitrogen deposition data for July and August 2014 used in this study can be downloaded from . We use the NASA collection 6 MODIS Global Burned Area Product MCD64CMQ to estimate burnt area in each year, which is available to download from . Nitrogen content in peatlands was taken from Hugelius et al. (2020) 46 .

Smoke from a Siberian wildfire may have transported enough nitrogen to parts of the Arctic Ocean to amplify a phytoplankton bloom, according to new research from North Carolina State University and the International Research Laboratory Takuvik (CNRS/Laval University) in Canada. The work, which appears in Communications Earth & Environment, sheds light on some potential ecological effects from Northern Hemisphere wildfires, particularly as these fires become larger, longer and more intense. In the summer of 2014, satellite imagery detected a larger than normal algal bloom in the Laptev Sea, located in the Arctic Ocean approximately 850 kilometers (528 miles) south of the North Pole. "For a bloom that large to occur, the area would need a substantial influx of new nitrogen supply, as the Arctic Ocean is nitrogen-depleted," says Douglas Hamilton, assistant professor of marine, earth and atmospheric sciences at NC State and co-first author of a paper describing the work. Hamilton was formerly a research associate at Cornell University, where the research was conducted. "So we needed to figure out where that nitrogen was coming from." First, the researchers looked at the "usual suspects" for nitrogen input, such as sea ice melt, river discharge and ocean upwelling, but didn't find anything that would account for the amount of nitrogen necessary for the bloom to occur. But during that same time period, exceptionally large wildfires in Siberia, Russia, located directly upwind of the bloom, had burned approximately 1.5 million hectares (or about 3.5 million acres) of land. So the researchers turned their attention to atmospheric composition. They used the Community Earth System Model (CESM), a computer model that can simulate what happens to emissions from natural and human sources as they enter and leave the atmosphere. The model was fed information about wind, temperature and atmospheric composition—including the composition of wildfire smoke—from the time period in question. The model simulations showed that during late July and August of 2014—when the bloom was detected and the Siberian wildfire was burning—nitrogen deposition from the atmosphere was almost double that of the preceding and following years. "The wildfires were located in rapidly warming boreal regions, which have a lot of peat in the thawing permafrost," Hamilton says. "Peat is very nitrogen rich and the smoke from the burning peat was hypothesized as the most likely source of much of the additional nitrogen." "We've known that fires can impact phytoplankton blooms, though it is unexpected to see something like this in the Arctic Ocean," says Mathieu Ardyna, co-first author and CNRS researcher at the International Research Laboratory Takuvik (CNRS/Laval University). "Most likely, since fires are locality-specific and difficult to predict, blooms like this won't be the norm—but when these wildfires do occur the nutrients they bring in could lead to sustained or multiple blooms." The researchers' next steps could include reviewing the historical satellite record and further characterizing the chemical composition of the particles within the smoke to get a clearer picture of how wildfires like these might impact different ecosystems. "A one-off bloom like this won't change ecosystem structure, but both Siberia and high arctic Canada are getting more wildfires," Hamilton says. "So it may be interesting to explore potential downstream effects if fire activity and nutrient supply remain high."

10.1038/s43247-022-00511-9

Chemistry

Fully identified—the pathway of protons

Jifu Duan et al. Crystallographic and spectroscopic assignment of the proton transfer pathway in [FeFe]-hydrogenases, Nature Communications (2018). DOI: 10.1038/s41467-018-07140-x Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-018-07140-x

https://phys.org/news/2018-11-fully-identifiedthe-pathway-protons.html

Abstract The unmatched catalytic turnover rates of [FeFe]-hydrogenases require an exceptionally efficient proton-transfer (PT) pathway to shuttle protons as substrates or products between bulk water and catalytic center. For clostridial [FeFe]-hydrogenase CpI such a pathway has been proposed and analyzed, but mainly on a theoretical basis. Here, eleven enzyme variants of two different [FeFe]-hydrogenases (CpI and HydA1) with substitutions in the presumptive PT-pathway are examined kinetically, spectroscopically, and crystallographically to provide solid experimental proof for its role in hydrogen-turnover. Targeting key residues of the PT-pathway by site directed mutagenesis significantly alters the pH-activity profile of these variants and in presence of H 2 their cofactor is trapped in an intermediate state indicative of precluded proton-transfer. Furthermore, crystal structures coherently explain the individual levels of residual activity, demonstrating e.g. how trapped H 2 O molecules rescue the interrupted PT-pathway. These features provide conclusive evidence that the targeted positions are indeed vital for catalytic proton-transfer. Introduction [FeFe]-hydrogenases represent one of natures’ most effective classes of redox enzymes catalyzing the reversible reduction of protons to dihydrogen (H 2 ) at turnover frequencies of up to 9000 s -1 1 , 2 , 3 . Most [FeFe]-hydrogenases favor proton reduction while [NiFe]-hydrogenases are usually more biased toward H 2 oxidation 4 . With their low catalytic over-potential 5 , [FeFe]-hydrogenases represent excellent models for a regenerative and likewise economically feasible H 2 production. Their active center (“H-cluster”) can be structured into a standard [4Fe-4S]-cluster ([4Fe] H ) and a diiron site ([2Fe] H ). The latter is uniquely coordinated by three carbonmonoxide (CO) and two cyanide (CN − ) ligands. They stabilize the cofactor in its protein environment 6 and fine-tune its redox features 7 . An azadithiolate ligand (adt) further bridges the proximal (Fe p ) and the distal (Fe d ) iron center, which are differentiated according to their location relative to the [4Fe] H -cluster. To achieve the extraordinary high turnover frequencies of [FeFe]-hydrogenases 8 , it can be implied that proton transfer (PT) is facilitated by distinct and optimized pathways. PT pathways span large distances through protein scaffolds, e.g. to enable proton-coupled electron transfer or proton translocation 9 , 10 , 11 , 12 . They usually comprise of a succession of protonatable or polar residues and protein-bound water molecules aligned at hydrogen-bonding distance 13 , 14 . Based on the crystal structures of [FeFe]-hydrogenases CpΙ from Clostridium pasteurianum and DdH from Desulfovibrio desulfuricans , several putative PT pathways have been discussed 15 , 16 , 17 , 18 . Theoretical studies suggest that the most probable PT pathway comprises of strictly conserved residues E282, S319, E279, and C299 (from surface to H-cluster, numbering corresponds to CpI), including two protein-bound water molecules (Wat826, Wat1120 4XDC 19 , chain B) located between E279 and C299 15 , 16 , 20 . This pathway ends with C299 located in hydrogen-bonding distance to the amine head-group of the adt-ligand. Its identity and importance as a proton relay was verified in comparative studies on cofactor variants of HydA1 from Chlamydomonas reinhardtii 21 . Cornish and co-workers could show that amino-acid substitutions along the putative PT pathway dramatically decreased the catalytic activities. In particular, their study indicated a participation of surface-exposed residue E282 in catalytic PT 15 . In a second study they suggested a regulative function for positions R286 and S320 in the PT of CpI 22 . Furthermore, Morra and co-workers described that the pH optimum of variant C298D of [FeFe]-hydrogenase CaI from Clostridium acetobutylicum (corresponding to C299D in CpI) is shifted from pH 8 to pH 7, indicative for the involvement of C298 in PT 23 , 24 . Although several studies were conducted, immediate experimental evidence for the relevance of residues in the putative PT pathway is missing, leaving an essential aspect of enzymatic performance opaque. In this study, site-directed mutagenesis (SDM) is used to investigate the PT pathway of two [FeFe]-hydrogenases, namely CpI and HydA1, which represent the largest (M3) and smallest (M1) type of monomeric [FeFe]-hydrogenases, respectively 25 . Most of the 22 SDM variants show strongly affected H 2 release activities and pH optima. For 11 of these variants the crystal structure is solved, which facilitates a correlation of individual structural features (i.e. hydrogen-bonding distances) and catalytic performance. This provides insight into the minutiae of PT on the molecular level. Catalytically hampered SDM variants are analyzed by in situ attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy. When flushed with H 2 , these enzymes are found to adopt a key intermediate of hydrogen turnover, the recently described H hyd state 26 , 27 , 28 , 29 , 30 . H hyd accumulation under H 2 clearly correlates with the diminished PT efficiency of the enzyme 29 . Herein we provide complementary kinetic, structural, and spectroscopic data, which allow to verify the PT pathway discussed above as the key route of catalytic PT in [FeFe]-hydrogenases. Results H 2 release assays and pH-dependent enzyme activities For both CpI and HydA1, 11 SDM variants were generated to target residues along the putative PT pathway applying conservative and non-conservative exchanges (Fig. 1 ). Conservative exchanges (e.g. E → D) maintain the functional group of the targeted position, but due to other structural differences in the substitute residue, will affect the precise spatial placement and configuration of the functional group. In a highly ordered system such as the well-distanced H-bond chain of an evolutionarily optimized PT pathway, this should at least affect the efficiency of the functional aspect. Non-conservative exchanges (e.g. E → Q/A) delete the functional group entirely and therefore prohibit these substitute residues from rescuing the targeted function. For wild-type (wt) enzyme, H 2 release activities of about 860 (HydA1) and 2600 (CpI) µmol H 2 per mg per min were measured and defined as 100% activity 29 , 31 , 32 , 33 . Corresponding substitutions in CpI and HydA1 had similar effects on H 2 -release activity, except for variants E282Q CpI (8.2%) and E144Q HydA1 (0.4%). The strongest impacts were achieved when replacing E279 CpI for A or Q, or C299 CpI for A or S, resulting in activities <1%. This is well in line with the overall trend of an increasing impact of substitutions along the PT pathway in the direction from surface to H-cluster with C299D being the only exception (Fig. 1b ). In general, conservative amino-acid exchanges had less dramatic effects than non-conservative substitutions. However, in some cases a non-conservative exchange to alanine retained a surprisingly large fraction of activity. Variants E282D CpI , C299D CpI , R286A CpI , and E282A CpI were only mildly affected and showed residual activities between 30 and 90%. For the same PT pathway position, different substitutions can have a dramatically different impact as exemplified by position E144 HydA1 /E282 CpI . Here the exchange to glutamine diminished enzyme activity to only 0.4–8%, while the non-conservative exchange to alanine and the conservative exchange to aspartic acid retained between 46 and 81% activity. Further, there is a general trend for substitutions of the more surface-exposed PTP positions (R286, E282, and even S319) in CpI to have slightly lower impact on enzymatic activity as compared to the corresponding HydA1 variants. It might suggest for CpI a more open or flexible access to the PT pathway, which can tolerate variations slightly better than HydA1. This is especially obvious in the light of the 20-fold difference in the relative activity of E282Q CpI , as compared to its HydA1 counterpart E144Q. Fig. 1 Putative PT pathway in CpI and H 2 evolution activities of SDM variants. a The PT pathway of CpI (PDB: 4XDC 19 ) is presented as stick structure with individually colored residues while the [4Fe-4S]-cluster and water molecules are shown as spheres. The substitutions applied in this study for individual positions are shown in parentheses below the respective position labels. b H 2 -production activity of SDM variants targeting the putative PT pathway in CpI and HydA1. H 2 production activities of PT pathway variants determined at pH 6.8 are presented in % relative to the respective wild-type activity. Activity bars for different variants of the same position exhibit corresponding basic colors but different shades. Relative activities higher than 5% are shown in the upper part of the discontinuous scale. Wild-type CpI and HydA1 exhibit activities of 2576 ± 107 and 862 ± 46.5 µmol H 2 per mg per min, respectively. The bars represent mean values from at least three independent measurements, including standard deviations. Details on the in vitro assay are presented in Supplementary Fig. 1 Full size image To link the role of individual residues in the putative PT pathway to substrate/product (H + ) transfer, we probed the pH-activity profiles of variants with a high enough level of residual activity in terms of H 2 -release and H 2 -oxidation activity. As shown in Fig. 2a , native HydA1 is most active from pH 7 to pH 8 while wild-type CpI (Fig. 2b ) clearly reaches its highest activity at about pH 8. Variant CpI-Y572A served as a negative control, with an amino-acid substitution outside of the putative PT pathway and features the pH-dependency profile of wild-type enzyme (see Supplementary Fig. 2a ). In contrast to this behavior, the pH optima of all variants targeting the putative PT pathway were shifted to lower pH values, indicating that the limited PT efficiency can at least be partially rescued by an increased proton concentration. Fig. 2 pH-activity profiles of wt-CpI and wt-HydA1 compared to selected SDM variants. a HydA1 and E144A; b wt-CpI and variants of position E282; c wt-CpI and variants of position E279; d wt-CpI and selected variants of positions R286, S319, and C299. H 2 production activities were determined for different buffers, covering a pH gradient between 5 and 9 in steps of 0.5 pH units. Relative values correspond to % of maximum activity obtained throughout the entire pH gradient. Black plots indicate the relative pH-dependent activities of CpI and HydA1 wild-type enzymes. All values are mean values ± standard deviations from at least three independent measurements Full size image Just as observed for catalytic activity, different variants of the same position in the putative PT pathway can cause significantly deviating shifts in pH optimum. In case of E282 CpI, a substitution for aspartic acid shifted the optimum only slightly to pH 7.5 while an exchange to alanine or glutamine rendered the enzyme most active at pH 7 and 6, respectively (Fig. 2b ). In most cases, the extent of the shift correlates to an overall loss in H 2 release activity, however this has not been observed in all cases. The conservative variant E279D retains 10–30% of H 2 release activity but exhibits a significantly stronger down-shift in pH optimum than the largely inactive variant E279Q (Fig. 2c ). As shown in Fig. 2 and Supplementary Fig. 2 , no significant differences between CpI and HydA1 and their respective SDM variants were observed, underlining the universal impact of the putative PT pathway. To probe the relevance of the putative PT pathway for catalytic proton release, we also investigated the pH-dependency of H 2 oxidation. It is known from electrochemistry experiments with different [FeFe]-hydrogenases that the enzyme exhibits an increasing H 2 -oxidation rate with increasing buffer pH 6 . Accordingly, the pH range used in this assay was extended to pH 10. As shown in Supplementary Fig. 3 , H 2 -uptake activity of native HydA1 generally enhanced with decreasing H 3 O + concentrations, exhibiting a nearly fivefold rate-increase per pH unit between pH 6 and 8. After an intermediate drop between pH 8 and 9 the H 2 -oxidation activity further increased to nearly 20.000 µmol H 2 per mg per min between pH 9 and 10. The absolute activities of SDM variants were significantly diminished compared to wild-type HydA1, reaching a pH optimum of at best 6% (R148A). Up to pH 9, variant R148A HydA1 showed a wild-type-like trend for the pH-activity profile of relative H 2 -oxidation activity, while instead of a second increase between pH 9 and 10, the activity strongly declined, suggesting that this second increase is connected with the titration of the guanidine base of R148. The pH-activity profile of E144A HydA1 was quite similar to wild-type HydA1, despite a flattening out of the local maximum at around pH 8. The latter suggests that this local activity maximum is depending on the presence of the surface-exposed Glu residue. S189A and C169D only showed single activity peaks at pH 8 or 9, respectively with very low H 2 -uptake rates of 1–2%, compared to wild type. Infrared spectroscopy We recently showed that decreased proton release efficiency in the presence of H 2 leads to an accumulation of the hydride state, H hyd 29 . This H-cluster intermediate carries a terminal hydride and represents the first catalytic intermediate after heterolytic cleavage of H 2 26 , 27 , 28 , 29 . In previous studies we exploited this behavior to demonstrate that positions C169 HydA1 /C299 CpI and E141 HydA1 /E279 CpI contribute to catalytic PT 29 . Non-conservative substitutions to alanine that interrupted the PT pathway led to an enrichment of H hyd instead of adopting a mix of reduced redox species (see below). Here this approach was applied to link PT activity of residues in the putative PT pathway to the steady-state equilibrium of redox species in the presence of H 2 . Employing ATR-FTIR spectroscopy, we probed CpI and HydA1 wild-type protein and the respective SDM variants at pH 8 under either N 2 or H 2 (Fig. 3 ). When flushed with N 2 , auto-oxidized HydA1 wild type and SDM variants uniformly exhibit peaks around 1964, 1939, and 1802 cm −1 characteristic for the H ox state 34 . HydA1 and CpI wild-type enzymes adopt a mixture of reduced states when exposed to H 2 , predominantly H red /H sred and H red ´ 34 , 35 . In contrast, most variants populate the H hyd state that can be enriched in wild-type enzyme only at pH 4 (see Supplementary Fig. 4 ). Similar results were achieved for the corresponding variants of CpI as shown in Fig. 3c, d . HydA1 variants C169D, S189A, E144D, and R148A, as well as corresponding CpI variants show wild-type-like spectra by adopting a mixture of reduced states instead of H hyd (Fig. 3b ). With the exception of S189A HydA1 /S319A CpI , these variants retain a higher level of H 2 -release activity (Fig. 1b ), which explains why in these cases H hyd cannot be trapped at pH 8; just as wild-type enzyme, these variants populate H hyd when titrated to significantly lower pH values (Supplementary Fig. 4 ). Interestingly, variant C169D and the corresponding variant of CpI (C299D) seem to be incapable of accumulating H hyd under any of the conditions applied here. Fig. 3 Infrared spectra of wild types and PT pathway variants. The frequency regime of the H-clusters’ CO ligands is shown (2000–1785 cm −1 ). For ATR-FTIR spectroscopy, the buffer was set to pH 8 and the rehydrated samples were purged with 100% H 2 for 5 min. a , c Auto-oxidation in absence of H 2 (i.e. purged with N 2 ) was exploited to likewise enrich for all examined proteins the oxidized resting state, H ox (gray bands). Some CpI variants tend to accumulate H ox H in parallel with H ox (e.g. C299A bands at frequencies 1975/1953/1809) 34 . b , d When shifting from N 2 to H 2 the spectrum of wild-type protein changes to different fractions of reduced species including H red (cyan) and H sred (red) as well as H red ´ (magenta). Most PT pathway variants accumulate H hyd (blue) instead or in addition to a mix of reduced states. For precise state-specific vibrational signals of CpI and HydA1 see Supplementary Table 7 Full size image Protein crystallography For eight CpI variants (holoenzyme) and three HydA1 variants (apo-enzyme) protein crystals were obtained. Their structures were solved and refined to resolutions of 1.45–2.76 Å, allowing us to gain insight into the structural consequences of SDM. Corresponding to earlier crystal structure data, for all CpI variants a space group of P1 2 1 1 was observed with two copies in the asymmetric unit cell (chain A and B) 19 , 25 . Crystallization of HydA1 exclusively succeeded for apo-protein, which carries the [4Fe] H -cluster but lacks the [2Fe] H -site (Supplementary Fig. 5 ). However, comparisons of crystal structures of CpI apo- and holo-protein with apo-HydA1 demonstrated that a lack of cofactor does not affect the configuration of the putative PT pathway 19 , 36 . Variants of HydA1 apo-protein were crystallized in space group P3 2 2 1, with a single chain in the asymmetric unit. Overall, SDM did not induce unspecific structural changes as deduced from superposition with wild-type enzyme and corresponding root-mean square deviations of Cα atoms (Supplementary Table 2 ) and the the H-cluster was largely present for all CpI variants (Supplementary Table 5 ). In the following, local structural changes in the putative PT pathway will be described. We start from the surface-exposed residues R286 CpI and E282 CpI , will continue addressing the median positions S319 CpI and E279 CpI , and end with C299 CpI in the vicinity of the H-cluster. According to their close inter-residue distance of 2.8 Å, R286 CpI may function as a salt bridge partner of deprotonated E282 CpI (Fig. 4 , right and Supplementary Figs. 6 , 7 ) and thus could be involved in the PT mechanism. R286 CpI is further part of an extensive hydrogen-bonding network, which includes the carboxyl group of E282 CpI , histidine residues H565 CpI and H569 CpI , and surface-bound H 2 O molecules. In case of variant R286A significantly fewer water molecules are observed in the same region (Supplementary Fig. 8 ). We therefore assume that the loss of the guanidine group in R286A destabilizes the network of proton-accepting/donating surface-bound H 2 O molecules near the entrance to the PT pathway. However, as the essential chain of PT pathway residues is not directly affected, the influence of this exchange on catalytic activity is comparatively weak (Fig. 1b ). Fig. 4 Structural features of SDM variants targeting the putative PT pathway in CpI. Structures of nine SDM variants are superimposed as cartoon-loop models together with the 4XDC 19 wild-type structure. No unspecific differences are observed. For each variant an enlargement of its electron density map in the putative PT pathway and the corresponding sticks model has been aligned with the structure of wild-type protein (H 2 O molecules and carbon atoms colored in black). For the CpI proteins local structural differences near the site of mutagenesis are depicted (for the HydA1 variants see Supplementary Fig. 5 ). Simulated annealing omitting maps ( F o − F c ) were contoured at 3 σ except for E282D, which was contoured at 1.9 σ due to its comparatively low resolution. Chain B provides a more flexible N terminus but a more rigid H-domain where both the PT pathway and the active center are located 19 . Therefore, all the structural information of CpI was derived from chain B if not stated otherwise Full size image In E282A, two water molecules occupy the space of the missing carboxylic acid group (chain B, Wat717 and Wat974) (Fig. 4 , top middle and Supplementary Fig. 13a ). The distances between these water molecules and the hydroxyl group of S319 are 2.5 and 3.1 Å (see Supplementary Fig. 7 ), respectively. They are close enough to rescue the PT activity between surface water and S319 CpI further downstream the PT pathway. In the crystal structure of apo-E144A HydA1 , one water molecule remains in a very similar position (Supplementary Fig. 5 ). It is therefore not surprising that both variants exhibit 50% wild-type activity (Fig. 1 ). In variant E282Q CpI , the glutamine residue is potentially stabilized by two hydrogen bonds (H-bonds) unrelated to the putative PT pathway thus, blocking PT and rendering the enzyme largely inactive (Supplementary Fig. 7 ). Additionally, the two different conformations of Q282 CpI in chain A and B (Supplementary Fig. 7 ) indicate structural flexibility at the entrance of the PT pathway, which may support the residual activity measured for this variant. In variant S319A CpI , the carboxyl group of E282 CpI is slightly shifted outward, probably due to the lack of the hydroxyl group at position 319 CpI . In wild-type enzyme S319 CpI acts as H-bond donor to E282 CpI and drags its carboxy group further into the PT pathway (Fig. 4 ). In S319A CpI an unprecedented water molecule (Wat735) is located between E282 and A319, and the nearby water molecule Wat873 of wild-type CpI (chain B) is missing here, suggesting a translocation enabled by the unoccupied space of the missing hydroxyl group of position 319 and the slight outward shift of E282 (see Supplementary Fig. 12c ). Although the distance to the carboxyl group of E279 amounts to 5.8 Å, the presence of Wat735 may explain the dramatically diminished yet still detectable H 2 release activity of variant S319A CpI (6%). The conformational differences between E279Q CpI and wild-type CpI are insignificant, except that the carboxamide group of Q279 CpI is slightly twisted relative to the original carboxyl group (Fig. 4 ). In the corresponding HydA1-variant E141Q, the glutamine residue precisely adopts the conformation of the glutamic acid residue in HydA1 wild type (Supplementary Fig. 5 ). For the corresponding alanine variant of both, HydA1 and CpI, shifts of the conserved water molecule in the putative PT pathway can be observed. While glutamine occupies the entire space of the native carboxyl group and thus clearly interrupts the putative PT pathway, an exchange to alanine may support additional H 2 O molecules to bridge the gap. However, in contrast to corresponding substitutions at the surface-exposed glutamate E282 CpI , in neither E141A HydA1 (Supplementary Fig. 5 ) nor E279A CpI 25 (Fig. 4 ) additional H 2 O molecules are observed. This explains why exchanges of the median glutamate to either alanine or glutamine lead to residual activities of below 1% (Fig. 1 ). In case of C299D CpI , the carboxyl group of the side chain establishes H-bonds (2.7 Å) with the amine group of [2Fe] H and the “conserved” water molecule Wat708 (corresponding to Wat826 in chain B of 4XDC 19 ) (Fig. 4 ). Unsurprisingly, protein samples of C169D HydA1 and C299D CpI both retain a rather high level of H 2 release activity (30–80%, Fig. 1 ). This is not the case when substituting cysteine for alanine, which for both enzymes leads to a complete loss of catalytic activity. Interestingly, the crystal structure of C299A CpI reveals that the removal of the thiol group causes an additional H 2 O molecule to occupy the vacant space. While the additional H 2 O is expected to rescue PT activity, it obviously does not restore enzymatic activity. A summary of the characteristic experimental data gained for each enzyme variant is presented in Supplementary Table 1 . Discussion Fast proton transfer between bulk solvent and the H-cluster is a precondition for the high catalytic turnover rates of [FeFe]-hydrogenases. Mainly based on theoretical studies, four potential PT pathways have been discussed in literature 15 , 16 , 17 , 18 , 20 , 37 , 38 . The putative trajectory examined in this study has been favored as key catalytic PT pathway. However, unique experimental evidence for this assumption was lacking. A characterization of PT does not only help to understand the catalytic mechanism but will further contribute valuable parameters for the design of bio-inspired inorganic catalysts 20 . For artificial hydrogen catalysts, proton supply is often rate-limiting, e.g. due to the lack of a defined proton relay, thus making it necessary to add strong acids and adjust a very low solvent pH 39 . In this study, corresponding sets of SDM variants of CpI and HydA1 were generated to uncover the individual contribution of residues to PT in the putative PT pathway of [FeFe]-hydrogenases 25 , 40 . The complementary data gained from pH-activity profiles, crystal structure analysis, and ATR-FTIR spectroscopy unanimously show that the residues of this pathway are key for the supply and release of protons during catalytic turnover. The crystal structures of wild-type CpI and its SDM variants reflect the H ox state 19 , which is commonly accepted to be the resting state of hydrogen turnover. To obtain a general survey on the putative hydrogen-bonding pattern in H ox , the distances between consecutive PT partners in wild-type CpI and the nine SDM variants were derived from the corresponding structures (Fig. 5 and Supplementary Fig. 7 ). On the basis of CpI crystal structure 4XDC 19 , PROPKA empirically predicted p Ka values of 8.6 and 3.5 for E279 and E282, respectively (pH 8) 41 . The relatively high p Ka of E279 CpI reflects its hydrophobic environment compared to the surface-exposed E282 CpI and suggests E279 CpI to be largely protonated while E282 CpI probably resides in the deprotonated state. Starting at the H-cluster, possible hydrogen-bonding interactions can be assumed for adt-NH/C299 CpI and C299 CpI /Wat826, both of which exhibit comparatively large distances of 3.5 and 3.2 Å 42 . Molecular dynamics simulations proposed a low H-bond occupancy between adt-NH and the thiol group of C299 CpI and a stronger contact between E282 CpI and R286 CpI to be preconditions for H 2 oxidation starting from H ox 20 . This proposal cannot be confirmed by our crystallographic data, which show the C299 thiol group in an intermediate orientation. However, we cannot rule out the possibility that such differences between theoretical data and structural information are influenced by the non-physiological crystallization conditions or low temperature crystal storage in liquid nitrogen, which might favor certain configurations in the PT pathway. The short distance of 2.5 Å between Wat826 and E279 CpI suggests a strong H-bond, in contrast to the adjoined pair E279 CpI /S319 CpI , for which a distance of 3.6 Å again indicates a fairly weak interaction. The putative hydrogen bond between S319/E282 can be estimated to be of moderate strength (2.9 Å). E282 CpI potentially interacts with at least one surface-bound H 2 O molecule (Wat990). Finally, arginine R286 CpI (calculated p Ka: 12.4) may serve as a putative salt bridge or hydrogen-bonding partner at close distance to E282 CpI (2.8 Å). The combination of calculated p K a values and H-bond distances described in this study suggests for H ox the H-bond pattern depicted in Fig. 5 . Fig. 5 Hydrogen-bonding pattern in the catalytic PT pathway of the H ox state. H-bond pattern of H ox according to p Ka values of the residues calculated for the structure of wild-type CpI (4XDC 19 ) via PROPKA 41 (see red numbers in parentheses). The p Ka of the adt-ligand in H ox is derived from a previous study 58 . The blue numbers indicate the distances between neighboring positions of the PT pathway. The arrow labeled “SB” indicates a presumptive salt bridge contact between R286 and deprotonated E282 Full size image Electron densities in the omit maps of crystal structures for wild-type CpI and most of the SDM variants examined here are unambiguously oriented (Fig. 6 ) and show low displacement factors (Supplementary Table 8 ). This is in contrast to quantum mechanics/molecular mechanics simulations by Long et al. 38 who proposed a conformational bi-stability for the two glutamic acid residues E282 and E279 in the catalytic PT pathway. Our data rather favor a model in accordance with the Grotthuss mechanism, based on simultaneous deprotonation/protonation events according to a bucket line between strictly orientated residues and water molecules 43 . However, a bi-stability of the glutamate residues under turnover conditions might still be a valid interpretation, as it could also be regarded as switching between two different hydrogen-bonding patterns. In both cases, the direction of PT would be solely determined by the redox state of the [2Fe] H -cluster (see Supplementary Fig. 9 and Supplementary Discussion ). Fig. 6 Influence of SDM on the proton transfer mechanism during H 2 -uptake in CpI. Effects of SDM on the proposed proton transfer mechanism for selected CpI variants: C299A ( a ), E279A ( b ), E282A ( c ), and E282Q ( d ). H 2 -binding induces a shift in the H-bond pattern (from mode 1 to mode 2) and initiates the catalytic mechanism during which the H-bond pattern repeatedly shifts between modes 1 (blue) and 2 (green) while promoting a stepwise proton release via the PT pathway (for details see Supplementary Fig. 9+14 ). The p Ka values of adt-ligand at different redox states are derived from previous studies 49 , 58 . The mutated residues and hydrogen atoms from substrate were highlighted as red. The green double arrow indicates a putative salt bridge contact. In c (E282A), the pink shading area indicates a slowed-down but still functioning proton transfer. Protons presented in blue close to the [4Fe] H sub-cluster originate from the recently described regulatory PT pathway 34 , 59 , 60 , which is independent of substrate/product transfer Full size image The FTIR data presented here were recorded under steady-state conditions applying changes in gas atmosphere and pH. As protons are reactants in hydrogen turnover (H 2 ⇌ 2H + + 2e − ), it is plausible to assume that mutagenesis in the catalytic PT pathway influences the equilibrium of H-cluster species. This was exemplified above for those variants that accumulate H hyd . A similar effect on the dynamic equilibrium of catalytic states has been observed for [NiFe]-hydrogenases 44 , 45 . The substitution of E17 for glutamine in the putative PT pathway of soluble hydrogenase I (SHI) of Pyrococcus furiosus disabled proton-coupled electron transfer (PCET) between two catalytic states (Ni a -C and Ni a -S) 44 , which was indicative by an accumulation of intermediates Ni a -I 1 and Ni a -I 2 . For [FeFe]-hydrogenases, the hydride state has been demonstrated to accumulate under H 2 -oxidation conditions if proton efflux is severely restricted. This may be the result of an oversaturation of the native PT pathway due to the enhanced proton pressure, i.e. low bulk pH 29 , 34 . It can further be the result of [2Fe] H derivatization or eliminating protonatable side chains in the catalytic PT pathway 29 . The hydride state has been shown to exhibit an uncharged azadithiolate ligand (adt-NH) 37 , 46 that implies a transient intermediate with a protonated ammonium cation, e.g. as proposed by Reijerse et al. (H hyd H + ) 27 . Accordingly, oxidation of H hyd can be explained by PCET that relies on a functional release of protons. All PT pathway variants specifically react with H 2 and it can be concluded that at least one proton is always injected into the hydrogen-bonding network, despite the compromised PT pathway. Single deprotonation appears possible as the next proton-binding site downstream C299 CpI would be Wat826/Wat1120, which could form a Zundel-ion-like configuration upon protonation 47 (H 5 O 2 + , Fig. 5 and Supplementary Fig. 12a ). Neither of the two H 2 O molecules has been directly affected by mutagenesis. It may be concluded that PT is only blocked if a protonation of Wat826/Wat1120 yield a Zundel-ion-like configuration, stabilized due to increased p K a in comparison to H 3 O + . The favorable position of the acidic asparagine side chain in the structure of C299D CpI effectively connects adt-NH of the [2Fe] H moiety with the conserved water molecule Wat708 in the otherwise intact PT pathway, rendering this variant significantly active. This comparatively high level of absolute activity and the fact that both corresponding variants, C169D HydA1 and C299D CpI , exhibit significant shifts in their pH-dependent activity optimum to the acidic range might explain why this type of variant refuses to accumulate H hyd even at pH 4, while instead effectively continuing the turnover process as evident according to the comparatively dominant infrared bands for the reduced states H red and H sred (1890 and 1882 cm −1 for C169D) recorded at pH 4 (see Supplementary Fig. 4 ). In case of variant C299A CpI , the first deprotonation step (H hyd H + ⇌ H hyd + H + ) seems impossible but Fig. 3 shows that this variant very effectively accumulates H hyd under H 2 . Thus, an alternative proton acceptor has to be assumed. The additional H 2 O molecule (Wat962), which is trapped in the space of the missing thiol group of variant C299A would be a plausible proton acceptor (Fig. 6 and Supplementary Fig. 13b ). As it is well-positioned to bridge the adt-ligand (distance 3.4–3.7 Å) and Wat826 (distance 3.6–3.7 Å), it seems surprising that we are unable to measure any significant catalytic activity. This could be due to an unfavorable p Ka difference among the Zundel-ion-like complex, the uncharged adt-ligand and glutamic acid E279 Cp 48 . The p Ka calculated for the adt-ligand in H hyd is significantly larger than the one determined for H ox in mimics of the [2Fe] H -cluster 49 . However, no re-protonation of H hyd is observed. We assume that the proton is trapped in a Zundel-ion-like configuration with a dangling H 2 O molecule Wat962 downstream of position A299 that prevents E279 from being re-protonated (Fig. 6a , C299A). Upon oxidation of H 2 , proton release via the H-cluster may restrict re-protonation of H hyd and steady-state accumulation of H hyd H + . Recently, an alanine was identified at the position homologous to C299 CpI in the newly described sensory [FeFe]-hydrogenase HydS of Thermotoga maritima , which showed very low H 2 release and oxidation activities (<5% of HydA1) 50 . However, HydS is clearly more active than the C → A variants of CpI and HydA1 suggesting a slightly different situation for HydS. Nevertheless, the very low activity level of this sensory-type [FeFe]-hydrogenase overall agrees very well with our results. For variant E279A CpI the deprotonation of the [2Fe] H -cluster, which produces the hydride state would lead to the formation of the Zundel-ion-like configuration. However, due to the large distance between Wat875 and S319 (up to 4.5 Å) the proton cannot proceed any further. As a consequence, H hyd with its deprotonated adt-NH ligand would be stabilized in presence of H 2 (Fig. 6b , E279A). In case of the corresponding exchange at the surface-exposed glutamate in E282A CpI , two H 2 O molecules (chain B: Wat717 and Wat974) invading from the solvent are able to bridge the gap of the missing carboxyl group (Fig. 6c , E282A and Supplementary Fig. 13a ) thereby rescuing a large fraction of H 2 -release activity (~50%). The fact that E282A CpI is rescued by an H 2 O moiety while in E279A CpI and C299A CpI the gap in the catalytic PT pathway cannot be closed, is probably connected to the superior accessibility at the protein surface and the specific functional requirements (such as e.g. p Ka) of the position to be rescued within the PT pathway chain. This illustrates the significance of individual residues for the catalytic PT pathway, which appears to increase from surface to catalytic center. However, the lack of H 2 release activity resulting from the non-conservative exchange in SDM variant E282Q CpI (Fig. 6d , E282Q), which prohibits H 2 O access due to steric reasons demonstrates the necessity of a protonatable position at the entrance of the catalytic PT pathway (for SDM variants E279Q and S319A see Supplementary Figs. 10 - 11 ). Supplementary Table 9 provides a survey of the presumptive functions of all positions in the PT pathway, according to the conclusions drawn here. The multilayered approach followed here, comprising enzyme kinetics, ATR-FTIR spectroscopy, and crystal-structure analysis, unambiguously verifies the herein proposed PT pathway to be the main trajectory for substrate/product transfer in [FeFe]-hydrogenases. From surface to active center the impact of substitutions along the PT pathway increases, with the inner part between E279 CpI and C299 CpI being the most vital positions for basic PT function. A model of the PT mechanism in line with the presented data further suggests a major role for the H 2 O-cluster and a Zundel-ion-like configuration comprising Wat826 and Wat1120 in regulating PT pathway function. External H 2 O molecules can rescue PT function (1) if provided sufficient space and accessibility, (2) if gaps to neighboring PT pathway positions do not exceed H-bond distance (≤4 Å), and (3) if the overall p Ka-gradient of the PT chain is not severely imbalanced. How PT is precisely coupled and synchronized with electron transfer between [2Fe] H and [4Fe] H remains to be further elucidated, possibly by a time-resolved characterization of the catalytic mechanism under sub-turnover conditions. The variants described here are dramatically slowed down in PT efficiency and provide excellent models to resolve the individual steps. In conclusion, the herein presented data give valuable insights into the molecular parameters that enable and tune PT in redox active proteins, thus providing useful guidelines for de novo catalyst design. Methods Site-directed mutagenesis Expression constructs were generated using the QuikChange ® Site-Directed Mutagenesis Kit from Stratagene using either pET21b- HydA1Cr or pET21b- CpI as template and corresponding mismatch primers 29 , listed in Supplementary Table 6 . The resulting constructs for the expression of SDM variants were verified via sequencing (3130xl Genetic Analyzer; Applied Biosystems) and used for the transformation of Escherichia coli expression host strain BL21 (DE3) Δ iscR 51 , which was kindly provided by Patrik R. Jones. Protein expression and maturation Handling of expression cultures and enzyme variants was entirely done under strictly anaerobic conditions in a glove box (Coy laboratory) atmosphere of 98.5% N 2 and 1.5% H 2 . HydA1 and CpI proteins were expressed in absence of maturases HydE, -F, and -G 52 as described previously and therefore lack the catalytically essential [2Fe] H sub-cluster in the original state after purification. Cell disruption was achieved by ultrasonication. Cell debris was separated from the supernatant by ultracentrifugation and filtration (pore size 0.2 µm). Protein purification was done employing strep-tactin/strep-tag II (IBA GmbH) affinity chromatography. HydA1 and CpI apo-proteins were maturated in vitro by adding the synthetic mimic of the native [2Fe] H -complex (Fe 2 [µ-(SCH 2 ) 2 NH] (CN) 2 (CO) 4 [Et 4 N] 2 ) ([2Fe H ] MIM ) at a 10-fold molar excess 31 . [2Fe H ] MIM was synthesized following literature procedures and kept in 100 mM K 2 HPO 4 /KH 2 PO 4 buffer (pH 6.8) at −80 °C 53 . After an incubation period of 1 h at 25 °C, size-exclusion chromatography was employed to remove redundant complex, using NAP-5 columns (GE Healthcare), which were equilibrated with 100 mM Tris–HCl (pH 8) containing 2 mM sodium dithiolate (NaDT) prior to use. Proteins were concentrated using 30 kDa Amicon Ultra centrifugal Filter units (Merck Millipore) and stored anaerobically at −80 °C. SDS-polyacrylamide gel electrophoresis and Bradford assays were employed to verify protein purity and to calculate the resulting protein concentration. Activity assays H 2 -evolution rates of enzyme variants (0.4–4 µg, depending on the level of residual activity) were determined employing a standard in vitro assay, comprising 100 mM NaDT as sacrificial electron donor and 10 mM methyl viologen as electron mediator in 0.1 M K 2 HPO 4 /KH 2 PO 4 , pH 6.8. The reaction volume of 2 ml was prepared in air-tight suba-seal vessels and purged with argon for 5 min prior to the H 2 -production period of 20 min in a shaking water bath, kept at 37 °C. The H 2 -production yield was determined by analyzing 400 µl of the sample headspace via gas-chromatography (GC-2010, Shimadzu). For the determination of pH-activity profiles, four different buffers were used and adjusted to individual pH values according to their respective buffer range, covering pH 5–10 with a resolution of 0.5 pH units: pH 5–6.5 (200 mM MES-NaOH, 2-( N -morpholino) ethanesulfonic acid); pH 7–7.5 (200 mM MOPS-NaOH, 3-( N -morpholino) propanesulfonic acid); pH 8–9 (200 mM Tris–HCl); and pH 10 (200 mM CAPS-NaOH, N-cyclohexyl-3-aminopropanesulfonic acid). pH-dependent H 2 -uptake activity was monitored performing a colorimetric assay at 25 °C with enzyme amounts between 0.5 ng and 1 µg in an atmosphere of 1 bar H 2 , using 10 mM benzyl viologen (Sigma-Aldrich) as electron acceptor ( ε 600 : 10 mM −1 cm −1 ) 6 . Crystallization and structural determination of variants Vapor diffusion hanging drop and sitting drop methods were applied under anaerobic conditions at 277 K to crystallize the SDM variants using protein concentrations between 10 and 15 mg ml −1 . Details of the crystallization conditions are summarized in Supplementary Table 3 . Mounting was carried out after 5–10 days of crystal growth and crystals selected for data collection were flash-frozen and stored in liquid N 2 . Diffraction data were collected at 100 K in different beamlines as indicated in Supplementary Table 3 and processed using XDS 54 . Phenix 55 and Coot 56 were employed for molecular replacement (starting models for CpI and HydA1 were 4XDC 19 and 3LX4 36 respectively) and structural refinement. The details of crystallographic data gained for each of the variants are summarized in Supplementary Table 4 . Infrared spectroscopy Wild-type and SDM variants of HydA1 and CpI were probed by in situ ATR-FTIR spectroscopy 29 , 34 , 57 using a rapid-scan Tensor 27 spectrometer (Bruker Optik, Germany) equipped with a three-reflection ZnSe/silicon crystal ATR cell (Smith Detection, USA). The spectrometer was kept under anaerobic conditions in a vinyl glove box (Coy Laboratories, USA) under a water-free atmosphere of 99% N 2 and 1% H 2 . All experiments were performed at room temperature, on hydrated films and in the dark. The oxidized state H ox was populated under 100% N 2 ambient partial pressure while the reduced states (H red /H sred , H red ′, and H hyd ) were observed in the presence of H 2 exclusively. All spectra shown in Fig. 3 were mathematically corrected for the broad combination band of liquid H 2 O at around 2120 cm −1 and normalized to unity. Reporting Summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The coordinates and structure factors for all structures are deposited in the PDB as 6GLY , 6GLZ , 6GM0 , 6GM1 , 6GM2 , 6GM3 , 6GM4 , 6GM5 , 6GM6 , 6GM7 , and 6GM8 . Further data supporting findings of this study are available from the corresponding authors upon reasonable request. A Reporting Summary for this Article is available as a Supplementary Information file .

In their catalytic center, hydrogenases manufacture molecular hydrogen (H2) from two protons and two electrons. They extract the protons required for this process from the surrounding water and transfer them – via a transport chain – into their catalytic core. The exact proton pathway through the hydrogenase had as yet not been understood. "This transfer pathway is a jigsaw piece, crucial for understanding the interplay of cofactor and protein which is the reason why biocatalysts are so much more efficient than hydrogen-producing chemical complexes," explains Dr. Martin Winkler, one of the authors of this study from the Photobiotechnology research group at RUB. In order to figure out which of the hydrogenase building blocks are involved in proton transfer, the researchers substituted them individually. They replaced each either by an amino acid with a similar function or by a dysfunctional amino acid. Thus, 22 variants of two different hydrogenases were created. Subsequently, the researchers compared those variants with regard to different aspects, including their spectroscopic properties and their enzyme activity. "The molecular structures of twelve protein variants, which were solved using X-ray structure analysis, proved particularly informative," says Winkler. Amino acids with no function shut down hydrogenases Depending on where and how the researchers had changed the hydrogenase, hydrogen production became less efficient or stopped altogether. "Thus, we ascertained why some variants are severely impaired in terms of enzyme activity and why others are hardly impaired at all – against all expectations," says Martin Winkler. The closer to the catalytic centre the replaced amino acids were located, the less able the hydrogenase was to compensate for these modifications. If building blocks with no function were embedded in sensitive locations, hydrogen production was shut down. "The thus generated state resembles an oversaturation due to proton stress where protons as well as hydrogen are simultaneously introduced into the hydrogenase," elaborates Martin Winkler. "In the course of our project, we were for the first time able to stabilise and analyse this highly transient state that we had already encountered in experiments." Valuable baseline information This study has made it possible to assign the functions of individual amino acids to the proton transfer pathway for the enzyme group of [FeFe] hydrogenases. "Moreover, it provides valuable information on the molecular mechanism of proton transfer by redox-active proteins and the structural requirements thereof," concludes Thomas Happe.

10.1038/s41467-018-07140-x

Medicine

Researchers uncover ion channel trio that mediates painful heat sensing

A TRP channel trio mediates acute noxious heat sensing, Nature (2018). nature.com/articles/doi:10.1038/nature26137 Journal information: Nature

http://nature.com/articles/doi:10.1038/nature26137

https://medicalxpress.com/news/2018-03-uncover-ion-channel-trio-painful.html

Abstract Acute pain represents a crucial alarm signal to protect us from injury 1 . Whereas the nociceptive neurons that convey pain signals were described more than a century ago 2 , the molecular sensors that detect noxious thermal or mechanical insults have yet to be fully identified 3 , 4 , 5 , 6 . Here we show that acute noxious heat sensing in mice depends on a triad of transient receptor potential (TRP) ion channels: TRPM3, TRPV1, and TRPA1. We found that robust somatosensory heat responsiveness at the cellular and behavioural levels is observed only if at least one of these TRP channels is functional. However, combined genetic or pharmacological elimination of all three channels largely and selectively prevents heat responses in both isolated sensory neurons and rapidly firing C and Aδ sensory nerve fibres that innervate the skin. Strikingly, Trpv1 −/− Trpm3 −/− Trpa1 −/− triple knockout (TKO) mice lack the acute withdrawal response to noxious heat that is necessary to avoid burn injury, while showing normal nociceptive responses to cold or mechanical stimuli and a preserved preference for moderate temperatures. These findings indicate that the initiation of the acute heat-evoked pain response in sensory nerve endings relies on three functionally redundant TRP channels, representing a fault-tolerant mechanism to avoid burn injury. Main Nociceptors involved in acute heat sensing have cell bodies in the dorsal root and trigeminal ganglia and extend their sensory endings into the skin, mucosa and internal organs 4 , 6 . In these nociceptor endings, a rise in temperature induces the firing of action potentials, which propagate towards the dorsal horn or sensory nucleus in the brain, where the sensory information is relayed to second-order neurons 4 . Heat-activated nociceptors are characterized by the expression of the heat- and capsaicin-activated TRP channel TRPV1 7 . However, although ablation 8 or silencing 9 of TRPV1-expressing neurons in mice result in almost complete absence of heat nociception, Trpv1 -knockout mice display only minor deficits in acute noxious heat sensing 10 , 11 , indicating that heat-sensitive nociceptors express additional molecular heat sensors. Although various other ion channels, including several other TRP channels and calcium-activated chloride channels of the Anoctamin/TMEM16 family 4 , 6 , have been put forward as molecular heat sensors, mice deficient in these various channels showed either no or relatively mild deficits in acute heat sensing 12 , 13 , 14 , 15 . Thus, the molecular basis of acute heat sensing has remained unresolved 4 . The heat-sensitive TRP channel TRPM3 is expressed in a large subset of TRPV1-expressing sensory neurons, and Trpm3 −/− knockout mice exhibit reduced heat sensitivity at cellular and behavioural levels 14 , 16 . To investigate the consequences of combined elimination of TRPV1 and TRPM3, we analysed the heat responsiveness of Trpm3 −/− Trpv1 −/− double knockout (DKO M3/V1 ) mice. Using calcium imaging, we probed the thermal and chemical responsiveness of sensory neurons isolated from the trigeminal ganglia in primary cultures. Whereas responses to the TRPM3 agonist pregnenolone sulfate and to the TRPV1 agonist capsaicin were largely eliminated, the proportion of heat-responsive neurons was reduced only slightly compared to that of wild-type mice, with approximately 40% of the DKO M3/V1 neurons still exhibiting a robust heat-induced calcium response ( Fig. 1a ; Extended Data Fig. 1a ; see Methods for identification of heat responders). We used a skin-nerve preparation to measure action potentials upon stimulation of the receptive fields of single nerve fibres 17 , and recorded robust heat-evoked responses in 33% (6/18) of the mechanosensitive C fibres from DKO M3/V1 mice, as compared to 49% (17/35) in wild-type mice ( P = 0.38; Fisher’s exact test; Extended Data Fig. 1b ). At the behavioural level, DKO M3/V1 mice exhibited vigorous heat avoidance responses in the tail-immersion and hot-plate assays, although withdrawal latencies were prolonged compared to the wild type ( Extended Data Fig. 1c ). Thus, combined elimination of TRPM3 and TRPV1 results in only very mild deficits in heat responsiveness, implying the existence of one or more additional heat sensors in somatosensory neurons. Figure 1: TRPA1 mediates TRPM3- and TRPV1-independent heat responses. a , b , Responses to heat (45 °C) and AITC (50 μM) in sensory neurons from DKO M3/V1 mice in the absence and presence of the TRPA1 antagonist HC030031 (100 μM). WT, wild-type; PS, pregnenolone sulfate; Caps, capsaicin. c , Example of a HC030031-sensitive heat-activated inward current in an AITC-sensitive DKO M3/V1 neuron. d , Mean (± s.e.m.) current densities for experiments as in c ( n = 9 cells). e , f , Heat and ligand responses in sensory neurons from TKO and indicated DKO mice. Scale bars in a , b , e , f , 60 s/100 nM. g , Percentages of heat-responding sensory neurons in mice of the indicated genotypes, in the absence or presence of HC030031. The experiments in a , b , e , f and summary data in g are representative of the following numbers ( n ) of tested trigeminal neurons from N mice: wild-type: n = 296, N = 6; DKO M3/A1 : n = 613, N = 5; DKO V1/A1 : n = 725, N = 7; DKO M3/V1 : n = 1,353, N = 9; TKO: n = 720, N = 11; WT+HC030031: n = 208, N = 3; DKO M3/V1 + HC030031: n = 293, N = 3; DRG neurons from wild-type: n = 133, N = 3; TKO: n = 181, N = 3. *** P < 0.00001; Fisher’s exact test with Holm–Bonferroni correction. PowerPoint slide Full size image We observed that most residual heat-sensitive neurons in the DKO M3/V1 mice also responded to allyl isothiocyanate (AITC, 50 μM; Fig. 1a ), a pungent chemical that at this concentration causes selective activation of TRPA1 18 . Moreover, in wild-type heat-sensitive neurons, we found extensive overlap of functional expression of TRPA1 with both TRPM3 and TRPV1 ( Extended Data Fig. 1d ). To test whether TRPA1 contributes to residual heat sensitivity in DKO M3/V1 mice, we first evaluated the effect of the selective TRPA1 antagonist HC030031, and found that it caused an almost complete loss of heat-induced calcium responses in DKO M3/V1 sensory neurons ( Fig. 1b, g ). Accordingly, we measured heat-induced inward currents in a subset (9 out of 23) of DKO M3/V1 sensory neurons; these currents were inhibited by HC030031 and correlated with cellular responsiveness to AITC ( Fig. 1c, d ). Next, we produced Trpv1 −/− Trpm3 −/− Trpa1 −/− triple knockout (TKO) mice ( Extended Data Table 1 ), and compared them with the different double knockout (DKO) mouse lines for these three TRP channels (DKO M3/V1 , DKO M3/A1 ( Trpm3 −/− Trpa1 −/− ) and DKO V1/A1 ( Trpv1 −/− Trpa1 −/− ) mice). Sensory neurons isolated from TKO mice showed an almost complete loss of heat responses, which contrasts sharply with the presence of more than 40% heat responders in wild-type mice and in the three DKO mouse lines ( Fig. 1e–g ). Several lines of evidence indicate that the loss of heat sensitivity in TKO neurons is specifically due to the combined loss of the three TRP channels. Most importantly, reintroduction via transient transfection of TRPV1, TRPM3 or TRPA1 into TKO neurons restored sensitivity to heat and to the respective channel agonists ( Fig. 2a–c, f ). Conversely, heat responses in wild-type sensory neurons were suppressed by an inhibitor cocktail consisting of the TRPA1 antagonist HC030031, the TRPV1 antagonist AMG9810, and the TRPM3 antagonist isosakuranetin ( Fig. 2d–f ). Together, these results indicate that TRPA1, TRPV1 and TRPM3 have critical but redundant roles in heat transduction: functionality of at least one of the three channels is necessary and sufficient to sustain robust heat responses in sensory neurons. Figure 2: Properties of TKO sensory neurons. a – c , Heat- and ligand-induced calcium responses of TKO trigeminal neurons transfected with TRPV1, TRPM3 or TRPA1. d , e , Responses of wild-type trigeminal neurons to heat in the absence ( d ) and presence ( e ) of an inhibitor cocktail of the TRPV1 antagonist AMG9810 (5 μM), the TRPM3 antagonist isosakuranetin (5 μM) and the TRPA1 antagonist HC030031 (100 μM); after the heat stimulus and wash-out of the inhibitor cocktail, responses to the indicated agonists (pregnenolone sulfate, 40 μM; AITC, 50 μM; capsaicin, 1 μM) were conserved. Scale bars in a – e , 60s/100 nM. f , Percentages of heat-responding sensory neurons in transfected TKO neurons and in wild-type neurons, in the absence and presence of the inhibitor cocktail. The experiments in a – e and summary data in f are representative of the following numbers ( n ) of tested trigeminal neurons from N mice. TKO mock: n = 277, N = 5; TKO + TRPV1: n = 130, N = 5; TKO + TRPM3: n = 189, N = 4; TKO + TRPA1: n = 289, N = 6; wild-type control: n = 276, N = 6; wild-type + inhibitor cocktail: n = 199, N = 5. PowerPoint slide Full size image Other aspects of sensory neuron biology were unaltered in the TKO mice. First, functional expression of the cold-activated channel TRPM8 was preserved in the trigeminal ganglia of TKO mice ( Extended Data Fig. 2a–c ). Compared to wild-type neurons, TKO neurons showed a reduction in the fraction of cold responders ( Extended Data Fig. 2c ), consistent with the contribution of TRPA1 to cold sensing 19 . Second, RNA sequencing (RNA-seq) analysis of isolated sensory ganglia from wild-type, TKO and the different single and double knockout mouse lines for TRPA1, TRPM3 and TRPV1 did not reveal any global changes in mRNA expression profile ( Extended Data Fig. 2d ). In particular, the mRNA expression levels of other channels implicated in heat sensing, and of a previously defined set of about 200 operational components of sensory neurons 20 , were not significantly altered in the TKO mice ( Extended Data Fig. 2e ). Considering that heat-sensitive neurons constitute a large fraction of the total sensory neuronal population in wild-type mice (about 60%; Fig. 1g ), these results argue against important alterations at the transcriptional level that would provide an obvious explanation for the striking deficit in heat sensing in the TKO neurons. A recent study 21 identified a population of TRPM3- and TRPV1-negative heat-sensitive sensory neurons, and concluded that these neurons rely on TRPM2 for their heat responses. An important piece of evidence to support the involvement of TRPM2 was the finding that TRPM3- and TRPV1-independent heat responses were enhanced by H 2 O 2 , a known TRPM2 activator. However, as TRPA1 is also sensitive to H 2 O 2 22 , TRPA1 may contribute to the H 2 O 2 -sensitive heat responses in sensory neurons. Using DKO M3/V1 sensory neurons, we confirmed strong potentiation of TRPM3- and TRPV1-independent heat responses by H 2 O 2 , including in heat-insensitive neurons, which gained heat sensitivity in the presence of H 2 O 2 ( Extended Data Fig. 3a, f ). However, these potentiated heat responses occurred exclusively in AITC-sensitive neurons, and were fully suppressed in the presence of HC030031 ( Extended Data Fig. 3a, b, f ). Likewise, H 2 O 2 failed to enhance the heat responses in TKO neurons: heat-insensitive TKO neurons never gained heat sensitivity in the presence of H 2 O 2 ( Extended Data Fig. 3c ), and the few residual heat responders (about 3%) among TKO neurons were not potentiated by H 2 O 2 ( Extended Data Fig. 3d, f ). Notably, the residual heat responses in TKO neurons were also not affected by the TRPM2 antagonist 2-APB ( Extended Data Fig. 3e ). Overall, these data do not provide support for the notion that TRPM2 contributes significantly to heat responses in somatosensory neurons, and instead demonstrate that TRPA1 mediates H 2 O 2 -sensitive, TRPV1- and TRPM3-independent heat responses in these cells. Consistent with earlier research 23 , we were unable to reliably measure heat responses in non-neuronal CHO cells expressing mouse TRPA1 ( Extended Data Fig. 4a, d ). However, about 40% of the CHO-TRPA1 cells showed a robust heat-evoked Ca 2+ response following a brief preincubation with H 2 O 2 ( Extended Data Fig. 4b, d ), and these heat responses were fully suppressed by HC030031 ( Extended Data Fig. 4c, d ), in line with a recent report showing that the temperature sensitivity of TRPA1 is strongly dependent on the redox state 24 . Together, our results indicate that the heat responsiveness of mouse TRPA1 is highly dependent on the cellular environment: under control conditions, expression of TRPA1 induces robust heat responses in sensory neurons ( Figs 1a, c, d , 2c ) but not in non-neuronal cell lines ( Extended Data Fig. 4 ); in both cellular environments, TRPA1-dependent heat responses are strongly enhanced by H 2 O 2 ( Extended Data Figs 3 , 4 ). In this context, a recent study of TRPA1 in the planarian flatworm Schmidtea mediterranea (smed-TRPA1) demonstrated that H 2 O 2 and reactive oxygen species are rapidly (within seconds) produced as a result of heat exposure, leading to smed-TRPA1 activation and TRPA1-dependent acute heat avoidance 25 . Our results also do not exclude the possibility that heat-induced TRPA1 activation occurs downstream of an as yet unknown heat sensor. Next, we evaluated the heat responses of single C and Aδ fibres innervating the skin of wild-type, DKO and TKO mice. The proportion of C fibres that responded to a heat ramp to 50 °C was reduced in TKO mice ( Fig. 3a ). We further characterized the thermal response characteristics of heat-sensitive C fibres from wild-type, DKO and TKO mice, by analysing their thermal thresholds and peak firing rates ( Fig. 3b, c ). Overall, the five genotypes did not differ significantly with respect to the thermal threshold ( P = 0.07, one-way ANOVA; Fig. 3b ). The peak firing rates of C fibres showed a bimodal distribution, and, accordingly, heat-sensitive C fibres were classified as either high frequency or low frequency, depending on whether their peak firing rate was above or below 23 Hz, respectively ( Extended Data Fig. 5 ). Out of 23 C fibres from TKO mice, none was classified as high-frequency heat-sensitive, which compares to 23% high-frequency heat-sensitive C fibres in wild-type mice (8/39; P = 0.021, two-tailed Fisher’s exact test) and between 11 and 40% in the different DKO mice ( Fig. 3c ). Notably, TKO C fibres did not differ significantly from wild-type or DKO C fibres with respect to mechanical thresholds and conduction velocities, indicating that the deficit in the TKO C fibres is specific for heat, without affecting general neuronal excitability ( Fig. 3d, e ). In addition to C fibres, a subset of mouse Aδ fibres also exhibits responses to heat. Whereas approximately 10% of wild-type Aδ fibres respond to the heating ramp, no TKO Aδ fibres out of 41 tested responded to heat ( Fig. 3f ). Histological analysis did not reveal any significant difference in the nerve bundles innervating the hindpaw skin of TKO mice ( Extended Data Fig. 6 ). Overall, these results indicate that combined ablation of TRPA1, TRPV1 and TRPM3 causes a striking deficit in the heat sensitivity of Aδ and C fibres. Figure 3: Heat sensitivity of sensory fibres innervating the skin of TKO mice. a , Percentage of mechanosensitive C fibres innervating the paw skin from wild-type, DKO and TKO mice that show a response to heat (48 °C) (* P = 0.041, ** P = 0.0057, # P = 0.037; Fisher’s exact test). b , Thermal threshold for activation of heat-sensitive C fibres. Horizontal lines in b , d , e indicate mean. c , Peak firing frequency of heat-sensitive C fibres during a heating stimulus. The dotted line at 23 Hz indicates the border between low frequency and high frequency firing (see Extended Data Fig. 5 ). * P = 0.021, ** P = 0.00086; one-way ANOVA with Tukey’s post-hoc test. d , e , Mechanosensitivity ( d ) and conduction velocity ( e ) are unaltered in C fibres from the different genotypes. Black dots overlaying TKO data points mark the heat-sensitive C fibres. f , Percentage of mechanosensitive Aδ fibres in mice that show a heat response. * P = 0.05; Fisher’s exact test. ND, not done. Number of tested fibres: wild-type C, n = 35, Aδ, n = 31; DKO M3/A1 C, n = 17, Aδ, n = 14; DKO M3/V1 C, n = 18, Aδ, n = 13; DKO V1/A1 C, n = 10; TKO C, n = 23, Aδ, n = 41. PowerPoint slide Full size image Next, we compared TKO mice with wild-type mice and the different DKO mouse lines with respect to acute heat nociception. In both the tail-immersion and hot-plate assays, the TKO mice showed a severe deficit in heat avoidance behaviour, to such a degree that the majority of TKO animals reached the cut-off times without withdrawal response to a range of noxious temperatures over 45 °C ( Fig. 4a–c ; Extended Data Figs 7 , 8a, b ; Supplementary Video 1 ). In comparison, the different DKO mice all exhibited vigorous withdrawal responses to noxious heat, albeit with overall longer latencies in the case of the DKO M3/V1 and DKO V1/A1 mice ( Fig. 4a–c ; Extended Data Fig. 7 ). Note that, by using a cut-off latency set at four times the mean response latency of wild-type mice, we did not observe superficial signs of acute skin burns immediately following the hot-plate or tail-immersion experiments in mice of any genotype. However, visual re-inspection of the animals three days after the 57 °C tail-immersion experiments revealed severe tissue damage (scarring, necrosis) in the tails of all TKO mice, which was never observed in wild-type mice or any of the DKO mice ( Extended Data Fig. 8d–f ). Thus, TKO mice lack the acute heat response that is necessary to avoid burn injury. Figure 4: Selective loss of acute heat nociception in TKO mice. a , b , Withdrawal latencies of male mice in the hot-plate ( a ) and tail-flick ( b ) assays. Dotted lines indicate cut-off times. Results obtained at additional temperatures and statistical analyses are provided in Extended Data Fig. 7 . Horizontal lines in a , b , d – g indicate mean. c , Percentage of tested animals (WT: N = 16; DKO M3/A1 : N = 12; DKO V1/A1 : N = 15; DKO M3/V1 : N = 12; TKO: N = 16) that did not withdraw their tail before the cut-off. d , Withdrawal latencies of wild-type ( N = 13–14) and TKO ( N = 15) mice to smooth plastic and jagged metal clips. P = 0.50 and 0.86, respectively; two-sided Student’s t -test. e , Sensitivity of wild-type ( N = 15) and TKO ( N = 15) mice to calibrated von Frey hairs ( P = 0.09; two-sided Student’s t -test). T 50 , 50% paw withdrawal threshold. f , Withdrawal latencies of wild-type ( N = 19) and TKO ( N = 12) mice in the cold plantar test ( P = 0.76; two-sided Student’s t -test). g , Number of icilin-induced wet dog shakes in wild-type ( N = 12) and TKO ( N = 11) mice ( P = 0.20; two-sided Student’s t -test). h , i , Representative experiments depicting the positions of a wild-type ( h ) and a TKO ( i ) mouse on a thermal gradient. Border zones of the gradient are red (>45 °C) and blue (<10 °C). j , Mean presence time ± s.e.m. in the different temperature zones during the first and second 30-min periods ( N = 10 mice for each genotype). k , Two-temperature choice test with control plate set at 30 °C and test plate set at 30 °C or 45 °C. Mean percentage ± s.e.m. of the total time spent on the test plate ( N = 9 mice for each genotype; P = 0.77 at 30 °C, P = 0.16 at 45 °C; two-sided Student’s t -test). Both genotypes showed statistically significant avoidance at 45 °C ( P = 0.001; one-sample t -test versus 50%). PowerPoint slide Source data Full size image The absence of a rapid withdrawal response to noxious heat was not due to a general lack of sensitivity to aversive stimuli, as TKO mice showed wild-type-like pain responses in the tail clip assay ( Fig. 4d ), normal mechanosensitivity of their hindpaws in the von Frey assay ( Fig. 4e ; Extended Data Fig. 8c ), and an unaltered withdrawal latency in the cold plantar assay 26 ( Fig. 4f ). Likewise, the number of icilin-induced wet dog shakes, a measure of TRPM8-dependent cooling-induced behaviour, was not significantly different between wild-type and TKO mice ( Fig. 4g ). In addition to reflex avoidance responses to noxious cold or hot temperatures, mice exhibit thermotaxis behaviour towards their preferred ambient temperature, typically in the moderately warm (25–35 °C) range. To test whether TKO mice exhibited alterations in thermal preference, mice were allowed to move freely on a 1.2-m-long thermal gradient ranging from 5 to 50 °C for 60 min. Wild-type and TKO mice showed extensive exploratory behaviour and exhibited a similar mean distribution on the gradient ( Fig. 4h–j ). Overall, we did not detect any significant differences between wild-type, TKO and the different DKO mice with respect to the total covered distance on the gradient or the peak occupied temperature ( T peak ; Extended Data Fig. 9 ). Mice of all genotypes explored the full length of the gradient, including the border zones (marked in blue (below 10 °C) and red (above 45 °C) in Fig. 4h, i ), and the combined time spent in these two zones was similar between genotypes ( Extended Data Fig. 9c ). However, whereas wild-type and the different DKO mice showed a preference for the blue (cold) versus the red (hot) border zone, this preference was not observed in the TKO mice ( Extended Data Fig. 9a ). In the two-temperature choice test, where animals were given the choice between a control plate at 30 °C and a test plate at 45 °C, animals of all genotypes showed a similar preference for the control plate ( Fig. 4k ; Extended Data Fig. 9c ). A striking difference between the genotypes was found when we analysed the timing of individual visits to the test plate: the duration of the longest visit to the 45 °C plate was significantly longer for TKO mice than for wild-type or the different DKO mice ( Extended Data Fig. 9b ). A further experiment with the test plate set at 50 °C was prematurely terminated, as we observed that several TKO mice would stay for more than 60 s on the test plate, risking burn injury. Together, these data demonstrate that thermal preference is largely preserved in TKO mice, despite the strong deficit in acute heat-induced pain. These results contrast with the phenotype of mice in which all TRPV1-positive neurons were ablated, which not only lack noxious heat responses but are also indifferent to temperatures between 30 and 50 °C in the two-temperature choice assay 8 , 27 . The ability of TKO mice to discriminate temperatures in the 30–50 °C range may depend on the very small but consistent subset (about 3%) of TRPV1-, TRPM3- and TRPA1-independent heat-responsive sensory neurons ( Fig. 1g ; Extended Data Fig. 3 ), or on thermal stimulation of skin keratinocytes that then signal to sensory neurons via diffusible messengers 28 . The subset of low-frequency heat-sensitive C fibres, which is conserved in the TKO mice ( Fig. 3 ), are potential candidates to convey such keratinocyte-dependent signals. Finally, heat avoidance behaviour may be secondary to changes in core body temperature. In particular, the preoptic area of the hypothalamus contains temperature-sensitive neurons that detect small increases in local brain temperature and initiate both physiological heat loss mechanisms 29 and cold-seeking behaviour 30 . Although the nociceptive neurons involved in acute pain signalling in mammals were described more than a century ago 2 , the molecular mechanisms whereby these neurons detect harmful signals have remained largely unresolved. The TKO mouse represents, to our knowledge, the first demonstration in mammals of elimination of the pain response to a noxious physical stimulus at the level of the transducer ion channels. The presence of three redundant molecular heat-sensing mechanisms with overlapping expression in nociceptor neurons represents a powerful fail-safe mechanism that ensures avoidance of noxious heat even under conditions where the function of one or two heat sensors is compromised, for instance due to channel desensitization or naturally occurring inhibitory ligands. We speculate that similar redundancy may operate in mammals for the detection of other noxious physical stimuli. Methods Animals Wild-type C57BL/6J and Trpa1 −/− Trpm3 −/− , Trpa1 −/− Trpv1 −/− , Trpm3 −/− Trpv1 −/− and Trpa1 −/− Trpm3 −/− Trpv1 −/− mice on a C57BL/6J background, aged 10–14 weeks, were used for all experiments. Male mice were used for behavioural experiments, unless stated otherwise; primary neurons and skin-nerve preparations were obtained from both male and female mice. Mice were housed in a conventional facility at 21 °C on a 12-h light–dark cycle with unrestricted access to food and water. All experiments using animals were approved by the KU Leuven Ethical Committee Laboratory Animals under project numbers P021/2012 and P265/2015. Isolated sensory neurons After mice were killed by CO 2 inhalation, trigeminal ganglia were excised, washed in neurobasal A medium (Invitrogen) supplemented with 10% fetal calf serum (basal medium), and then incubated for 45 min at 37 °C in a mix of 1 mg/ml collagenase (Gibco) and 2.5 mg/ml dispase (Gibco). Digested ganglia were gently washed twice with basal medium and mechanically dissociated by passage through syringes fitted with increasing needle gauges. Neurons were seeded on poly- l -ornithine/laminin-coated glass-bottomed chambers (Fluorodish, WPI) and cultured overnight at 37 °C in 5% CO 2 in B27 (Invitrogen) supplemented neurobasal A medium, containing 2 ng/ml GDNF (Invitrogen) and 10 ng/ml NT4 (Peprotech). For transfection experiments, trigeminal ganglia neurons were grown for 4 days before transfection with 0.5 μg cDNA encoding mouse TRPA1, TRPV1 or TRPM3 in the bicistronic pCAGGS-IRES–GFP vector, using Lipofectamine 2000 (Invitrogen) as a transfection reagent. Neurons were kept at 37 °C in 5% CO 2 for 3 days after transfection before calcium imaging. Cell culture CHO cells stably expressing mouse TRPA1 (CHO-A1 cells, kindly provided by A. Patapoutian) were cultured in DMEM containing 10% fetal bovine serum, 2% glutamax (Invitrogen), 1% non-essential amino acids (Invitrogen) and 200 μg/ml penicillin/streptomycin at 37 °C in a humidity-controlled incubator with 5% CO 2 23 . The cell line tested negative for mycoplasma contamination. Cell line authentication was not performed, but TRPA1 expression was verified based on AITC responsiveness. Calcium imaging Changes in intracellular calcium concentration were monitored using ratiometric Fura-2-based fluorimetry. Cells were loaded with 2 μM Fura-2-acetoxymethyl ester (Alexis Biochemicals) for 30 min at 37 °C. Fluorescence was measured during alternating illumination at 340 and 380 nm using either a Cell M (Olympus) or Eclipse Ti (Nikon) fluorescence microscopy system, and absolute calcium concentration was calculated from the ratio of the fluorescence signals at these two wavelengths ( R = F 340 / F 380 ) as [Ca 2+ ] = K m × ( R − R min )/( R max − R ), where K m , R min and R max were estimated from in vitro calibration experiments with known calcium concentrations. The bath solution contained (in mM): 138 NaCl, 5.4 KCl, 2 CaCl 2 , 2 MgCl 2 , 10 glucose, and 10 HEPES, pH 7.4. Allylisothiocyanate, pregnenolone sulfate, and capsaicin were obtained from Sigma-Aldrich and dissolved in bath solution from a 1,000× stock solution in DMSO. At the end of each experiment, cells were subjected to a depolarizing solution containing 50 mM KCl, and non-responsive cells were excluded from analysis. The following procedure was used in sensory neurons to distinguish stimulus-induced responses from background variations in calcium concentration or calcium-independent temperature-induced changes in Fura 2 fluorescence. We calculated the time derivatives of the calcium concentration, F 340 and F 380 as well as the s.d. of these derivatives in the absence of any stimulus. A positive response was noted when the following criteria were simultaneously met: 1) the stimulus caused an increase of d[Ca 2+ ]/d t exceeding 3 × s.d.; 2) d F 340 /d t > 3 × s.d.; and 3) d F 380 /d t < 3 × s.d. For CHO-A1 cells, a positive heat response was defined as a response that exceeded the mean response in the presence of HC030031 by at least five times the s.d., as described elsewhere 29 . Whole-cell patch-clamp recordings Membrane currents in isolated sensory neurons were measured in the whole-cell configuration at a constant holding potential of −40 mV using an EPC-10 amplifier and PatchMaster Software (HEKA Elektronik). The extracellular solution contained (in mM): 140 NaCl, 4 KCl, 2 CaCl 2 , 2 MgCl 2 , 10 HEPES (pH 7.4 with NaOH), supplemented with 100 nM tetrodotoxin (TTX) to block voltage-gated sodium channels. The internal solution contained (in mM): 140 CsCl, 1 MgCl 2 , 1 EGTA, 10 HEPES and 5 tetraethylammonium chloride (pH 7.2 with CsOH). Skin-nerve recordings Single-fibre recordings from cutaneous C and Aδ fibres of the saphenous nerve were obtained using an isolated skin-nerve preparation as described previously 17 . A skin flap of the hind paw hairy skin and the innervating saphenous nerve were excised and fixed in an organ bath, corium side up. The chamber containing the skin tissue was continuously superfused with pre-warmed (30 °C) synthetic interstitial fluid (SIF) solution, consisting of (in mM) 107.8 NaCl, 3.5 KCl, 26.2 NaHCO 3 , 1.67 NaH 2 PO 4 , 9.64 Na-gluconate, 0.69 MgSO 4 , 7.6 sucrose, 5.05 glucose, and 1.53 CaCl 2 . The pH was buffered to 7.4 using carbogen (95% oxygen and 5% carbon dioxide). The nerve trunk was placed on a mirror in a separate chamber filled with paraffin oil, desheathed and divided under microscopic control until a unit with a single distinguishable receptive field was obtained. Initially, mechanosensitive receptive fields were mapped using a blunt plastic rod. Once a distinct receptive field was identified, electrostimulation was applied to the receptive field through a metal microelectrode to determine the fibre’s conduction velocity. Values of conduction velocity were used for fibre classification with a cut-off criterion of <1.4 m/s for unmyelinated C fibres. A marking technique, in which latency shifts are provoked through simultaneous application of a mechanical and electrical stimulus to the receptive field, was then applied, ensuring recording from a distinct single fibre. To assess the sensory properties of these mechanically sensitive fibres and for further classification into subclasses, the mechanical, cold and heat sensitivity of the individual fibres was tested. First, the mechanical threshold of the fibre was characterized using calibrated von Frey monofilaments ranging from 1 mN to 128 mN in a geometric scale. Subsequently, a rounded plastic ring (~8 mm diameter) was placed above the receptive field. The volume of fluid within the ring was evacuated and cold (by fluid replacement with cold buffer at 4 °C), followed by radiant heat (ramp of 30–50 °C delivered by a halogen lamp with an 8-mm focus beam) stimuli were applied. Heat thresholds were taken as the temperature at which the second spike occurred. Data were acquired using the Dapsys data acquisition system (B. Turnquist, Bethel University, USA). Behavioural tests Considering that we tested animals of five different genotypes, with a priori unknown somatosensory properties, we were not able to perform a reasonable power analysis to determine the sample size for behavioural experiments. Therefore, sample sizes were chosen to be consistent with those reported in similar behavioural studies. In nociceptive assays, mice from different genotypes were tested for all the different stimuli, so no randomization of experimental groups was necessary. In the tail-immersion assay, mice were immobilized in aluminium foil, which allowed free tail movement. The tip of the tail (one-third of the length) was immersed in a water bath maintained at 45, 48, 50, 52, or 57 °C. The latency to a nociceptive response (withdrawal of the tail) was measured. To prevent acute tissue damage, the tail was removed from the bath immediately after a nociceptive response or upon reaching a cut-off time, which was set at four times the mean response latency of wild-type mice. In the hot-plate assay, mice were individually confined in a Plexiglas chamber on a metal surface set at 50, 55 or 58 °C and the latency to a nociceptive response (licking or shaking of hind paws, jumping) was measured. To prevent tissue damage, mice were removed from the hot plate after a nociceptive response or upon reaching a cut-off time of three times the mean response latency of wild-type mice. In the thermotaxis assay, mice were individually tracked for 60 min in a thermal gradient apparatus (Bioseb), which consisted of a controlled and stable temperature gradient from 5 to 50 °C on an aluminium floor. Tracking was performed using a video camera-based system and the time spent in two-degree zones was determined. In the two-temperature choice assay, mice were placed in a chamber containing two identical, adjacent floor platforms (Bioseb) with one set to 30 °C and the other set to 45 or 50 °C. Mice were free to explore for 10 min, and the total time spent on each surface and the number of times an animal moved to an adjacent platform were analysed. In the tail clip assay, either a plastic mini-peg (Maped) or metal crocodile clamp (Teishin, Japan), exerting a force of 5 N, was applied to the root of the tail, and the latency to a nociceptive response was measured. In the Von Frey test, mice were individually placed in a chamber with a wire mesh floor, and allowed to habituate for 15 min before testing. Mechanical responses were tested by stimulating the middle plantar surface of the hind paw with von Frey monofilaments using the up-and-down method, starting with 1 g. Biting, licking, or withdrawal during or immediately following the 3-s stimulus were considered as a positive response. In the rotarod assay, mice were initially placed on the stationary rod. After habituation, rotation was started at 4 r.p.m. (rotations per minute), accelerating over a 5-min period to 40 r.p.m. This was done in three sessions during four trial days and one test day. The time taken for the mouse to fall from the rod was recorded. In the elevated maze test, mice were placed in a plus-shaped apparatus with two open and two enclosed arms, connected by a central platform, which was elevated from the floor. Mice were individually placed in the centre of the maze and observed for 5 min. The time spent in the open and enclosed arms was recorded using ANY-maze tracking software (Stoelting). To evoke icilin-induced wet dog shakes, mice were intraperitoneally injected with 50 mg/kg icilin, which induces robust shivering and shaking in mice. The incidence of these wet dog shakes was counted for 20 min after icilin administration. In the cold plantar test, mice were acclimated on a 4-mm-thick glass plate, and then the withdrawal latency was measured when a compressed dry ice pellet was held against the glass surface underneath the hindpaw. The latency to withdrawal was tested three times for each hindpaw with an interval of ~15 min, and averaged to obtain a single value per mouse 26 . RNA-seq After mice were killed by CO 2 inhalation, DRGs were isolated and kept in RNAlater Stabilization Reagent (Qiagen) at −20 °C until further processing. Total RNA was extracted using the RNeasy Mini Kit (Qiagen). The sequencing libraries were prepared according to the standard Illumina TruSeq mRNA stranded sample preparation protocol and each library was sequenced on an Illumina HISeq 2500 according to the manufacturer’s recommendations, generating 50-bp single-end reads at the Genomics Core Leuven. Reads were aligned using the open-source, splice-aware tool Tophat (v2.0.13). Read counts per feature (genes or exons) were done by HTSeq (v0.5.3p3). Count data were analysed to identify differentially expressed genes using DESeq2 (v1.01.1) at FDR 10%. In situ hybridization Mice were transcardially perfused via the left ventricle with 25 ml PBS followed by 25 ml neutral buffered formalin (10% (w/v), pH 7.4). Subsequently, trigeminal ganglia were dissected and immersion-fixed in neutral buffered formalin for 6 h. Thereafter, trigeminal ganglia were transferred to PBS for 1 h, and then to 70% ethanol until paraffin embedding. Paraffin-embedded tissue was cut into 4-μm-thick sections using an RM2235 rotary microtome (Leica), and probed with digoxigenin-labelled antisense Trpm8 RNA probes. Antisense RNA probes were generated using the MAXIscript SP6/T7 Transcription Kit (ThermoFisher Scientific), using a 204-bp cDNA fragment of mouse Trpm8 (accession number NM_134252.3) corresponding to nucleotides 1779-1982 (exons13 and 14). Hybrid molecules were detected using the TSA Plus Fluorescein Kit (Perkin Elmer) according to the manufacturer’s instructions. Nerve bundle anatomy The plantar skin of the mouse hind paw was fixed in 4% paraformaldehyde and further processed for staining. Four-micrometre sections were incubated for 30 min in H 2 O 2 (3%) and antigen retrieval was performed in 0.01 M citrate buffer (pH 6, 90 °C, 1 h). Slides were gradually cooled down and incubated in blocking buffer containing 2% bovine serum albumin (Sigma-Aldrich), 1% non-fat dry milk (Nestlé), 0.1% Tween (Merck), complemented with 1:25 normal goat serum (X0907, Dako), before overnight incubation with the primary antibody (4 °C, 1 μg/ml rabbit polyclonal anti-PGP9.5, Z5116, Dako). Slides were washed and blocked again with blocking buffer (15 min), after which they were incubated with the biotinylated swine anti-rabbit secondary antibody (30 min, 1:400; Z0196, Dako) supplemented with normal mouse serum (1:25; Dako). Staining was initiated by the addition of horseradish peroxidase-conjugated streptavidin (30 min, 1:1,000; P0397, Dako) followed by 3-amino-9-ethylcarbazole substrate-chromogen (25 min). Sections were counterstained with Maeyer haematoxylin and mounted in glycerin jelly. For each mouse, the nerve bundle areas of 20 sections were evaluated with an average distance of 0.1 mm between the cross sections. Statistical analysis Data analysis was performed using Origin software (v8.6-9.0; OriginLab). Group data are represented as mean ± s.e.m. from n cells or fibres or N animals. Tests used for statistical comparison between groups are indicated in the text and figure legends. Whenever parametric statistics were performed, normality was assessed using the Shapiro–Wilk test and variances were checked to ensure that they were similar between groups being compared. If conditions for parametric tests were not met, non-parametric tests were used. The investigators were not blinded to allocation during experiments. Analysis of data from calcium imaging, skin-nerve and behavioural experiments was performed without knowledge of the genotype of the cells, tissues or mice. Data availability Sequencing data that support the findings of this study have been deposited in the NCBI Sequence Read Archive (SRA) and are accessible under the BioProject ID PRJNA431738 . All other relevant data sets are included within the article or are available from the corresponding authors on reasonable request. Accession codes Primary accessions Sequence Read Archive PRJNA431738 Change history 02 May 2018 Please see accompanying Publisher correction ( ). In this Letter, the trace is missing in Fig. 1e. This error has been corrected online.

Researchers at VIB and KU Leuven have uncovered a trio of complementary ion channels in sensory neurons that mediate detection of acute, harmful heat. Having three redundant molecular heat-sensing mechanisms provides a powerful fail-safe mechanism that protects against burn injuries. The seminal findings have been published today in Nature. Although the sensory neurons involved in acute pain signaling in mammals were described more than a century ago, the molecular mechanisms whereby these neurons detect harmful signals have remained largely unresolved. A research team jointly led by prof. Thomas Voets (VIB - KU Leuven) and prof. Joris Vriens (KU Leuven) used genetic knockout models to pinpoint which molecular partners are involved. "We already knew several potential molecular heat sensors, but none of them, when deactivated, resulted in severe loss of acute noxious heat sensing," explains Ine Vandewauw, postdoctoral scientist in the lab of Thomas Voets. The researchers started by eliminating two different heat-activated TRP ion channels, including one known to be also activated by capsaicin, the active component of chili peppers. But this only resulted in very mild deficits in heat sensing. Interestingly, most residual heat-sensitive neurons in the double knockout mice also responded to allyl isothiocyanate, responsible for the pungent sensation of mustard, radish and wasabi. This chemical selectively activates a third TRP channel, which prompted the scientists to go one step further and generate a triple knockout. Mice with all three TRP channels eliminated showed a complete loss of heat-induced pain responses. Reintroduction of the receptors via transient transfection restored sensitivity to heat, and conversely, heat responses could also be suppressed by an inhibitor cocktail for all three TRP channels. The signaling was specific for the pain response to heat, as the animals responded normally to other painful stimuli such as cold, pressure or pinpricks, and their overall thermal preference was not affected. This triple knock out mouse represents the first demonstration in mammals of elimination of the pain response to a physical stimulus at the level of the signal-transducing ion channel. "Acute pain in response to heat is a crucial alarm signal in all mammals," explains Thomas Voets. "The presence of three redundant molecular heat-sensing mechanisms with overlapping expression in pain-sensing neurons creates a powerful fail-safe mechanism. It ensures we avoid dangerous heat, even if one or even two heat sensors are compromised." Next, the researchers want to investigate how these channels can be targeted to treat chronic pain. Thomas Voets says, "Millions of people worldwide suffer from ongoing, burning pain caused for instance by nerve damage or inflammation, and the currently available drugs to treat chronic pain often don't work well or cause addiction. In such conditions, the three heat-activated TRP channels can get deregulated, signaling painful heat even when there is no risk of burning. By developing new drugs that specifically temper the activity of these molecular heat detectors, we hope to be able to provide effective and safe means to treat chronic pain in patients."

nature.com/articles/doi:10.1038/nature26137

Physics

New research unearths obscure and contradictory heat transfer behaviors

Suixuan Li et al, Anomalous thermal transport under high pressure in boron arsenide, Nature (2022). DOI: 10.1038/s41586-022-05381-x Journal information: Nature

https://dx.doi.org/10.1038/s41586-022-05381-x

https://phys.org/news/2022-11-unearths-obscure-contradictory-behaviors.html

Abstract High pressure represents extreme environments and provides opportunities for materials discovery 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 . Thermal transport under high hydrostatic pressure has been investigated for more than 100 years and all measurements of crystals so far have indicated a monotonically increasing lattice thermal conductivity. Here we report in situ thermal transport measurements in the newly discovered semiconductor crystal boron arsenide, and observe an anomalous pressure dependence of the thermal conductivity. We use ultrafast optics, Raman spectroscopy and inelastic X-ray scattering measurements to examine the phonon bandstructure evolution of the optical and acoustic branches, as well as thermal conductivity under varied temperatures and pressures up to 32 gigapascals. Using atomistic theory, we attribute the anomalous high-pressure behaviour to competitive heat conduction channels from interactive high-order anharmonicity physics inherent to the unique phonon bandstructure. Our study verifies ab initio theory calculations and we show that the phonon dynamics—resulting from competing three-phonon and four-phonon scattering processes—are beyond those expected from classical models and seen in common materials. This work uses high-pressure spectroscopy combined with atomistic theory as a powerful approach to probe complex phonon physics and provide fundamental insights for understanding microscopic energy transport in materials of extreme properties. Main Understanding materials properties under high pressure is critical to many areas, from planetary science, geophysics and condensed-matter theory, to new chemical bonding and superconductivity 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 . Over 100 years of literature studies of high pressure thermal transport, from the first experiment 9 , to additional analysis 10 , 11 , 12 , 13 , the lattice thermal conductivity of crystals has been found to increase monotonically with hydrostatic pressure, unless a phase transition takes place. From a theoretical perspective 9 , 14 , the crystal lattice symmetry remains under hydrostatic pressure, and therefore such a universal increase of thermal conductivity can be well understood as a result of volume shrink under pressure that leads to increased atom density and atomic bonding strength. Classical theories 15 , 16 , 17 , 18 , 19 use parameterized models to link with the Debye temperature, bulk modulus and Grüneisen parameters to predict that thermal conductivity must increase with pressure 20 . Although a quantum mechanical picture for this governing rule is missing, it has shown consistency with all experimental reports in the literature so far. Here we measured high-pressure thermal transport in a recently discovered semiconductor material, cubic boron arsenide (BAs), and observed anomalous pressure dependence of the thermal conductivity that has not, to our knowledge, been measured in any other materials 9 , 10 , 11 , 12 , 13 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 (see summary in Fig. 1a ). On the basis of our experimental characterizations, and in combination with ab initio atomistic theory, we attribute this anomalous behaviour to competitive heat conduction channels that are inherent and arise from the unique bandstructure and high-order phonon physics. We verified ab initio calculations and the unique competition mechanisms from interactive high-order anharmonic scattering processes that are beyond understanding using classical theories. Fig. 1: Anomalous thermal conductivity of BAs under high pressure compared to literature studies. a , Summary of thermal conductivities of non-metal crystals as a function of hydrostatic pressure (CsCl 21 , CsBr 21 , NaCl 22 , NaI 22 , quartz 23 , mica 24 , ice 25 , MgSiO 3 26 , diamond 27 , MgO 28 , BN 29 , GaAs 30 ). Results from existing literature indicate a ubiquitous increase in lattice thermal conductivity under a single phase, whereas BAs is an exception. b , Zinc blende crystal structure of cubic BAs. c , High-resolution transmission electron microscopy image showing atomically resolved lattices of BAs. Scale bar, 2 nm. Full size image BAs with high thermal conductivity To experimentally measure the high-pressure transport behaviour, we synthesized high-quality, single-crystal BAs and performed ultrafast pump–probe measurements under high pressure and temperature dependence. Cubic BAs (Fig. 1b ) is a semiconductor that was recently discovered to have an ultrahigh thermal conductivity 31 , 32 , 33 , 34 beyond most common materials and demonstrates high promise for electronics thermal management 35 , 36 . Importantly, these recent theoretical studies indicate that BAs can be an exception and showcase material for higher-order phonon anharmonicity. Ab initio theory expects anomalous thermal transport at high pressure 37 ; however, this has yet to be experimentally explored. Here we synthesized BAs crystals through chemical vapour deposition in a customized three-zone furnace as described in our recent reports 32 , 38 . The high crystal quality of BAs was carefully verified with structural characterizations (Fig. 1c ) and measurement of a room-temperature (298 K) thermal conductivity of approximately 1,300 W m − 1 K −1 (ref. 32 ). Our experimental set-up uses a diamond anvil cell (DAC) to provide high pressure, which was installed in a cryostat chamber for temperature control; pump–probe and optical measurements are applied to measure the sample through the optical window (Fig. 2a ). The DAC can generate tens of GPa of pressure on the loaded BAs crystals, as shown in Fig. 2b . Importantly, we performed synchrotron X-ray diffraction measurements up to 32.6 GPa and verified that BAs crystal remains stable in the cubic phase and that there is no phase transition under all pressures. We use pump–probe measurements based on ultrafast optics (Fig. 2c ) to perform in situ characterizations of the phonon properties under varied pressures and temperatures. The ultrafast set-up is configured for varied measurements, as discussed in the following. Fig. 2: Experimental set-up for in situ ultrafast optical measurements under high pressure. a , Schematic of DAC loaded with the sample for pump–probe measurements. b , BAs crystal in the DAC sample chamber from top view. Scale bar is 100 μm. c , Schematic of ultrafast pump–probe optics set-up, configurable as picosecond laser ultrasonic (PLU), Brillouin scattering, and time-domain thermoreflectance experiments. EOM, electro-optic modulator; BIBO, bismuth borate crystal. Full size image Pressure-dependent phonon bandstructure To understand the anomalous high-pressure behaviour of BAs, we characterize its pressure-dependent evolution of the phonon bandstructure. First, to examine the energy shift of optical phonons, we performed in situ Raman spectroscopy on BAs under pressure up to about 32 GPa. During Raman measurement, the energy of the incident light is absorbed by the crystal lattice and creates lattice vibrations as optical phonons; through energy transfer, the frequency of the scattering light is shifted. The peak position in the Raman spectra corresponds to the optical phonon frequency at the zone centre, owing to energy and momentum conservation. Our measured Raman spectra of BAs crystals under different pressures and the two-dimensional mapping result are shown in Fig. 3a,b , respectively, which clearly verify a shift of the Raman peaks towards higher frequency with increased pressure. Under ambient pressure, BAs exhibits two Raman peaks at 20.9 and 21.6 THz, which arise from the two natural boron isotopes 11 B and 10 B (ref. 32 ). When the pressure increases to approximately 32 GPa, the measurement shows that these two peaks shift to 24.8 and 26.3 THz, respectively. For comparison with the experimental results (symbols), we performed ab initio calculations (lines) of the Raman peak (Fig. 3c ) and found good consistency for the optical band evolution: with the increase of pressure, the interatomic bonding becomes stronger, owing to the smaller interatomic distance, which pushes the atomic vibrations to higher frequencies. In addition, high pressure increases the energy splitting between longitudinal optical (LO) and transverse optical (TO) branches. As predicted by ab initio theory in Fig. 3c , we attribute this to LO–TO splitting due to the long-range Coulomb interaction that increases the frequencies of long-wavelength LO modes. Under high pressure, a greater number of electrons are attracted towards the boron atoms owing to the reduced interatomic distance, which leads to larger LO–TO splitting. The agreement between experiments (symbols) and ab initio calculations (lines) verifies that the optical phonons shift to higher energy and split, indicating an increased phonon bandgap under high pressure. Fig. 3: Experimental measurements and ab initio calculations for the pressure-dependent phonon bandstructure evolution. a , Raman spectra of BAs measured at varied pressures from ambient to 32 GPa. b , Two-dimensional mapping of the pressure-dependent Raman data of BAs. c , Experimental data of the Raman peak frequencies in comparison with ab initio calculations, as a function of pressure. d , Schematic of picosecond laser ultrasonic (PLU) measurement. e , Two-dimensional mapping of the pressure-dependent ultrasonic data along the <111> direction of BAs. f , Experimental data of the longitudinal sound velocity of BAs by PLU and Brillouin scattering measurements, in comparison with ab initio calculations, as a function of pressure. Full size image Next, to examine the energy shift of the acoustic phonon branches of BAs, we performed picosecond laser ultrasonic (PLU) and Brillouin scattering measurements under high pressure and verified the phonon hardening effect. In our PLU experiment 39 (Fig. 3d ), a pump laser generates a picosecond acoustic pulse and the acoustic wave propagates inside the BAs sample at the speed of sound (that is, the acoustic phonon velocity). When the acoustic wave reaches the interfaces of the BAs sample, the wave is partially reflected and creates sound echoes. The echoes are then detected by the following probe laser when returning to the surface. The delay between the pump and probe pulse is the interval time (Δ t echo ) for a round trip of the acoustic wave inside the BAs sample with a thickness d . Therefore, the acoustic phonon velocity ( v ) is calculated by $$v=2d/{\Delta t}_{{\rm{e}}{\rm{c}}{\rm{h}}{\rm{o}}}$$ (1) Figure 3e shows our measurement data of a two-dimensional mapping of sound echoes under varied pressures and ultrafast time dependence with picosecond resolution. The two echoes are represented as a high-amplitude signal (yellow). The first echo corresponds to reflection by the interface between the top surface of BAs crystal and a thin-film metal transducer (deposited on top of BAs) 39 . The second echo corresponds to reflection by the bottom surface of BAs (Fig. 3d ); Δ t echo represents the time interval between these two echoes. As shown in Fig. 3e , a clear reduction of Δ t echo , and thus an increase of the sound velocity, is measured with increased pressure. We have also measured the Brillouin frequencies under pressures that provide consistent sound velocities. The sound velocity is measured as 8,150.0 m s −1 (longitudinal phonons along the <111> direction) under ambient pressure, and increases to 9,209.4 m s −1 at 29.3 GPa. Quantitatively, the pressure-dependent sound velocity from our PLU measurement (squares) and Brillouin scattering measurement (triangles) are plotted in Fig. 3f , together with our ab initio calculations (line). The experimental and ab initio theoretical results are in good agreement, verifying the phonon hardening and evolution of acoustic phonon branches under high pressure. To directly measure the evolution of the phonon bandstructure, we performed an inelastic X-ray scattering (IXS) experiment on BAs under high pressure. The IXS spectrum across the reciprocal vector space is characterized by scattering peaks that determine the corresponding phonon energy. In Fig. 4a , IXS experimental results (circles) of BAs phonon dispersion under hydrostatic pressures (ambient pressure, 18 GPa, and 32 GPa), are plotted together with the Raman measurement at the centre of the Brillouin zone (triangles), which clearly verify the ab initio calculation (lines). Fundamentally, the phonon bandstructure of BAs evolves under increased pressure with several key features (Fig. 4a ): (i) Optical phonon branches move up resulting in an increased bandgap. For example, the optical phonon frequency at the centre of the Brillouin zone (Γ point) increases from 20.9 to 24.8 THz, and the energy gap between the acoustic and optical phonon branches (X point) is measured to increase from 9.2 to 12.8 THz with higher pressure. (ii) For acoustic branches, phonon hardening with increased acoustic sound velocity is expected under pressure, thus expanding their total energy space. For example, the longitudinal acoustic velocity has a substantial increase, measured from 8,150.0 to 9,277.1 m s −1 , consistent with theory, as shown in Fig. 4a . This leads to a larger energy difference between different acoustic branches 31 under higher pressure: the longitudinal acoustic phonon frequency (K point) is measured to increase from 8.8 to 10.1 THz with higher pressure. Fig. 4: Experimental measurements and ab initio theory for phonon dispersion and thermal conductivity as a function of pressure and temperature. a , Experimentally measured phonon dispersion from inelastic X-ray scattering (IXS; circles) and Raman spectroscopy (triangles) overlaid on the ab initio calculations (lines) under hydrostatic pressures of ambient, 18 GPa and 32 GPa. 3ph and 4ph indicate three-phonon and four-phonon processes, respectively. b , Phonon scattering physics: schematic examples for three-phonon scattering (combination process) and four-phonon scattering (redistribution process). c , Typical time-domain thermoreflectance data: thermal reflectance phase signal versus time for BAs under ambient pressure (black circles), 8.5 GPa (red circles) and 16.5 GPa (blue circles), fitted by thermal transport model (solid lines). To illustrate the measurement accuracy, the dashed lines indicate ±10% changes in the thermal conductivity values. d , Measurement results of thermal conductivity for varied pressures (0–30 GPa) and temperatures (200–300 K). Dashed lines are theoretical results from ab initio calculations. The shaded background regions represent measurement uncertainty. Full size image Competition of three- and four-phonon processes Such phonon bandstructure evolution under high pressure will affect all heat conduction channels that contribute to the thermal conductivity, but their pressure-dependent effects differ, owing to varied fundamental phonon processes. By modern atomistic theory, thermal energy in semiconductors and insulators is mainly carried by lattice vibrations and thermal conductivity is calculated based on scattering between different phonon modes, that is, anharmonicity 14 . For each scattering event, multiple phonons can participate, and the three-phonon processes represent the lowest-order anharmonicity (Fig. 4b ). For the past several decades, the thermal conductivity of solids has been usually considered in terms of the three-phonon scattering processes; the effects of higher-order anharmonicity such as the four-phonon processes were found to be negligible for most materials, and their low scattering probability was assumed by past literature according to the momentum and energy conservation requirements 14 , 40 , 41 . However, owing to the unique phonon bandstructure of BAs, both three- and four-phonon processes can be important at room temperature and make interactive contributions to thermal conductivity 37 , 42 , 43 . We also note that BAs has a unique large bandgap (approximately 10 THz at ambient pressure and further increases at higher pressure) between the acoustic (a) and optical (o) phonon branches. Such a large a–o bandgap makes optical phonons inaccessible for most three-phonon processes 31 , but critically involved for most four-phonon processes, as dictated by the energy and momentum conversation laws. Now a paradox exists in the consideration of high-pressure effects on thermal conductivity, owing to contributions from these competitive conduction channels. On one hand, for the three-phonon processes only: high pressure increases the energy separation (expands the total energy spreading) of the acoustic branches, and so ramps up the scattering phase space and thus increases the probability for three-phonon processes (that is, exclusive aaa processes in BAs) 31 . On the other hand, in consideration of the four-phonon processes only: high pressure increases the a–o bandgap and suppresses four-phonon scattering, which mostly involves optical phonons (for example, aaoo or aaao processes), as energy matching across a larger a–o bandgap becomes more challenging. Together, the effects of high pressure on BAs crystals affect different conduction channels in an interactive and competitive way, which is very different from common materials where only the three-phonon processes dominate (four-phonon processes are negligible) and the a–o bandgap is small. Here we consider high pressure as a tuning knob to study such interactions: in BAs, the strength of these different processes should vary with pressure—that is, for different pressure ranges, the dominant contribution channel to thermal conductivity varies. Therefore, with the unique competition mechanisms between three-phonon and four-phonon processes, ab initio theory serves as a powerful tool with which to capture the details of phonon scattering and predicts that the evolution under pressure can lead to anomalous pressure dependence of thermal conductivity 37 , 43 . Pressure and temperature dependence To quantitatively verify such interactive behaviours, we measured the thermal conductivity of BAs crystals as a function of pressure and temperature and compared the results with ab initio theory calculations. For thermal conductivity measurement, the ultrafast set-up is configured as time-domain thermoreflectance (TDTR) 32 , 35 . TDTR is well suited for measuring samples in a DAC under high pressure as no physical contact with the sample is required and the measurement can provide high spatial resolution, down to the micrometre size. The thermal conductivity of BAs at varied pressures from the ambient to approximately 30 GPa and temperatures were carefully measured. Figure 4c shows typical experimental data in which the ultrafast signal (that is, the time-dependent phase decay) was fitted to the thermal diffusive model to determine the thermal conductivity 32 , 35 . Our experimental results are presented in Fig. 4d and indeed show an anomalous non-monotonic trend with dependence on both pressure and temperature. At 300 K, the measurement shows that the thermal conductivity first monotonically increases from around 1,300 W m −1 K −1 at ambient pressure to approximately 1,500 W m −1 K −1 at 16.5 GPa, indicating that in this pressure range, the effect of the increased a–o bandgap on suppressing four-phonon processes dominates over the separation of acoustic branches on increasing three-phonon processes. From 16.5 to 29.3 GPa, the thermal conductivity of BAs is measured to monotonically decrease, indicating that in this pressure range, the four-phonon processes are weak and further suppressed by an increased a–o bandgap, whereas the increased energy separation of different acoustic branches enhances the three-phonon processes to dominate the thermal conductivity trend. We note that such non-monotonic pressure dependence exists for our measurement results at all different temperatures (Fig. 4d ), and the transition pressure (that is, for local maximum conductivity) shifts to a lower pressure (from 16.5 to 7.0 GPa) when temperature is reduced from 300 to 200 K. This is because at low temperature a greater number of phonons occupy lower energy states, making it less possible to access high-energy optical phonons across the a–o bandgap, which suppresses the four-phonon processes and the trend of the pressure dependence becomes determined by the three-phonon processes. Indeed, the experimental results (symbols) agree well with ab initio theory (dashed line; Fig. 4d ), and clearly verify that the atomistic origin of such anomalous pressure-dependent thermal conductivity results from the unique competition between three-phonon and four-phonon high-order anharmonic processes 37 . Conclusions and outlook Our study shows that the general rule of monotonic pressure dependence fails when the lowest-order interactions no longer dominate in energy transport—and these phase spaces could emerge under extreme conditions for many other materials systems. Therefore, we expect this study not only provide a benchmark result to revise current understanding, but also could impact established modelling predictions for extreme conditions such as the Earth’s interior when direct measurements are not possible, or standard techniques such as shockwave studies when reliable modelling of the window materials is critical. In addition, the discovery of novel materials with non-monotonic thermal conductivity bandwidth could enable innovative design for pressure-adapted thermal windows or thermal switches to control energy flow under high pressure. In summary, we have reported the observation of anomalous pressure-dependent thermal conductivity that breaks the generic rule for high-pressure heat conduction established over the past 100 years. Our study developed high-pressure transport experiments in combination with ab initio theory to provide microscopic understanding over the complex phonon physics. We measured the evolution of optical and acoustic phonon bands and thermal transport of BAs using Raman spectroscopy, PLU, Brillouin scattering, inelastic X-ray scattering and TDTR techniques up to 32 GPa under varied temperatures. Our experimental results show good agreement with ab initio calculations and revealed competing contributions to BAs thermal conductivity from the interactive three-phonon and four-phonon processes. This study provides fundamental insights into understanding microscopic energy transport in materials of extreme properties and opens up new opportunities in designing novel structures. Methods Chemical synthesis of cubic BAs crystals Cubic BAs crystals were synthesized by the chemical vapour transport method. High-purity boron and arsenic powders (99.9999%, Alfa Aesar) were ground using a mortar and pestle with a stoichiometric ratio of 1:2. The reaction sources were introduced into a quartz tube followed by evacuating under high vacuum (10 −5 torr) before flame-sealing. The quartz tube was placed into a customized three-zone reaction furnace with a 1,083 K hot zone, 1,058 K centre zone and a 1,033 K cold zone for the growth, before slowly cooling down to room temperature. Further details on the chemical synthesis of BAs can be found in our recent reports 32 , 38 . Structural characterization of BAs Transmission electron microscope (TEM) samples were prepared using a focused ion beam (FIB) machine (Nova 600, FEI) before transferred to the TEM holder (PELCO FIB Lift-Out, Ted Pella). The high-angle annular dark-field (HAADF) image was obtained by using aberration-corrected high-resolution scanning TEM (Grand ARM, JEOL, 300 kV). The atomic-resolution TEM image was processed with Gatan software. The optical image of BAs sample inside the high-pressure DAC set-up was obtained by the CCD-camera mounted on an optical microscope (Leica DM 4000M) with a 5× objective lens. High-pressure set-up A piston-cylinder-type DAC was used to generate ultrahigh pressure. BAs samples were loaded into the DAC with 400-μm diameter culet diamonds during the study. Rhenium gaskets were pre-indented to a thickness of roughly 60 μm followed by an approximately 180-μm-diameter aperture drilled at the centre. Silicone oil was used as the pressure-transmitting medium to apply the hydrostatic pressure uniformly to the sample. Ruby spheres were used to carefully calibrate and determine the pressure using its R 1 fluorescence peaks 44 with a pressure uncertainty less than 0.2 GPa. All the measurements including Raman spectroscopies, inelastic X-ray scattering, synchrotron X-ray diffraction, and thermal and phonon transport measurements were performed on samples inside the DAC. Raman spectroscopy Raman measurements were performed under high pressure using a confocal micro-Raman system (inVia, Renishaw) equipped with 633-nm laser excitation and 1,200 mm −1 grating. The polarized laser has backscattering geometry with a Leica DM2500 optical system. A ×50/0.75 objective lens was used with lateral spatial resolution of 0.5 μm. For surface-enhanced Raman measurement, gold nanoparticles (~5 nm) were used on the BAs samples to enhance the signal resolution. Synchrotron X-ray experiments Synchrotron X-ray experiments were performed at the Advanced Light Source (ALS) at Lawrence Berkeley National Laboratory and the Advanced Photon Source (APS) at Argonne National Laboratory to understand the crystal structure and phonon dispersion of BAs under high pressure. Pressure-dependent X-ray diffraction (XRD) experiments were performed to verify the crystal phase structure of BAs on the beamline 12.2.2 of the Advanced Light Source at Lawrence Berkeley National Laboratory, with a focused monochromatic X-ray beam at 30 keV (ref. 45 ). The synchrotron XRD measurements from ambient pressure to 32.6 GPa (Extended Data Fig. 1a ) show that the diffraction peaks were identified and indexed to the cubic diffraction facets of the BAs crystal and no additional or disappearing peaks are observed, verifying the single cubic phase for all the pressures. The pressure-dependent lattice constant of BAs was determined from our synchrotron XRD measurements and is shown in Extended Data Fig. 1b (circles), in good agreement with ab initio theory (line). In addition, single-crystal XRD experiments (Extended Data Fig. 2 ) were conducted at both ALS and APS. Pressure-dependent IXS measurements were conducted to directly measure the phonon dispersion of BAs. The IXS measurements were performed on beamline 30-ID of the Advanced Photon Source at Argonne National Laboratory 46 , 47 . The experiment was carried out under a transmission geometry with a high energy resolution at ~1.5 meV by the monochromatic synchrotron-based IXS technique with the wavelength of the incident X-ray at 23.7 keV. The size of the beam was 10 μm in diameter. Helium gas was used as the pressure-transmitting medium 48 . The IXS spectrum is collected at varied Q-points characterized by scattering peaks that determine the corresponding phonon energy (Extended Data Fig. 3 ). The energy gap between the optic and acoustic branches is directly measured and shows a monotonic increase from 9.2 THz to 12.8 THz when the pressure increases from ambient to 32 GPa (Fig. 4a ). The IXS measurements verify the ab initio theory prediction of the phonon bandstructure evolution and the single cubic structure of BAs under high pressure. In situ thermal and phonon transport measurements Thermal and phonon transport experiments were performed in situ under varied high pressures using ultrafast optical techniques, including TDTR, PLU and Brillouin scattering. The TDTR technique is a non-contact optical method to measure thermal conductivity 32 , 35 , 49 , 50 , 51 , 52 , 53 . A mode-locked Ti:sapphire laser with a repetition rate of 80 MHz generates ultrafast femtosecond pulses and is split into a pump and a probe beam. Using a second harmonic generator, the pump beam is fixed at 400 nm by doubling its frequency, and the probe beam stays at 800 nm. A thin metal film is coated on the sample surface, which absorbs the pump laser energy to heat up the sample and generate an instantaneous temperature rise. A mechanical delay stage is used to change the arrival time of the probe beam. The surface temperature is detected by the probe beam. The transient temperature decay is continuously recorded versus the delay time between pump and probe with a subpicosecond temporal resolution. The temperature decay is fitted with a thermal model to determine the thermal conductivity of the sample. We have used TDTR to measure thermal conductivity of a wide range of materials including BAs 32 , 35 , 49 , 50 , 51 , 52 , 53 . The PLU technique 39 uses a pump beam to generate acoustic pulses propagating from the surface inside the sample and a probe beam to measure the round-trip time of the acoustic pulses across the sample. Sound echoes are detected and used to obtain sound velocity under varied high pressures. Ultrafast pump–probe optics was applied to measure the Brillouin frequency following literature settings 54 , 55 , 56 , 57 , 58 , 59 : the pump pulses launch a strain wave that propagates inside BAs, and the probe pulse is partially reflected at the acoustic strain owing to the local change of refractive index (Brillouin scattering) and measured as a function of delay time. The time-dependent probe signal carries Brillouin scattering oscillations (Extended Data Fig. 4 ) due to constructive or destructive interference with the travelling strain pulse, and the Brillouin frequency is determined through fast Fourier transform analysis. Ab initio theory for lattice dynamics More calculation details for ab initio phonon theory 27 , 31 , 32 , 39 , 42 , 43 , 60 , 61 , 62 , 63 , 64 and consistent calculation results 37 can be found in the recent papers. The phonon dispersion, sound velocity and lattice thermal conductivity are calculated using ab initio methods based on density functional theory. The phonon dispersion is determined through diagonalization of the dynamical matrix and the thermal conductivity is determined by solving the phonon Boltzmann transport equation through self-consistent iterations. The lattice vibrations are determined by the interatomic potential. The interatomic potential U can be expanded with respect to atomic displacement. $$\begin{array}{c}U={U}_{0}+\frac{1}{2}\sum _{\{l,b,\mu \}}{\Phi }_{{\mu }_{1}{\mu }_{2}}({l}_{1}{b}_{1};{l}_{2}{b}_{2}){u}_{{\mu }_{1}}({l}_{1}{b}_{1}){u}_{{\mu }_{2}}({l}_{2}{b}_{2})\\ \,+\frac{1}{3!}\sum _{\{l,b,\mu \}}{\Phi }_{{\mu }_{1}{\mu }_{2}{\mu }_{3}}({l}_{1}{b}_{1};{l}_{2}{b}_{2};{l}_{3}{b}_{3}){u}_{{\mu }_{1}}({l}_{1}{b}_{1}){u}_{{\mu }_{2}}({l}_{2}{b}_{2}){u}_{{\mu }_{3}}({l}_{3}{b}_{3})\\ \,+\frac{1}{4!}\sum _{\{l,b,\mu \}}{\Phi }_{{\mu }_{1}{\mu }_{2}{\mu }_{3}{\mu }_{4}}({l}_{1}{b}_{1};{l}_{2}{b}_{2};{l}_{3}{b}_{3};{l}_{4}{b}_{4}){u}_{{\mu }_{1}}({l}_{1}{b}_{1}){u}_{{\mu }_{2}}({l}_{2}{b}_{2}){u}_{{\mu }_{3}}({l}_{3}{b}_{3}){u}_{{\mu }_{4}}({l}_{4}{b}_{4})\\ \,+\,...\,\end{array}$$ (2) where U 0 is the equilibrium potential and the summation is performed over all numbered indices. u μ ( lb ) denotes the atomic displacement of the b th atom in the l th unit cell from its equilibrium position along the μ direction. The first-order derivatives are zero as they are calculated at equilibrium. \({\varPhi }_{{\mu }_{1}{\mu }_{2}}({l}_{1}{b}_{1};{l}_{2}{b}_{2})\) are the second-order interatomic force constants (IFCs). From the second-order IFCs, the phonon frequency of given wave vectors q can be calculated by diagonalizing the dynamical matrix, and the group velocity of the phonon mode λ can be calculated from v λ = ∂ ω λ /∂ q . The sound velocity is the group velocity at the Brillouin zone centre. \({\varPhi }_{{\mu }_{1}{\mu }_{2}{\mu }_{3}}({l}_{1}{b}_{1};{l}_{2}{b}_{2};{l}_{3}{b}_{3})\) and \({\varPhi }_{{\mu }_{1}{\mu }_{2}{\mu }_{3}{\mu }_{4}}({l}_{1}{b}_{1};{l}_{2}{b}_{2};{l}_{3}{b}_{3};{l}_{4}{b}_{4})\) are the third- and fourth-order IFCs, respectively, which are used to determine the three- and four-phonon scattering rates, respectively. The lattice thermal conductivity tensor is given as $${\kappa }^{\alpha \beta }=\frac{1}{N}\sum _{\lambda }{C}_{\lambda }{v}_{\lambda }^{\alpha }{F}_{\lambda }^{\beta }$$ (3) where λ ≡ ( q , s ) labels a phonon mode with wavevector q and polarization s . α and β represent the direction in the Cartesian coordinate system. N is the number of q -points in the mesh of the Brillouin zone. C λ and \({v}_{\lambda }^{\alpha }\) are the volumetric specific heat and the group velocity along the α direction of phonon mode λ , respectively. \({F}_{\lambda }^{\alpha }\) measures the deviation of the phonon distribution ( n λ ) from the equilibrium ( \({n}_{\lambda }^{0}\) ) as driven by the temperature gradient ∇ T , \({n}_{\lambda }={n}_{\lambda }^{0}+(-\partial {n}_{\lambda }^{0}/\partial T){{\bf{F}}}_{\lambda }\cdot \nabla T\) . For most materials, the deviation of n λ will not be large, and so the scattering rates \({\tau }_{\lambda }^{-1}\) for each individual phonon mode λ can be calculated by keeping background phonons in equilibrium, and in this case F λ = τ λ v λ . But for high thermal conductivity materials, n λ is usually driven far away from equilibrium by temperature gradient, so the deviation of n λ for all the phonon modes should be simultaneously considered by determining F λ from the Boltzmann transport equation through self-consistent iteration. In the Boltzmann transport equation, a phonon flux is driven by temperature gradient and balanced by phonon scatterings, $${{\bf{v}}}_{\lambda }\cdot {\rm{\nabla }}T\frac{{\rm{\partial }}{n}_{\lambda }}{{\rm{\partial }}T}={\left(\frac{{\rm{\partial }}{n}_{\lambda }}{{\rm{\partial }}t}\right)}_{\text{scattering}}$$ (4) The scattering term on the right-hand side of equation ( 3 ) describes the rate of change of n λ owing to phonon scatterings, and is the summation of different types of scattering processes, $${\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{scattering}}={\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{3ph}}+{\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{4ph}}+{\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{iso}}$$ (5) where \({\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{3ph}}\) , \({\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{4ph}}\) and \({\left(\frac{\partial {n}_{\lambda }}{\partial t}\right)}_{\text{iso}}\) represent the rate of change of n λ due to three-phonon, four-phonon and isotope scattering, respectively, and can be determined by quantum perturbation theory 65 . The only input of this ab initio approach is the IFCs calculated from density functional theory 66 , 67 . The procedure to obtain the IFCs is as follows. First, an irreducible displacement set is generated based on a real-space 256-atom supercell. For each displacement configuration, the forces acting on atoms were calculated by density functional theory using the Quantum ESPRESSO package 68 , 69 , and on that basis we build up force–displacement sets. Then the IFCs are extracted by fitting the displacement-force sets using the ALAMODE package 70 . All the density functional theory calculations in this work are based on the projector-augmented wave approach. We use PZ pseudopotentials (B.pz-n-kjpaw_psl.0.1.UPF and As.pz-n-kjpaw_psl.0.2.UPF from ) under a local density approximation to determine the IFCs and use PBEsol pseudopotentials (B.pbesol-n-kjpaw_psl.1.0.0.UPF and As.pbesol-n-kjpaw_psl.1.0.0.UPF from ) under a generalized gradient approximation to determine the theoretical lattice constants in Extended Data Fig. 1b . The Monkhorst–Pack grids for electronic structure calculations are 6 × 6 × 6 and 2 × 2 × 2 for an eight-atom unit cell and 256-atom supercell, respectively. The convergence threshold for self-consistency is 10 −11 . The kinetic energy cut-offs for electronic wavefunctions are 120 Ry and 100 Ry for the PZ and PBEsol pseudopotentials, respectively. Data availability The data that support the plots within this paper and findings of this study are available from the corresponding author upon request.

UCLA researchers and their colleagues have discovered a new physics principle governing how heat transfers through materials, and the finding contradicts the conventional wisdom that heat always moves faster as pressure increases. Up until now, the common belief has held true in recorded observations and scientific experiments involving different materials such as gases, liquids and solids. The researchers detailed their discovery in a study published last week by Nature. They have found that boron arsenide, which has already been viewed as a highly promising material for heat management and advanced electronics, also has a unique property. After reaching an extremely high pressure that is hundreds of times greater than the pressure found at the bottom of the ocean, boron arsenide's thermal conductivity actually begins to decrease. The results suggest that there might be other materials experiencing the same phenomenon under extreme conditions. The advance may also lead to novel materials that could be developed for smart energy systems with built-in "pressure windows" so that the system only switches on within a certain pressure range before shutting off automatically after reaching a maximum pressure point. "This fundamental research finding shows that the general rule of pressure dependence starts to fail under extreme conditions," said study leader Yongjie Hu, an associate professor of mechanical and aerospace engineering at the UCLA Samueli School of Engineering. "We expect that this study will not only provide a benchmark for potentially revising current understanding of heat movement, but it could also impact established modeling predictions for extreme conditions, such as those found in the Earth's interior, where direct measurements are not possible." According to Hu, the research breakthrough may also lead to retooling of standard techniques used in shock wave studies. Thermal conductivity measured from in-situ spectroscopy experiment showing the activity slowing down under high pressure. Credit: The H-Lab/UCLA Similar to how a sound wave travels through a rung bell, heat travels through most materials by way of atomic vibrations. As pressure squeezes closer together the atoms inside a material, it enables heat to move through the material faster, atom by atom, until its structure breaks down or transforms to another phase. That is not the case, however, with boron arsenide. The research team observed that heat started to move slower under extreme pressure, suggesting a possible interference caused by different ways the heat vibrates through the structure as pressure mounts, similar to overlapping waves canceling each other out. Such interference involves higher-order interactions that cannot be explained by textbook physics. The results also suggest that the thermal conductivity of minerals can reach a maximum after a certain pressure range. "If applicable to planetary interiors, this may suggest a mechanism for an internal 'thermal window'—an internal layer within the planet where the mechanisms of heat flow are different from those below and above it," says co-author Abby Kavner, a professor of earth, planetary and space sciences at UCLA. "A layer like this may generate interesting dynamic behavior in the interiors of large planets." To achieve the extremely high-pressure environment for their heat-transfer demonstrations, the researchers placed and compressed a boron arsenide crystal between two diamonds in a controlled chamber. They then utilized quantum theory and several advanced imaging techniques, including ultrafast optics and inelastic X-ray scattering measurements, to observe and validate the previously unknown phenomenon. Mechanical engineering graduate students Suixuan Li, Zihao Qin, Huan Wu and Man Li from Hu's research group are the study's co-lead authors. Other authors are Kavner, Martin Kunz of Lawrence Berkeley National Laboratory and Ahmet Alatas of Argonne National Laboratory.

10.1038/s41586-022-05381-x

Chemistry

New material created to clean up fossil fuel industry

Zichun Wang et al. Acidity enhancement through synergy of penta- and tetra-coordinated aluminum species in amorphous silica networks, Nature Communications (2020). DOI: 10.1038/s41467-019-13907-7 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-019-13907-7

https://phys.org/news/2020-02-material-fossil-fuel-industry.html

Abstract Amorphous silica-aluminas (ASAs) are widely used in acid-catalyzed C-H activation reactions and biomass conversions in large scale, which can be promoted by increasing the strength of surface Brønsted acid sites (BAS). Here, we demonstrate the first observation on a synergistic effect caused by two neighboring Al centers interacting with the same silanol group in flame-made ASAs with high Al content. The two close Al centers decrease the electron density on the silanol oxygen and thereby enhance its acidity, which is comparable to that of dealuminated zeolites, while ASAs with small or moderate Al contents provide mainly moderate acidity, much lower than that of zeolites. The ASAs with enhanced acidity exhibit outstanding performances in C–H bond activation of benzene and glucose dehydration to 5-hydroxymethylfurfural, simultaneously with an excellent calcination stability and resistance to leaching, and they offer an interesting potential for a wide range of acid and multifunctional catalysis. Introduction Silica-alumina materials, particularly crystalline zeolites and amorphous silica-aluminas (ASAs), are among the most popular solid acids that have been widely commercialized as efficient and environmentally friendly catalysts in the petrochemical industry 1 , and in bio-refinery 2 . These materials can provide Brønsted acid sites (BAS) with tunable density and strength, which facilitates the optimization of the surface acidity to promote a series of important industrial chemical reactions, through the formation of surface complexes or transition states by proton transfer from BAS to reactants 3 , 4 , 5 , 6 , such as to initialize C–H activation for hydrocarbon conversions 7 , 8 , 9 , 10 , 11 , 12 . Zeolites with strong Brønsted acidity, are of increasing importance in various sustainable processes, in the fields of biomass conversion, CO 2 capture and conversion, air-pollution remediation, and water purification 13 . For instance, zeolites can efficiently catalyze the redox disproportionative conversion of biomass-derived sugars into α-hydroxy acids 14 , and they are more active than ASA catalysts, albeit the latter facilitate improved molecular diffusion in the porous network 15 . The lower performance of ASA in many catalytic applications is widely attributed to their moderate Brønsted acidity 1 , which fostered recent works on the discovery of ASAs with increased Brønsted acidity 5 , aiming at expanding their applications in a broad range of fields. The formation of BAS in silica-aluminas is based on aluminum centers distributed in the silica framework or network 6 , 16 , 17 , 18 , 19 , 20 , as (i) a tetra-coordinated aluminum species (Al IV ), replacing a Si 4+ atom in the zeolite framework, can introduce a negatively charged framework oxygen to be balanced by a proton, resulting in bridging OH groups 6 , 17 , 18 , and (ii) the interaction between Al IV atoms and neighboring silanols in the silica network can provide acidic OH groups, acting as BAS in ASA 5 , 21 , 22 , 23 , 24 . In crystalline zeolites, it is well accepted that increasing the Si/Al ratio can enhance the BAS strength by increasing the overall electronegativity 25 . However, the amount of BAS (e.g. bridging SiOHAl groups) is then significantly reduced, owing to the fewer Al sites in the framework. Alternatively, the introduction of extra-framework cations, such as Al 3+ and La 3+ , via dealumination or ion exchange can significantly improve the BAS strength due to a synergistic effect between Lewis acid sites (LAS) and nearby BAS 4 , 17 . Those solid acids have been widely applied in gas-phase cracking, such as fluid catalytic cracking processes 26 . However, extra-framework cations can easily leach out from solid acids during the liquid-phase reactions 27 . Additionally, the synergy between two nearby Al sites in the zeolite framework is impossible due to the absence of Al-O-Al linkage based on Löwenstein’s rule 28 . Enhancing the BAS strength in ASAs still remains a significant challenge. Although a surface bridging SiOHAl model has been proposed for the formation of BAS on ASA 21 , 29 , the strength of the BAS on ASA is generally lower than that on crystalline zeolites 1 , since the amorphous structure of ASAs weakens the Al-O bonds (2.94–4.43 Å) 24 , compared to those in the crystalline zeolite framework (1.88–2.0 Å) 30 . Two models have been proposed to account for BAS generation in ASA: (i) a flexible coordination between the Al atom and the neighboring silanol oxygen atom 5 , 23 , and (ii) a pseudo-bridging silanol (PBS) with a nearby Al atom 24 , 31 , 32 . In both models, it was proposed that a Lewis acidic Al center interacts with a nearby silanol group, withdrawing electron density from the hydroxyl O atom to enhance the acid strength of the hydroxyl proton. Notably, these models mainly account for the formation of moderate BAS. Nevertheless, the presence of an additional Al species in the vicinity of the SiOH site may further enhance the acid strength of ASA via a synergistic effect, which has not yet been evidenced to the best of our knowledge. In this work, the synergy between Al species in the ASA network has been studied using solid-state NMR spectroscopy and atom probe tomography (APT). The 27 Al double-quantum single-quantum (DQ-SQ) through-space homonuclear correlation (D-HOMCOR) NMR experiments allow us to probe 27 Al- 27 Al proximities by applying recoupling sequences that restore the dipolar interaction between neighboring 27 Al spins 33 , 34 , 35 , 36 . Unlike for crystallized materials, the investigation of the location and distribution of Al atoms is impossible in ASAs by routine characterization methods, such as X-ray diffraction (XRD) and high-resolution transmission electron microscopy (HRTEM), due to their amorphous structure. Atom probe tomography (APT) can provide quantitative three-dimensional (3D) information on elemental distributions in catalyst nanoparticles at the atomic scale 37 , 38 . It has recently been employed with sub-nanometer-scale resolution on zeolite-based catalysts to establish their structure–composition–property relationships 39 , 40 , 41 , 42 , 43 . Here, we apply APT for the same purpose in ASAs 44 , 45 , 46 , 47 . The combined investigations of APT on Al distribution and 27 Al and 1 H DQ-SQ NMR experiments reveal, for the first time, the existence of a synergy between Al species in the ASA network. This synergy can significantly enhance the acid strength of ASA, as demonstrated by the H/D exchange with deuterated benzene. The beneficial effect of the enhanced acidity is demonstrated by an example of the liquid-phase catalytic dehydration of glucose to 5-hydroxymethylfurfural, which is an important building block in the production of various valuable chemicals, such as liquid alkanes, biofuels, and furan derivatives 48 . Results Local structure of SA/10 The ASA materials were prepared by flame spray pyrolysis as described in Supplementary Methods and they are designated as SA/x, where x = 10 or 50 represents the percentage of Al atoms with respect to the total amount of Al and Si atoms in the precursor. APT has been applied to show the distribution of Al within the SA/10 as shown in Supplementary Fig. 1 and Supplementary Movie 1 in the Supplementary Information . The tomographic reconstruction qualitatively shows a homogeneous distribution of Al, Si and O (Supplementary Fig. 1b, c, d ), where each sphere represents the 3D position of an individual atom. In agreement with recent energy-dispersive X-ray atom mapping investigations 49 , the APT reveals a homogeneous dispersion of Al species in the silica network, similar to that observed in well-developed crystalline zeolites 42 , 43 . Due to the lack of evidence for the existence of zeolite-like bridging OH groups in ASA, Al atoms in SA/10 can be expected either bridged to SiO (SiOAl) or located nearby the silanol groups (SiO(H)∙∙∙Al), as reported in literature 16 , 21 . Supplementary Fig. 2a displays the single-pulse 1D 27 Al MAS spectrum of SA/10. It exhibits three peaks at 50, 30, and 4 ppm, assigned to tetra- (Al IV ), penta- (Al V ) and hexa- (Al VI ) coordinated Al sites, respectively 16 , 50 . In the DQ-SQ 2D spectra, as shown in Fig. 1 , autocorrelation diagonal peaks resonating at the frequencies (2 ν , ν ) along the indirect and direct dimensions, respectively, indicate the proximities between nuclei with identical isotropic shifts, ν , whereas the cross-peak pairs resonating at frequencies ( ν a + ν b , ν a ) and ( ν a + ν b , ν b ) specify proximities between nuclei resonating at distinct frequencies, ν a and ν b , in the 1D spectra. The 27 Al DQ-SQ 2D spectrum (Fig. 1a ) exhibits a weak diagonal peak at (60, 30) ppm, which indicates proximities between Al V sites in SA/10. No other obvious 27 Al correlation signal could be detected, which was confirmed by the 27 Al slices that were extracted at the corresponding shifts of these correlation signals (Supplementary Fig. 2e–i ). The 1 H MAS spectrum of SA/10 shown in Supplementary Fig. 2b exhibits a single peak resonating at 1.9 ppm, which is ascribed to the silanol protons. The 1 H DQ-SQ 2D spectrum shows a single strong signal at (3.8, 1.9) ppm (Fig. 1b ). This peak indicates the spatial proximity of many silanol groups on the surface. Fig. 1: DQ-SQ D -HOMCOR 2D NMR spectra of SA/10. The spectra of subfigures a , c and b , d are for 27 Al and 1 H NMR spectra, respectively, acquired at 18.8 T with a MAS frequency of ν R = 20 kHz. The spectra of subfigures a and b were acquired for SA/10 dehydrated at 723 K for 12 h under vacuum, whereas the spectra of subfigures c and d were dehydrated and loaded with ammonia, and then evacuated at 393 K for 1 h to remove the weakly physisorbed molecules. Full size image The presence of BAS has been further confirmed by the adsorption of a basic molecular probe, ammonia. The 1 H spectrum of SA/10 loaded with ammonia (see Supplementary Fig. 2d ) exhibits a peak at δ 1H = 6.7 ppm, which is ascribed to ammonium ions 6 . The presence of an auto-correlation peak at (13.4, 6.7) ppm in the 1 H DQ-SQ spectrum of Fig. 1d indicates that the ammonium protons are dipolar coupled and hence, do not exhibit isotropic motion in the sample. The cross-peaks at (8.6, 1.9) and (8.6, 6.7) ppm in the same spectrum demonstrate mainly that some ammonia ions stay adsorbed near the SiOH groups. The peak at δ 1H = 2.6 ppm is assigned to ammonia adsorbed at LAS 51 , 52 , 53 , 54 . The autocorrelation peak at (5.2, 2.6) ppm in Fig. 1d indicates that the protons of adsorbed ammonia are dipolar coupled and hence that these molecules do not exhibit isotropic motions. Conversely, the 27 Al 1D (Supplementary Fig. 2c ) and 2D D -HOMCOR (Fig. 1c ) spectra are not significantly modified by the adsorption of ammonia. In particular, we only detect a weak Al V diagonal peak in the 2D D -HOMCOR spectrum, and no other correlation signal could be detected as observed with dehydrated SA/10 (Fig. 1a ). The 13 C NMR signal of SA/10 loaded with CH 3 13 COCH 3 probe molecule resonates at δ 13C = 213 ppm 5 . This value is similar to those observed in most ASA samples 22 , but is much smaller than that of zeolite H-ZSM-5 ( δ 13C = 223 ppm) 55 . This chemical shift δ 13C value is commonly utilized to evaluate the strength of acid sites in solid acids, e.g., a larger δ 13C value indicates a higher acid strength 6 . Hence, the BAS of SA/10 exhibits a moderate acidity. These sites have been described as SiOH groups in the proximity of one Al IV or Al V site 16 . Local structure of SA/50 The APT reconstructions of SA/50 in Fig. 2 , Supplementary Fig. 3 , and Supplementary Movie 2 show the 3D distributions of Al, Si and O species. Visually, the Si and O species are homogeneously distributed, similarly to those observed in Supplementary Fig. 1 for SA/10, however, the Al atoms are distributed rather heterogeneously. The comparison of these APT reconstructions with those in SA/10 (Supplementary Fig. 1 ) shows a higher Al density in SA/50. Additionally, a radial distribution function (RDF) 56 , 57 was calculated to further evaluate the clustering tendency of Al (Supplementary Note 1 ). The RDF of Al in SA/10 shows no significant positive or negative correlation, which is in a good agreement with the random Al distribution in the silica network without significant clustering in SA/10. The RDF of Al in SA/50 has a high positive correlation indicating that more and more Al species are close to each other. Indeed, at high Al concentration, more Al species in the network lead to shorter average Al-Al distance in SA/50 than in SA/10. Hence, more than one Al center can be expected in the proximity of a SiOH group in SA/50. Fig. 2: 3D-APT reconstruction of two isolated SA/50 nanoparticles. a All atoms. b Enlargement of the top SA/50 nanoparticle in a , showing only c Al, d Si and e O. The 3D-APT reconstruction of the bottom nanoparticle of a is shown in Supplementary Fig. 3 . Full size image The 27 Al 1D NMR spectrum of SA/50 (Supplementary Fig. 4a ) indicates the presence of Al IV , Al V and Al VI species in that sample. Moreover, the relative amount of Al V sites is higher in SA/50 than in SA/10 (compare Supplementary Figs. 2a and 4a ). Various correlation peaks were detected in the 27 Al DQ-SQ 2D spectrum of dehydrated SA/50 as shown in Fig. 3a , providing information about the proximities between the different Al sites. The most intense peak is the diagonal one of Al V site at (60, 30) ppm (Fig. 3d ). When normalized by the number of transients and the Al molar fraction, the intensity of that peak is 3-fold higher for SA/50 than for SA/10 in the spectrum of Fig. 1a . This higher intensity indicates a shorter average distance between the closest Al V sites in SA/50 than in SA/10, in line with the APT data (Supplementary Note 1 ). The pair of intense cross-peaks at (85, 55) and (85, 30) ppm also indicates that a significant amount of Al V species ( δ 27Al = 30 ppm) is close to Al IV ones ( δ 27Al = 55 ppm) (Fig. 3e ). An Al V -Al VI correlation is also detected at (34, 4) and (34, 30) ppm (Fig. 3c ). The weak cross-peaks at (59, 55) and (59, 4) ppm detected in Fig. 3d also indicate the proximity between Al IV and Al VI species. The weak diagonal signals at (8, 4) or (110, 55) ppm point to proximities between two Al VI or two Al IV sites, respectively. However, owing to the low density of Al VI and Al IV species, these peaks are very weak, as shown in the corresponding slices of Fig. 3b, f , respectively. Fig. 3: 27 Al DQ-SQ 2D NMR spectrum. a 27 Al 2D NMR spectrum recorded at 18.8 T with ν R = 20 kHz of SA/50 dehydrated at 723 K for 12 h under vacuum. b – f Rows extracted from the 2D spectrum corresponding to the various auto-correlations and cross peaks. All rows are plotted with the same intensity scale. Full size image The 1 H 1D NMR spectrum of dehydrated SA/50 (Supplementary Fig. 4b ) is dominated by the signal of SiOH groups resonating at 1.9 ppm. The shoulder at 1.1 ppm is assigned to non-acidic terminal Al VI OH protons 16 . The 1 H DQ-SQ spectrum in Supplementary Note 2 exhibits an intense autocorrelation peak at (3.8, 1.9) ppm, indicating close proximity between silanol protons. As mentioned above, the RDF data in Supplementary Note 1 indicates that almost all network Si atoms are close to Al ones. Furthermore, it has been shown for SA/50 using 1 H- 27 Al through-space correlation experiments at 18.8 T that the silanol protons are mostly close to Al IV and Al V sites 16 . Hence, the intense autocorrelation peak for silanol protons in the 1 H DQ-SQ 2D spectrum is consistent with the intense Al IV -Al V and Al V -Al V correlations detected in the 27 Al DQ-SQ 2D spectrum (Fig. 3d, e ). As already observed for SA/10, the loading of SA/50 with ammonia leads to the appearance of a signal of ammonium protons at 6.7 ppm in the 1 H 1D NMR spectrum. The relative intensity of this signal with respect to that of silanol is higher for SA/50 than for SA/10, which indicates the increased amount of BAS in SA/50 in agreement with previous studies 5 . The 1 H DQ-SQ 2D spectrum of SA/50 loaded with ammonia is shown in Supplementary Note 2 . The significant increase of the broad auto-correlation signal at (2.6, 5.2) ppm with increasing Al/Si ratio from 1/9 to 5/5 (compare Fig. 1d and Supplementary Note 2 ) allows assigning it to ammonia adsorbed on LAS 51 , 52 , 53 , 54 . Such assignment is supported by previously reported 2D 27 Al- 1 H through-space correlations of SA/10 and SA/50 loaded with ammonia 16 . In those spectra, the 1 H signal resonating at 2.6 ppm is mainly correlated with Al V sites, acting as LAS. After ammonia adsorption, the 27 Al DQ-SQ 2D spectrum is broadened (Supplementary Fig. 5 ). However, the majority of the Al IV -Al V and Al V -Al V correlations are still observed. In the 1 H DQ-SQ spectrum of dehydrated SA50 (Supplementary Note 2 ) no signal of Al(OH)Al groups at ca. 1.7–2.7 ppm could be detected 6 , indicating a low probability of Al-Al correlations originating from Al IV -OH-Al V and Al V -OH-Al V groups in alumina domains. Furthermore, these sites do not protonate ammonia and hence, the corresponding 27 Al correlation will not be significantly broadened in the presence of ammonia. Conversely ammonia can be protonated by surface BAS (SiOH with nearby Al) on ASAs, which explains the peak broadening observed in the 27 Al DQ-SQ 2D spectrum after ammonia adsorption (Supplementary Fig. 5 ). This broadening has been proposed by the synergy of Al IV -Al V and Al V -Al V spin pairs in the local structure of the same SiOH group (Al-SiOH-Al) with enhanced acid strength. Acidity enhancement by the synergy of nearby Al species In SA/10, as in most ASAs, only a moderate acidity strength ( δ 13C = 213 ppm probed with CH 3 13 COCH 3 ) was detected 5 . Figure 1 and Supplementary Fig. 1 show that the Al IV and Al V species are well-distributed on ASA without obvious correlations, e.g. SA/10. Moreover, bridging OH groups ( δ 1H = 3.6–5.2 ppm) were not detected in 1 H NMR 1D experiments (Supplementary Fig. 2b ). Therefore, the moderate BAS strength in SA/10 is proposed to be generated by one Al center interacting with the neighboring silanol group and in that way decreasing the electron density of the O atom, leading to the formation of one acid SiOH site (BAS), as shown in Fig. 4a . This arrangement is similar to the PBS model proposed in previous theoretical calculation studies, where a SiOH group electrostatically interacts with an acceptor Al center (Al IV or Al V ), but is not covalently bonded as bridging OH groups in zeolites 24 , 31 . Fig. 4: Proposed models for BAS on ASA generated by. a one Al center per SiOH for moderate BAS, and b two Al centers per SiOH group, leading to moderate or zeolitic acidity strengths. The acid strength is estimated by measuring the 13 C isotropic chemical shift of CH 3 13 COCH 3 , which is loaded on dehydrated samples on a vacuum line, followed by evacuation at room temperature for 1 h to remove weakly adsorbed molecules. Full size image In SA/50, two Al centers (two Al V or one Al V and one Al IV ) can be in the proximity of the same SiOH group. In a previous study, based on DNP (dynamic nuclear polarization) and first-principle calculations, several possible models for two or more Al centers in the vicinity of a SiOH group on ASA, prepared by chemical liquid deposition of SiO 2 on Al 2 O 3 , have been proposed by Valla and coworkers 21 . In ASAs, BAS are formed at the surface consisting of mixed alumina and silica and thus a wide distribution of Al species in the local structure of Si species can be expected, particularly in the Al-rich phase, as exemplified for ASAs prepared by chemical liquid deposition of SiO 2 on Al 2 O 3 21 . However, these models do not account for the enhanced Brønsted acidity of flame-made ASAs with high Al content, such as SA/50. Therefore, here we propose a structural model of BAS, in which Al IV and Al V sites or two Al V sites interact with the same SiOH group (see Fig. 4b ), most likely via pseudo-bridging OH groups as often proposed 24 , 32 , and withdraw electrons from the oxygen of the near SiOH group, thus enhancing its Brønsted acidity. The structural models shown in Fig. 4 bear similarities with oxygen tri- and tetra-coordinated clusters, which have been proposed earlier in ASA prepared by chemical liquid deposition of SiO 2 on Al 2 O 3 and aluminosilicates glasses 21 , 58 . The formation of oxygen tri- or tetra- coordinated clusters is driven by the increased ionicity of these materials at higher Al content due to the difference in electronegativity between Si and Al atoms 31 , 59 . However, one or more Al centers solely covalently bound to the Si atoms of the silanols are unable to protonate adsorbed ammonia, since the silanolate cannot be efficiently stabilized after deprotonation 31 , 60 . Proton transfer from silanol to the guest molecule can be promoted by the stabilization of the conjugated base (silanolate). The silanolate can be stabilized by a neighboring unsaturated Al center via PBS model 31 , 60 , 61 , 62 , such as proposed in Fig. 4a , which is often characterized by moderate acidity. Here, we propose a model as shown in Fig. 4b , for the formation of BAS on SA/50 with an acidity comparable to zeolites. Compared to Fig. 4a for SA/10, the second neighboring unsaturated Al center involved as an acceptor is able to further stabilize the formed silanolate, and thus to promote the proton transfer. Besides, the extra Al center(s) may further withdraw electrons from the O atom from neighboring SiOH that significantly enhances the BAS strength (up to δ 13C = 227 ppm) 5 , to a higher extent than that of zeolite H-ZSM-5 ( δ 13C = 223 ppm) 55 , similar to that of dealuminated zeolite H-Y ( δ 13C = 228 ppm) 17 . The proposed model is similar to those reported for zeolites where an ionic effect induced by extra-framework Al species can enhance the Brønsted acid strength of bridging OH groups 17 . Therefore, the synergy of two Al V or one Al IV and one Al V centers with nearby SiOH groups is expected to significantly enhance the acid strength of BAS in the ASA. In situ 1 H MAS NMR study on acidity enhancement of BAS In this work, we show that the proximity between one SiOH group and two Al sites can remarkably enhance the Brønsted acidity in ASAs 5 , 16 . The activation of the C-H bond in hydrocarbon conversion often requires solid acids containing strong BAS. The activation of the C-H bond in benzene, the simplest aromatic compound, has been extensively studied using H/D exchange experiments, which is of great importance to understand the alkylation processes of aromatic compounds 8 , 9 , 10 , 11 , 12 . The H/D exchange carried out between C 6 D 6 and surface BAS (bridging OH on zeolite H-ZSM-5 and acidic SiOH groups on ASA) was confirmed by 1 H solid-state NMR (Supplementary Fig. 6 ). On H-ZSM-5 zeolite, the three 1 H signals of H-ZSM-5 zeolite at 7.5, 4.0 and 1.8 ppm (see Supplementary Fig. 6a ) are assigned to hydrogen atoms bound to aromatic rings, bridging OH (SiOHAl) groups and terminal SiOH groups, respectively. After H/D exchange reaction, the intensity of SiOH groups remained unchanged while that of bridging OH groups decreased with increasing intensity of the aromatic hydrogens. This indicates that the H/D exchange occurred between the benzene- d 6 and the bridging OH groups (e.g. SiOHAl), rather than with the terminal SiOH groups ( δ 1H = 1.8 ppm). Conversely, no bridging OH groups could be observed at 3.5–5.2 ppm on SA/50 (see Supplementary Fig. 6b ) 6 , while the signal of protons bound to the aromatic rings was observed at 7.3 ppm. As shown in the stack plot of the 1 H MAS spectra recorded during H/D exchange of benzene- d 6 loaded over dehydrated SA/50 (Fig. 5a ), the intensity of the terminal SiOH groups ( δ 1H = 1.8 ppm) decreased as a function of time, while that of protons bound to aromatic rings ( δ 1H = 7.3 ppm) increased. This demonstrates that the H/D exchange occurred between C 6 D 6 and the acidic SiOH groups of ASA. It must be reminded that Lewis acid aluminum sites cannot exchange H with C 6 D 6 (see Supplementary Note 3 ). Fig. 5: Catalytic performance of ASA in H/D exchange with C 6 D 6 . 1 H MAS spectra recorded at 9.4 T of dehydrated catalysts, a stack plot spectra recorded during H/D exchange of C 6 D 6 loaded over dehydrated SA/50 at 313 K, with a loading of one benzene molecule per BAS; b Kinetics and H/D exchange rates k between deuterons bound to the aromatic rings of C 6 D 6 (99.6 %) and BAS at 313 K in H-ZSM-5 (top), SA/50 (middle) and SA/10 (bottom). Full size image The reaction rate, k , for the C 6 D 6 –SiOH exchange can be utilized to evaluate the relative strength of BAS in silica-alumina catalysts under the same conditions. As described in the Methods section, one molecule of C 6 D 6 per Brønsted acid site was quantitatively loaded, and thus, a higher k value indicates a higher acid strength 63 , 64 , 65 . The rate of the H/D exchange reaction over different catalysts upon loading one C 6 D 6 per BAS at 313 K was determined by fitting the evolution of the signal intensity of aromatic protons as a function of reaction time 9 , 12 . The obtained k values are shown in Fig. 5b . Evidently, SA/10 with mainly moderate BAS was virtually inactive in the reaction, resulting in a small k value, while zeolite H-ZSM-5 with strong BAS provided a much higher reaction rate ( k = 0.0072 min −1 ). A similar value, k of 0.0078 min −1 , was obtained with SA/50 under the same conditions, hinting to the existence of strong BAS in SA/50 with a strength comparable to that of zeolite H-ZSM-5. Considering the structural difference between SA/10 and SA/50, it shows that the proximity between more than one Al center and SiOH group in ASA with high Al content, such as SA/50, gives rise to BAS with zeolitic strength. Stability test of ASA catalysts Besides their enhanced surface Brønsted acidity, the high stability of ASAs under various conditions is crucial for efficient catalysis. Current ASAs were thus firstly tested in liquid-phase glucose dehydration to 5-hydroxymethylfurfural (HMF), requiring LAS for glucose isomerization to fructose and BAS for fructose dehydration 48 . The catalytic reaction results are summarized in Supplementary Note 4 , 5 , and Supplementary Fig. 7 . The reusability of ASA was tested with SA/50, which exhibited the best performance under the same conditions. After five recycle runs, no significant loss of catalytic activity could be observed (Supplementary Note 5 ). This is attributed to the high stability of SA/50 in the liquid phase glucose dehydration. In liquid phase reactions under heating, dealumination could cause the modification of surface acid sites. The comparison of Supplementary Fig. 12a, b demonstrates that aluminum species in SA/50 are stable without dealumination upon water treatment at 433 K for 2 h (Supplementary Note 6 ). Therefore, the synergy of Lewis acidic Al V and enhanced Brønsted acidity renders these ASAs promising bifunctional catalysts for Brønsted-Lewis acid-catalyzed reaction, such as the glucose dehydration to HMF. It is noteworthy to point out the high stability of the BAS with enhanced strength when they are exposed to high temperature (1073 K) calcination (regeneration temperature in fluid catalytic cracking, Supplementary Fig. 12c ) or to a liquid-phase conversion of glyceraldehyde in ethanol (Supplementary Fig. 12d ). Two or more Al centers nearby SiOH groups in the amorphous silica network did not leach out from the catalysts after five recycle uses in a batch reaction with ultrasonic washing (Supplementary Fig. 12d ), as confirmed by the lack of detectable Al species in the reaction mixture. Conversely, a similar treatment caused strong leaching of extra-framework aluminum species in zeolites, which exhibited enhanced acidity of BAS via the synergy of extra-framework aluminum and BAS. This leaching led to a significant activity loss of these zeolites (Supplementary Note 7 ) as confirmed by 1D 27 Al NMR experiments (Supplementary Note 8 ). Discussion In conclusion, a remarkable synergy between two Al centers (Al V -Al IV or Al V -Al V ) close to the same SiOH group has been evidenced in flame-made amorphous silica-alumina (ASA) by 2D 27 Al and 1 H DQ-SQ NMR experiments, and analysis of the 3D spatial elemental distribution of Al and Si by APT. The study revealed that compared to the widely accepted model of one Al center per SiOH group with moderate strength ( δ 13C = 213 ppm) 5 , two proximate Al centers can strongly decrease the electron density from a neighboring silanol oxygen and thereby can significantly boost its acid strength (with δ 13C = 227 ppm for CH 3 13 COCH 3 ) to a value higher than that of H-ZSM-5 ( δ 13C = 223 ppm) 55 , or even reaching that of dealuminated zeolite HY ( δ 13C = 228 ppm) 17 . These BAS with zeolitic strength have been evidenced by comparative H/D exchange experiments with C 6 D 6 . Furthermore, the synergy between BAS with zeolitic strength and LAS afforded a much higher HMF yield (38%) than catalysts with moderate BAS strength (e.g. SA/10 and [Al]MCM-41). The achieved yield was comparable to that realized with metal-doped zeolites (33%) at a higher temperature. The present study highlights a promising route for generating BAS with zeolitic strength and high stability on ASAs, which could facilitate improved catalytic performances in a wide range of applications, including acid and multifunctional catalysis. Methods APT sample preparation and measurement method A drop of the diluted dispersed ASA nanoparticles in methanol (≈0.01 mol/L) was placed onto a Si flat wafer, which was covered by a 150-nm thick protective Cr layer in Leica EM ACE600. Needle-shaped APT specimens were prepared from the Si flat sample by a site-specific lift-out procedure using a FEI G4 CX focused ion beam (FIB)/scanning electron microscope 66 . The APT experiments were conducted on a CAMECA LEAP 5000 XR instrument equipped with an ultraviolet laser with a spot size of 2 µm and a wavelength of 355 nm. The detection efficiency of this state-of-the-art microscope is ca. 54%. Data were acquired in laser pulsing mode at a specimen temperature of 50 K, with a target evaporation rate of 3 ions per 1000 pulses, a pulsing rate of 200 kHz, and a laser pulse energy of 50 pJ. The APT data were reconstructed and analyzed using the commercial IVAS 3.6.14™ software. NMR experimental details Before each experiment, the samples in glass tubes were dehydrated at 723 K for 12 h at a pressure lower than 10 −2 bar. Subsequently, the samples were transferred into the MAS NMR rotors under dry N 2 inside a glove box. These ammonia-loaded samples were prepared by dehydrated samples loaded with ammonia on a vacuum line, followed by evacuation at 393 K for 1 h to remove weakly physisorbed molecules. All 1 H and 27 Al NMR spectra were recorded on a Bruker Avance III 800 MHz spectrometer equipped with 3.2 mm MAS rotors spinning at 20 kHz. For 1 H DQ-SQ 2D experiments, the 1 H DQ coherences were excited and reconverted by applying the symmetry-based R \(12_2^5\) scheme 67 , which reintroduces the 1 H- 1 H dipolar interactions under MAS. The 1 H radio frequency (rf) amplitudes for the π/2 pulse and R \(12_2^5\) scheme were equal to ν 1 = 75 and 60 kHz, respectively. The length of the excitation recoupling scheme was equal to that of the reconversion and ranged from 250 to 300 μs, depending on the experiment. 1 H DQ-SQ 2D spectra resulted from averaging 32 to 128 transients with recycle delay of 1 to 5 s, resulting in a total experimental time of 2 to 4 h. During 27 Al DQ-SQ 2D experiments, selective central transition (CT) π/2 and π-pulses of 8 and 16 μs, that is, an rf amplitude of about 10 kHz, were applied. The 27 Al two-spin DQ coherences were excited and reconverted by applying the BR \(2_2^1\) pulse sequence 36 , which reintroduces the 27 Al− 27 Al dipolar interactions under MAS. The lengths of the excitation and reconversion periods were equal and ranged from 800 to 1200 μs, depending on the experiment. The rf amplitude applied during the BR \(2_2^1\) pulse sequence was 6.6 kHz, which corresponds to a nutation frequency of 20 kHz for the 27 Al CT. Furthermore, the Hyper–Secant (HS) scheme was applied before the BR \(2_2^1\) excitation 68 , in order to enhance the 27 Al CT polarization by saturating the satellite transitions 69 , 70 . HS employed a shaped pulse lasting 4 ms with an rf field amplitude of 16 kHz and a frequency sweep of 20 kHz around an offset of 200 kHz with respect to the CT. 27 Al DQ-SQ 2D spectra resulted from averaging 14,400 and 3200 transients for SA/10 (Fig. 1 ) and SA/50 (Fig. 3 ) with recycle delay of 0.2 s, resulting in a total experimental time of 25.3 and 5.7 h, respectively. The 1 H isotropic chemical shifts were referenced to tetramethylsilane using the resonance of adamantane (1.83 ppm) as a secondary reference, whereas the 27 Al ones were referenced to 1 M solution Al(NO 3 ) 3 . In situ 1 H MAS NMR Spectroscopy of H/D exchange with C 6 D 6 1 H MAS NMR spectra of H/D exchange with C 6 D 6 was carried out on a Bruker Avance III 400 WB spectrometer at the Larmor frequency of 400.1 MHz with 4 mm MAS rotors spinning at 8 kHz. Spectra were recorded after single-pulse π/2 excitation with repetition times of 20 s and 8 scans. Prior to measurements, all samples were dehydrated at 723 K in vacuum (pressure <10 −2 bar) for 12 h in glass tubes. The density of BAS on all dehydrated samples was determined by quantitative 1 H MAS NMR experiments using NH 3 as probe molecules. The total number of BAS was calculated based on the BAS density and weight of the sample. A known amount of dehydrated samples was transferred into the MAS rotors under dry nitrogen gas inside a glove box, sealed and utilized for in-situ loading on a vacuum line. The loading pressure of benzene- d 6 (99.6%, Cambridge Isotope Laboratories, Inc.) was calculated and controlled according to the total number of BAS and known volume of the vacuum line to ensure one molecule of benzene- d 6 per BAS. Then the sample was cooled down by liquid nitrogen till nearly no pressure could be detected. Subsequently, the loaded samples in the MAS rotors were kept 10 min at room temperature under dry nitrogen gas inside a glove box for better diffusion. The H/D experiments were performed by heating the MAS rotor at 313 K in a variable-temperature probe for 1 H MAS NMR investigations. The concentration of protons bound to the aromatic rings was calculated as the ratio between the integrated intensity of the aromatic 1 H signal and the number of Brønsted acid sites. The rate k of the H/D exchange between the deuterated molecules and the acidic OH groups (BAS) is described by an exponential relationship 9 $$I\left( t \right) = I\left( \infty \right)\left[ {1 - b{\mathrm{exp}}\left\{ { - kt} \right\}} \right]$$ (1) where I ( t ) and I ( \(\infty\) ) denote the intensities of the 1 H MAS NMR signal of the aromatic rings at the observation time t and t → + \(\infty\) in the equilibrium state, respectively. The b parameter describes the exchange at t = 0, which corresponds to the start of the H/D exchange experiment, i.e. when the temperature was increased from ca. 293 K to that of the reaction. Data availability Raw data are available from the authors upon reasonable request.

Researchers at the University of Sydney have created a new material that has the potential to reduce CO2 emissions released during the refinement process of crude oil by up to 28 percent. Silica-alumina materials are among the most common solid acids that have been widely commercialised as efficient and environmentally-friendly catalysts in the petrochemical and bio-refinery industries. In a world first, a team of researchers at the University of Sydney led by Associate Professor Jun Huang, have produced a new amorphous silica-alumina catalyst with stronger acidity than any other silica-alumina material created before. "This new catalyst can significantly reduce the amount of CO2 emitted by oil refineries, which has the potential to make the fossil fuel industry much greener and cleaner," Associate Professor Huang from the Faculty of Engineering and Sydney Nano Institute said. A significant amount of carbon is emitted during the refinement of crude oil to produce products like petroleum, gasoline and diesel. Estimates suggest 20 to 30 percent of crude oil is transferred to waste and further burnt in the chemical process, making oil refineries the second largest source of greenhouse gases behind power plants. Credit: University of Sydney Silica-aluminas with strong Brønsted acidity—a substance that gives up or donates hydrogen ions (protons) in a chemical reaction—are becoming increasingly important to various sustainability processes, including the fields of biomass conversion, CO2 capture and conversion, air-pollution remediation, and water purification. "Renewable energy is important to achieving a more sustainable energy supply, but the reality is that we will still be reliant on fossil fuels in the foreseeable future. Therefore, we should do all we can to make this industry more efficient and reduce its carbon footprint while we transition to renewable energy sources "This new catalyst offers some exciting prospects, if it were to be adopted by the entire oil refinery industry, we could potentially see a reduction of over 20 percent in CO2 emissions during the oil refinement process. That's the equivalent of double Australia's crude oil consumption, over 2 million barrels of oil per day." "The new catalyst also has the potential to develop the biomass industry. We can now look to biomass material like algae to be part of sustainable energy solutions." The next steps for the researchers are to work on manufacturing the new catalyst at a large, industrial scale.

10.1038/s41467-019-13907-7

Physics

Researchers use laser light to transform metal into magnet

Mark S. Rudner et al. Self-induced Berry flux and spontaneous non-equilibrium magnetism, Nature Physics (2019). DOI: 10.1038/s41567-019-0578-5 Journal information: Nature Physics

http://dx.doi.org/10.1038/s41567-019-0578-5

https://phys.org/news/2019-09-laser-metal-magnet.html

Abstract When a physical system is governed by statistical or dynamical equations possessing certain symmetries, its stationary states can be classified into phases according to which of those symmetries are preserved, and which are broken 1 , 2 . Near equilibrium, the properties of the system’s collective excitations reflect the symmetries of the underlying phase and thereby provide means for detecting these phases 3 , 4 . Here, we show that, in driven systems, the collective modes may take on a separate life, exhibiting their own spontaneous symmetry-breaking phenomena independent of the underlying equilibrium phase. We illustrate this principle by demonstrating a mechanism through which a non-magnetic interacting metal subjected to a linearly polarized driving field can spontaneously magnetize. The strong internal a.c. fields of the metal driven close to its plasmonic resonance 5 , 6 enable Berryogenesis: the spontaneous generation of a self-induced Bloch band Berry flux. The self-induced Berry flux supports and is sustained by a chiral circulating plasmonic motion that breaks the mirror symmetry of the system. This non-equilibrium phase transition may be of either continuous or discontinuous type. Berryogenesis can occur in a wide variety of multiband metals with high-quality plasmons, as available in present-day graphene devices 7 , 8 , 9 . Main Interacting many-body systems may exhibit collective modes of excitation whose emergent properties are unlike those of the constituent particles of the system. Importantly, such modes may host strong internal fields associated with the restoring force underlying the system’s collective oscillations. For example, plasmons have recently gained wide attention for their ability to resonantly enhance applied electric fields by many orders of magnitude 5 , 6 . Here we propose that new collective mode phases may arise due to feedback in which a system’s internal fields modify its microscopic properties, in turn altering its response to the driving field. Time-dependent driving by laser or microwave fields has recently emerged as a powerful tool for dynamically controlling the non-equilibrium properties of quantum matter 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 . As we discuss here, such fields provide the means to access non-equilibrium spontaneous symmetry breaking associated with novel collective mode phase transitions. To demonstrate this principle, we show how feedback due to the internal fields of a metallic disk driven at a frequency close to its natural dipole resonance (Fig. 1a ) gives rise to multistability and spontaneous symmetry breaking. When excited, the resonance may produce rotating fields that modify the system’s Bloch band Berry curvature (Fig. 1b ). The resulting amplitude-dependent Berry flux splits the resonance for right- and left-handed polarizations. This feedback between field strength, field chirality and Berry flux ultimately leads to a phenomenon we call Berryogenesis: the spontaneous appearance of a chiral circulating plasmonic motion together with a non-vanishing Berry flux in a non-magnetic material driven by a linearly polarized field (Fig. 1c ). Fig. 1: Spontaneous generation of Berry flux via plasmonic internal fields. a , When excited, the dipole mode of a metallic disk hosts an internal electric field, E int ( t ), that oscillates at the drive angular frequency, ω d . The corresponding motion is decomposed into right- and left-circulating amplitudes, \({\cal{Z}}_ +\) and \({\cal{Z}}_ -\) , respectively. b , Electronic Bloch functions at the Fermi energy E F (represented by pseuodospinors for graphene, in valleys K and K′) are dressed by an off-resonant rotating electric field with ħω d < 2 E F ; the dressing gently cants the pseudospins out of the equatorial plane. In this way, a.c. internal fields dynamically alter the system’s electronic properties, giving rise to feedback. c , Stability diagram, with stable (unstable) solutions of equation ( 5 ) indicated by black solid (red dashed) lines. For small amplitudes of a linearly polarized drive, the response is linearly polarized ( \(\overline {\cal{F}} = 0\) ). Above threshold, the linearly polarized solution becomes unstable and the system spontaneously develops a right- or left-handed circulation. Parameters: E F = 160 meV, ħω d = 100 meV, ħω 0 = 100.125 meV, Q = 75. Full size image The threshold driving amplitudes that are needed to induce Berryogenesis are modest, and can be achieved in current high-quality graphene devices 7 , 8 , 9 with readily available terahertz to mid-infrared drives. While we use graphene for illustration due to its simplicity and favourability for near-term experiments, the mechanism we discuss is general and may apply to a wide variety of materials (for example, transition metal dichalcogenides and multiband semimetals). The ease with which large plasmon-enhanced internal fields can be accessed in graphene 6 , 9 makes it a natural platform for plasmonic non-equilibrium spontaneous symmetry breaking. We now demonstrate how a non-vanishing Berry flux can be induced in a non-magnetic metal subjected to an off-resonant a.c. field. As a concrete example, we consider electrons in graphene at finite doping, with Fermi energy E F > 0. Taking E F as the natural energy scale, we express the Hamiltonian for electrons in the K valley of graphene as $${\cal{H}}_{\rm{K}}(t) = E_{\rm{F}}[\tilde{\mathbf{k}} - \tilde{\mathbf{A}}(t)]\cdot {\boldsymbol{\sigma }},\quad \tilde{\mathbf{A}}(t) = \frac{{ev}}{{cE_{\rm{F}}}}{\mathbf{A}}(t)$$ (1) where v is the Fermi velocity of graphene, c is the speed of light, − e < 0 is the electron charge, σ = ( σ x , σ y ) is a vector of Pauli matrices and \(\tilde{\mathbf{k}} = {\mathbf{k}}/k_{\rm{F}}\) is the normalized 2D wavevector, with ħvk F = E F . The vector potential A ( t ) describes the (time-varying) electromagnetic field. Importantly, A ( t ) may arise both due to an externally applied driving field, and due to time-dependent internal fields of the system when it is driven out of equilibrium. The Hamiltonian \({\cal{H}}_{{\rm{K}}^\prime}(t)\) for electrons in the K′ valley is the same as that in equation ( 1 ), with σ x → − σ x in σ . For illustration, we focus on a monochromatic field, A ( t ), oscillating at frequency ω . Below, it will be convenient to work in the basis of circularly polarized fields, \({\mathbf{A}}(t) = \frac{1}{2}({\mathbf{A}}_{\rm L} + {\mathbf{A}}_{\rm R})e^{ - i\omega t}{\mkern 1mu} + {\mkern 1mu} \text{c.c.}\) , where the left- and right-handed components are given by \({\mathbf{A}}_{\rm L} = A_{\rm L}(\widehat {\mathbf{x}} + i\widehat {\mathbf{y}})/\sqrt 2\) and \({\mathbf{A}}_{\rm R} = A_{\rm R}(\widehat {\mathbf{x}} - i\widehat {\mathbf{y}})/\sqrt 2\) . We analyse the induced Berry flux by studying the Floquet–Bloch band structure arising from equation ( 1 ) with the time-periodic field A ( t ) (see Methods ). Crucially, we consider frequencies with ħω < 2 E F , such that the main action of A ( t ) in equation ( 1 ) is to perturbatively modify the system’s Bloch wavefunctions at the Fermi energy through off-resonant hybridization with states in the valence band 10 (see Fig. 1b and discussion below). In the presence of a linearly polarized drive, | A L | = | A R |, the Hamiltonians \({\cal{H}}_{\rm{K}}(t)\) and \({\cal{H}}_{\rm{K}^{\prime}}(t)\) are each invariant under reflections across the polarization axis of the drive (see Supplementary Information ). This mirror symmetry implies that the time-averaged Berry flux \(\overline {\cal{F}}\) associated with the Floquet Fermi sea must vanish (see refs. 23 , 24 and Supplementary Information ). However, a chiral field with | A L | ≠ | A R | breaks the mirror symmetry, generating a term proportional to σ z in the system’s effective Floquet Hamiltonian. The σ z term causes the pseudospinor Floquet eigenstates to cant out of the equatorial plane of the Bloch sphere (see Fig. 1b ), resulting in a finite Berry flux, \({\cal{F}} \ne 0\) (refs. 15 , 17 , 19 , 20 , 21 ). In Fig. 2a we show the d.c. Berry flux induced by an off-resonant field, as a function of the (scaled) left- and right-circulating field amplitudes \(\tilde A_{\rm L}\) and \(\tilde A_{\rm R}\) . In this regime, \(\overline {\cal{F}}\) exhibits a characteristic saddle-like shape \(\overline {\cal{F}} \approx \beta (|\tilde A_{\rm L}|^2 - |\tilde A_{\rm R}|^2)\) , as may be anticipated by second-order perturbation theory. Here β is an \({\cal{O}}(1)\) dimensionless prefactor; for the parameters in graphene used in Fig. 2a , we find β ≈ 2.3. Berry flux generation thus provides a sensitive detector for chiral motion and, as we discuss below, naturally couples with plasmonic dynamics. Fig. 2: Berry flux generation and nonlinear plasmon dynamics. a , Time-averaged Berry flux induced by a harmonic drive with angular frequency ω d and (dimensionless) left and right circular polarization amplitudes \(\tilde A_{\rm L}\) and \(\tilde A_{\rm R}\) , respectively (see equation ( 1 )). In the off-resonant regime, the induced flux is approximately given by \(\overline {\cal{F}} \approx \beta (|\tilde A_{\rm L}|^2 - |\tilde A_{\rm R}|^2)\) , where \(\beta \sim {\cal{O}}(1)\) depends on ħω d / E F . The surface shown is for ħω / E F = 0.625. b , Steady-state Berry flux for a graphene disk driven by a circularly polarized driving field. Stable and unstable steady states are indicated by black solid and red dashed lines, respectively. Parameters: E F = 160 meV, ħω d = 100 meV, ħω 0 = 100.75 meV, Q = 100. Full size image When a plasmonic mode is excited near resonance, A ( t ) in equation ( 1 ) may easily be dominated by internal fields associated with the oscillating charge density in the system 5 , 6 . Importantly, the plasmon resonance itself is sensitive to the development of Berry flux; this feedback provides the crucial source of plasmonic nonlinearity that we study. To illustrate this phenomenon, we consider the dipolar mode of a circular electronic disk (Fig. 1a ). We describe the plasmon dynamics using centre of mass (COM) coordinates for position, { r ( t )} = ({ x ( t )}, { y ( t )}), and momentum, { p ( t )} = ({ p x ( t )}, { p y ( t )}). Here { ⋅ } denotes an average over the electronic distribution function. The resulting plasmon equations of motion (derived from the kinetic equation 25 , 26 , 27 ) are given by: $$\frac{{{\rm d}\{ {\mathbf{r}}\} }}{{{\rm d}t}} = \frac{{\{ {\mathbf{p}}\} }}{m} - \frac{{{\cal{F}}[{\mathbf{E}}_{{\mathrm{tot}}}]}}{{\hbar n_0}}\widehat {\mathbf{z}} \times e{\mathbf{E}}_{{\mathrm{tot}}}(t)$$ $$\frac{{{\rm d}\{ {\mathbf{p}}\} }}{{{\rm d}t}} = - m\omega _0^2\{ {\mathbf{r}}\} - \gamma \{ {\mathbf{p}}\} - e{\mathbf{E}}_{{\mathrm{drive}}}(t)$$ (2) where ω 0 is the angular frequency of the bare plasmon resonance of the disk (that is, for \({\cal{F}} = 0\) ), m is the plasmon mass, n 0 is the electron density and γ is the damping rate of the plasmon mode. We include γ in the equation of motion to account for the plasmon damping that inevitably arises in real devices. However, dissipation is not essential for multistability or spontaneous symmetry breaking in this system. The fields E tot ( t ) and \({\mathbf{E}}_{{\mathrm{drive}}}(t) \propto {\rm e}^{ - i\omega _{\rm d}t}\) are the total and the (monochromatic) driving electric fields, respectively. Here ω d is the angular frequency of the drive. Importantly, for \({\cal{F}} \ne 0\) , the cross product in the first line of equation ( 2 ) couples the modes with linear polarizations along \(\widehat {\mathbf{x}}\) and \(\widehat {\mathbf{y}}\) . The nonlinearity of the system arises due to the dependence of E tot on the plasmonic motion, { r ( t )} (see below). Within our mean-field approach that focuses on the COM motion of the plasmon dipole, any spatial dependence of the internal and external fields is integrated out to obtain the effective electric field acting on the COM: \(- e{\mathbf{E}}_{{\mathrm{tot}}}(t) = - e{\mathbf{E}}_{{\mathrm{drive}}}(t) - m\omega _0^2\{ {\mathbf{r}}\}\) , where \(- m\omega _0^2\{ {\mathbf{r}}\}\) is the restoring force acting on the COM. Close to resonance, the plasmonic internal field is enhanced by a factor ω 0 / γ = 2 Q relative to the driving field, where Q is the quality factor of the resonance. For large Q (refs. 7 , 8 , 9 ), the total electric field may thus be well approximated by \(e{\mathbf{E}}_{{\mathrm{tot}}}(t) \approx m\omega _0^2\{ {\mathbf{r}}(t)\}\) ; for simplicity, we make this replacement in the analysis below. Our approach (equation ( 2 )) is designed to capture the time-periodic steady-state motion of the system, where the Berry flux \({\cal{F}}\) is determined self-consistently from the (periodic) oscillating external and internal fields present in the steady state. Transient dynamics, including fluctuations in the vicinity of the phase transition 28 , are beyond the scope of this work. Noting that the time-averaged part of the Berry flux plays the most essential role in altering the character of the plasmon dynamics, below we replace \({\cal{F}}\) by its time-averaged value, \(\overline {\cal{F}}\) , in equation ( 2 ). This approximation preserves the crucial nonlinearity associated with the dependence of \(\overline {\cal{F}}\) on { r ( t )}, and enables a detailed analytical treatment; see Methods for details of a fully self-consistent numerical solution. We now solve for the steady-state plasmonic oscillations described by equation ( 2 ) with \({\cal{F}}\) replaced by \(\overline {\cal{F}}\) . Feedback due to self-generated Berry flux arises via the saddle-like dependence of \(\overline {\cal{F}}\) on \(\tilde A_{\rm L}\) and \(\tilde A_{\rm R}\) (see Fig. 2a ). To capture this interplay, we decompose the steady-state motion in terms of right-circulating (+) and left-circulating (−) amplitudes, \({\cal{Z}}_ \pm ^{(0)}\) (see Methods and Supplementary Information ). We identify the internal field contribution to A ( t ) in equation ( 1 ) by using \({\mathbf{E}}_{{\mathrm{tot}}}(t) = - \frac{1}{c}\frac{\partial }{{\partial t}}{\mathbf{A}}(t)\) , together with the replacement \(e{\mathbf{E}}_{{\mathrm{tot}}}(t) \approx m\omega _0^2\{ {\mathbf{r}}\}\) . In this way, we obtain the self-generated (d.c.) Berry flux: $$\overline {\cal{F}} = f(|{\cal{Z}}_ - ^{(0)}|^2/l^2,|{\cal{Z}}_ + ^{(0)}|^2/l^2),\quad l^{ - 1} = \frac{{vm\omega _0^2}}{{E_{\rm F}\omega _{\rm d}}}$$ (3) where f is the (dimensionless) saddle function of Fig. 2a , and l defines an intrinsic length scale of the system. Transforming equation ( 2 ) to the circular polarization basis, and defining \({\cal{E}}_{{\mathrm{drive}}}^ \pm (t) = \frac{1}{{\sqrt 2 }}[E_{{\mathrm{drive}}}^x(t) \pm iE_{{\mathrm{drive}}}^y(t)] = {\cal{E}}_ \pm ^{(0)}{\rm e}^{ - i\omega _{\rm d}t}\) , we find that the amplitudes \({\cal{Z}}_ \pm ^{(0)}\) are given by: $${\cal{Z}}_ \pm ^{(0)} = \frac{{ - e{\cal{E}}_ \pm ^{(0)}/m}}{{[ - \omega _{\rm d}^2 + \omega _0^2 \pm \kappa \overline {\cal{F}} \omega _{\rm d}] - i\gamma [\omega _{\rm d} \mp \kappa \overline {\cal{F}} ]}}$$ (4) where \(\kappa = m\omega _0^2/(\hbar n_0)\) . Equation ( 4 ) demonstrates how a d.c. Berry flux \(\overline {\cal{F}}\) modifies the disk’s dipole resonance 27 . Importantly, because \(\overline {\cal{F}}\) in equation ( 4 ) depends on \({\cal{Z}}_ \pm ^{(0)}\) via equation ( 3 ), the system may exhibit multistability. For demonstration, we first consider the simple case of a circularly polarized drive, \({\cal{E}}_ + ^{(0)} = 0\) , \({\cal{E}}_ - ^{(0)} = E_{{\text{r.m.s.}}}\) . Here, the steady-state motion captured by equation ( 4 ) is itself circularly polarized: there is no mixing between left- and right-hand polarizations. In Fig. 2b , we show the corresponding solutions of equation ( 4 ), using the parametrized form \(\overline {\cal{F}} = \beta (|\tilde A_{\rm L}|^2 - |\tilde A_{\rm R}|^2) = \beta l^{ - 2}(|{\cal{Z}}_ - ^{(0)}|^2 - |{\cal{Z}}_ + ^{(0)}|^2)\) . We track the bistability via the induced (d.c.) Berry flux, \(\overline {\cal{F}}\) , as it provides a sensitive measure of the amplitude of circular motion. Note that for a circularly polarized driving field in equation ( 1 ), the induced Berry flux \({\cal{F}}\) is in fact time independent: \({\cal{F}}(t) = \overline {\cal{F}}\) . In the absence of driving, the graphene disk possesses zero Berry flux, \({\cal{F}} = 0\) . As shown in Fig. 2b , as the amplitude of the circularly polarized drive is increased from zero, a finite Berry flux is generated. Strikingly, when the drive amplitude is strong enough, the induced Berry flux exhibits bistability: two distinct steady-state Berry fluxes (corresponding to two stable steady-state amplitudes for \({\cal{Z}}_ - ^{(0)}\) in equation ( 4 )) may arise for the same drive amplitude. For even stronger driving, only the solution with a large self-generated contribution to the Berry flux remains. Bistability arises from the fact that the (self-induced) Berry flux splits and shifts the plasmon resonances of the disk. Consider a weak external drive, with frequency ω d slightly red-detuned from the bare resonance ω 0 . The drive induces a small-amplitude circular motion of the plasmon dipole, which correspondingly generates a small Berry flux, \(\overline {\cal{F}}\) . Due to the non-vanishing \(\overline {\cal{F}}\) , the resonance shifts downward, towards the frequency of the drive. As the resonance approaches ω d , the amplitude of the response increases, thereby amplifying \(\overline {\cal{F}}\) and bringing the drive even closer to resonance. This feedback provides the mechanism for bistability. We now turn to the main phenomena of interest: spontaneous symmetry breaking and Berry flux generation. To illustrate spontaneous circulation of the plasmonic mode, we return to equation ( 4 ), with a linearly polarized drive \({\cal{E}}_ + ^{(0)} = {\cal{E}}_ - ^{(0)} = E_{{\mathrm{r.m.s.}}}\) . When \({\cal{F}} = 0\) , the equation of motion ( 2 ) for the driven system has a mirror symmetry about the linear polarization axis of the drive (see Supplementary Information ). However, as we now discuss, a spontaneous chiral circulating plasmonic motion can break this mirror symmetry, leading to the spontaneous generation of \({\cal{F}} \ne 0\) . We characterize spontaneous chirality in the plasmon motion by a magnetization order parameter: \(\eta \equiv |{\cal{Z}}_+^{(0)}|^2 - |{\cal{Z}}_-^{(0)}|^2\) . A non-zero value of η indicates the presence of an internal field with a net right-handed ( η > 0) or left-handed ( η < 0) rotation; as discussed above, such a circulating field induces a finite Berry flux. With finite \({\cal{F}}\) induced, the shifting of resonances reinforces and amplifies the circulating motion. This feedback is the driving force for Berryogenesis. For the simple parametrized saddle form considered above, the d.c. Berry flux is a function of η alone: \(\overline {\cal{F}} = - \beta l^{ - 2}{\mkern 1mu} \eta\) . Subtracting the expressions for \(|{\cal{Z}}_ - ^{(0)}|^2\) and \(|{\cal{Z}}_ + ^{(0)}|^2\) given in equation ( 4 ), we obtain an algebraic relation for the steady-state magnetization: $$\eta \left[ {1 + 4\nu \omega _{\rm d}(\omega _{\rm d}^2 + \gamma ^2 - \omega _0^2)\frac{{|eE_{{\text{r.m.s.}}}/m|^2}}{{D_ + D_ - }}} \right] = 0$$ (5) with \(D_ \pm = [\omega _0^2 - \omega _{\rm d}^2 \mp \nu \omega _{\rm d}\eta ]^2 + \gamma ^2(\omega _{\rm d} \pm \nu \eta )^2\) and ν = κβ / l 2 . Equation ( 5 ) can be expressed as a fifth-order polynomial in η , and may exhibit several solutions (see Figs. 1c and 3a ). For small amplitudes of the linearly polarized drive, the system responds with linearly polarized oscillations ( η = 0). As the drive amplitude is increased, a bifurcation is encountered where the linearly polarized solution becomes unstable to fluctuations and the system spontaneously acquires a magnetization ( η ≠ 0). The appearance of such pitchfork bifurcations is consistent with symmetry considerations 29 . Fig. 3: Spontaneous magnetization in the presence of a linearly polarized drive. a , Stability diagram in the discontinuous transition regime. Stable (unstable) solutions of equation ( 5 ) are indicated by black solid (red dashed) lines. The green circles indicate self-consistent solutions to equation ( 2 ) in which the full time dependence of \({\cal{F}}(t)\) is calculated from the self-induced Floquet band structure. Parameters: E F = 160 meV, ħω d = 100 meV, ħω 0 = 100.5 meV, Q = 100. b , The phase transition to the magnetized state can be either continuous (pink region) or discontinuous (blue region), depending on the detuning and damping rate. No instability occurs in the region to the right of the dashed line. Full size image We confirm the validity of the results above, which were obtained by replacing \({\cal{F}}\) by \(\overline {\cal{F}}\) in equation ( 2 ), by numerically obtaining self-consistent solutions of equation ( 2 ) including the full time dependence of \({\cal{F}}(t)\) (see Methods ). As shown in Fig. 3a , the time-averaged magnetization obtained from these simulations (green dots) agrees well with the results of our analytical treatment (solid lines). The type of phase transition (discontinuous versus continuous) is controlled by the detuning of the drive, ( ω d − ω 0 )/ ω 0 , and the damping rate, γ . The character of the transition can be straightforwardly extracted from the η dependence of the expression in brackets in equation ( 5 ) (see Supplementary Information ). As summarized in Fig. 3b , we find that spontaneous magnetization may occur only for \((\omega _{\rm d}^2 + \gamma ^2 - \omega _0^2) < 0\) . For small negative detunings, discontinuous transitions are favoured at low damping (high Q ). Note that spontaneous symmetry breaking persists for γ = 0. Inspecting equation ( 5 ) and Fig. 3b for γ = 0, we see that in the absence of damping the transition occurs on the red-detuned side of the resonance, and is always of discontinuous type. The phenomena described above are enabled by strong internal plasmonic fields, which for a high-quality resonance can exceed the driving field by several orders of magnitude. The threshold driving amplitude for Berryogenesis is therefore controlled by the plasmonic quality factor, along with other device parameters. For the high quality factors ( Q > 100) recently achieved in graphene plasmonic devices 8 , we expect Berryogenesis to be achievable at moderate driving powers of order 30 W cm −2 at a frequency of approximately 25 THz (see Supplementary Information ). The spontaneous magnetization produced above threshold can be detected via numerous experimental probes 30 , 31 , 32 . We estimate for disk sizes of order 100 nm that magnetic fields exceeding several hundred nanotesla can be generated, and can be detected using precision superconducting quantum interference devices or diamond nitrogen–vacancy centre-based magnetometers 30 , 31 (see Supplementary Information for full estimate). Throughout this work, we have focused on feedback arising from self-generated Berry flux. At high excitation amplitudes, nonlinear dissipation and other sources of nonlinearity may also arise. Crucially, we work at frequencies outside the particle–hole continuum, ħω d < 2 E F , where both direct absorption from the drive (which leads to heating) and the decay of a single plasmon into a particle–hole pair are forbidden by Pauli exclusion. To suppress the nonlinear contributions to damping and heating, it is furthermore beneficial to work at lower frequencies where two- or three-photon processes are also blocked; for the simulations in this work, we used 2 E F > 3 ħω d , guaranteeing that the rates of these intrinsic dissipative processes are small throughout the parameter range we studied. Furthermore, we have checked that driving-induced anisotropies are also small throughout this regime, and are not expected to significantly affect the threshold for Berryogenesis (see Supplementary Information ). Berryogenesis is a ‘self-Floquet’ process through which the collective motion of an electronic system causes it to reconstruct its own band structure, yielding dramatic effects including non-equilibrium spontaneous symmetry breaking. Strikingly, plasmonic magnetism arises from a dynamical bistability induced by the chiral motion of (spinless) plasmons that breaks the mirror symmetry of the system; this mechanism is in stark contrast to conventional ferromagnetism that arises from (spinful) exchange physics. Looking ahead, we anticipate that other types of non-equilibrium phase transitions 33 , including complex spatiotemporal dynamics in extended systems 34 , may be triggered by analogous feedback mechanisms. This work opens new prospects for exploiting the near-field compression of electromagnetic fields in metals to realize novel non-equilibrium phases of matter. Methods Floquet band structure and Berry flux In the main text, we describe how an (internal or external) a.c. electric field may modify the electronic spectrum and Bloch band Berry curvature of a metallic system. Here we describe how we calculate the time-dependent Berry flux induced by a monochromatic a.c. field. For each value of the crystal momentum k , we find two Floquet state solutions to the time-dependent Schrödinger equation 35 , 36 : \(|\psi _{{\mathbf{k}}\alpha }(t)\rangle = {\rm e}^{ - i\varepsilon _{{\mathbf{k}}\alpha }t}|\Phi _{{\mathbf{k}}\alpha }(t)\rangle\) , where α = ± is the Floquet band index, \(\varepsilon _{{\mathbf{k}}\alpha }\) is the corresponding quasienergy and \(|\Phi _{{\mathbf{k}}\alpha }(t)\rangle\) is a periodic function of time, with period \(T = 2{\mathrm{\pi}}/\omega\) . Due to the off-resonant nature of the a.c. field, states at the Fermi surface are only weakly modified; we assume that their populations map smoothly onto the Floquet states. The periodic part of the Floquet eigenstate, \(|\Phi _{{\mathbf{k}}\alpha }(t)\rangle\) , typically involves frequency components at many harmonics of the drive frequency \(\omega\) . As a result, the Berry connection \({\cal{A}}_{{\mathbf{k}}\alpha }(t) = \langle \Phi _{{\mathbf{k}}\alpha }(t)|i\nabla _{\mathbf{k}}|\Phi _{{\mathbf{k}}\alpha }(t)\rangle\) , and ultimately the net Berry flux in the (dressed) conduction band 37 , \({\cal{F}}(t) = {\oint} {\rm d} {\mathbf{k}}\cdot {\cal{A}}_{{\mathbf{k}} + }(t)\) , will be periodic functions of time. (Here the integral is taken around the Fermi surface, which for weak dressing maps smoothly from equilibrium onto the Floquet bands.) In the analytical treatment in the main text, we focus on the steady-state (d.c.) part of the induced Berry flux, \(\overline {\cal{F}} = \frac{1}{T}{\int}_0^T {\rm d} t{\mkern 1mu} {\cal{F}}(t)\) , which provides the driving force for Berryogenesis. Through fully self-consistent numerical simulations, we show that including the a.c. part of \({\cal{F}}\) does not significantly alter our conclusions. Self-consistent solutions of the steady-state time evolution In the main text, we provide a detailed analysis of the steady states of the nonlinear dynamics described by equation ( 2 ), under the approximation that the time-periodic Berry flux \({\cal{F}}(t)\) is replaced by its time-averaged (d.c.) part \(\bar {\cal{F}}\) . From a physical point of view, this approximation is motivated by the fact that it is the d.c. part of \({\cal{F}}\) that signifies a net chirality in the system, and which we expect to be responsible for the instability towards a magnetized state. From a technical point of view, this approximation introduces a vast simplification: working in a complex representation with \({\mathbf{E}}_{{\mathrm{drive}}}(t) = {\mathbf{E}}_{{\mathrm{drive}}}^{{\mathrm{(0)}}}{\kern 1pt} {\rm e}^{ - i\omega _{\rm d}t}\) , by suppressing the time-dependent harmonics in \({\cal{F}}\) , we ensure that equation ( 2 ) supports solutions where \(\{ {\mathbf{r}}(t)\} = {\mathbf{r}}^{(0)}{\rm e}^{ - i\omega _{\rm d}t}\) and \(\{ {\mathbf{p}}(t)\} = {\mathbf{p}}^{(0)}{\rm e}^{ - i\omega _{\rm d}t}\) exhibit purely monochromatic oscillations. (Note that the physical solutions are given by the real parts of these quantities.) This simplification allows us to extract the time dependence \(\sim {\rm e}^{ - i\omega _{\rm d}t}\) from all variables, and solve a (nonlinear) algebraic equation for the steady-state amplitudes of the left- and right-circulating components of \(\{ {\mathbf{r}}(t)\}\) , \({\cal{Z}}_ \pm ^{(0)} = \frac{1}{{\sqrt 2 }}[x^{(0)} \pm iy^{(0)}]\) . These solutions are given in equation ( 4 ) of the main text. To support our conclusions and to demonstrate that the time-dependent harmonics in \({\cal{F}}(t)\) do not significantly change the behaviour of the system, we also performed self-consistent numerical simulations of the full equations of motion (equation ( 2 )). The simulations were performed as follows: 1. We first initialize the system with values of position and momentum, \({\mathbf{r}}(0)\) and \({\mathbf{p}}(0)\) , as well as an initial guess for the Berry flux \({\cal{F}}\) . The Berry flux \({\cal{F}}\) is specified by its d.c. part \(\bar {\cal{F}}\) and a list of up to \(n\) harmonics, \({\cal{F}}_1 \ldots {\cal{F}}_n\) , such that \({\cal{F}}(t) = \bar {\cal{F}} + ({\cal{F}}_1{\rm e}^{ - i\omega _{\rm d}t} + \ldots {\cal{F}}_n{\rm e}^{ - in\omega _{\rm d}t} + {\text{c.c.}})\) . The value of \(n\) is taken large enough to ensure convergence. 2. Next we numerically solve equation ( 2 ) with the supplied form of \({\cal{F}}(t)\) for a large number of periods of the drive (typically of order 100), such that the system reaches a time-periodic steady state. 3. From the last several periods of evolution, we extract the left- and right-handed components of the motion via \({\cal{Z}}_ \pm (t) = (x \pm iy)/\sqrt{2}\) . From the Fourier transform of \({\cal{Z}}_ \pm (t)\) , we extract the values at its peaks centred around frequencies \(\omega _d\) , \(2\omega _d\) and so on. 4. We use the harmonics extracted from \({\cal{Z}}_ \pm (t)\) to construct the time-periodic internal electric field and associated vector potential produced by the motion via \(e{\mathbf{E}}_{{\mathrm{int}}} = m\omega _0^2{\mathbf{r}}\) and \({\mathbf{E}}_{{\mathrm{int}}} = - \frac{1}{c}\frac{\partial }{{\partial t}}{\mathbf{A}}_{{\mathrm{int}}}\) . 5. We numerically compute the Floquet band structure of the system using the total time-periodic field \({\mathbf{A}}_{{\mathrm{tot}}} = {\mathbf{A}}_{{\mathrm{drive}}} + {\mathbf{A}}_{{\mathrm{int}}}\) . We obtain a new time-periodic Berry flux \({\cal{F}}(t)\) by integrating the Berry connection \({\cal{A}}_{{\mathbf{k}} + }(t) = \langle \Phi _{{\mathbf{k}} + }(t)|i\nabla _{\mathbf{k}}|\Phi _{{\mathbf{k}} + }(t)\rangle\) around the Fermi surface, \({\cal{F}}(t) = {\oint} {\rm d} {\mathbf{k}} \cdot {\cal{A}}_{{\mathbf{k}} + }(t)\) . Here \(|\Phi _{{\mathbf{k}} + }(t)\rangle\) is the time-periodic part of the Floquet state at crystal momentum k in the upper (+) Floquet band, derived from the original conduction band on the non-driven system. 6. Finally we return to step 1 and initialize the solver for equation ( 2 ) with the final position and momentum of the previous iteration, and a new guess for \({\cal{F}}(t)\) . This procedure is iterated until the Berry flux \({\cal{F}}(t)\) produced by the motion agrees with the form that was used to compute it (that is, until the change in \({\cal{F}}(t)\) from one iteration to the next falls below a convergence threshold). To improve the stability of the code, we introduce an interpolating factor \(\zeta\) such that the initial guess for the Berry flux for iteration \(i\) + 1, \({\cal{F}}^{(i + 1)}\) , is computed by interpolating between its value on iteration \(i\) and the new value computed in step 5 above: \({\cal{F}}_n^{(i + 1)} = (1 - \zeta ){\cal{F}}_n^{(i)} + \zeta {\cal{F}}_n\) . Here \({\cal{F}}_n\) is the n th harmonic extracted from the Fourier transform of \({\cal{F}}(t)\) computed in step 5 above. In the simulations shown, we used \(\zeta = 0.3\) . For each value of the driving amplitude, we obtained convergence of all harmonics in \({\cal{F}}\) to better than 1 part in \(10^6\) . We observe that, throughout the parameter regime studied, the higher harmonics of \({\cal{F}}\) decay very quickly with the order of the harmonic. (As a typical order of magnitude, we observe \({\cal{F}}_1/\bar {\cal{F}}\sim 10^{ - 3}\) .) Therefore, we obtain rapid convergence with respect to the number of harmonics retained (in the simulations, we keep the track the values of the first 5 harmonics of \({\cal{F}}\) ). The procedure above was used to compute the green points shown in Fig. 3a . The good agreement with the solid curves confirms the validity of our analytical treatment based on the d.c. part of \({\cal{F}}\) . Data availability The data that support the plots within this paper and other findings of this study are available from the corresponding authors upon reasonable request.

Pioneering physicists from the University of Copenhagen and Nanyang Technological University in Singapore have discovered a way to get non-magnetic materials to make themselves magnetic by way of laser light. The phenomenon may also be used to endow many other materials with new properties. The intrinsic properties of materials arise from their chemistry—from the types of atoms that are present and the way that they are arranged. These factors determine, for example, how well a material may conduct electricity or whether or not it is magnetic. Therefore, the traditional route for changing or achieving new material properties has been through chemistry. Now, a pair of researchers from the University of Copenhagen and Nanyang Technological University in Singapore have discovered a new physical route to the transformation of material properties: when stimulated by laser light, a metal can transform itself from within and suddenly acquire new properties. "For several years, we have been looking into how to transform the properties of a matter by irradiating it with certain types of light. What's new is that not only can we change the properties using light, we can trigger the material to change itself, from the inside out, and emerge into a new phase with completely new properties. For instance, a non-magnetic metal can suddenly transform into a magnet," explains Associate Professor Mark Rudner, a researcher at the University of Copenhagen's Niels Bohr Institute. He and colleague Justin Song of Nanyang Technological University in Singapore made the discovery that is now published in Nature Physics. The idea of using light to transform the properties of a material is not novel in itself. But up to now, researchers have only been capable of manipulating the properties already found in a material. Giving a metal its own "separate life," allowing it to generate its own new properties, has never been seen before. By way of theoretical analysis, the researchers have succeeded in proving that when a non-magnetic metallic disk is irradiated with linearly polarized light, circulating electric currents and hence magnetism can spontaneously emerge in the disk. Researchers use so-called plasmons (a type of electron wave) found in the material to change its intrinsic properties. When the material is irradiated with laser light, plasmons in the metal disk begin to rotate in either a clockwise or counterclockwise direction. However, these plasmons change the quantum electronic structure of a material, which simultaneously alters their own behavior, catalyzing a feedback loop. Feedback from the plasmons' internal electric fields eventually causes the plasmons to break the intrinsic symmetry of the material and trigger an instability toward self-rotation that causes the metal to become magnetic. Technique can produce properties 'on demand' According to Mark Rudner, the new theory pries open an entire new mindset and most likely, a wide range of applications: "It is an example of how the interaction between light and material can be used to produce certain properties in a material 'on demand.' It also paves the way for a multitude of uses, because the principle is quite general and can work on many types of materials. We have demonstrated that we can transform a material into a magnet. We might also be able to change it into a superconductor or something entirely different," says Rudner. He adds: "You could call it 21st century alchemy. In the Middle Ages, people were fascinated by the prospect of transforming lead into gold. Today, we aim to get one material to behave like another by stimulating it with a laser." Among the possibilities, Rudner suggests that the principle could be useful in situations where one needs a material to alternate between behaving magnetically and not. It could also prove useful in opto-electronics—where, for example, light and electronics are combined for fiber-internet and sensor development. The researchers' next steps are to expand the catalog of properties that can be altered in analogous ways, and to help stimulate their experimental investigation and utilization.

10.1038/s41567-019-0578-5

Medicine

Checkpoint blockade may be key for immunity to malaria

Samarchith P Kurup et al, Regulatory T cells impede acute and long-term immunity to blood-stage malaria through CTLA-4, Nature Medicine (2017). DOI: 10.1038/nm.4395 Journal information: Nature Medicine

http://dx.doi.org/10.1038/nm.4395

https://medicalxpress.com/news/2017-10-checkpoint-blockade-key-immunity-malaria.html

Abstract Malaria, caused by the protozoan Plasmodium , is a devastating mosquito-borne disease with the potential to affect nearly half the world's population 1 . Despite mounting substantial T and B cell responses, humans fail to efficiently control blood-stage malaria or develop sterilizing immunity to reinfections 2 . Although forkhead box P3 (FOXP3) + CD4 + regulatory T (T reg ) cells form a part of these responses 3 , 4 , 5 , their influence remains disputed and their mode of action is unknown. Here we show that T reg cells expand in both humans and mice in blood-stage malaria and interfere with conventional T helper cell responses and follicular T helper (T FH )–B cell interactions in germinal centers. Mechanistically, T reg cells function in a critical temporal window to impede protective immunity through cytotoxic-T-lymphocyte-associated protein-4 (CTLA-4). Targeting T reg cells or CTLA-4 in this precise window accelerated parasite clearance and generated species-transcending immunity to blood-stage malaria in mice. Our study uncovers a critical mechanism of immunosuppression associated with blood-stage malaria that delays parasite clearance and prevents development of potent adaptive immunity to reinfection. These data also reveal a temporally discrete and potentially therapeutically amenable functional role for T reg cells and CTLA-4 in limiting antimalarial immunity. Main CD4 + T helper cells are essential for control of malaria in mouse models of the disease 6 . We show that the frequencies of activated T helper cells increased in mice with malaria and in a cohort of Malian children ( n = 15; Supplementary Table 1 ) diagnosed with febrile malaria ( Supplementary Figs. 1c and 2 ). T helper cell depletion after established infection with the normally nonlethal Plasmodium yoelii 17XNL ( P. yoelii ) resulted in uncontrolled parasitemia and death in infected mice ( Supplementary Fig. 1a,b ). Unlike CD4 + T cell responses to most other infections in mice (for example, Listeria monocytogenes ), expansion of pathogen-specific T helper cells (defined as CD49d + CD11a hi CD4 + ) 7 in response to P. yoelii infection is distinctly biphasic. Specifically, the frequencies and total numbers of pathogen-specific T helper cells increased after P. yoelii infection and then temporarily fell or plateaued before rising again prior to parasite clearance ( Supplementary Fig. 1d–f ). Although the mechanisms underlying this unique hiatus in T helper cell expansion were unknown, we speculated that it was due to the widely acknowledged immunosuppression that occurs in blood-stage malaria 5 . This notion was reinforced by our observation of increased numbers and frequencies of FOXP3 + CD4 + T reg cells, a key immunosuppressive cell population, during the blood stage of malaria in humans and mice ( Fig. 1a–d ). We observed that higher parasite densities were associated with higher frequencies of T reg cells in humans; also, chloroquine treatment to decrease parasitemia reduced T reg cell frequencies and increased T helper cell frequencies in mice ( Supplementary Fig. 3a–c ). The role of T reg cells in malaria remains controversial 5 , 8 , 9 : some independent studies suggest that T reg cells suppress protection 10 , 11 , 12 , while others imply that they enhance it 13 , 14 . Importantly, these studies manipulated T reg cells (most of which express CD25) before or shortly after Plasmodium infection using anti-CD25 antibody–mediated depletion 10 , 11 , 13 , 14 , 15 or the (more precise) Foxp3 –diptheria toxin receptor (DTR) system 16 , 17 . Although variations in approaches may underlie these inconsistencies, the timing of T reg cell targeting could be a critical consideration for understanding malaria. Here we observed that expansion of T reg cells in P. yoelii –infected mice preceded or coincided with the hiatus in T helper cell responses, suggesting a causal relationship beginning ∼ 10 days post infection (d.p.i.). To test this proposition, we depleted both circulating and lymphoid 18 , 19 , 20 T reg cells in P. yoelii –infected Foxp3 -DTR ( Supplementary Fig. 4a,b ) and C57BL/6 ( Supplementary Fig. 5a ) mice with diphtheria toxin 21 and anti-CD25 antibody 10 , respectively, beginning at 9 d.p.i. T reg depletion interrupted the hiatus in T helper cell responses, restored expansion of Plasmodium -specific T helper cells and substantially accelerated control of P. yoelii infection ( Fig. 1e,f and Supplementary Fig. 5b,c ). In contrast, T reg depletion at 0 and 2 d.p.i. in Foxp3 -DTR mice resulted in the death of P. yoelii –infected mice ( Supplementary Fig. 4c ). Figure 1: T reg cells expand and modulate T helper cell responses and immunity to malaria. ( a ) Longitudinal frequencies of FOXP3 + T reg cells among the total CD4 + T cell population in peripheral blood mononuclear cells (PBMCs) from a cohort of children in Mali ( Supplementary Table 1 ) before, during, and after infection with acute febrile malaria. Each connected line indicates a unique subject. ( b ) Kinetics of activated (CD49d + CD11a hi ) T helper or FOXP3 + T reg cell frequencies among the total CD4 + T cell population in circulation in P. yoelii –infected C57BL/6 mice. ( c , d ) Absolute numbers of FOXP3 + CD4 + T reg cells in spleen ( c ) and lymph nodes (LN) ( d ) at the indicated time points in P. yoelii –infected C57BL/6 mice. ( e , f ) Frequencies of activated CD4 + T helper cells among the total CD4 + T cell population in circulation ( e ) and parasitemia ( f ) at various time points following infection with P. yoelii in C57BL/6 or Foxp3 -DTR mice treated with diphtheria toxin (Rx DT) or PBS (Rx PBS) at 9 and 11 d.p.i., as indicated by the pink label below the x axis. All experimental data represent one of at least three separate experiments with five mice per group and are presented as mean ± s.e.m. Statistical analysis was performed by comparing the indicated groups using one-way ANOVA with Bonferroni correction in a ( F 3,52 = 18.89) or the groups of Foxp3 -DTR mice treated with diphtheria toxin or PBS and C57BL/6 mice treated with diptheria toxin using two-way ANOVA with Tukey's correction in e ( F 2,60 = 28.6) and f ( F 2,53 = 62.17). ** P ≤ 0.01. Full size image To further investigate the opposing contributions of T reg and T helper cells in the control of parasitemia during the hiatus in T helper cell responses in P. yoelii –infected mice, we took advantage of the differential expression of high- and low-affinity IL-2 receptors by these cell populations. Specifically, beginning at 9 d.p.i., we treated mice with either IL-2–JES6 antibody complexes that signal through the high-affinity IL-2 receptor CD25 and amplify CD25 + T reg cells or IL-2–S4B6 antibody complexes that selectively expand pathogen-specific T helper cells by signaling through the low-affinity IL-2/IL-15 receptor, which is expressed by activated T helper cells 22 . Increasing the frequencies of T reg cells further dampened pathogen-specific T helper cell responses, resulting in higher parasitemia and death, whereas increasing pathogen-specific T helper cell frequencies resulted in better control of infection ( Supplementary Fig. 5d–f ). Together, these data suggest that T reg cells suppress T helper cell responses during a critical window of time in blood-stage malaria, compromising control of acute infection. There are two major mechanisms through which T reg cells counter T helper cell responses in the context of infection: IL-10-mediated inhibition and CTLA-4-mediated repression of co-stimulation by antigen-presenting cells (APCs) 23 . In mouse malaria, T reg cells transcriptionally upregulate both IL-10 and CTLA-4 (ref. 16 ). In accordance with another study 16 , blocking IL-10 at 9 d.p.i. failed to alter the T helper cell response and the course of parasitemia ( Supplementary Fig. 6 ). However, at 9 d.p.i. in P. yoelii –infected mice, T reg cells exhibited enhanced upregulation of CTLA-4 in comparison to T reg cells from mice with acute infection with influenza or vaccinia virus ( Fig. 2a,b ). Additionally, elevated amounts of soluble CTLA-4, potentially cleaved from the surface of T cells, were detectable in blood plasma and spleen lysates throughout the course of P. yoelii infection ( Fig. 2c ). Of note, of all the T reg cells, the percentage expressing CTLA-4 in P. yoelii –infected mice was substantially higher than the percentage of T helper cells that had detectable CTLA-4 expression ( Fig. 2d ) 16 . In accordance with these mouse data, longitudinal analyses of samples from humans without malaria obtained at the end of the dry season in Mali and samples taken subsequently at diagnosis of febrile malaria showed increased frequencies of circulating CTLA-4 + T helper cells and CTLA-4 + T reg cells in blood-stage malaria; these frequencies returned to preinfection levels after treatment with an antimalarial drug ( Fig. 2e,f and Supplementary Fig. 7a ). Febrile malaria in humans was associated with higher frequencies of T reg cells positive for Helios (a marker of superior suppressive function) 24 , 25 , Helios + CTLA-4 + T reg cells, and CTLA-4 + follicular T reg (T FR ) cells 26 , 27 ( Supplementary Fig. 7b–d ). Together, these data suggest that T reg cells may modulate T helper cells, and possibly humoral immunity to blood-stage malaria, through CTLA-4. Figure 2: CTLA-4 expression is enhanced in malaria and is integral to T FH –B–T FR cell interactions in GCs. ( a ) Representative histograms quantifying CTLA-4 expression in FOXP3 + T reg cells in C57BL/6 mice infected with influenza A virus (IAV), vaccinia virus (VacV), or P. yoelii at 9 d.p.i. Closed histograms represent isotype controls. Inset gates and numbers correspond to the proportions of CTLA-4 + FOXP3 + T reg cells. ( b ) Summaries of the proportions of CTLA-4 + FOXP3 + T reg cells shown in a . Boxes and whiskers depict the limits of the data distribution in each group, with upper and lower whiskers representing the range of data distribution, the box perimeters representing its 25 th and 75 th percentiles, and the midline showing median of the distribution. ( c , d ) Kinetics of CTLA-4 expression in spleen or serum ( c ) and of the proportion of splenic FOXP3 + T reg cells or T helper cells expressing CTLA-4 relative to the total T reg cell or T helper cell population, respectively ( d ), during the T helper cell hiatus after P. yoelii infection in C57BL/6 mice. The dashed line in c is the threshold of detection. ( e , f ) Proportions of T helper ( e ) and T reg ( f ) cells expressing CTLA-4 in a cohort of children in Mali before, during, and after acute febrile malaria as compared to healthy controls (US control). ( g ) Representative pseudocolored images of fluorescently labeled sections from the spleen (left, middle) and lymph node (right) of a C57BL/6 mouse with a resolving P. yoelii infection (21–27 d.p.i.). T FH –B–T FR cell clusters in the nascent GCs are enclosed by dashed circles in the left panel; a magnified view of an encircled GC from the left panel is shown in the middle panel. A single B cell and CTLA-4 + T FR cell in apposition are shown in the right panel. Markers and their colors are indicated in the figure. All experimental data represent one of at least three experiments with five mice per group and are presented as mean ± s.e.m. Confocal microscopy images represent at least five regions observed in nine sections examined in three separate mice from three separate experiments. Statistical analysis was performed by comparing the indicated groups using two-way ANOVA with Bonferroni correction in b ( F 2,14 = 25.08), two-tailed Student's t -test in d , or one-way ANOVA with Tukey's correction in e ( F 3,52 = 18.70) and f ( F 3,52 = 18.89). * P ≤ 0.05, ** P ≤ 0.01. Full size image Humoral immunity depends on efficient T FH –B cell cooperation in secondary lymphoid organs and is perhaps the most important component of the acquired immunity that controls blood-stage malaria 28 , 29 . To examine the potential of T reg cells and CTLA-4 to interfere with humoral immunity against malaria, we examined their relationships with the T FH –B cell interactions in germinal centers (GCs) by confocal and intravital microscopy in mice. Within the GCs in secondary lymphoid organs, CD4 + T helper cells and GL-7 + B220 + B cells formed discrete clusters of interaction after P. yoelii infection ( Supplementary Video 1 ). These clusters were composed of CTLA-4 + T FR and T FH cells 30 in close apposition with GC plasmablasts or B cells ( Fig. 2g and Supplementary Video 2 ). Expression of neuropilin-1 (NRP-1) or FOXP3 distinguished T FR cells from T FH cells 31 . In the context of infection, CTLA-4 receptors expressed on T cells bind to B7 ligands on APCs and limit immune responses by competitive inhibition of B7–CD28 co-stimulatory interactions 32 . As B cells are the primary APCs that sustain T FH cell responses 33 and dictate protective antibody responses in malaria, we investigated whether GC B cells interacted with CTLA-4 during the course of P. yoelii infection. In accordance with this notion, CTLA-4 + T FR cells directly associated with GC B cells ( Fig. 2g and Supplementary Video 3 ) during P. yoelii infection, and CTLA-4 was detectable at the T FR –B cell interface ( Supplementary Video 4 ). Moreover, intravital imaging showed that individual T FR cells transiently interacted with multiple B cells in GCs ( Supplementary Video 5 ), suggesting an explanation for how the relatively few T FR cells could effectively modulate the GC reaction. We failed to detect other potential CTLA-4-driven suppressive mechanisms of T reg cells 32 , including induced idoleamine 2,3-dioxygenase (IDO) production ( in vitro ) in B cells or discernable transendocytosis of B7 molecules ( in vitro or in vivo ) from B cells after P. yoelii infection (data not shown). These observations suggest that the CTLA-4 expressed on or released from T FR cells might directly bind B7 ligands on the surface of GC B cells, restricting productive co-stimulation of T FH cells and limiting production of antibody-secreting plasma cells and memory B cells. Thus, blocking CTLA-4–B7 interactions might augment humoral immunity and clearance of blood-stage malaria. Our precise definitions of T reg cell kinetics, CTLA-4 expression dynamics, and the timing of GC reactions during the course of P. yoelii infection suggest that CTLA-4–B7 interactions may be meaningfully targeted during the hiatus in T helper cell expansion to improve immune responses and parasite clearance. Hence, P. yoelii –infected C57BL/6 mice were treated with CTLA-4-blocking (anti-CTLA-4) or IgG control antibodies at the onset of the hiatus in the expansion of pathogen-specific T helper cells ( Fig. 3a ). Similar to T reg cell depletion, therapeutic blockade of CTLA-4 truncated the hiatus and enhanced the total numbers of CD4 + T cells, pathogen-specific T helper cells, follicular CD4 + T cells, and T FH cells in the spleen as compared to treatment with IgG control antibodies ( Fig. 3b–e ). Hypothetically, CTLA-4 blockade could target T reg cells, T helper cells, or both ( Fig. 2d ). However, CTLA-4 blockade failed to further improve the T helper cell response in T reg cell–depleted mice, suggesting only a minor role for CTLA-4 expressed on T helper cells ( Supplementary Fig. 8 ). Unlike in tumor models 34 , anti-CTLA-4 treatment during the course of malaria did not deplete T reg cells or T FR cells in the spleen ( Fig. 3f and Supplementary Fig. 9 ). We observed a corresponding increase in the total numbers of splenic B cells, plasmablasts, GC B cells, GC plasmablasts, and P. yoelii –specific, protective antibody titers in serum 35 of anti-CTLA-4-treated mice ( Fig. 3g–k ). Inhibiting GC B cell formation with anti-CD40L treatment 33 prevented the revival of T helper cell responses after CTLA-4 blockade, indicating that T reg cell interaction with GC B cells likely mediated the repression of T helper cell responses ( Supplementary Fig. 10 ), although effects on other APCs remain possible. CTLA-4 blockade resulted in considerably reduced splenomegaly ( Fig. 3l ), a hallmark of P. yoelii infection, and improved splenic architecture, with distinct T cell zones, B cell follicles, and GC reactions visible shortly after treatment ( Fig. 3m ). Therapeutic blockade of CTLA-4 dramatically accelerated control of P. yoelii infection in C57BL/6 and BALB/c mice compared to control IgG treated mice and partially rescued BALB/c mice (40% survival) from lethal Plasmodium berghei ANKA infection ( Fig. 4a–c ). However, CTLA-4 blockade before or after the critical window of expansion of T reg cells did not resolve the T helper cell hiatus or accelerate control of P. yoelii infection ( Supplementary Fig. 11 ), in accordance with some previous results 17 , 36 . We previously showed that blockade of programmed cell death 1 protein (PD-1) and lymphocyte-activation gene 3–encoded protein (LAG-3) signaling beginning at 14 d.p.i., during the post-hiatus revival of T helper cell responses, resulted in accelerated clearance of P. yoelii infection in mice 7 . In contrast, blocking PD-1 and LAG-3 signaling starting at 9 d.p.i., during the T reg cell–mediated interruption in T helper cell responses, did not improve immunity or parasite clearance, indicating a minimal contribution of these pathways to the dampening of immune responses during this critical interval ( Supplementary Fig. 12 ). Of note, stimulation of OX40 signaling starting at 7 d.p.i. can also improve immunity against blood-stage P. yoelii by enhancing T helper cell responses 37 . Together, these results reinforce the notion that immunomodulation in blood-stage malaria is based on multiple molecular pathways that may be dominant during discrete time windows over the course of the infection. Figure 3: CTLA-4 blockade enhances CD4 + T cell and B cell responses, GC reaction, and antibody titers after P. yoelii infection. ( a ) A schematic of therapeutic blockade in P. yoelii –infected C57BL/6 mice; i.v., intravenous injection; i.p., intraperitoneal injection. ( b – k ) Total numbers of CD4 + T helper ( b ), Plasmodium -specific CD49d + CD11a hi CD4 + T helper ( c ), CXCR5 + ICOS + PD-1 + CD4 + follicular T ( d ), CXCR5 + ICOS + PD-1 + FOXP3 − CD4 + T FH ( e ), and CXCR5 + ICOS + PD-1 + FOXP3 + CD4 + T FR ( f ) cells, CD19 + B220 + B cells ( g ), CD138 + IgD − CD19 + B220 + plasmablasts ( h ), CD95 + GL7 + CD19 + B220 + GC B cells ( i ), and CD95 + GL7 + CD138 + IgD − CD19 + B220 + GC plasmablasts ( j ) in spleen at various time points and the relative titers of P. yoelii merozoite surface protein (MSP 1–19 )-specific antibody in serum at 18 d ( k ) following P. yoelii infection in C57BL/6 mice with or without CTLA-4 blockade as depicted in a . Data are presented as mean ± s.e.m. at each time point or serum dilution and represent one of three separate experiments, each with at least five mice per group. Statistical analysis was performed by comparing the treatment and control groups at the indicated time points in b – j or serum dilutions in k with two-tailed Student's t -tests. * P ≤ 0.05, ** P ≤ 0.01. ( l , m ) Representative images of the gross appearance at 18 d.p.i. of spleens from P. yoelii –infected C57BL/6 mice with or without CTLA-4 blockade ( l ) and H&E-stained or pseudocolored fluorescently labeled sections of those spleens at the indicated time points ( m ). Markers and their colors are indicated in the figure in m . All microscopy images represent at least five regions observed in nine sections examined from three separate mice from three separate experiments. Full size image Figure 4: Therapeutic blockade of CTLA-4 enhances immunity to malaria. ( a – c ) Percentages of parasitemia at the indicated time points in C57BL/6 ( a ) and BALB/c ( b , c ) mice infected with P. yoelii ( a , b ) or P. berghei ( c ) with or without CTLA-4 blockade as depicted in Figure 3a . The time of treatment is indicated by the pink label under the x -axis. ( d ) Parasitemia at the indicated time points in P. yoelii –infected C57BL/6 mice that received serum (100 μl) at 0 d.p.i., obtained from donor mice treated as in a at 56 d.p.i. ( e , f ) Parasitemia at the indicated time points ( e ) and survival ( f ) in mice from a heterologously challenged with P. berghei at 56 d.p.i. All data are representative of one of at least three separate experiments; each experiment began with five mice per group. Data are presented as mean ± s.e.m. Statistical analysis was performed by comparing the corresponding treatment and control groups at the indicated time points with two-tailed Student's t -tests in a – e or a chi-squared test in f . * P ≤ 0.05, ** P ≤ 0.01. †, death of one mouse in the corresponding group. Full size image CTLA-4 blockade might be a less realistic independent treatment option for malaria in endemic areas because of its requirements for frequent dosages, parenteral administration, and precise timing; its potential for toxicity; and its currently prohibitive costs. Nevertheless, why humans fail to generate potent adaptive immunity to subsequent Plasmodium infections, which may also involve multiple species of the parasite 4 , 38 , is a major unresolved issue in malaria pathogenesis. To investigate the role of CTLA-4 in limiting the generation of long-term immunity, we transferred sera obtained at 56 d.p.i. from C57BL/6 mice that cleared P. yoelii infection with or without CTLA-4 blockade into C57BL/6 recipients with fresh (0 d.p.i.) or established (10 d.p.i.) P. yoelii infections. Recipients of sera from anti-CTLA-4-treated mice, which contained elevated amounts of P. yoelii –specific antibodies ( Fig. 3k ) but no detectable residual anti-CTLA-4 antibodies ( Supplementary Fig. 13a ), exhibited significantly lower parasitemia than recipients of sera from mice without anti-CTLA-4 treatment ( Fig. 4d and Supplementary Fig. 14 ). To test whether anti-CTLA-4 therapy aided long-term and perhaps species-transcending immunity, we rechallenged the C57BL/6 mice originally infected with P. yoelii and cured with or without CTLA-4 blockade, using the lethal P. berghei ANKA strain at 56 or 100 d.p.i. CTLA-4 blockade during P. yoelii infection resulted in durable, CD4 + T cell–driven immunity against P. berghei ANKA and significantly improved long-term survival of the mice ( Fig. 4e,f and Supplementary Fig. 13b–d ). Taken together, these findings suggest that CTLA-4 expression by T reg cells is potentially a major mechanism in the limitation of acquired, cross-species immunity to malaria. Direct evidence of a role for T reg cells and CTLA-4 in limiting protection from Plasmodium reinfection in humans can only be obtained through clinical trials and, like the translation of checkpoint blockade from animal models to human cancer immunotherapy 39 , 40 , the precise pathway forward for research in malaria treatment must be carefully defined. However, our findings provide important mechanistic insights to consider while evaluating evidence-based interventions to target host immunity for improved control of malaria. Malaria is a global health threat, with close to 200 million clinical cases and >500,000 deaths reported annually 1 . Therefore, it is critically important to understand how Plasmodium protozoa circumvent effective immune responses in humans 2 . Here we build on detailed studies of immune-cell dynamics during the blood stage of malaria in humans and mice to show how T reg cells can act in a discrete temporal window through CTLA-4 to suppress T helper cell and humoral immune responses. Thus, T reg cells may function as an essential component of the immunoregulation observed in blood-stage malaria to inhibit clearance of acute infection and development of long-term sterilizing immunity to future infections. Methods Malian and US blood donors. The Ethics Committee of the Faculty of Medicine, Pharmacy and Dentistry at the University of Sciences, Technique, and Technology of Bamako and the National Institute of Allergy and Infectious Disease of the National Institutes of Health (NIAID/NIH) Institutional Review Board approved the study in Mali (protocol no. 11-I-N126). Informed consent was obtained from the parents or guardians of participating children. The field study, described in detail elsewhere 2 , was conducted in Mali, where malaria transmission is seasonal. Blood samples were obtained from children at their healthy, uninfected baseline before the malaria season, during their first febrile malaria episode of the ensuing malaria season, and 7 d after treatment of this initial episode. Blood samples of healthy US adults were obtained from the NIH blood bank for research use after written informed consent was obtained from all study participants enrolled in a protocol approved by the NIH Institutional Review Board (protocol no. 99-CC-0168). Mice and pathogens. Female C57BL/6, BALB/c and Foxp3 -DTR (B6.129(Cg)- Foxp3 tm3 ( DTR / GFP ) Ayr /J) mice aged 6–8 weeks were obtained from the Jackson Laboratories. Foxp3 -GFP reporter mice 41 (a gift from S. Perlman, University of Iowa) and Bcl6 -RFP reporter mice 42 (a gift from S. Crotty, La Jolla Institute for Allergy and Immunology) were crossbred to generate Foxp3 -GFP× Bcl6 -RFP.B6 mice. All mice were housed at the University of Iowa Animal Facilities at the appropriate biosafety levels and were subjected to studies approved by the University of Iowa Animal Care and Use Committee. Mice were inoculated with 0.8–1.2 × 10 6 P. yoelii 17XNL– or 8 × 10 5 P. berghei ANKA–infected erythrocytes (originally obtained from the Insectary Core Facility at New York University or clone 234 obtained from Imperial College, London, respectively); 1 × 10 5 colony-forming units (CFU) of actin-assembly-inducing protein (ActA) from L. monocytogenes (strain DP-L1942) 43 i.v.; 2 × 10 4 tissue culture infectious dose (TCID) 50 of influenza A virus (PR8) intranasally; or 5 × 10 6 plaque-forming units (PFU) of vaccinia virus (Western Reserve) epicutaneously on the ear. Flow cytometry. In human studies, PBMCs were isolated and stained, and FACS analysis was performed as described previously 44 . FACS reagents were purchased from BioLegend, BD Biosciences or eBiosciences and included antibodies to human CD3 (catalog no. UCHT1), CD4 (catalog no. SK3), CD45RO (catalog no. UCHL1), CD45RA (catalog no. HI100), CXCR5 (catalog no. MU5UBEE), CXCR3 (catalog no. G025H7), PD-1 (catalog no. EH12.2H7), HLA-DR (catalog no. LN3), CD38 (catalog no. HIT2), Foxp3 (catalog no. 259D), Helios (catalog no. 22F6), CTLA-4 (catalog no. 14D3), and Ki-67 (catalog no. 20Raj1). In mouse studies, parasitemia frequencies were determined by flow cytometry as described 45 . To phenotype lymphocytes from spleen, lymph nodes, or blood, single-cell suspensions were stained on the surface with mouse antibodies to CD16/32 (catalog no. 2.4G2), CD4 (catalog no. RM4-5), CD49d (catalog no. R1-2), CD11a (catalog no. M17/4), PD-1 (catalog no. RMP1-30), CXCR5 (catalog no. L138D7), ICOS (catalog no. C98.4A), CD19 (catalog no. 6D5), GL-7 (catalog no. GL7), CD95 (catalog no. 15A7), CD138 (catalog no. 281-2), IgD (catalog no. -26c), B220 (catalog no. RA3-6B2), CD80 (catalog no. 16-10A1), or CD86 (catalog no. GL-1), or intracellularly with antibodies to CTLA-4 (catalog no. U10-4B9), IDO1 (catalog no. 2E2/IDO1), or FOXP3 (catalog no. FJK-16s) obtained from BioLegend, eBioscience, or BD Biosciences. FOXP3 staining kit (eBiosciences) was used for intracellular staining with the manufacturer's instructions. Multicolor flow cytometry was performed on a BD LSRFortessa and results were analyzed with FlowJo software (Tree Star). Transendocytosis assay. As a modification of the assay described elsewhere 46 , 47 , Foxp3 -eGFP + T reg cells and Foxp3 -eGFP − T helper cells (as controls) were sorted using FACS from P. yoelii –infected (9 or 12 d.p.i.) Foxp3 -eGFP donor mice and plated in a 1:2 ratio with LPS-matured dendritic cells (DCs) for 3 h in the presence of bafilomycin A. CD80/86 loss by DCs or gain by T reg and T helper cells were assessed by surface or intracellular staining, respectively, and flow cytometry. Therapeutic regimens. The following are the dosages of various reagents used in mice, along with the appropriate IgG controls or the diluent: (i) anti-CTLA-4 monoclonal antibody (mAb) (catalog no. UC10-4F10-11, BioXcell) at 500 μg per mouse, i.p., at 9, 11, 13, 15 and 17 d.p.i., (ii) anti-IL-10 mAb (catalog no. JES5-2A5, BioXcell) at 100 μg per mouse, i.p., at 9, 11, 13, 15 and 17 d.p.i., (iii) anti-CD25 mAb 48 (catalog no. PC61.5; a gift from S. Varga, University of Iowa) at 500 μg per mouse, i.p., at 9, 11, 13, 15 and 17 d.p.i., (iv) IL-2–anti-IL-2 (catalog no. S4B6, mAb, PeproTech/ATCC) or IL-2–anti-IL-2 (catalog no. JES6-1A12, mAb, PeproTech/ATCC) complexes, made as previously described 46 at 1.5 μg per mouse, i.p., at 9, 11, 13, 15 and 17 d.p.i., (v) anti-CD40L mAb 33 (MR-1; a gift from T. Waldschmidt, University of Iowa) at 1 mg per mouse, i.v., at 9 and 11 d.p.i., (vi) anti-CD4 mAb (GK1.5, BioXcell), i.p., at 400 μg per mouse at 9 and 11 d.p.i., (vii) DT (Sigma-Aldrich) at 1 μg per mouse, i.p., at 9 and 11 d.p.i., (viii) anti-PDL-1 (catalog no. 10F.9G2, BioXcell) and anti-LAG-3 (prepared from hybridoma clone C9B7W (tested negative for mycoplasma), a gift from D.A.A. Vignali) at 100 μg per mouse each, i.p., at 9, 11, 13, 15 and 17 d.p.i. and (ix) chloroquine (CQ) at 10 mg per kg bodyweight, i.p., at 7, 9, 11, 13, 15 and 17 d.p.i. Microscopy. Spleen or lymph node sections collected from mice were fixed, stained, and imaged using the Zeiss LSM 710 laser scanning confocal microscope as described in detail previously 7 . Direct fluorochrome-conjugated antibodies to CD4 (catalog no. GK1.5), B220 (catalog no. RA3-6B2), GL-7 (catalog no. GL7), CD138 (catalog no. 281-2), NRP-1 (catalog no. 761705), CTLA-4 (catalog no. U10-4B9) or FOXP3 (catalog no. 150D) from BioLegend or R&D Systems were used to stain the sections. The cryosections were permeabilized with 1% Triton X-100 (Fischer Bioscience) in intracellular staining for CTLA-4 and FOXP3. For intravital confocal microscopy, Foxp3 -GFP × Bcl6-RFP.B6 mice were injected with B220–Alexa Fluor 647 (Biolegend, 50 μg, i.v.) 14 h before imaging. Mice were anesthetized with ketamine and xylazine (87.5 and 12.5 mg per kg body weight, respectively) and placed with an exposed spleen in dorsal recumbency on the microscope base in a continuously heated (37 °C) enclosed chamber (Leica). A custom suction tissue window apparatus (VueBio) was placed on the spleen with 20–25 mm Hg of negative pressure to immobilize the tissue against a fixed coverslip. Images were acquired on a Leica SP8 Microscope (Leica) using a 25×, 0.95 NA water-immersion objective with coverslip correction. High-resolution confocal stacks of 30–54 xy sections sampled with 1-μm z spacing were acquired at an acquisition rate of 40 frames per second to provide image volumes of 170/388 × 170/388 × 30–54 μm 3 . Sequences of image stacks were transformed into volume-rendered, 4D time-lapse videos with Imaris software (Bitplane). ELISA. MSP 1-19 -specific antibodies in sera were detected as described previously 45 . Results are presented as the average endpoint titer, with absorbance readings at 450 nm. To quantify CTLA-4 in tissues, homogenates of whole, weighed spleens or sera were tested with the DuoSet mouse CTLA-4 ELISA kit (R&D Systems) according to the manufacturer's protocol. Residual anti-CTLA-4 antibody in mouse serum was detected using anti–hamster IgG (Poly4055, BioLegend) and quantified using anti-CTLA-4 mAb (UC10-4F10-11, BioXcell) as standard. Statistical analyses. For data from human subjects and mice, data were compared using paired or unpaired Student's t -tests, chi-squared tests or ANOVA as appropriate. Bonferroni adjustments ( t -tests) and Tukey's corrections (ANOVA) were applied to give a more precise confidence interval (of at least 95%) for differences among the groups in single or multiple comparisons, respectively. All analyses were performed in Prism 6.0h (GraphPad Software). Data availability. Data are available from the authors on reasonable request. A Life Sciences Reporting Summary is available. Additional information Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

A molecule that prevents the immune system from attacking cancer may play a similar role with malaria. A new study by researchers at the University of Iowa Carver College of Medicine shows that targeting the molecule at the right time during infection allows mice to quickly clear malaria. Importantly the treated mice also develop lasting immunity to malaria. The molecule is a checkpoint protein called anti-cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4). In cancer, CTLA-4 is a target for new immunotherapy drugs that prevent cancer from supressing the immune system. In the new study, published in Nature Medicine, the UI researchers show that CTLA-4 is expressed and released by a subset of immune cells called regulatory T cells (Tregs) that are involved in immune suppression. "People have long known that malaria is associated with a huge immunosupressive response, but no one knew the mechanism," say Samarchith (Sam) Kurup, PhD, a UI assistant research scientist and first author on the study. "The CTLA-4 molecule interferes with appropriate immune activation," explains John Harty, PhD, UI professor of microbiology and immunology and senior study author. "Specifically, it interferes with the function of several types of T helper cells whose job it is to drive immune responses against malaria." Malaria is a serious, life-threatening disease across most of the developing world. There are 200 million cases of malaria each year and almost 500,000 malaria-related deaths. The malaria parasite, which is transmitted by mosquito bites, persists in the blood for long periods. Critically, although there are drugs that can cure the infection, humans do not develop immunity to infection, which means a person can get reinfected with malaria year after year. Multiple infections eventually make a person resistant to the severity of the disease symptom, but when the person - usually a young child - experiences their first few infections, the disease can be deadly. Malaria meddles with immune responses In most infectious diseases, a group of immune cells called T helper cells increases in number, cooperates with B cells to make antibodies that clear the infection, then leaves behind memory T and B cells to defend against reinfection with the same pathogen. That doesn't happen in malaria. Instead, there is a period where the T helper cell expansion stalls. During that critical time period, the UI reseachers showed that the Treg population increases. "Our observation that the Tregs went up when the T helper cells stopped going up showed a timing relationship that suggested the possibility of a functional relationship," says Harty, who also is a UI professor of pathology. When the UI team eliminated Treg cells in mice with blood-stage malaria infections, the expansion of the T helper cells did not plateau; they kept expanding and cleared the infection faster. Further experiments revealed that Tregs suppress the normal immune response by continuously expressing and shedding the CTLA-4 molecule, which interferes with the normal immune function of T helper cells. It also interferes with another set of T helper cells (follicular T helper cells) whose job it is to help make antibodies against malaria. Malaria prevents the host body (human or mouse) from developing lasting immunity. A new University of Iowa study, led by microbiology professor John Harty, has homed in on a potential culprit of this immunosuppression. The molecule, CTLA-4, is produced by a type of immune cell called a Treg cell. Blocking CTLA-4 at the right time during blood-stage infection allows mice to quickly clear malaria. Importantly the treated mice also develop lasting immunity to malaria.The video, captured with intravital confocal immunofluorescence microscopy, shows Treg cells (green) delivering CTLA-4 to B cell (blue/ magenta)-Helper T cell (red) clusters in the spleen of a mouse. The CTLA-4 molecule inhibits efficient production of antibodies against malaria.If this pathway works in humans as it does in mice, blocking CTLA-4 might be a way to improve malaria treatment and boost immunity to reinfection. The study was published Sept. 11 online in Nature Medicine. Credit: Samarchith Kurup/ Scott Anthony, dept. of Microbiology and Immunology, University of Iowa. "Tregs meddle with these processes through the CTLA-4 molecule," Harty says. Blocking CTLA-4 at the right time during blood-stage infection cured mice of the infection and promoted immunity against reinfections. It even provided protection against challenge from another deadlier malaria parasite. Previous work in Harty's lab found that blocking a different checkpoint protein called PDL1 at a later point in malaria infection also improved the host immune response to malaria. The new work shows that the CTLA-4 pathway is in play at an earlier stage in malaria infection, and shows that the two pathways don't overlap. "Both pathways impede the appropriate activation of the immune system, but in different ways, targeting different interactions, and at different time points," Harty says. "The more we understand how and when these pathways are operating, the better chance we have to rescue them." Of mice and men Harty and Kurup are quick to point out that findings in mice often do not translate easily to human patients, but access to unique human data may help determine if the CTLA-4 findings are relevant in humans. For about a decade, Harty has collaborated with Peter Crompton, an National Institutes of Health (NIH) scientist who works with malaria patients in the African country of Mali. The clinic where Crompton's colleagues work tracks around 700 children year after year. During each wet season, when malaria is endemic, children are diagnosed and treated for malaria, and the team collects blood samples for immunological studies. "This has been a very potent resource for us," Harty says. "When we look at blood samples from the same infection timeframe that we investigated in the mouse, we see some of the same immune changes (expansion of Tregs and upregulation of CTLA-4) are also happening in humans. That does not prove that everything is the same, but at the level of resolution that we have, there is some reasonable similarity." "Practically, we have shown there is a pathway that can be targeted, and although CTLA-4 blockers that are available as cancer immunotherapies are too costly and impractical to use for malaria, there may be other parts of this immunological pathway that could be targeted using other drugs or small molecules, to produce the same effect," Kurup says. The malaria parasite is adept at developing resistance to antimalarial medicines that target it directly, as has happened time and time again. The UI approach focuses on modulating or improving the immune response of the host. "The parasite can't become resistant to that," Harty says.

10.1038/nm.4395

Biology

Wild monkeys use loud calls to assess the relative strength of rivals

Marcela E. Ben?tez et al. Evidence for mutual assessment in a wild primate, Scientific Reports (2017). DOI: 10.1038/s41598-017-02903-w Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-017-02903-w

https://phys.org/news/2017-06-wild-monkeys-loud-relative-strength.html

Abstract In aggressive interactions, game theory predicts that animals should assess an opponent’s condition relative to their own prior to escalation or retreat. Despite the benefits of such mutual assessment , few studies have been able to reject simpler assessment strategies. Here we report evidence for mutual assessment in a wild primate. Gelada ( Theropithecus gelada ) males have conspicuous loud calls that may function as a signal of male quality. “Leader” males with harems putatively use loud calls to deter challenges from non-reproductive “bachelor” males. By contrast, leader males pose no threat to each other and congregate in large groups for a dilution effect against bachelors. In playback experiments and natural observations, gelada males responded to loud calls according to both their own and their opponent’s attributes. Although primates routinely classify others relative to themselves using individual attributes, this represents some of the first direct evidence for mutual assessment in primate signaling contests. Introduction Limited resources lead animals into contests. Because aggressive contests are costly, game theory predicts that contestants will assess the costs and benefits of a particular contest before escalating 1 , 2 . Contestants with a high ability to compete (i.e., high resource holding potential – RHP) should escalate the contest, while those with a low ability to compete (low RHP) should withdraw. Despite the simplicity of this prediction, there is enormous debate about how animals make these decisions 3 . It stands to reason that a contestant should gather information about their opponent’s condition and compare that to their own ( mutual assessment 2 ). However, many empirical studies find it difficult to reject “simpler” assessment strategies 4 such as self-assessment (relying solely on one’s own condition 5 ), or opponent-only assessment (relying solely on a rival’s condition 6 ). For example, when an inferior contestant withdraws from an aggressive contest with an opponent, the contestant may indeed be using mutual assessment, or they may simply be withdrawing because the damage incurred was too high. Therefore, a “cumulative” self-assessment strategy is difficult to distinguish from a “sequential” mutual assessment one 4 . Non-contact displays – such as those involving animal signals – avoid this problem entirely 7 . Contestants do not accumulate sufficient costs during displays for a cumulative assessment strategy to operate. Therefore, measuring receiver responses based on the relative quality of the signaler and the receiver makes for a strong test of mutual assessment. However, within the vast literature documenting receiver responses to putative signals based on the quality of the signaler 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , only a handful of studies also examined receiver responses based on the relative quality of the signaler and the receiver 7 , 19 , 20 , 21 , 22 . Thus, the current evidence for many signals is only sufficient for identifying opponent-only assessment. In practice, however, it is likely that many of these taxa may be using mutual assessment. In non-human primates (hereafter, “primates”), there is very little evidence for mutual assessment based solely on the content encoded in a signal for two reasons. First, experimental manipulations are necessary. In one of the few captive experiments that sought to examine mutual assessment (e.g., manipulated scrotal color and staged contests in male vervet monkeys, Chlorocebus pygerythrus ), the results were inconclusive possibly because it is near impossible to recreate realistic scenarios that males typically encounter in the wild 22 . Second, primates tend to rely predominantly on social information that derives from individual recognition (not signals) to guide their interactions 23 . The use of this social information can be quite sophisticated (e.g., eavesdropping) 24 or quite simple (e.g., requiring nothing more than recognizing an individual) 23 . For example, in primate systems with linear dominance hierarchies, the dominance rank of an individual determines the outcome of most social interactions 25 , 26 , 27 . Typically, these interactions are not thought to be based on the relative Resource Holding Potential (RHP) of contestants at the time of the interaction, but rather based on social knowledge of contestants derived from a recent history of interactions with them 28 . However, results from previous research on mutual assessment have not been able to entirely distinguish whether primate subjects use individual identity or quality signals as the underlying basis for assessment, even in cases where playback experiments were able to manipulate the individual and/or the signal. For example, playback experiments in chacma baboons ( Papio ursinus ) have shown that subjects respond to loud call displays based on the relative RHP of contestants 19 , 25 , 26 . Specifically, based on simulated loud call contests, chacma males were more likely to enter vocal contests if the opponent’s rank was similar to their own 26 , they were able to discriminate between the relative ranks of “third party” contestants 25 , and they were more interested in calls that were manipulated to suggest a higher quality male 19 . Combined, these three studies suggest that mutual assessment may be occurring in chacma baboons. However, the first two studies 25 , 26 and half of the third study 19 were conducted entirely with familiar rivals. In a parallel study on chacma male grunts (soft vocalizations that are unlikely to carry information about the sender’s competitive ability), researchers also observed that the strongest responses occurred between similarly-ranked males 29 . Therefore, because known males were used, it is impossible to distinguish whether chacma male assessment hinged on individual recognition (knowing who is calling) or signal strength (knowing the quality of the caller). In the only experiment to use unknown callers, the subjects did not alter their responses according to their own rank – they all attended more strongly to the higher quality loud call 19 . Therefore, although these results are certainly suggestive, they fall just short of providing clear evidence of mutual assessment in the context of animal signals. Here we examined whether a close relative to chacma baboons, the gelada ( Theropithecus gelada ), uses mutual assessment when hearing loud calls from other males. Geladas present an unusually tractable system for experimentally studying assessment in primates. Geladas have a vocal signal that is used in male-male competition, allowing us to use playback experiments to disentangle various assessment strategies. Geladas’ reliance on vocal signals likely relates to their large, fluid social systems where, in contrast to closely-related species (including chacma baboons), they frequently interact with unfamiliar individuals 30 . Geladas are large-bodied, terrestrial primates that live in the high-montane grasslands of Ethiopia 31 . They congregate in a large, fluid, multi-level society composed primarily of harems (“reproductive units” – hereafter “unit”) comprising one harem-holding male (“leader male”), 1–12 related adult females and their offspring, and occasionally one or more subordinate males (“follower males”). Leader males (often joined by follower males) fiercely guard their harems from “bachelor males” that reside in all-male groups at the periphery of the larger aggregations of units 32 . Importantly, bachelor males gain reproductive access to females primarily by challenging and defeating a leader male 33 . By contrast, leader males pose no threat to each other 33 and frequently gather into large foraging aggregations 34 for a putative “dilution effect” against predators 35 and/or bachelors 36 . Leader males deter bachelors from challenging them by engaging in ritualized vocal displays that culminate in a series of loud calls 32 . These displays begin when a leader male approaches, threatens, and solicits a chase from a group of bachelor males 32 . The display itself does not immediately result in aggression between the leader and the bachelor males, but is thought instead to transmit information on the strength and/or condition of the initiating leader male to the recipient bachelor males 33 , 37 . The end of each display is punctuated with one or more bouts of loud calls by the leader male. While only one leader male is chased at a time, these displays elicit the attention from other males, and the loud calls themselves appear to be “contagious”; that is, after each display, between 2–13 leader males (and occasionally follower males as well) produce additional loud calls of their own 33 . Additionally, each display is often followed by subsequent displays from other leader males, with each male taking a turn (i.e., soliciting a chase and ending with a bout of loud calls), venturing away from his harem to engage with the bachelors and produce loud calls before returning to his females 33 . Bachelors do not produce loud calls during these displays 33 . Previous research in geladas reported that leader males that display more frequently were less likely to be targeted by bachelors, suggesting that the quantity of these displays serves to deter rivals 33 . But, in addition to the quantity of loud calls produced, recent evidence also suggests that the quality of these loud calls is important for rival assessment 37 . Specifically, the males with the highest RHP in gelada society (e.g., prime-aged, high-status males) utter the most calls per bout, produce calls that are the lowest in overall frequency measures, and exhibit the greatest vocal range 37 . Thus, the loud calls themselves appear to be honest signals of male RHP, and bachelor males could use RHP information encoded in these calls for identifying relatively low-quality males (e.g., old males, low-status males, exhausted males) 37 . By contrast, leader males do not assess bachelors. Leader males are always on defense, never offence, from bachelor males. However, leader males do have the potential to assess other leader males via these calls 37 . If indeed leader males rely on a putative dilution effect to avoid being challenged by bachelor males, then each leader’s position is secure only if they, themselves, have a higher RHP than the other leader males around them. Thus, leader males can use RHP information encoded in these calls for identifying situations when they are surrounded by relatively strong males (and are, thus, weaker by comparison). We used a playback experiment as well as observations of natural behavior in wild geladas to investigate the rival assessment strategy used by males. We examined male responses to both experimental and natural loud calls of varying quality. If gelada males rely only on self-assessment in male contests, we predicted that neither leaders nor bachelors would respond differently to low- and high-quality calls (Fig. 1a ). If gelada males rely on opponent-only assessment , we predicted that all subjects would attend more strongly to high-quality calls than low-quality calls regardless of their own status (Fig. 1b ; note that the direction of this response could also be reversed). However, if gelada males rely on mutual assessment in male contests (Fig. 1c ), we predicted that male subjects would respond to loud calls based on the combined information about themselves (i.e., their own status and/or RHP) and the quality of their rival (i.e., call quality). Specifically, we expected: (1) bachelor males to attend more to low-quality calls (a weak rival) because this represents a prime opportunity for a takeover; (2) leader males to attend more to high quality calls (a situation that weakens their relative security in the group, and one that indicates they might soon be challenged) – (3) particularly if they themselves have high RHP; (4) high RHP leader males to not just attend to high quality calls, but to advertise their own quality by subsequently participating in the display; and (5) females to not discriminate between call quality (because loud calls are used in male-male competition rather than female choice). Figure 1 Predictions for bachelor male ( a ) and leader male ( b ) responses to low- and high-quality simulated loud calls for three assessment strategies: self-assessment, opponent-only assessment, and mutual assessment. Full size image Results Do males attend differently to high- and low-quality calls based on their own status? We conducted a playback experiment on 60 adult geladas (20 females, 20 leader males, and 20 bachelor males) using previously recorded loud calls obtained during naturally-occurring displays between adult males (7 high-quality bouts and 7 low-quality bouts were used to construct 10 playback sets each containing a unique combination of one high- and one low-quality loud call bout from different males). Each subject heard both a high-quality loud call (one caller) and a low-quality loud call (a different caller). We visually recorded each subject’s response to each call type (randomized for order of presentation) and examined six response variables ( look duration, approach duration, latency to look, latency to approach, approach distance , and time to resume activity ), which were reduced using factor analysis. The factor analysis resulted in two latent factors, (1) an “approach” response, and (2) a “look” response, with Eigenvalues >1, together explaining 90.68% of the total variance. Factor 1 (“approach response”) accounted for 64.18% of the variance and loaded heavily on approach duration , approach distance , and latency to approach . Factor 2 (“look response”) accounted for 26.50% of the variance and loaded heavily on look duration , latency to look , and time to resume activity (Table 1 ). Table 1 Loadings from Factor Analysis. Full size table To examine whether social status and/or call quality (high or low) determined a subject’s response, we constructed two Linear Mixed Models (LMMs) with each factor score as the dependent variable. In each model, we included social status (leader, bachelor, or female), call quality (high or low), and an interaction between them as predictors and controlled for call order (fixed) and subject (random). To further assess if bachelors and leaders differentiated between call-quality, we conducted additional pairwise contrasts and adjusted the p-value accordingly. For Factor 1 (“approach response”), we found a significant interaction between bachelors and call quality (β = 0.25, s.e. = 0.11, t = 2.202, p = 0.032; Table 2 ). Bachelors approached low-quality calls significantly more than females and leaders (Fig. 2a ). Bachelors were also more likely to approach low-quality calls than high-quality calls (β = −0.27, s.e. = 0.08, t = −3.353, p = 0.014; Table 2 ), but neither leaders (β = 0.02, s.e. = 0.08, t = 0.199, p = 0.843) nor females (β = 0.02, s.e. = 0.08, t = 0.234, p = 0.816) differed in whether they approached either call type (Table 2 ). In fact, females rarely approached the speaker (Table S1 ). Table 2 Results from LMMs. Full size table Figure 2 Subject responses (Mean of factor scores + SEM) to simulated high and low quality loud calls from bachelors, leaders, and females. ( a ) Factor 1 is a composite score where larger values indicate a stronger “approach” response. ( b ) Factor 2 is a composite score where larger values indicate a stronger “look” response. See text for details. Full size image By contrast, leaders were more likely to look (Factor 2) towards high-quality calls than low-quality calls (β = 0.52, s.e. = 0.18, t = 2.856, p = 0.006; Fig. 2b ). Yet, neither bachelors (β = 0.18, s.e. = 0.18, t = 1.000, p = 0.321) nor females (β = −0.19, s.e. = 0.18, t = −1.042, p = 0.302) distinguished between call type in terms of looking time (Table 2 ). In general, both bachelors (β = 0.99, s.e. = 0.29, t = 3.389, p = 0.001) and leader males (β = 0.61, s.e. = 0.29, t = 2.110, p = 0.038) spent more time looking towards the speaker than females did. We found no effect of call order in either model (Factor 1; β = 0.05, s.e. = 0.05, t = 1.145, p = 0.257: Factor 2; β = −0.06, s.e. = 0.11, t = −0.568, p = 0.572). In addition to the Factor Analysis, we further examined overall response time, a measure of how long each individual spent investigating the source of the call ( look duration + move duration ), to assess whether status or call type affected the overall strength of a male’s response. Supporting the previous results, we found a significant effect of call quality for bachelors and a significant interaction between social status and call quality. Bachelors spent more time oriented towards loud call bouts of low-quality (low quality; β = 10.66, s.e. = 3.82, t = 2.787, p = 0.008; Fig. 3a,b ) while leader males spent more time oriented towards loud call bouts of high-quality (leader x low-quality; β = −16.35, s.e. = 5.41, t = −3.024, p = 0.004; Fig. 3a,b ). In general, bachelors and leaders did not differ in their overall total orientation time to high-quality loud calls (β = 0.93, s.e. = 5.92, t = 0.157, p = 0.876; Fig. 3 ). Figure 3 Status difference for males in overall response time to high- and low-quality calls. Figure 3 represents both ( a ) within subject differences for 20 bachelors and 20 leaders, and ( b ) mean total response time (+SEM) to different call types. Full size image Do males attend differently to high- and low-quality calls based on their own quality? Males’ responses to differences in call quality were based on their own categorical differences in status as a leader or bachelor. We additionally wanted to examine whether males further differentiated playback stimuli based on their own “quality” (i.e., using the quality of their own loud calls as a proxy for overall “quality” 37 ). The sample for this analysis (N = 11) was only a subset of the leader males used for the first analysis (we did not have recordings from all subjects, and generally only leader males produce loud calls 33 ). Note that because low-quality males rarely produce loud calls, the leader males included in this analysis disproportionately comprise males whose loud calls are mid- to high-quality. We predicted that the previous result was due mainly to the high-quality leader males responding to the high-quality call type. We established a call quality score for each subject’s loud calls in the same way that we determined high- and low-quality calls for the playback experiment. For each call type (low, high), we compared each subject’s overall response time to his own call quality score. In response to the simulated low-quality calls, we found no relationship between the subject’s call quality score and his overall response time (r s = 0.489, p = 0.127). However, in response to the simulated high-quality calls, leader males with high call quality scores themselves, responded more strongly than those with low quality scores (r s = 0.752, p = 0.008; Fig. 4 ). Figure 4 Overall response time (s) to the high-quality playback call in relation to the subjects own call quality. Full size image Are males more likely to join a loud call display when they hear calls of similar quality to their own? The previous two results suggest that we can predict a male’s response to the quality of a loud call by using experimental stimuli – bachelor males responded strongly to simulated low-quality calls and leader males responded strongly to simulated high-quality calls. Within leader males, high-quality leaders responded strongest to high-quality calls suggesting that leaders attend to both the quality of the caller and their own quality. Next, we wanted to determine if these same results hold true in natural observations of male contests. Loud call displays often serve as a catalyst for other unit males to join in with loud calls of their own. We predicted that high-quality leader males will be more likely to enter a loud call display when the display includes other males of high quality. To test these predictions, we used behavioral observations and loud call recordings from 20 unit males (16 leader males, 4 follower males) across 291 loud call displays, recording 423 loud calls from all 20 males. We then examined whether male A (subject) was more likely to participate in a display given that male B also displayed (binomial distribution). We included relative call quality (the difference between the call quality scores of both males), caller “familiarity” (using social network analysis), and leader/follower status in the models as fixed effects; and we included the identification of both males as random effects. Males were more likely to display with males of similar call quality scores to their own (β = −0.62, s.e. = 0.31, z = −2.042, p = 0.041; Table 3 ), with caller familiarity having little effect (β = 0.25, s.e. = 0.14, z = 1.808, p = 0.071; Table 3 ). Table 3 Results from General Linear Mixed Model (GLMM). Full size table Discussion In simulated signal displays, gelada males, but not females, discriminated between loud calls based on the acoustic quality of the signal as well as their own status and quality. Specifically, bachelor males – males that must compete to gain reproductive access to females – exhibited a stronger response to low-quality loud calls, while leader males exhibited a stronger response to high-quality loud calls. Furthermore, within leader males from whom we had loud calls (a sample biased towards mid- to high-RHP males), we found that higher-RHP males themselves (based on their call quality) responded more strongly to the high-quality stimuli than did lower-quality males. Finally, in natural observations, leader males were more likely to join loud call displays when their own calls were of similar quality to the other males involved in the display. In all three cases, a male’s response to other males’ loud calls was based on both their own RHP and that of the caller (coded into the quality of the signal). Taken together, these findings support the hypothesis that gelada males use a mutual assessment strategy, rather than a self- or opponent-based one. These data provide some of the first evidence for a mutual assessment strategy using signals for a non-human primate. Although bachelor males attended to both high- and low-quality loud call bouts (Fig. 2b ), they only approached the hidden speaker (“escalated”) when they were played the low-quality call (Fig. 2a ). In playback studies, approach behaviors (e.g., approach distance, approach rate, latency to approach) represent more “intense” measures of interest in the signal than looking time alone 38 , 39 , 40 , 41 . This is especially true in the study of aggressive signals where approaching the source of the call is a relatively high-cost response, as it implies an interest in engaging the caller 42 , 43 . In support of this, males that approached the speaker reached (or passed) the source of the call (mean approach distance 32.7 m, mean speaker distance 28.9 m), and 73% of these approaches were accompanied with visual and vocal threats. When confronted with a potentially weak rival, bachelors may benefit from an escalated response (i.e., an approach) because successful challenges can result in reproductive access to females 33 , 44 , 45 . By contrast, when confronted with a potentially strong rival, bachelors may suffer severe (and possibly fatal) costs from an escalated response 32 . Our results suggest that bachelor males assess the quality of leader males by attending to these loud calls, and they use information gleaned from these calls to make decisions about which males to challenge and which to avoid. By contrast, leader males rarely approached the speaker regardless of call quality (Fig. 2a ). A strong approach is especially risky for a leader male as it requires him to leave his female unattended with bachelor males in close proximity. However, leader males did spend more time attending to the high-quality calls compared to the low-quality ones (Fig. 2b ), with the strongest responses deriving from the leader males exhibiting the highest-RHP (as measured by their own loud call quality, Fig. 4 ). The motivation for leader males to attend to (and, engage in) call displays presumably derives from the need to showcase their own quality in the midst of bachelors. The large aggregations of geladas (sometimes numbering over 1200 individuals) have been hypothesized to create a “dilution effect” against predators 35 , but also against bachelors 36 . Indeed, at least one feature of loud call displays (how often a leader male participated in displays 33 ) was negatively associated with his likelihood of takeover. Therefore, leader males should broadcast their loud calls when they “compare well” to displaying males around them. In support of this, leader males were more likely to participate in natural loud call displays when their call quality was similar to the males calling around them. Similar results have been reported for chacma baboons ( Papio ursinus ) where males were more likely to loud call with males of similar dominance rank 25 . However, the male baboons were likely using social knowledge (monitoring other males’ ranks and attending to acoustic cues of identity) to assess one another, not signals 19 , 25 . Indeed, mutual assessment using social knowledge appears routine among primates. In primate societies, interactions are structured by dyadic properties such as relative rank or kinship that require animals to account for the behavior of other individuals in relation to themselves . The novelty of our finding is in demonstrating that primates can use signals to perform mutual assessments while interacting with completely unfamiliar individuals. This use of signals to guide interactions is particularly useful in geladas where their large groups and fission-fusion social system 46 require males to consistently monitor the quality of unfamiliar males. Importantly, bachelors, by successfully defeating the resident harem-holding male, become leaders. As males transition from bachelors to leaders, the information an individual pays attention to is likely to change with this change in status. Other primate species have been shown to monitor changes in other individuals’ dominance ranks and social relationships over time 47 . In chacma baboons, for example, males track temporary changes in the status of other males’ consortship but, once again, the results of this study were likely based on identity information, not signals 48 . In the case of geladas, information acquired from quality signals may be the only way to successfully navigate such large social groups of unknown conspecifics. Unlike males, gelada females did not differentiate between high- and low- quality calls. Indeed, they rarely attended to either call (Fig. 2a,b ). There has been considerable debate as to whether loud calls in primates evolved to attract mates or to deter competitors 49 , 50 . One of the strengths of this study is that both females and males were tested within the same design. Our results indicate that gelada loud calls evolved as a signal for assessing rivals and not attracting mates. One promising avenue for future research will be to assess how group dynamics influence assessment strategies in social animals in a natural context 23 . For example, if the composition of social groups is dynamic, we might expect males to rely on information gleaned from signals rather than individual recognition and social knowledge when assessing rivals. More studies that combine experiments with natural observations of assessment behavior are necessary to understand the role of assessment strategies in social animals. Methods Study site and subjects Research was conducted on a population of wild geladas living in the Simien Mountains National Park, Ethiopia from Feb-Dec 2013. The University of Michigan Gelada Research Project has been collecting long-term behavioral and demographic data on this population since January 2006. All males were individually recognizable and habituated to observers on foot (approach distance <3 m). Methods include a combination of playback experiments and behavioral observations. We have adhered to the Guidelines for the Use of Animals in Research and the Institutional Animal Care and Use Committee guidelines at the University of Michigan and all field research was conducted with permissions from the appropriate offices in Ethiopia. Do males attend differently to high- and low-quality calls based on their own status? We conducted a playback experiment on 60 adult geladas (20 females, 20 leader males, and 20 bachelor males). To increase our sample size, we included both known and unknown individuals in this experiment. All unknown individuals were identified using morphological features to ensure they were not used in subsequent experiments. Playback stimuli Playback stimuli comprised previously-recorded loud calls obtained during naturally-occurring signaling contests between adult males. Loud calls were recorded using a Sennheiser ME-66 directional microphone and a Marantz PMD 660 digital recorder. Loud call bouts were only used as playback stimuli if they were complete (no calls were missed during the recording) and devoid of background noise and interruptions. We audibly and visually inspected calls using Avisoft SASLab Pro (Avisoft Bioacoustics, Berlin, Germany) acoustic software for acoustic disturbances (e.g., background noise). At the time of the experiment, we had 157 loud call bouts from 50 prime-age males that fit this criteria (e.g., free of background noise). For geladas, loud call bouts generally consist of a series of two-syllable “ee-yow” calls (2–9 calls per bout). Previously, we found support for the hypothesis that the entire bout (and not just the individual calls within the bout) functions as a quality signal 37 . We selected a total of 14 loud call bouts as playback stimuli: 7 high-quality bouts and 7 low-quality bouts, to construct 10 playback sets, each containing a unique combination of one high- and one low-quality loud call bout from two different males. We determined call quality by comparing calls along several parameters we had previously found to differ with age and status, fundamental frequency and number of calls per bout 37 . To assess these parameters, we conducted a spectrogram analysis in Avisoft with a fast Fourier transformation size of 1024 points (frequency range: 22 kHZ; frequency resolution: 43 Hz time resolution: 2.903 ms; 100% frame). For the 157 loud call bouts, we examined the distribution for both parameters and chose loud call bouts that were at the extremes of these distributions. Playback design We presented each subject with one of the playback sets comprising both a high- and a low-quality loud call bout. Because natural occurrences of loud calls in geladas generally occur when leader males encounter bachelor males, experiments were only conducted when both bachelors and leader males were present on a given day. To simulate a natural loud call contest, the calls were played from the direction of the bachelors (when the subject was a unit individual) or from the direction of the units (when the subject was a bachelor male). For each trial, we placed a Bose Roommate II portable speaker approximately 25–50 m (M = 28.96 m, SD = 7.95 m) from the subject. The speaker was hidden behind a physical barrier (i.e., tree, rock, or bush), and completely obscured from the subject’s view. All subjects were observed for 15 minutes prior to the start of the playback experiment; and experiments were only conducted if (1) the subject was sitting (e.g., feeding or resting) for at least 2 minutes prior to the start of each call, and (2) the subject was oriented away from the speaker. The experiment used a within-subjects design, in which subjects heard both a high-quality loud call bout and a low-quality loud call bout in each trial (to simulate a loud call contest between many males). The second call was played 5 minutes after the first call to allow subjects to return to an initial resting state. Subjects generally returned to an initial resting state within 1 minute after hearing the first call. We played each set of calls (n = 10 unique sets) to 6 subjects each: two bachelors, two leaders, and two females. To combat any order effect, we counterbalanced the order in which the high- and low-quality bouts were played across leaders, bachelors, and females. No subject heard any of the calls in the set prior to his or her experimental trial, and each trial was separated by at least 10 days for individuals in the same band. Prior to all trials, we noted the identity of the subject, the location of the speaker relative to the subject, the subject’s initial state (feeding or resting), the experimental playback set used, and the order of calls heard. During each trial, one observer played the loud calls from a loudspeaker using an MP3 player (Apple ipod touch 3 rd generation). A second experimenter with a Kodax PlaySport (Z × 5) HD video camera, positioned herself 5–10 m in front of the subject, with the speaker hidden to the left or right of the subject. All subjects were video-recorded continuously from 15 seconds prior to the first call to 5 minutes after the second call. For each individual, we matched his or her state (feeding or resting) and distance to the speaker between the first and second call – in some cases, moving the speaker to a new hiding spot the appropriate distance away. For playback trials on unit individuals (leaders and females), we pre-designated two different subjects prior to the start of the trial: a unit male from one unit and a unit female from a different unit at least 40 m away. In such cases, we placed the speaker between the two subjects (from the direction of the bachelors) to ensure their visual trajectories towards the stimuli were not overlapping. Each subject was filmed and scored independently. Experimenters were in contact via two-way radios, and if any of the conditions were not met, we aborted the experiment immediately. We conducted a total of 60 successful playback trials and an additional 22 trials were aborted prior to completion. Playback responses All videos were scored on a computer with a frame-by-frame analysis using Adobe Premier (Adobe Systems, Inc.) by two independent observers. Prior to video analyses, playback videos were cut to contain only the response to one loud call bout within a set. All files were then renamed and randomized such that observers were blind to the identity of the subject (i.e., whether he was a unit male or a bachelor male – it was impossible to hide whether the subject was a female) and the condition (i.e., whether it was a high- or low-quality bout). Reliability for all measurements between the two observers was greater than 95% (M = 97%, SD = 1.2%). We measured 6 different response variables: (1) duration of time spent looking towards the speaker ( look duration ), (2) duration of time spent moving towards the speaker ( approach duration ), (3) latency of look response ( latency to look ), (4) latency of approach response ( latency to approach ), (5) total distance moved towards the speaker ( approach distance ), and (6) total time to return to initial resting state ( resume activity ). Look duration measured the time a subject spent oriented toward the speaker while stationary. When a subject oriented toward the speaker while moving towards it, we recorded the response as approach duration . For both duration responses, we measured the total duration of all responses until the subject returned to their initial state for up to 1 minute after playback onset. Time to return to initial state was assessed once an individual spent at least 15 seconds feeding or resting without orienting towards the speaker. We did not record responses after subjects returned to their initial state as such responses were overly influenced by other individuals within the group (unit or bachelor group). We subtracted any time spent looking or moving towards the direction of the speaker during the 15 seconds prior to the onset of the trial. Latency to look and/or latency to approach were measured as the time from the onset of the playback stimuli until the onset of the subject’s first look and/or movement towards the speaker. Due to the high mobility of the group during feeding, if a subject did not look or move within the first minute after the onset of the stimulus, we assigned the subject’s latency as 60 seconds. We also recorded approach distance for all movement toward the speaker in the first minute after the onset of each playback stimulus. Playback analyses To remove redundancy between response variables, we reduced response variables into latent factors using Factor Analyses (FA 51 ) with a varimax rotation with SPSS (v.22.0.0.0). We accepted all factors with eigenvalues greater than 1.0, which produced two factors associated with approach and looking behavior (Table 1 ). To assess differences due to call quality and status, we constructed two LMMs, one for each factor score from the FA as the dependent variable with status (leader, bachelor, or female), call quality (high or low), and an interaction between both as predictor variables. In each model, we controlled for call order (fixed effect) and subject (random intercept). To assess whether bachelor’s or leaders differed in their response to high and low-quality calls, we conducted two additional a priori contrast (Table 2 ). We corrected for multiple testing using a Bonferroni adjusted alpha level of 0.025 (0.05/2). In addition to the FA, we examined whether males differed in overall response towards high and low-quality calls. We calculated “ overall response time” for each male by summing the time he spent looking and approaching the speaker ( look duration + approach duration ). We constructed a third LMM with overall response time as the dependent variable, subject as a random factor, status and call quality as fixed effects and an interaction between status and call quality. Since call order was not a significant factor in any of the previous LMMs, we excluded it from this model. For these models (and all subsequent models), we determined the statistical significance of the full model by comparing its fit using likelihood tests with that of a null model including only the intercept and the random effect (Table 2 ). We conducted all model analyses in R v.3.2.0 using the “lmer” function in the lme4 packages v.1.1–11 52 and contrasts using the lsmeans package v. 2.5–5 53 . We visually inspected each model using a Q-Q plot, histogram of residuals, and scatter plot of fitted versus residual values. Residual values for all models were normally distributed. Do males attend differently to high- and low-quality calls based on their own quality? We assigned each male a call quality score by examining 12 acoustic parameters related to frequency (e.g., fundamental frequency) and temporal measures (e.g., call duration). Given our previous results that lower frequency calls are energetically-costly to produce 37 , we established a call quality score based on the factor analysis (i.e., Factor 1, spectral measures ). We focused our analysis on the first calls given within a bout (n = 122) as these calls are the lowest in frequency measures (and presumably the highest quality). We calculated a mean for the spectral measure scores to establish a call quality score for each male. Because calls that are lower in spectral measures were higher in quality, we multiplied the call quality score by −1, so that a high call quality score represents a high-quality call. We then ran two Spearman’s rank-order correlations, comparing each male’s response time to the high- and low-quality playback stimuli to his own call quality score. Are males more likely to join a loud call display when they hear calls of similar quality to their own? We collected all-occurrence behavioral sampling and recorded 423 loud calls across 291 different loud call displays from 20 males across the study period. For all displays, we recorded the identity of all known males that participated in the display as well as the males present in the group that did not participate. We conducted acoustic analyses on all calls in the same way as described above. Again, we focused our analysis on the first calls given within a bout (n = 122) and established a call quality score for each male. To control for a subject’s “familiarity” with the caller, we used all proximity data between the subject and the caller in a social network analyses 34 . We constructed an undirected, weighted network based on male-male association. In this network, males were represented by nodes and the edge weight was given by an association index. This index was calculated as: $$\begin{array}{c}Association\,index\,males\,A,B=\underline{\#\,of\,times\,male\,A\,seen\,with\,B}\\ \quad \quad \phantom{\rule{5em}{0ex}}\phantom{\rule{4em}{0ex}}\quad \quad minimum\,\#\,of\,times\,male\,A\,or\,male\,B\,seen\end{array}$$ where the numerator is the total number of times males A and B were seen together in the same group, divided by the minimum number of times we observed either A or B in the same group 34 . The association index ranges from 0 (if two individuals were never seen together) to 1 (if they were always seen together). From this network, we used the Louvain community identification algorithm to assign males to “cliques” within their social network. Males associated into two distinct cliques (N = 15 and N = 21 males respectively) with a modularity coefficient of 0.011. Males were considered to be “familiar” with each other if they were assigned to the same clique, and “not familiar” if assigned to different cliques. To assess if relative call quality or caller familiarity influenced the likelihood that a male would participate in these vocal displays, we conducted a GLMM with a binomial distribution. For each subject, we examined the dyadic calling relationship with other males. The outcome variable in our model was the likelihood that male A participated in a display, given that male B also displayed. This was modeled as the count of successes , the number of times male A and male B displayed together, offset by the count of failures , the total number of times male A or B displayed (but not both), given that both males could have displayed (e.g., were both present in the group on that day). We included relative call quality and caller familiarity in the models as fixed effects. Relative call quality was calculated for each dyad by taking the absolute value of the difference between the call quality scores of both males. The smaller the difference between the two call quality scores, the closer the males were in relative call quality . Caller familiarity was established for each dyad from the social network analysis. Males were considered to be “familiar” with each other if they were assigned to the same clique, and “not familiar” if assigned to different cliques. We controlled for the identity of both males by including their identification as random variables in the model. Although the majority of calls were given by leader males, occasionally subordinate follower males engaged in these displays. Because leaders are more likely to display than followers, we controlled for status of both males in the model. We compared the full model to a null model, which included only the intercept and random effects. The social network analysis and GLMM were conducted in R 3.2.4 using igraph 54 and lme4 52 packages respectively. Data availability Datasets and R scripts generated and analyzed during the current study are available from the corresponding author on reasonable requests. Playback videos are also available upon request.

Gelada males—a close relative to baboons—pay attention to the loud calls of a rival to gain information about his relative fighting ability compared to themselves, a new study indicated. Researchers at the University of Michigan, Georgia State University and Princeton University found evidence that gelada males decide to escalate contests with their opponents based on their own condition relative to the condition of their opponent. They appear to do this by using the acoustic quality of the loud calls of their rivals—long-distance vocalizations that carry honest information about the fighting ability of the caller. There has been much debate on specifically how animals make competitive decisions during contests. Game theoretical models predict that animals should assess an opponent's condition relative to their own condition prior to engaging in combat to avoid costly fights they are unlikely to win, a strategy known as mutual assessment. Despite the benefits of such mutual comparisons ("I am stronger than him"), remarkably few studies have been able to reject much simpler assessment strategies such as self-assessment ("I am strong and should fight") and opponent-only assessment ("he is strong and so I should not fight"). Researchers say one approach for distinguishing these strategies is to use animal displays (rather than aggressive contests) to examine how animals make informed decisions about rivals. "Particularly for quality signals that contain honest information on the condition of its bearer, signals used in animal displays offer an ideal situation for examining mutual assessment because they are low cost and allow for experimental manipulation," said Marcela Benítez, a postdoctoral research associate at Georgia State University and the study's lead author. In geladas, harem-holding "leader" males engage in loud call displays to deter challenges from "bachelor" males, who must compete with leaders to gain reproductive access to females. Supporting a mutual assessment strategy, gelada males responded to loud calls of different quality (in both playback experiments as well as in natural observations) according to attributes of themselves and their opponent. "Previous studies in wild primates have shown that they use mutual assessment, but this was between animals that knew one other," said Jacinta Beehner, U-M associate professor of psychology and anthropology. "They see Kevin and they remember that they beat him in a previous fight. The novelty of our finding is that we have shown that primates can do this even for completely unfamiliar individuals—using signals." Although primates routinely classify others relative to themselves using individual attributes, this represents some of the first direct evidence for mutual assessment in primate signaling contests, Benítez said. The findings appear in Scientific Reports.

10.1038/s41598-017-02903-w

Physics

Wriggling microtubules help understand coupling of 'active' defects and curvature

Perry W. Ellis et al, Curvature-induced defect unbinding and dynamics in active nematic toroids, Nature Physics (2017). DOI: 10.1038/nphys4276 Journal information: Nature Physics

http://dx.doi.org/10.1038/nphys4276

https://phys.org/news/2017-10-microtubules-coupling-defects-curvature.html

Abstract Nematic order on curved surfaces is often disrupted by the presence of topological defects, which are singular regions in which the orientational order is undefined. In the presence of force-generating active materials, these defects are able to migrate through space like swimming microorganisms. We use toroidal surfaces to show that despite their highly chaotic and non-equilibrium dynamics, pairs of defects unbind and segregate in regions of opposite Gaussian curvature. Using numerical simulations, we find that the degree of defect unbinding can be controlled by tuning the system activity, and even suppressed in strongly active systems. Furthermore, by using the defects as active microrheological tracers and quantitatively comparing our experimental and theoretical results, we are able to determine material properties of the active nematic. Our results illustrate how topology and geometry can be used to control the behaviour of active materials, and introduce a new avenue for the quantitative mechanical characterization of active fluids. Main Defects are singular regions where the characteristic order in a material is undefined. The human hand, for example, is lined by parallel ridges and associated defects, which play a pivotal role in fingerprint identification. More broadly, defects play a significant role in physics and materials science; for example, defect proliferation is the mechanism behind the celebrated Kosterlitz–Thouless–Halperin–Nelson–Young (KTHNY) phase transition 1 , 2 , 3 , and defect entanglement can add orders of magnitude to the stiffness of normally soft liquid crystals 4 , 5 . Furthermore, since defects can serve as preferential sites for chemical linking, they can be used to help assemble hierarchical structures that hold promise for both photonic and biosensing applications 6 , 7 . Remarkably, when defects are combined with force-generating active materials, the structural complexity comes to life, giving rise to a spectacular variety of self-regulated behaviours, such as self-sustained oscillations and spontaneous formation of morphological features, including kinks, protrusions 8 and undirected motility 8 , 9 . These remarkable observations have acted as a stimulus for investigating how biological functionality emerges from the interplay between activity, the geometry of the system and the structure of the internal phase 10 , 11 , 12 . In this context, defects that could be present in active materials could be harnessed to achieve life-like functionality, since defects are extremely sensitive to the intrinsic geometry of the space they inhabit. This is clearly exemplified in a nematic liquid crystal composed of anisotropic mesogens aligned along a common direction, referred to as the nematic director n , on a sphere. As you ‘cannot comb a hairy ball’, there will necessarily be irreducible singularities on the surface; however, the type and number of singularities depends on the energetics of the system 13 . As it was theoretically predicted 6 , 14 and then experimentally observed 15 , such a system attains its lowest energy configuration when nematic order is disrupted at the location of four point defects arranged on the vertices of a tetrahedron. Around each defect the nematic director rotates by π. Each defect is then said to carry an s = +1/2 topological charge, which is defined as the winding number of n along a path encircling the defect. From this definition, it is evident that s is quantized and can take on only a discrete set of values; for a nematic, s must be an integer multiple of 1/2. A formal statement of the constraint on the total topological charge on a closed (compact and without boundary) surface is given by the Poincaré–Hopf theorem 16 , which forces the total topological charge to be equal to the surface Euler characteristic χ = 1/(2π)∫ K d A (ref. 16 ), where K = ( R 1 R 2 ) −1 is the Gaussian curvature, with R 1 and R 2 the principal radii of curvature at a point on the surface 16 . For a sphere of radius R , there are four +1/2 defects, consistent with having K = R −2 everywhere and χ = 2. The connection between defects and the space they inhabit is predicted to go beyond topology. The theory of defects on curved surfaces connects the topology of the confining surface to the free energy associated to a given defect configuration via the Gaussian curvature of the surface 13 . Thus, even though there have been new discoveries examining defective structures on spherical surfaces 15 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , the constant Gaussian curvature implies that only the size of the sphere will have any impact on the defect structure; this is confirmed by the size-dependent onset of grain-boundary scars in colloidal crystals on the surface of emulsion droplets 18 , 19 . Even when the nematic mesogens are active and the four +1/2 defects on the sphere become mobile, the defect dynamics can be explained without any mention of Gaussian curvature 8 . Despite theoretical interest in the interplay between varying Gaussian curvature and topological defects, with few exceptions 24 , 25 , 26 , there has been little experimental work with defects on surfaces where the Gaussian curvature is non-constant. More notably, there has been no experimental work where the defects have the option to explore regions with both positive (sphere-like) and negative (saddle-like) Gaussian curvature. In this case and for nematic order, the topological charge is expected to unbind into +1/2 and −1/2 defects (see Fig. 1a ), which in turn are attracted by regions of like-sign Gaussian curvature 27 , 28 , 29 . In this Article, we explore this experimentally for the first time using an active nematic liquid crystal comprised of microtubules and kinesin, constrained to lie on a toroidal surface (see Fig. 1b ), which has positive Gaussian curvature on the exterior, negative Gaussian curvature on the interior, and is such that χ = 0. This implies that a nematic liquid crystal on a torus can be either defect-free or populated by the same amount of positive and negative defects. In contrast to conventional nematics, active nematics are driven out of equilibrium as the active stress makes the uniformly aligned state unstable to distortions 30 , 31 . Whereas strong spatial confinement can suppress formation of excess defects 8 , we perform our experiments in the more commonly found turbulent regime. In this regime, the active nematic is characterized by a large number of ±1/2 defects, which are dynamically created and annihilated over time 9 , 32 , 33 , 34 , 35 . Despite such chaotic and highly non-equilibrium dynamics, we demonstrate that on average, topological defects unbind and segregate in regions of opposite Gaussian curvature. Notably, the chaotic dynamics also transforms the topological charge from a discrete to a continuous variable. A numerical integration of the equation of motion of active nematic defects 8 , 36 , 37 confirms our experimental results, further illustrating that the degree of defect unbinding can be selectively controlled and even suppressed by tuning the system activity. In addition, contrary to equilibrium predictions 27 , 28 , 29 , 38 , 39 , we find that this active unbinding depends only on the local geometry and is independent of system size and aspect ratio. Furthermore, by using topological defects as microrheological tracers and quantitatively comparing our experimental and theoretical results, we provide the first ever reported estimate of the Frank elastic constant, the active stress, and the defect mobility of a microtubule–kinesin active nematic liquid crystal. Overall, our results not only confirm the theory of topological defects on curved surfaces, but also demonstrate the surprising phenomenology that arises from adding activity to the interplay between geometry, topology, and order. Our work thus provides insights into the physics of partially ordered active matter and introduces a new avenue for the quantitative mechanical characterization of active fluids. Figure 1: Curvature-induced defect unbinding for a nematic on the surface of a torus. a , Plot of topological charge versus integrated Gaussian curvature as φ increases for the torus with four defect pairs shown in the upper-left quadrant of the plot. On the upper left schematic, the K > 0 region is red and the K < 0 region is blue. The visible +1/2 and −1/2 defects are marked by triangles and circles, respectively. The red and blue curves on the plot correspond to integrating over the K > 0 and K < 0 regions of the torus, respectively. Specifically, for the K > 0 region we integrate over the half-circle from θ 1 = −π/2 to θ 2 = π/2, while for the K < 0 region we integrate over the half-circle from θ 1 = π/2 to θ 2 = 3π/2. For both regions we also integrate from φ = 0 to φ = φ 0 with φ 0 ranging from 0 to 2π. The resultant curves are stepped, reflecting the discrete nature of the topological charge. Note that the total integrated Gaussian curvature in the K > 0 region is 4π; this is the exact amount of n -rotation around the four +1/2 defects, as n rotates by π along a path encircling a +1/2 defect. Similarly, the total n -rotation of the four −1/2 defects is equal to the total integrated Gaussian curvature in the K < 0 region of the torus. Lower-right schematic: cross-section of a torus obtained by rotating along . The aspect ratio ξ = R 0 / a , with R 0 the ring radius and a the tube radius, sets the slenderness of the torus: ( r , θ , φ ) are a set of toroidal coordinates: ( r , θ ) are the polar coordinates in the plane, and φ is the azimuthal angle. b , Raw confocal image of an active nematic toroid imaged at a single instant in time, viewed at an angle in order to highlight the geometry. The data has been false-coloured by height for additional clarity. The scale bar is 250 μm. Full size image Our system consists of water-based toroidal droplets containing the active nematic and stabilized with an oil-based yield-stress fluid as the continuous phase 40 . We characterize the toroidal droplets by their size and by their slenderness or aspect ratio, ξ = R 0 / a , with R 0 the radius of the central circle and a the tube radius, as schematically shown in Fig. 1a . The active nematic is comprised of microtubules that are bundled together and driven out of equilibrium by clusters of kinesin molecular motors powered by adenosine triphosphate (ATP) 8 , 9 , 35 . The system is further endowed with an ATP regeneration system, phosphoenol pyruvate and pyruvate kinase/lactic dehydrogenase, and a depletant, poly-ethylene glycol (PEG), which causes the microtubules to assemble on the toroidal surface, where they form a nematic liquid crystal. We then image the lower half of the toroidal droplet over time using confocal microscopy until the activity ceases, which typically happens between 6 and 12 h after making the toroidal droplet. An example of a confocal stack over an azimuthal section of the droplet is shown in Fig. 2a . To determine the director and locate the defects, we first project the stack onto the xy -plane, as shown in Fig. 2b , and then perform coherence-enhanced diffusion filtering (CEDF) to find the coherence direction at each pixel, defined as the direction where spatial intensity fluctuations are weakest 41 . The coherence direction in our case directly corresponds to the ‘molecular director’, u , at each pixel. From the image shown in Fig. 2b , we then obtain the u -map shown in Fig. 2c , where black represents a molecular director oriented along and white a molecular director oriented along . From the u -map, we calculate the tensor order parameter Q = (1/ N )∑ i =1 N ( u i u i T − I /2), with I the identity matrix, at each pixel 42 . We do this by averaging over a region within a five-pixel radius. Diagonalizing this tensor results in Q = S ( nn T − I /2), which naturally provides the director and the scalar order parameter, S , at every pixel. To find defects, we search for pixels of low S and quantify the director rotation in a path encircling the pixel. Our procedures yield the director map and the position and associated charge of the topological defects for each confocal snapshot ( Fig. 2d ; see Methods ). Figure 2: Quantifying curvature-induced defect unbinding for an active nematic confined to the surface of a toroidal droplet. a , 3D confocal stack of a portion of a nematic toroid at a given time. b , Maximum intensity projection along of the data in a . Scale bar is 200 μm. c , Coherence direction for the data in b . The orientation is periodic in π and measured clockwise from the horizontal, with black representing 0 and white representing π. d , The director n and associated +1/2 (filled triangle) and −1/2 (filled triangle down) defects obtained from c . e , Gaussian curvature of the surface seen in a . f , Number of +1/2 (filled triangle) and −1/2 (filled triangle down) defects over time in a region of the torus. The error bar for each point is 5%. We calculate this error by comparing the results from our defect-finding routine to results from counting the defects in the director field manually in a set of regions randomly sampled over all time frames in all of our experiments. g , Time-averaged topological charge in a region versus the integrated Gaussian curvature of that region for five experiments with (filled circle) ξ = 1.6, a = 275 μm, (filled triangle) ξ = 2.0, a = 372 μm, (filled diamond) ξ = 2.4, a = 268 μm, (filled square) ξ = 5.9, a = 200 μm, and (filled triangle down) ξ = 6.6, a = 167 μm. The error bars on the data points are the standard error of the mean. The line is the weighted average of a linear fit to the data for each individual torus. It has a slope of 4.3 ± 0.7 and an intercept of 0.01 ± 0.02; the error in both the slope and the intercept is the standard error of the mean. Full size image We next use the time-averaged three-dimensional (3D) confocal data to obtain the height of our toroidal surface for each ( x , y ) position in the confocal stack. To calculate the Gaussian curvature we reconstruct the Weingarten matrix L ij = ∇ i N ⋅ e j , with N the surface normal and e j a basis vector on the tangent plane, mapping a small displacement on the tangent plane to the corresponding variation of the normal vector: Δ N i = L ij Δ r j , with Δ r = Δ r j e j an arbitrary tangent displacement. We then compute the Delaunay triangulation 43 of our surface to obtain both the displacement vector associated to pairs of points in our surface around a given point, and the corresponding surface normals ( Supplementary Fig. 2 ), and robustly fit the components of the Weingarten matrix using an iteratively reweighted least-squares routine (IRLS) 44 , 45 , 46 . The local Gaussian curvature is obtained as the determinant of the Weingarten matrix 47 . An example of this procedure for the azimuthal section of the toroidal droplet in Fig. 2a , is shown in Fig. 2e . From the fitting weights of the final iteration of the IRLS routine, we recalculate N at each point and use this final value to obtain the local area element: , with g the determinant of the metric 16 (see Methods ). From these measurements, we determine the number of +1/2 and −1/2 defects in a region of area A within the toroid, N A + and N A − , as a function of time, as shown for a part of a typical data set in Fig. 2f (see also Supplementary Video 1 ), obtain the time-averaged topological charge in that region, , and correlate it with the integrated Gaussian curvature of that region. We find that is linear with 1/(2π)∫ K d A , as shown in Fig. 2g , where we plot data from toroids with a range of ξ and a . The slope of the curve is positive, consistent with the curvature-induced defect unbinding predicted theoretically. However, due to the large defect number and their motion, the topological charge approaches a continuous distribution and does not exhibit the jumps that characterize the ground state depicted in Fig. 1a . Furthermore, depends only on the integrated Gaussian curvature and is independent of ξ and a . This implies that the unbinding depends only on the local geometry and is insensitive to the global size and shape of the system. Note that this is in direct contrast to equilibrium simulations and theory, which all predict dependence on both ξ and a (refs 27 , 28 , 29 , 38 , 39 ). We also note that the line in Fig. 2g goes through zero, indicating that regions with ∫ K d A = 0 have , which is a topological requirement for the entire toroid. In our case, however, we see that this is true irrespective of the region we consider, provided that the region has vanishing integrated Gaussian curvature. This implies that a region with ∫ K d A = 0 is representative of the entire toroid, in contrast to equilibrium nematics, where, because of the modest number of defects and lack of significant motion, different regions with the same net integrated Gaussian curvature will generally enclose a different topological charge. Thus, our results suggest that the presence of activity in our system is playing a role akin to temperature in equilibrium systems. In this case, when sufficiently close to the nematic–isotropic phase transition, which is predicted to be continuous in two-dimensional systems 48 , the nematic elasticity will vanish smoothly, eventually allowing thermal fluctuations to mobilize the defects and produce results qualitatively similar to adding activity to the nematic. To gain further insight about our experimental findings, we perform numerical simulations ( Fig. 3a ). Defects are modelled as massless particles on the torus, whose position r i and orientation p i are governed by the following equations of motion: where v 0 is the speed associated to the self-propulsion of the +1/2 defects, μ is a mobility coefficient, F ij and V are, respectively, the force resulting from inter-defect interactions and the potential energy associated to the interaction between the defect and the local Gaussian curvature, and ζ i t and η i r are uncorrelated translational and rotational noises, and p i ⋅ p i ⊥ = 0. Note that, in contrast to the +1/2 defects, −1/2 defects do not self-propel; this difference results from the polar structure of the nematic director around the core of +1/2 disclinations compared with the three-fold symmetry of −1/2 disclinations 36 , 37 ( Supplementary Fig. 3 ). Both F ij and V can be calculated from the Frank free energy, F = k F /2∫ d A | ∇ n | 2 , with k F a two-dimensional effective elastic constant: F i j = 4 π 2 k F s i s j ∇ r i G ( r i , r j ) ) and V = k F ∫ d AG ( r , r i ) K ( r ), where G ( r i , r j ) is the Laplacian Green function on the torus (see Methods ). In the turbulent regime discussed here, the speed v 0 of the +1/2 defects, is given by v 0 ≍ αl a / η , where α is the active stress experienced by the microtubules as a consequence of the forces exerted by the motors, η is the viscosity, and is the active length scale resulting from the competition between active and elastic stresses 32 , 34 . For large l a , the nematic elasticity dominates and activity slightly affects the equilibrium configuration of the nematic director. In contrast, for small l a , the opposite is true and the active stress is able to induce appreciable distortions in the director field; in this regime, the active flow becomes turbulent, and creation and annihilation of defects is expected 32 . As a result, the active length scale is proportional to and of the same order of magnitude as the average distance between defects, , with N the total number of defects and A torus = 4π 2 aR 0 the total area of the torus 32 . Thus, the dimensionless distance between defects λ ′ = λ / a and the dimensionless speed v 0 ′ = v 0 a /( μk F ) specify all the relevant material parameters of the active nematic on a given torus. Figure 3: Numerical simulations show that curvature-induced defect unbinding for active nematics is sensitive to defect velocity and defect density. a , Schematic of a torus with ξ = 3 and 250 defect pairs on the surface ( λ ′ = 0.49) at a single moment in time. b , Plot of the time-averaged topological charge in a region versus the integrated Gaussian curvature of that region for λ ′ = 0.49 and (filled square) v 0 ′ = 10, (filled circle) v 0 ′ = 20, (filled triangle) v 0 ′ = 30, (filled triangle down) v 0 ′ = 40, (filled diamond) v 0 ′ = 60. c , C ′ versus defect velocity for (filled square) λ ′ = 0.69, (filled circle) λ ′ = 0.49, (filled triangle) λ ′ = 0.40, (filled triangle down) λ ′ = 0.34, (filled diamond) λ ′ = 0.31, (filled star) λ ′ = 0.28. The horizontal black line corresponds to the experimental C ′ = 4.3. Full size image For fixed λ ′, we find that is a linear function of 1/(2π)∫ d AK , consistent with the experimental results. We also find that the slope of the line, C ′, decreases as v 0 ′ increases, as shown in Fig. 3b , eventually approaching zero for sufficiently high v 0 ′. At this point, activity dominates over the elastic forces associated with inter-defect and defect-curvature interactions, confirming that, from this perspective, the role of activity is reminiscent of the expected role of thermal fluctuations close to the isotropic–nematic phase transition in passive nematics. However, we emphasize that, in equilibrium, both +1/2 and −1/2 defects would randomly explore the toroid as a result of thermal motion, whereas in our case, only the +1/2 defects exhibit a directed motion reminiscent of persistent random walkers 49 . The observed behaviour of C ′ with v 0 ′ is maintained irrespective of λ ′, as shown in Fig. 3c . Experimentally, we find C ′ = 4.3 ± 0.7 (see Fig. 2g ). We plot this value in Fig. 3c as a horizontal line. To match the experimental slope and the simulation results, we need λ ′. We determine this quantity from the time-averaged defect number density and the mean Gaussian curvature in that area, 〈 K 〉 A = 1/ A ∫ K d A ( Fig. 4a ). We find λ using the value of corresponding to 〈 K 〉 A = 0, since this value can be taken as representative of the whole torus. We further take the average of similar measurements for all experimental toroids and find λ ′ = 0.3 ± 0.1. We then return to Fig. 3c and find that the horizontal line corresponding to C ′ = 4.3 crosses the simulation curve for this value of λ ′, highlighted by the dashed vertical line, for a speed v 0 ′ = 81 ± 2. Figure 4: Defect number fluctuations depend on defect velocity. a , Time-averaged defect density in a region of area A versus the mean Gaussian curvature of that region for an experiment with ξ = 1.6, a = 275 μm. Error bars are the standard error of the mean. b , Probability of finding a certain number of defects in a given region for an experiment with ξ = 1.6, a = 275 μm. The blue curve is a fit of the data to a Gaussian. c , Relative RMS defect number fluctuations in a region versus the time-averaged number of defects in the region for experiments with (filled square) ξ = 1.6, a = 275 μm, (filled circle) ξ = 2.0, a = 372 μm, (filled triangle) ξ = 2.4, a = 268 μm, (filled triangle down) ξ = 5.9, a = 200 μm, and (filled diamond) ξ = 6.6, a = 167 μm. The error bars are standard errors of the mean. The line is an average of a linear fit to the data for each torus on a log–log scale, with a slope 0.53 ± 0.04 and a value for the exponential of the intercept equal to 1.02 ± 0.04. The errors are standard errors of the mean. d , Probability of finding a certain number of defects in a given region for a numerical simulation with λ ′ = 0.49, v 0 ′ = 50. The blue curve is a fit of the data to a Gaussian. e , Relative RMS defect number fluctuations in a region versus the time-averaged number of defects in the region for simulations with (filled square) v 0 ′ = 4, (filled circle) v 0 ′ = 50, (filled triangle) v 0 ′ = 80. f , Slope of the relative RMS defect number fluctuations versus plotted against the dimensionless defect velocity. Error bars are the standard error of the mean. The vertical solid line corresponds to the value v 0 ′ = 81. Full size image We confirm our estimate of v 0 ′ by considering the defect number fluctuations in a given area of our toroids (see Fig. 2f ). We find that the defect number distribution, P ( N A ), is Gaussian, as shown in Fig. 4b . Furthermore, the relative defect number fluctuations, , obtained from the width of the distribution, scale as , regardless of ξ and a , as shown in Fig. 4c . Similar results are obtained in the numerical simulations, as shown in the corresponding Fig. 4d, e . In the simulations, however, we can correlate , which is the slope of the line obtained from fitting versus , with v 0 ′, as shown in Fig. 4f . Using v 0 ′ = 81, plotted as a vertical line in Fig. 4f , we find , shown with a dashed line in the same figure. Experimentally, we obtain 1.02 ± 0.04, consistent with the numerical calculations. The agreement between the experiments and the theory reveals that the interaction between curvature and topological charge depends only on two parameters: the dimensionless mean distance between defects, λ ′, and the dimensionless speed of the +1/2 defects, v 0 ′, which relates to the intrinsic activity of the system. We emphasize that both of these parameters reflect local interactions that do not depend on the global size or geometry of the system. Intriguingly, we find that because of the large number of defects and their mobility, the topological charge approaches a continuous distribution. In addition, our results confirm that in the turbulent regime and without the influence of a boundary, the mean distance between defects acts as a proxy for l a . The agreement between the experiment and the theory combined with the fact that λ ≍ l a prompts us to use the topological defects as microrheological tracers and perform ‘topological microrheology’ to estimate the material parameters of our active material. From our confocal data, we measure the typical value of the +1/2 defect speed, v 0 ≍ 1.5 μm s −1 , use an estimate of the nematic viscosity, η = 13 Pa s from ref. 33 , and take λ = l a in the expressions for v 0 and l a , to estimate α ≍ 250 mPa and k F,3D ≍ 1.6 × 10 −9 N. These values are the first estimates of their kind. In addition, from the nematic layer thickness, l layer ≍ 15 μm, the value v 0 ′ = 81 and a typical a ≍ 250 μm, we estimate k F = k F,3D l layer ≍ 2.4 × 10 −14 N m and μ ≍ 200 m N −1 s −1 . Notably, topological microrheology is in principle independent of geometry and solely dependent on the fact that the nematic is in the turbulent regime and away from boundaries. To illustrate this, we consider experiments using an active nematic formulation identical to ours depleted onto a flat surface 50 . From the data for 140 μM ATP in that paper, we find an average v 0 ≍ 1 μm s −1 and λ ≍ 50 μm. Due to the system size and the prevalence of defects, boundary effects are likely unimportant. We obtain α ≍ 260 mPa and k F,3D ≍ 7 × 10 −9 N, consistent with the estimates obtained from our data. Topological microrheology should also be valid for other types of two-dimensional active nematics, such as confined bacterial suspensions in a lyotropic liquid crystal 34 . Our results introduce a new framework to explore the mechanical properties of active fluids and suggest that partially ordered active matter can be guided and controlled via gradients in the intrinsic geometry of the underlying substrate. Methods Sample preparation. Tubulin purification and preparation of kinesin–streptavidin complexes were done by the Dogic Group 8 , 9 . Microtubule polymerization 8 , 9 was performed by us prior to mixing the active samples. The final active samples were mixed 8 , 9 with an ATP concentration of 144 μM and the addition of 2% w/v Pluronic F127, a PEG-polypropylene oxide-PEG triblock copolymer sold by BASF, so that the active mixture could be used in conjunction with the yield-stress material employed to stabilize the toroidal droplets. The yield-stress material was composed of 84% w/w DC-9041 (Dow Corning), a silicone elastomer, diluted with 16% w/w polydimethylsiloxane oil with viscosity 10 cSt (Clearco). To generate toroidal droplets, we inject the active solution into a rotating cuvette containing the yield-stress material 40 . The resulting droplets are generally symmetric, and are stable provided the yield-stress is larger than the surface tension stress 40 . The stable droplet was left for ∼ 4–6 h until the microtubules had depleted sufficiently to form a nematic at the interface between the active mixture and the yield-stress fluid. Imaging. Image stacks were taken continuously on a Nikon A1R confocal microscope using a 10× objective until the activity ceased. Given the time spent in generating the droplets, this results in a typical measurement with 150–300 frames taken over 3–5 h. We emphasize we take images only of the lower half of the toroid; due to refraction effects, we cannot image the upper portion of the toroid. We image as much of the lower half as possible, balancing the time required to take an image stack versus the size of the image stack. Image analysis. All image analysis was done using custom code written in MATLAB. Our algorithms allow us to completely characterize the system for each measurement we take. We start with an intensity image from the confocal stack and extract the director using an anisotropic filtering scheme 41 . We then find the defects from the director field and the scalar order parameter. We use the time-averaged intensity at every point in the confocal stack to determine the location of the surface and then use a fitting technique from the computer vision literature 45 , 46 to measure the surface curvature. We are thus able to completely determine the director, identify the topological defects and their charge, and completely determine the surface curvature. Finding the director and the defects. In the intensity projection of a confocal stack, as seen for an instant in time in Supplementary Fig. 1a and Fig. 2b , n is given by the direction of the microtubule bundles. This direction is obtained from the greyscale image by finding the direction along which the intensity fluctuates the least for each pixel using a technique called coherence-enhanced diffusion filtering (CEDF) 41 . To begin with, the original intensity I is denoised using a Gaussian blur of standard deviation σ and side length 6 σ − 1 to lessen contributions to the intensity fluctuations from random noise, giving us the blurred intensity I σ ( Supplementary Fig. 1b ). Next, we calculate the gradient tensor for each pixel: Since the microtubule bundles have head–tail symmetry, we cannot use the gradient vector alone to define the bundle orientation as the gradient vector has a defined head and tail. The rank-2 gradient tensor is symmetric and is the same whether it is constructed with the gradient vector or the negative gradient vector. We now define the coherence direction of a rank-2 tensor as the direction of the eigenvector associated with the smallest eigenvalue. The coherence direction represents the direction along which the spatial intensity fluctuations are the weakest; its orientation is defined on the interval [0°, 180°]. The coherence direction of the gradient tensor is sensitive to pixel-length intensity fluctuations ( Supplementary Fig. 1c ); however, intensity variations defining the bundle direction occur on a longer length scale than a single pixel. Thus, we average once more to remove small-scale fluctuations in the coherence direction. Here, we perform a component-wise average of the gradient tensor for each pixel to find the structure tensor for each pixel. This operation can be written as: where K ρ is a Gaussian filter with standard deviation ρ , where ρ should be about the size of the relevant coherence feature in the image. If ρ is too small, the coherence direction of the structure tensor at each pixel will resemble those of the gradient tensor, whereas if ρ is too large the desired coherence features will be washed out by the averaging. We take this output of the coherence direction of the structure tensor for each pixel as the local ‘molecular director’, u , representing the local orientation of the active nematic (see Supplementary Fig. 1d and Fig. 2c ). From u , we find the tensor nematic order parameter, with I the identity matrix, by averaging u over all points in a five-pixel radius of the point of interest. Diagonalizing Q gives: providing n and S for each pixel. This is shown in Supplementary Fig. 1e and Fig. 2d . The defects are found by selecting pixels with S < 0.1 and then measuring the n -rotation around the point of interest. We do this by numerically calculating s = (1/2π) ∮ d n /d t ⋅ d t anticlockwise along the edge of a 5 px × 5 px plaquette centred on the point of interest, where t is the arclength parameter along the edge of the plaquette. We take the point of interest to be a ±1/2 defect if s ∈ ±[0.49,0.51] (see Supplementary Fig. 1f and Fig. 2d ). Due to the discrete nature of our data it is possible to ‘miss’ defects, especially when pairs of defects are close to each other. We calculate the error by comparing the results from our defect-finding routine to results from counting the defects in the director field manually in a set of regions randomly sampled over all time frames in all of our experiments. We find this error to be roughly 5%; we also find that the error is random, as the average charge in the randomly sampled regions calculated with our defect routine or calculated with the manual routine converge to the same value as more regions are considered. Measuring the Gaussian curvature. We measure the Gaussian curvature of a triangulated surface by fitting the elements of the Weingarten matrix, L ij = ∇ i N ⋅ e j , where N is the unit surface normal and e j is the tangent vector in the j th direction, as K = det( L ) (ref. 47 ). L relates an arbitrary displacement in the tangent plane, Δ r = Δ r j e j , of a surface with the corresponding change in the unit normal along the displacement according to: For a point of interest we consider local displacement vectors and the corresponding change in the unit surface normal vector, and then fit the components of L robustly according to equation (6) using an iteratively reweighted least-squares (IRLS) routine 45 , 46 . Let r 0 be an example point of interest in the triangulated surface, as seen schematically in Supplementary Fig. 2 . First we make an initial fit of L using a least-squares fit to serve as an input into the IRLS routine. We consider all points within d 1 of r 0 and use the surface normal at r 0 , N 0 , to transform all the points under consideration into the tangent plane of r 0 . We estimate N for each point in the region of interest using the average of the normal vectors of the adjacent faces in the triangulation. Next we use this initial estimate and calculate Δ r and Δ N for every possible pair of points, then use a regular least-squares fit to get an initial estimate of the components of the extended Weingarten matrix Λ according to: where we have added M 1 and M 2 to L to form Λ and capture the full change in Δ N along Δ r . We will use this additional information to eventually re-estimate N 0 . Note that the pairs of points considered do not need to contain the point of interest; this reduces the influence of error in r 0 . This can be seen schematically for the pair of points r 1 and r 2 with surface normals N 1 and N 2 , and displacement vector Δ r 12 , for the point of interest r 0 in Supplementary Fig. 2 . We now choose all points within d 2 of r 0 , where d 2 > d 1 , and as before, transform into the tangent plane and calculate the associated Δ N and Δ r . We perform an IRLS fit, where the weights for each iteration are based on the residuals of the previous iteration, with weights for the first iteration calculated using the initial estimate of Λ . Explicitly, for the residuals from iteration p , , where i and j index the pair of points under consideration, we use the weighting function given by the Geman–McClure M-estimator 44 to calculate a set of fitting weights w for the following iteration, where ζ ( p ) = median{ bγ ( p ) }, with γ ( p ) a vector of all the residuals under consideration, and b an input tuning parameter governing the influence of the residuals on the weights as well as governing the outlier rejection criterion; in our case b = 2. We also include a geometric contribution to the fitting weights m ij defined as: where again i and j index the pair of points under consideration, and C is a normalization constant such that ∑ i , j m ij = 1. The geometric contribution ensures that the points far away from r 0 have a reduced contribution to the curvature at r 0 . The final fitting weight for each pair for the ( p + 1)th iteration is m ij w ij ( p +1) . To determine convergence, we set a tolerance on the difference between iterations of the cost function from the Geman–McClure estimator, Once the fit has converged, we calculate the Gaussian curvature K from the final value of L . Next, we use the weights from the final fitting iteration p to re-estimate N 0 . Let N i (0) be the initial estimate of the surface normal at r i such that N 0 ( p ) is the re-estimated vector at the point of interest, Note that here we have used only pairs of points that include r 0 . This re-estimation uses the fitted curvature to determine what the surface normal should be. We finally transform N 0 ( p ) back into the laboratory frame of reference and calculate the determinant of the metric, which gives the local area element . Numerical calculations. Following the approach introduced in ref. 8 , we describe nematic defects as self-propelled particles interacting via elastic forces and torques. Fuelled by the activity of the motors, the highly distorted configuration of the nematic director around an isolated disclination builds up a non-uniform active stress σ active ∼ α nn T , where the constant α embodies all the contributions associated with the kinesin activity. In the presence of a +1/2 disclination such an active stress drives the flow shown in Supplementary Fig. 3a . The director field surrounding the defect is then advected by such a self-generated flow, which results in a motion of the defect core at constant speed v 0 ≍ αl a / η , where η is the fluid viscosity and , with k F Frank’s elastic constant, is the active length scale resulting from the competition between active and elastic stresses. Due to their threefold rotational symmetry, −1/2 disclinations do not self-propel ( Supplementary Fig. 3b ), but still interact with each other and with the +1/2 disclinations. In addition, both positively and negatively charged disclinations experience a position-dependent force that tends to localize them in regions of like-sign Gaussian curvature 13 . Let r ( x 1 , x 2 ) be the position of a defect lying on a generic surface parametrized via the pair of coordinates ( x 1 , x 2 ). Furthermore, let g ν = ∂ r / ∂x ν , with ν = 1,2, be a basis of covariant vectors on the tangent plane, so that g μν = g μ ⋅ g ν is the surface metric tensor. If the vectors g μ are orthogonal, the velocity of the particle moving at a constant speed on the surface can be expressed as: with e ν = g ν / | g ν |, ψ the angle in the tangent plane measured from the e 1 direction, and v 0 a constant ( Supplementary Fig. 3c ). In local coordinates, equations (1) reads: where A ν = e 1 ⋅ ( ∂ e 2 / ∂x ν ) is the spin connection and ζ = ζ ν e ν and η are delta-correlated random variables representing possible translational and rotational fluctuations. The axisymmetric torus can be parametrized in via the coordinates θ and φ . Thus: In these coordinates, equations (14) and (15) become: We focus on equations (17) and (18), which govern the dynamics of the defect core. The elastic energy E is given, in the one elastic constant approximation, by: If the distortion of the nematic director results solely from the defects, using the theory of topological defects on curved surfaces 13 , it is possible to cast equation (20) in the form: where G ( r , r ′) is the Laplacian Green function, K is the Gaussian curvature, and ρ is the topological charge density. In the presence of N nematic disclinations of topological charge s i , with i = 1, …, N , The total topological charge is, in turn, subject to the global constraint imposed by the Poincaré–Hopf theorem: which implies that the torus is always populated by the same number of positive and negative disclinations. The Laplacian Green function on an axisymmetric torus has been calculated in refs 13 , 28 : where 〈 G 0 ( r , ⋅ )〉 indicates a spatial average with respect to the dotted variable and G 0 ( r , r ′) is given in ref. 28 : where: is a complex coordinate resulting from conformally mapping the torus onto the complex plane, where The function ϑ 1 ( u | τ ) = ϑ 1 ( u , q ), with q = e iπ τ , is the Jacobi theta function 51 , defined as: Combining equation (21) with equation (25) and using equation (24) results in: The first term represents the interaction energy of the defects, while the second term corresponds to the potential energy due to the substrate Gaussian curvature; it is given by: Finally, the third term collects all energy contributions that do not depend on the defect positions (that is, defect self and core energies and the distortion energy due to the Gaussian curvature alone). The forces acting on the defects are calculated by differentiating the Coulomb-like energy in equation (29). This yields: where Re[ ⋅ ] and Im[ ⋅ ] stand for the real and imaginary parts, respectively, and we have introduced the simplified notation: Numerical simulation details. Simulations are performed by numerically integrating equations (17)–(19). For simplicity, we neglect the elastic torque between defects. This hypothesis is justified since, in the chaotic regime, the defect angular motion is essentially random. To render equations (17)–(19) dimensionless, we rescale all distances by a and time by the timescale associated with the relaxation of the nematic director, τ = a 2 /( μ t k F ). Gaussian curvature is then expressed in units of a −2 , while the defect speed v 0 is measured in units of a / τ . Defects of opposite charge are annihilated when they come within the defect core radius of each other, the latter set to r c = 25( a / τ )Δ t , where Δ t is the time step used for numerical integration. To keep the number of defects constant, the annihilated pair is reintroduced at a random position on the torus as a dipole with separation sufficiently high to prevent immediate re-annihilation of pairs ( ≍ 200 r c ) and random orientation. The +1/2 defect is oriented such that it will self-propel away from the −1/2 defect. Measuring topological charge and Gaussian curvature. In the experiment, we consider regions of our toroids of a given area A , numerically integrate the Gaussian curvature in the region, and then obtain the time-averaged topological charge in the region. We numerically integrate the Gaussian curvature by summing the Gaussian curvature at each pixel in the region weighted by the square root of the determinant of the metric, , where g comes from equation (12) and ( i , j ) index pixels contained in the region. We find that the time-averaged topological charge in a region on our toroids depends only on the integrated Gaussian curvature and is insensitive to the area of the region under inspection as well as to the local magnitude of the curvature, provided we average long enough. This is seen in the plots of the time-averaged topological charge for three different regions on a toroid with vanishing integrated Gaussian curvature and different areas in Supplementary Fig. 4 . In each case, the time-averaged charge converges to its final value well within the measuring period. Note that, in general, smaller areas require longer averaging times; however, our areas are always large enough such that the time-averaged charge converges within the measuring window. This insensitivity to the local magnitude of curvature implies that our results are not dependent on the aspect ratio of the toroid, the system size, or even the local shape changes of our toroidal surface. Given an active nematic formulation, the only parameter that matters is the integrated Gaussian curvature of the region under consideration. In the simulations, we consider narrow ‘ribbons’ and integrate the Gaussian curvature over all φ from 0 to 2π and a narrow band of width Δ θ centred on a given θ value. As with the experiments, the time-averaged charge in the ‘ribbon’ converges to a steady-state value well within the simulated time frame. Each region in the experiment or ribbon in the simulation that we consider yields a single point on the plot of time-averaged charge versus integrated Gaussian curvature in Fig. 2g and Fig. 3b , respectively. Code availability. The code used to analyse the images and perform the numerical calculations is available from the corresponding authors upon reasonable request. Data availability. The data that support the findings of this study are available from the corresponding authors upon reasonable request. Additional Information Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Imagine a tiny donut-shaped droplet, covered with wriggling worms. The worms are packed so tightly together that they must locally line up with respect to each other. In this situation, we would say the worms form a nematic liquid crystal, an ordered phase similar to the materials used in many flat panel displays. However, the nematic phase formed by the worms is filled with tiny regions where the local alignment is lost - defects in the otherwise aligned material. In addition, because the worms are constantly moving and changing their configuration, this nematic phase is active and far from equilibrium. In research reported in the journal Nature Physics, scientists from the Georgia Institute of Technology and Leiden University in The Netherlands have described the results of a combined theoretical and experimental examination of such an active nematic on the surface of donut-shaped - toroidal - droplets. However, the researchers didn't use actual worms, but an active nematic composed of flexible filaments covered with microscopic engines that are constantly converting energy into motion. This particular active material, originally developed at Brandeis University, borrows elements of cellular machinery, with bundles of rod-like microtubules forming the filaments, kinesin motor proteins acting as the engines, and ATP as the fuel. When this activity is combined with defects, the defects "come to life," moving around like swimming microorganisms and exploring space - in this case, exploring the surface of the toroidal droplets. By studying toroidal droplets covered by this active nematic, the researchers confirmed a longstanding theoretical prediction about liquid crystals at equilibrium, first discussed by Bowick, Nelson and Travesset [Phys.Rev. E 69, 041102 (2004)] that nematic defects on the curved surface of such droplets will be sensitive to the local curvature. However, since the active nematic used in this work is far from equilibrium, the researchers also found how the internal activity changed and enriched the expectations. "There have been predictions that say defects are very sensitive to the space they inhabit, specifically to the curvature of the space," said Perry Ellis, a graduate student in the Georgia Tech School of Physics and the paper's first author. "The torus is a great place to investigate this because the outside of the torus, the part that looks locally like a sphere, has positive curvature while the inner part of a torus, the part that looks like a saddle, has negative curvature." "The quantity that characterizes a defect is what we call its topological charge or winding number," said Alberto Fernandez-Nieves, a professor in Georgia Tech's School of Physics and another of the paper's co-authors. "It expresses how the alignment direction of the nematic liquid crystal changes as we go around the defect. This topological charge is quantized, meaning that it can only take values from a discrete set that are multiples of one-half. " In these experiments, each defect has a topological charge of +1/2 or -1/2. To determine the charge and location of every defect, Ellis observed the toroidal droplets over time using a confocal microscope and then analyzed the resulting video using techniques borrowed from computer vision. The researchers found that even with the molecular motors driving the system out of equilibrium, the defects were still able to sense the curvature, with the +1/2 defects migrating towards the region of positive curvature and the -1/2 defects migrating towards the region of negative curvature. In this new work, the scientists took a step forward in understanding how to control and guide defects in an ordered material. "We have learned that we can control and guide partially ordered active matter using the curvature of the underlying substrate," said Fernandez-Nieves. "This work opens opportunities to study how the defects in these materials arrange on surfaces that do not have constant curvature. This opens the door for controlling active matter using curvature." An unexpected finding of the study was that the constant motion of the defects causes the average topological charge to become continuous, no longer taking only values that are multiples of one-half. "In the active limit of our experiments, we found that the topological charge becomes a continuous variable that can now take on any value," said Fernandez-Nieves. "This is reminiscent of what happens to many quantum systems at high temperature, where the quantum, discrete nature of the accessible states and associated variables is lost. Instead of being characterized by quantized properties, the system becomes characterized by continuum properties." Ellis' observations of the droplets compared well with those of numerical simulations done by Assistant Professor Luca Giomi and postdoctoral researcher Daniel Pearce at the Instituut-Lorentz for Theoretical Physics at the Universiteit Leiden in The Netherlands. "Our theoretical model helped us decipher the experimental results and fully understand the physical mechanism governing defect motion," said Pearce, "but also allowed us to go beyond the current experimental evidence." Added Giomi: "Activity changes the nature of the interaction between defects and curvature. In weakly active systems, defects are attracted by regions of like-sign Gaussian curvature. But in strongly active systems, this effect becomes less relevant and defects behave as persistent random-walkers confined in a closed and inhomogeneous space". There are many examples of active systems driven by internal activity, including swimming microorganisms, bird flocks, robot swarms and traffic flows. "Active materials are everywhere, so our results aren't limited to just this system on a torus," Ellis added. "You could see the same behavior in any active system with defects." The research sets the stage for future work in active fluids. "Our results introduce a new framework to explore the mechanical properties of active fluids and suggest that partially ordered active matter can be guided and controlled via gradients in the intrinsic geometry of the underlying substrate," the authors wrote in a summary of their paper.

10.1038/nphys4276

Earth

Study: Rural ranchers face less access to water during drought than urban counterparts

Melissa Haeffner et al, Investigating environmental migration and other rural drought adaptation strategies in Baja California Sur, Mexico, Regional Environmental Change (2018). DOI: 10.1007/s10113-018-1281-2

http://dx.doi.org/10.1007/s10113-018-1281-2

https://phys.org/news/2018-02-rural-ranchers-access-drought-urban.html

Abstract This paper explores the relationship between specific household traits (region of residence, head of household occupation, financial diversity, female level of education, land and animal ownership, social capital, and climate perception) and choice of specific adaptation strategies used by households in two sites in Baja California Sur, Mexico, during a severe drought from 2006 to 2012 using survey data and key informant interviews. We analyzed the co-occurrence of household traits adopting different drought adaptation strategies, then applied Qualitative Comparative Analysis (QCA) to examine the relationship between traits and strategies and integrated interview data to understand how rancheros perceive associations. We found evidence of diversity among households within the larger cultural group, both in the types of resources they have available and in the adaptation strategies they select. However, the most robust finding across the analyses appeared to be urban access; that is, the more a household was able to access urban services including piped water, the less likely they were to have used one of the drought adaptation strategies under study. These findings suggest that social structure and public investments are stronger predictors of smallholder adaptation rather than individual household traits. We also found that rancheros seem to rely less on traditional environmental migration to adapt to drought and rather settle in key watershed zones. We call for targeted policies to address inequities to access fresh water, including urban water, during drought times for the benefit of overall watershed health and the sustainability of rural ranchero livelihoods as they evolve to respond to climatological and economic change. Access provided by Universität des es, -und Working on a manuscript? Avoid the common mistakes Introduction The increase in drought risk due to a progressively drier climate in the subtropical latitudes over parts of North America (Cook et al. 2010 ) translates to a growing need to understand how households will adapt to drought changes, and why households respond in certain ways. Furthermore, novel approaches are increasingly necessary to better understand complexity in household responses to these environmental changes (Kok et al. 2016 ; Okpara et al. 2016 ). This paper analyzes the relationship between specific household characteristics and their decisions to adopt drought adaptation strategies. It does so by combining multiple methods: analyzing the co-occurrence of household characteristics with adoption (or not) of drought adaptation strategy, and complementing this information with qualitative comparative analysis and the thematic analysis of qualitative interview data. Vulnerability research has highlighted specific household traits that enable the ability to adopt drought adaptation strategies, including financial diversity (de Janvry and Sadoulet 2001 ), the amount and type of land and livestock owned (Massey et al. 2010 ), membership in social organizations (a mechanism to draw on social capital and social networks for help) (Narayan and Pritchett 1999 ), female education (Blankespoor et al. 2010 ), and access to fresh water sources (Gray 2009 ). Previous research has also identified the main coping and adapting strategies that smallholder rural households take during drought: changing farm practices, finding off-farm work, and migrating. Changing farm practices and finding off-farm work have been shown to be important survival mechanisms among smallholder farmers (for example, Laube et al. 2012 ; Mardero et al. 2015 ). Others move the entire household or single household members away from the environmental threat (for example, Afifi et al. 2013 ; Alscher 2010 ; Gray and Mueller 2012 ; Leighton 2011 ; McLeman and Ploeger 2012 ), here defined as environmental migration. Footnote 1 Persistent environmental degradation might also undermine watershed health, reducing the economic benefit of staying for households who depend on immediate ecosystem services whether or not these households attribute their decision to the changing environmental conditions (IOM 2007 ). Climate model scenarios for Baja California Sur Mexico, predict that precipitation will decrease by as much as 30% by 2050 (Cavazos and Arriaga-Ramírez 2012 ). How households, and specifically rural households in this area, are likely to respond to extended droughts has not been well studied to date. Anecdotal evidence suggests that the rural households of Baja California Sur, known as rancheros sudcalifornianos, were more mobile in the past—they were able to move from one grazing area to another or switch to fishing on the coast to wait out the drought. The question of whether or not rancheros can maintain ranching as their main economic source and their lifestyles within a changing environmental and social context remains to be answered. In this paper, we assess whether and how specific household characteristics (traits) in Baja California Sur influence household adoption of four specific drought adaptation strategies: (1) environmental migration (direct), (2) household member migration (indirect), (3) changing farm practices, and (4) finding off-farm work. We assess these relationships by analyzing the co-occurrence of specific traits in cases of adoption versus not adoption of a drought adaptation strategy by analyzing how different trait configurations lead to adoption/not adoption of specific adaptation strategies, and by comparing these with an analysis of trait and adaptation strategy themes embedded in the in-depth interviews. Methods Study area The northern Mexican states registered significant drops in precipitation starting in 2006 and peaking in 2011–2012—the worst drought in Mexico in the previous 70 years (CONAGUA 2013 ). Baja California Sur recorded the most serious decrease in rainfall, by 70% ( ibid ). Mexico declared a federal drought emergency in three municipalities in Baja California Sur in 2011 where rural rancheros live ( ibid )—two of which are studied here (Figure 1 Site Map, Online Resources). The rural area adjacent to the state capital of La Paz runs along the corridors of the Novillo and Trincheras mountain ranges, which receive the most rain per year in this region and is considered the main recharge zone for the city’s only aquifer. Through discussions with local practitioners, it was estimated that the total population in 2010 in the Sierra catchments was approximately 750 persons in a 1417-km 2 area (Niparajá 2014 ). The other rural area, San Javier, is adjacent to the municipal capital of Loreto with 131 inhabitants according to the 2010 Census (INEGI 2010 ). San Javier is the location of an historic Spanish Jesuit mission site and attracts some international tourism in addition to a small-scale agriculture economy (e.g., cattle, palms, onions, sugarcane, honey, etc.). Ranchero survey A household survey targeted rural households exposed to a severe, federally declared drought from 2006 to 2012 (see Online Resources for the full survey in English and Spanish). Sample sizes for each location were calculated to achieve a 95% confidence level with a +/− 5% margin of error. Because rancheros did not have internet access or addresses for mailing, participants were contacted in their homes using a door-to-door dispersed sampling method. Rancheros were selected based on the following criteria: were at home and available at the time of survey, consented, and were located within the boundaries of either the Sierra (La Paz) or San Javier (Loreto) watersheds. At the time, there was not an updated map (online or otherwise) showing current homes nor were either areas designed on a grid, making it impossible to know the true household locations for random sampling. However, in high dense areas, every third household was approached and extreme efforts were made to reach the most isolated areas including homes without road access. A trained team of 12 local students conducted the surveys in Spanish who read the question and answer choices aloud and marked the participants’ answer. Survey development took into account previous research assessing the relationship between specific household characteristics and adoption of drought adaptation strategies. In order to systematically review previous work, we searched Web of Science, Science Direct, JSTOR, and EBSCOHost using the following keywords: “adaptive capacity index,” “adaptive capacity assessment,” “drought adaptation,” and “environmental migration.” A total of 13 peer-reviewed journal articles were used to select indicators for the household survey, summarized in Table 1 . The six household traits we elicited, together with location of household, whether the head of household ranches as their primary occupation, and whether the household perceived changes in weather patterns, were considered to influence the ability and the household decision-making with respect to drought strategy. Table 1 Household traits and drought coping/adaptation strategies Full size table To ensure cultural sensitivity, experts from a local university and two local non-profit organizations Footnote 2 reviewed the survey to assess the cultural appropriateness and wording of the survey questions. Once accepted, we piloted the survey with two ranchero households within the two municipalities of choice. Table 1 reports the symbol, meaning, and descriptions of the variables used in this study. Ranchero interviews In addition to the surveys, 11 ranchero groups (some individual, some groups of neighbors) were selected for in-depth interviews on their drought experiences in Baja California Sur. Interviewees were selected according to the following criteria: spread of geographic location including proximity to urban centers, broad age range to capture differing drought memories and adaptation choices related to stage in the life cycle, and a mix of males and females to capture potential gender-specific roles in drought response. A total of 18 rancheros participated in the conversations, six groups in the La Paz region and five groups in the Loreto region consisting of five women and 13 men ranging in age from 29 to 96 years. Two locals were hired to conduct the interviews in the Spanish dialect specific to rural Baja California Sur in the presence of the first author. One interviewer was an employee of a water conservation non-governmental organization (NGO) in the area, and the other worked informally with a local sustainable ranching NGO. Some respondents spoke to us alone while others were more comfortable speaking to us with a spouse or neighbors, dynamically taking their cues from each other. The interviews were semi-structured in that questions were not asked verbatim to each respondent. Interviewers were trained to guide the open-ended conversation towards drought experience, drought response, and available resources for responding to the drought. Interviewers were also trained to prompt for place, age, and gender details (e.g., urban access, drought memory, gender roles such as carrying water, etc.). Interviews typically lasted from 30 min to an hour. The interviews were audio-recorded with permission and two bilingual speakers transcribed the recordings in both English and Spanish. English versions were thematically coded using the household traits (region, female education, head of household occupation, access to surface water for human consumption, climate perception, social capital, financial diversity, land ownership, and livestock ownership) and four adaptation strategies (environmental migration, household migration, changing farm practices, and acquiring off-farm work) using NVIVO Pro 11.4.1 for Mac (see Online Resources Table 2 for summary). Illustrative remarks are included to provide deeper meaning to the results. Analysis The relationships between household traits and strategies are assessed via the following methods: (1) visualization of trait co-occurrence in relation to the adoption of specific strategies using household survey data, (2) assessing configurations of traits that lead to specific strategies via qualitative comparative analysis (QCA) using household survey data, and (3) thematic coding of traits and strategies of purposively sampled interviews. While co-occurrence can give insights on what type of traits are more frequently associated with adoption and non-adoption of adaptation strategies, the use of QCA allows us to assess which traits are the most “important” ones in shaping household decisions to adopt or not a specific drought strategy. We complement these methods with insight from the in-depth interviews to allow for a deeper understanding of the phenomenon we are analyzing. For both co-occurrence analysis and QCA, we assess presence and absence of a specific trait in relation to the adoption or non-adoption of a specific adaptation strategy (see Table 4 , Online Resources). Co-occurrence analysis highlights the traits which are more likely to co-occur in the presence (or absence) of a specific adaptation strategy. Following a similar approach, Baggio et al. ( 2016b ) analyzed the co-occurrence of institutional design principles and their relationship to successful common pool resource regimes, while Rocha et al. ( 2015 ) analyzed co-occurrence of drivers in marine regime shifts. This type of analysis is purely visual and can be thought of as a first step in identifying the relationship between trait configurations and strategies. To gain a deeper knowledge of the trait configuration—drought response relationship—we complement the co-occurrence analysis with QCA (as in Baggio et al. 2016b ). The output of QCA is a list of all possible combinations of relevant conditions within the parameters of this study that lead to the designated outcome. Here, we assess whether household traits are necessary or sufficient for the adoption or non-adoption of a specific strategy (Ragin 1987 ; Ragin 2014 ). A trait, or a trait configuration, is necessary if it must be present for a certain outcome to occur. A trait, or a trait configuration, is sufficient if, by itself, it can produce a certain outcome (Ragin 1987 ). A trait or a trait configuration is both necessary and sufficient if it is the only cause of the outcome (see Online Resources for more information). QCA also allows us to assess the consistency, or the degree to which a relation of necessity or sufficiency is met in a given dataset, and coverage that provides a “degree” of relevance of the sufficiency and necessity conditions for an outcome to occur. Both metrics vary between 0 and 1: 0 indicates no consistency or coverage, and 1 indicates complete consistency/coverage. QCA allows us to identify the multiple, different, and non-exclusive trait configurations of rural households that lead to the adoption or non-adoption of different strategies. We focus on specific household traits relating to the drought strategies identified for study (if any), and not on the linear relation between variables. We also pay close attention to the sensitivity of QCA to missing information. Following Baggio et al. ( 2016b ), QCA is complemented by a reliability metric based on missing value imputation. The metric proposed in Baggio et al. ( 2016b ) allows us to build a range of reliability for the information presented. Finally, we complement the results with in-depth interviews to broaden our analysis beyond the binary variables that were necessary to establish co-occurrence and QCA. The integration of co-occurrence and QCA with in-depth interviews and knowledge of the system allows researchers to increase our understanding of system under study (Baggio et al. 2016b ; Barnett et al. 2016 ). Results Household survey results We attained a high response rate (90%) with the household survey, representing a total of 163 ranchero households and 657 individual family members. Of the 163 rancheros surveyed, the majority (87%) reported that they responded to the 2006–2012 drought in some way while 13% reported that they did nothing (see Table 1 , Online Resources). Among the strategies analyzed, changing farming practices and acquiring off-farm work were much more prevalent than either type of migration (> 50 vs. < 30%, respectively). Trait Co-occurrence We start our analysis of trait combinations by identifying co-occurrence of household traits under different coping/adaptation strategies. Here, we present only results for complete cases (see Online Resources for co-occurrence patterns using missing value imputation). First, we analyzed differences between households who did not respond to the drought versus those who did, independent of the specific type of strategy. Figure 1 shows households who did not employ the strategy under consideration (left side) and households who did (right side) with stronger co-occurrence represented in redder and weaker co-occurrence represented in bluer colors. Overall, we see that those who had strong co-occurrence of household traits (redder squares) were more likely to have adopted a drought adaptation strategy, and this was true for each individual strategy except for acquiring off-farm work (Fig. 1 ). Households that responded to the drought shown in Fig. 1b are characterized by the co-occurrence of financial diversity and land title as represented by red squares, and these co-occurred more strongly with livestock ownership and climate perception. Households in La Paz who had social capital and also perceived changes in the climate were slightly less likely to have responded to the drought (Fig. 1a, b ). Fig. 1 Frequency of household traits of co-occurrence by type of strategy. Strong associations between two traits are closer to 1 (1 indicating that two traits always co-occur) and are represented by darker red colors. Weak associations between two traits are closer to 0 (0 indicating that two traits never co-occur) and are represented by darker blue colors. For example, observing Fig. 1a , one can infer that households that lived in La Paz ( reg = 1) and stated that they perceived changes in the climate ( clim = 1) were less likely to adopt one of the drought adaptation strategies under study (darker red color where the row meets the column). On the other hand, the two traits co-occurred less frequently in case of households that adopted one of the adaptation strategies inferred by the darker blue color where the row meets the column (Fig. 1b ). The middle diagonal boxes represent the intersection where a trait on the x-axis is the same as the y-axis (essentially showing that it “co-occurs” with itself 100% of the time. Refer to Table 1 for abbreviations Full size image Environmental migration Fifteen percent of our sample self-reported migration as a direct drought strategy. Households who reported environmental migration were characterized by the co-occurrence of the following traits: female education, climate perception, and livestock ownership, and a stronger co-occurrence between climate perception and land title (Fig. 1d ). Meanwhile, households in the La Paz area who were aware of climate changes were less likely to have environmentally migrated (Fig. 1c ). Household member migration Twenty-nine percent of households surveyed had a member who migrated during the drought years 2006–2012, whether or not they attributed this movement directly to the drought. These families were more likely to have the combination of perceived climatic changes in the last 10 years, females with higher education, and livestock ownership (Fig. 1f ). Also, these households were more likely to have the combination of social organization membership (social capital) and livestock ownership (Fig. 1f ). On the other hand, households who perceived a change in climate patterns in the last 10 years and owned livestock, but also lived in the La Paz municipality, were less likely to have had any migration during the 2006–2012 drought. Changing farming practices More than half of the households (56%) reported changing farming practices as a drought strategy. Households who changed their farming practices were more likely to have the combination of the following traits: financial diversity, owning a title to the land, and owning livestock, or the co-occurrence of female education and livestock ownership as shown by the comparison of Fig. 1g, h . Living in the La Paz municipality and holding a land title co-occurred in households that did not change their farm practices as shown in Fig. 1g . Acquiring off-farm work Acquiring off-farm work was the most adopted strategy, although marginally (57 vs. 56% that adopted changing farming practices). Co-occurrence analysis indicates that the strongest distinction was that households living in La Paz with financial diversity were less likely to acquire off-farm work. Qualitative comparative analysis Figure 2 reports results of the relative trait importance in determining household decisions to adopt specific drought adaptation strategies. Trait position is given by the frequency in qualitative comparative analysis (QCA) solution sets relating to adoption and non-adoption of specific drought adaptation strategies (see Online Resources Tables 4 and 5 ). We report solution sets in which the configurations occur in over 50% of cases (consistency > 0.5). Trait position in Fig. 2 identifies discriminatory factors in the adoption of drought adaptation strategies. Traits placed in the upper left quadrant are positively associated with strategy adoption, while traits placed in the bottom right quadrant are negatively associated with strategy adoption. Traits placed in the other quadrants are neither positively or negatively associated with strategy adoption. Fig. 2 Positive association (top left quadrants) and negative association (bottom right quadrants) of household traits with each drought strategy. The top row shows complete cases, the middle row shows data when missing values are assumed 0 (absence of trait), and the bottom row shows data when missing values are assumed 1 (presence of trait). HH refers to “household” Full size image Figure 2 shows that owning a land title is positively related to the adoption of environmental migration even when accounting for missing cases. This means that owning land was often present in trait configurations that led to the adoption of the environmental migration strategy, and was, at the same time, absent when environmental migration was not chosen (top left quadrant). Ranching was also positively related to the adoption of environmental migration, albeit to a lesser extent and with lower reliability. Female education was positively correlated to having any member of the household migrate during the drought years. Living in the La Paz area was negatively associated with changing farming practices and acquiring off-farm work. No traits were positively associated with the adoption of drought adaptation strategies in general, while living in La Paz was negatively associated with adopting drought adaptation strategies in general. Interview data All respondents we approached agreed to an interview and agreed to be recorded, resulting in a 100% response rate. The interviews revealed that the cultural identification with rancheros was separate from the practice of ranching; in other words, they considered themselves to be rancheros even if they did not ranch. Several rancheros told us that they considered ranching to be too difficult to maintain, for example, as one ranchero said: “I am struggling. We struggle and suffer. There are two or three goats left to support me out of thirty or forty. And you can see that in most ranches, you can see the lack of animals, there are no animals, there is nothing (San Javier).” Table 2 in the Online Resources summarizes the topics that rancheros explored when discussing drought in the area, categorized by the assets we analyzed in the co-occurrence analysis and the QCA. Notably, no rancheros mentioned female education in any way, including, leaving the ranch to attend school, learning about drought strategies in school, or gender inequities in school attendance (Table 2 Online Resources, female education). Across the interviews, drought severity was mostly assessed by detriment to livestock (Table 2 Online Resources, livestock ownership). Climate perception and livelihoods In line with the descriptive statistics showing that 90% of the respondents were aware of the weather becoming more unpredictable in the last 10 years, the interviews also demonstrated that rancheros were very aware of how environmental conditions affects their livelihoods and ability to respond to drought. Some noted that the rains they expected in June did not fall until September and when they did fall, they were thunderstorms. Another rancher perceived more frequent hurricanes in the area (supported by other rancheros in the household survey open-ended questions). Further, some respondents told us that they chose land based on local knowledge, specifically in relation to weather patterns. For example, one respondent commented that rain was an important factor in their decision to migrate, in addition to social capital: “Because we had some relatives in the ejido (a system of cooperative land tenure) and it rains more here, there was more grass and water. There’s more life here, so to speak (La Paz region).” Changes in water availability and water distribution over time Many respondents also observed declines in both surface and groundwater sources (Table 2 , Online Resources) coinciding with climatological changes and urban extraction pressures on aquifers (Haeffner et al. 2016 ). Footnote 3 The interviews further revealed a perceived opposition between the efficiency values of the government and sufficiency ideals of rancheros in both regions, with respondents observing more government relief in the most recent drought than in previous years but not at sustainable rates. Those who were alive during the 1950 and 1970 droughts indicated that they did not receive any sort of government drought relief. Footnote 4 Specifically, older respondents remembered “none,” “none,” and “not much” drought relief from the government for the dry periods associated with 1953, before Hurricane Liza (pre-1976), and 1983, respectively, but all respondents recalled receiving some government assistance in 2011. Despite acknowledging this increase in assistance, one group in the La Paz region discussed this situation with one explaining that the government supplied 12 l of water per person per day in the last drought with one respondent retorting “I could barely shower!” and another lamenting “Barely enough to drink, even.” La Paz region residents talked about urban access to government water delivery and car transport of water from cities (Region row, Table 2 Online Resources) whereas Loreto region residents talked about local water infrastructure and neighbor sharing (Access to surface water row, Table 2 Online Resources). For example, a respondent told us: “When they brought water from La Paz, we had to transport it in a car (La Paz region).” Another ranchero in the La Paz region told us that even though he owned a truck, he could not afford the gas to go to town to get water. Houses in the La Paz region were also more likely to have indoor plumbing and water provided by a utility company. When conducting the survey in the La Paz municipality, we observed that the houses near the main highway were able to access the relatively more reliable urban pipe system (a tandeo system in which water is available every other day), and 21 of these households regularly used urban water for their animals. Migration outcomes From the interviews, we can further infer that seasonal migration was an important part of the ranchero adaptation portfolio. This might be why we saw a higher percentage of households who had a household member who migrated during the drought years but did not relate it directly to the drought. Tellingly, a popular response in the household survey to how long household members left the household was “va y viene,” (used in more than 66 write-in responses corresponding to 10% of all individuals considered) translated to “come and go.” This response was common whether the respondent moved between rural areas or to urban areas. When our interviewees did talk about migrating due to water scarcity, they talked about it in tandem with other driving factors, such as finding economic alternatives. According to one respondent, migration may be due to drought, even if not directly related to it: “They have left because there are no jobs here, no jobs at all. Because there are not any jobs here that pay very much. It used to rain a lot; there was water (La Paz region).” Another ranchero told us that while he used to migrate more readily in the past, costs were becoming too expensive to do so now: “There have been many droughts, I went to Puerto Escondido [a harbor town], because there was nothing here…The thing is, things weren’t as expensive as they are now, and now you can’t go anywhere… (San Javier).” Discussion Mixed methods: Integrating co-occurrence analysis, qualitative comparison analysis, and in-depth interviews Comparing the knowledge gained via in-depth interviews with co-occurrence analysis and QCA allowed us to better identify the key existing relationships between household trait and drought adaptation strategy adoption as well as why traits may influence such adoption. QCA provided the means to compare household trait make-up and drought responses while adding an important dimension to studying the variety of ways that rural households adapt to changing climatic conditions. QCA, however, required that complex factors to be collapsed to create binary variables. For example, owning livestock could have different effects depending on the type and number of livestock owned; land title could have different effects depending on the amount of area owned. Hence, by using QCA, we do lose some detailed information collected through the survey. The use of in-depth interviews reduces the limitations of QCA as they allow for complex factors to be better explained. Such interviews revealed the seriousness of drought consequences as well as knowledge gained about social capital, migration, and the importance of living closer to urban areas for water access, as all are critical traits that were identified in the in-depth interviews. According to our co-occurrence analysis, financial diversity and female education, together with owning livestock and land title, were more prevalent in households who responded to the drought. However, according to the QCA, the most important traits seemed to be land ownership and female education together with regional characteristics. Female education and owning land positively affected the decision to migrate: households that owned land titles with at least one female going to secondary school were more likely to choose to migrate. On the other hand, our interviews did not elicit any association between female education and drought adaptation. Furthermore, living in the La Paz municipality was negatively associated with household choices to change farming practices or finding off-farm work. Table 2 summarizes our findings for each of the methods used in this study. Table 2 Summary of trait influence on the adoption of specific drought adaptation strategy for each analysis Full size table Positive and negative definitions denote relationship, not normative assumptions. “Positive” means that the household trait is associated with the adoption of the drought adaptation strategy; “Negative” means that the household trait is associated with the non-adoption of the drought adaptation strategy. On the one hand, the co-occurrence and QCA results show that living in the La Paz area was negatively associated with decreased adoption of drought adaptation strategies. On the other hand, the interviews highlighted a situation where some La Paz area households were able to access urban water supplies, which seemed to increase their robustness to droughts. The importance of accessing urban water supply is highlighted by the fact that 21 households in the surveys, all in the La Paz region, told us that they regularly bring water from the city or use potable water from pipes for animal consumption. While our results are able to capture the relationship between rural households and drought adaptation strategies in Baja California Sur, the role of hard human-made infrastructure (i.e., canals, dams, piping of water, etc.) warrants further investigation. Increased reliance on such hard infrastructure may increase robustness in the short-term at the expense of reduced adaptive capacity in the future by changing (and possibly reducing) options available to the rancheros (Haeffner et al. 2016 ). Dependency may erode ranchero ability to adapt to more intense and frequent droughts, especially if urban water supplies should diminish or increase in cost, or regulatory barriers put in place that prevents rural access to urban water. This could especially become true given the current concerns of the ranchero community about rising costs and declining opportunities, unstable job market, and ultimately, the increased difficulty in maintaining farms and livestock. Finally, we found that while obtaining off-farm work was a popular choice for rancheros, environmental migration was not. A review of the open-ended and write-in answers to the survey questions revealed that the most popular head of household occupation was day-laborer. This type of low security employment is a short-term strategy to cope with changing social and environmental conditions but may add to overall vulnerability since those types of jobs depend on external factors like real estate markets (for new construction investments), the global tourism industry (for road building projects), federal funding (for disaster recovery), and the like. Given that the respondents in this study continue to identify as rancheros despite choosing alternative livelihoods to respond to drought, our findings suggest that, as Vadjunec et al. ( 2016 ) have proposed, smallholder identity may be as much of a state of mind as it is a certain number of hectares or livestock owned. Financial diversity was key for a household’s ability to create a buffer against drought according to the co-occurrence results, as well as enabling mechanisms that increased the ability of households to adopt other drought adaptation strategies such as migrating, or changing farm practices. The key role of financial diversity is not new, as its prominence is known within the literature on adaptive capacity (e.g., de Janvry and Sadoulet 2001 ; Liverman 1999 ; Massey et al. 2010 ). Owning livestock in this context can be thought to increase financial diversity if owners sell their animals as an income source. Although no one explicitly stated it in the interviews, it is likely that owning livestock allowed for households to “invest” in female education especially in the region of study, where secondary education is only available in urban centers (a high cost for households). Female education, whose benefits include increased social capital as girls are exposed to networks outside of the home, is considered key in increasing households’ overall adaptive capacity to respond to crisis (Eakin 2005 ; Gray 2009 ; Laczko and Aghazarm 2009 ; Massey et al. 2010 ). Specifically, research has shown that the number of educated girls and women in a household is associated with reduced flood and drought risk (Blankespoor et al. 2010 ). Social capital, and specifically social networks, has been shown to be an important factor for both migration decisions (Narayan and Pritchett 1999 ; Nelson et al. 2005 ) and community robustness (Baggio et al. 2016a ). The relationship between social capital and female education in this specific region could underlie the relation between female education and the decision to adopt migration strategies to adapt to drought, although this connection was not elicited in the interviews. Further, contrary to the robust literature on the role that social capital plays in adaptation in general (Narayan and Pritchett, 1999 ), our survey findings did not find strong associations between social capital and other household traits in relation to drought outcomes (but did find an association with household member migration for any reason). We used “membership in a formal organization” as a measure of social capital in the survey, a common indicator in the extant research literature (see for example, Narayan and Pritchett 1999 ; Vincent and Cull 2010 ). Perhaps this indicates a need to increase social capital among rural Baja California Sur households. On the other hand, this could suggest that the way that social capital is measured by researchers does not capture the types of social bonds in these communities, and future research could focus on culturally appropriate social capital measures designed by rancheros (for an example of social water management on mainland Mexico, see Navarro-Navarro et al. 2017 ). Possible justification for discordant results As we looked at ranchero drought adaptation from multiple angles, we found that some associations were emphasized using one lens while other associations were lost in the noise when looking at the data through a different lens. Indeed, our three-pronged data analysis process was designed to avoid confirmation bias of theoretical associations while exposing associations between traits and adaptation strategies that might not be immediately obvious at the household scale. Consider the fact that female education was related to household member migration strategies (a theoretical association) in both the co-occurrence and QCA analyses, but was not prominent in respondents’ telling of their drought experiences even when prompted. If we had only conducted interviews, we may have missed this important association and may have concluded that drought experience in Baja was gender neutral. While some might point to the inconsistency across analyses as a threat to validity and reliability, we believe that the information provides clues for further research, especially in the subjective observations of drought experience on the one hand and what we might be capturing when we think we are measuring female education on the other. Looking at the data through all three lenses also exposes the complexity of urban access shown by the La Paz region—we cannot conclude whether urban access alleviates or prevents households in the La Paz region to adopt drought strategies, only that there are strong relationships. This warrants further investigation as to the spatiality of inequity in access to urban water. Implications for water managers There is an opportunity here to re-imagine the narrative of rancheros as simply water end users or drought aid recipients to a partnership that recognizes ranchero settlement in key aquifer recharge zones. Drought insurance schemes are also scarce in this region but could provide needed economic recovery after drought disasters. A budget specifically earmarked for droughts could allow utilities to repair infrastructure in a timely manner to reduce lapses in urban water access, which we found to be important for many La Paz region households. And, while we acknowledge the climatic challenges of encouraging driving, short-term incentives such as a gas credits for car-pooling to transport water would allow rancheros more equitable access until long-term solutions are put into place. Indeed, potable water is already trucked to ranchero villages at the discretion of municipal utilities or commercial vendors in the La Paz region, but only household near major roads tend to have access to this urban water. It might be useful for municipal public utilities to work with city officials to review any restrictions, taxes, or laws preventing rural access to urban water, including exchange and sharing, especially in times of drought. Identifying where ranchero needs lie can help government policy makers address sufficiency gaps. Knowing this, government agencies could concentrate on options for households to be more flexible and secure in choosing drought adaptation strategies “in place,” such as supporting sustainable agriculture projects in the region, increasing transportation options to access city centers, providing access to insurance and credit schemes, increasing employment opportunities in order to increase the financial diversity of rural households, and setting aside land and corridors to facilitate seasonal migration for households and their livestock. Future studies on migration and other drought adaptation strategies can compare rural groups in other areas and their ability to migrate, adapt, and access urban amenities. Conclusion While our study was motivated by the goal of identifying which household traits, if any, were associated with which drought adaptation strategies households used, the most consistent finding across all analyses was that the more likely a household lived in the La Paz area (with more urban access), the less likely they were to have adapted to the last drought. Less consistently, we found that financial diversity when combined with land ownership and a perception that the weather had been more unpredictable in the last 10 years was associated with adapting to the last drought (in the co-occurrence analysis). What was not as surprising, however, was that since this cultural group has traditional ties to migrating during droughts, the QCA analysis found an association between having a rancher as the head of household and holding a land title (as opposed to the communal ejido system common in Mexico) with environmental migration out of the watershed. However, the qualitative work highlighted that even though rancheros talked about drought migration in the past, environmental migration was not a key strategy for rancheros in the 2006–2012 drought. Instead, rancheros in this study tended to adapt “in place” by changing their farm practices and/or acquiring off-farm work, but had varying assets available to maintain traditional livelihoods or conduct sustainable ranching practices. The qualitative data revealed the great difficulty rancheros have in sustainably maintaining their traditional ranching livelihoods under current conditions. Our study further suggests that spatial inequality exists between the two rural communities. In other words, living in the La Paz region was the one household trait that was important in all three data analyses, and with La Paz region households less likely to have taken any of the drought adaptation strategies under study. It is highly likely that access to fresh water during droughts exists along a gradient with more remote households lacking urban water infrastructure while being exposed to dry (or salt-contaminated) wells and dry streams. It is important to note, however, that the La Paz area interviewees reported that while they were receiving municipal water deliveries during the drought, that these were not sufficient. Future research should look at this phenomenon more closely to examine the interrelationship between water sources, water delivery, and household location, and the impacts of urban growth on the availability of water. If climate, economic, and urbanization trends continue, it appears highly unlikely that ranching can be a sustainable economic livelihood or cultural practice without intervention. Ranchero sudcalifornianos should ultimately be recognized as stewards of the land and supported as equal participants in sustainable water conservation. Notes We define environmental migration in accordance with the International Organization for Migration (IOM) ( 2007 ) “Environmental migrants are persons or groups of persons who, for compelling reasons of sudden or progressive changes in the environment that adversely affect their lives or living conditions, are obliged to leave their habitual homes, or choose to do so, either temporarily or permanently, and who move either within their country or abroad.” We define “migration” as crossing a specified boundary to establish residence for any period of time, in this case, across a watershed boundary. Experts were contacted at the local non-profit organizations Niparajá ( headquartered in the La Paz area and Raices Vivas ( /) in the San Javier area, and professors at the Universidad Autónoma de Baja California Sur. No respondents mentioned water quality issues such as saltwater intrusion, only quantity issues. Baja California Sur became a state of Mexico in 1974, which may have prevented previous aid distribution to the territory.

For farmers and ranchers in Mexico's southern Baja California peninsula during a six-year drought, the farther away they lived from urban areas, the more likely they were to have to make changes to cope with the dwindling supply of water, according to a Portland State University study. Melissa Haeffner, an assistant professor of environmental science and management at PSU's College of Liberal Arts and Sciences, said the findings highlight a rural-urban divide and show that ranchers' access to water was neither equal nor valued during the drought from 2006 to 2012. "Where people live and how close they are to the city and how well the city can deliver those services to households was unequal across the population and it had devastating effects for households who were not able to access those resources," said Haeffner, the study's lead researcher. The study recommends government agencies focus efforts on enacting policies and programs to better protect rural households during a drought, such as supporting sustainable agriculture projects in the region, increasing transportation options to access city centers and providing access to insurance and credit schemes. The study, published online in February in the journal Regional Environmental Change, surveyed 163 households from two municipalities: a rural area adjacent to the state capital of La Paz and the village of San Javier, high in the Sierra la Giganta mountains. Haeffner looked at whether the ranchers migrated or stayed in place but had to change their farming practices or find different work because of the drought. She found that most people reported changing their practices—reducing their herds, trading cows for goats or corralling them, as some examples—or finding other work that could sustain them. But the biggest finding was that those who lived closer to the city center or the main highway had better access to water deliveries than those in more remote areas—because of distance and transportation issues—but still less than their urban counterparts. Haeffner said city dwellers were supplied with 250 liters per person per day, while the rural families had to rely on twice-weekly deliveries that only became less reliable as the drought dragged on. To make matters worse, the wells and streams that the ranchers and their ancestors had relied on for centuries were either contaminated or dried up. Haeffner said the narrow-minded view that ranchers are only drought-aid recipients needs to be challenged. "If we think about their role in agriculture production and as people who are maintaining clean water for the city and the aquifer that serves the city, then we can think about how we support them and their livelihood in a completely different way," she said.

10.1007/s10113-018-1281-2

Nano

A new device with memorizing and forgetting functions like human brain is reported

Takeo Ohno, Tsuyoshi Hasegawa, Tohru Tsuruoka, Kazuya Terabe, James K. Gimzewski & Masakazu Aono, "Short-term plasticity and long-term potentiation mimicked in single inorganic synapses", Nature Materials (2011) Published online: 26 June 2011 doi:10.1038/nmat3054

http://dx.doi.org/10.1038/nmat3054

https://phys.org/news/2011-07-device-functions-human-brain.html

Abstract Memory is believed to occur in the human brain as a result of two types of synaptic plasticity: short-term plasticity (STP) and long-term potentiation (LTP; refs 1 , 2 , 3 , 4 ). In neuromorphic engineering 5 , 6 , emulation of known neural behaviour has proven to be difficult to implement in software because of the highly complex interconnected nature of thought processes. Here we report the discovery of a Ag 2 S inorganic synapse, which emulates the synaptic functions of both STP and LTP characteristics through the use of input pulse repetition time. The structure known as an atomic switch 7 , 8 , operating at critical voltages, stores information as STP with a spontaneous decay of conductance level in response to intermittent input stimuli, whereas frequent stimulation results in a transition to LTP. The Ag 2 S inorganic synapse has interesting characteristics with analogies to an individual biological synapse, and achieves dynamic memorization in a single device without the need of external preprogramming. A psychological model related to the process of memorizing and forgetting is also demonstrated using the inorganic synapses. Our Ag 2 S element indicates a breakthrough in mimicking synaptic behaviour essential for the further creation of artificial neural systems that emulate characteristics of human memory. Main Neuroplasticity, where changes in the strength of synaptic connections (or weights) are caused by memorization events, underlies the ability of the brain to memorize. STP is achieved through the temporal enhancement of a synaptic connection, which then quickly decays to its initial state. However, repeated stimulation causes a permanent change in the connection to achieve LTP; shorter repetition intervals enable efficient LTP formation from fewer stimuli. Although synaptic behaviour has been imitated by hardware-based neural networks such as hybrid complementary metal–oxide–semiconductor analogue circuits or other artificial neural devices 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , no inherent hardware-based memorizing ability has been implemented and is currently achieved through software programming. Recently, electrically induced switching phenomena have been shown in nanoscale ionic-based devices, which have been proposed as a basis for future non-volatile memories 21 , 22 , 23 . In non-stoichiometric materials with ionic and electronic conductivity, ion migration is coupled to reduction and oxidation processes that result in large electrical conductance changes. Artificial synaptic devices operated by ion migration have been reported 24 , and some of these devices 25 , 26 have shown evidence of spike-timing-dependent plasticity 27 characteristics. Spike-timing-dependent plasticity is an important memorization mechanism related to the synaptic strength of connections in biological circuits and synthetic devices that emulate the spike-timing-dependent plasticity model. They require precise control of the relative timing between the signals applied to the two electrodes, to mimic the pre- and post-synaptic potentials in biological systems. In previous work, we reported an atomic switch 7 , 8 , which is a two-terminal electroionics-based device, where formation and annihilation of a metallic atom bridge is controlled in a nanogap situated between two electrodes (typical current–voltage characteristics are presented in Supplementary Fig. S1 ). It uses a solid-state electrochemical reaction, which indicates the presence of some characteristics of memory formation related to a synaptic connection 28 , 29 . Namely, the switch showed two types of conductance state: one that rapidly fades away after weak signal inputs, analogous to STP, and a second long-lived stable state requiring a strong signal for erasure, conceptually analogous to LTP. Changes in conductance, including a transition between these two states, were observed on application of specific bias voltages, where conductance was determined by the history of previous input signals. As the adjustment of the timing of pre- and post-synaptic potential is not required, a new approach for emulating synaptic behaviour is available for exploration. However, a time dependence was not observed in that work, although it is an essential characteristic of neuroplasticity. Here, we report that the decay of conductance occurred in STP without any application of bias voltage, resulting in both time-dependent STP and LTP behaviour in a single device, which we find is dependent on stimulation rate. In a Ag 2 S inorganic synapse, we relate the temporal enhancement of conductance to STP, and find that it occurs before the complete formation of a metallic atomic-sized bridge, and that the decay in conductance is explainable by deformation of an incomplete bridge. Once a robust bridge is formed, the enhancement persists for a long period of time, that is, corresponding to LTP, as schematically illustrated in Fig. 1a . Moreover, we present a psychological model related to memorizing and forgetting, that was duplicated through a rehearsal process of the inorganic synapse by using the decay behaviour in measured electrical conductance. Figure 1: Inorganic synapse showing STP and LTP, depending on input-pulse repetition time. a , Schematics of a Ag 2 S inorganic synapse and the signal transmission of a biological synapse. Application of input pulses causes the precipitation of Ag atoms from the Ag 2 S electrode, resulting in the formation of a Ag atomic bridge between the Ag 2 S electrode and a counter metal electrode. When the precipitated Ag atoms do not form a bridge, the inorganic synapse works as STP. After an atomic bridge is formed, it works as LTP. In the case of a biological synapse, the release of neurotransmitters is caused by the arrival of action potentials generated by firing, and then a signal is transmitted as a synaptic potential. Frequent stimulation causes long-term enhancement in the strength of the synaptic connection. b , c , Change in the conductance of the inorganic synapse when the input pulses ( V =80 mV, W =0.5 s) were applied with intervals of T =20 s ( b ) and 2 s ( c ). The conductance of the inorganic synapse with a single atomic contact is 2 e 2 / h (=77.5 μS), where e is the elementary charge, and h is Planck’s constant. Full size image STP and LTP appeared during inorganic synapse operation, where input pulses with an amplitude ( V ) of 80 mV, a width ( W ) of 0.5 s and repetition intervals ( T ) of 2 s or 20 s were applied. When these input pulses were applied with a lower repetition rate, at intervals of 20 s, the system did not maintain the higher-conductance state of approximately one quantized channel (77.5 μS) that was achieved directly after each input pulse, but which decreases with time back to its initial low conductance value ( Fig. 1b ). This behaviour relates to the STP mechanism observed in biological synapses. The decay phenomenon appears after each input pulse, until the application of the next pulse. Importantly, this decay occurs without any application of voltage, a feature that is in contrast to other ionic devices, where an applied signal is required to cause a change in conductance 23 , 24 , 25 , 26 . The inorganic synapse also showed a long-lived transition to the higher-conductance state when the repetition rate of the stimulation pulses was increased, that is, when a shorter interval time ( T =2 s) between inputs was used, as shown in Fig. 1c . Here, a permanent transition to higher-conduction states is clearly observed with repeated application of input pulses, successfully mimicking the LTP mechanism of a biological synapse. The higher conductance range observed is over 77.5 μS, corresponding to a single atomic contact 30 , supporting a synaptic behaviour where STP is achieved by incomplete bridge formation whereas LTP is achieved by a complete atomic bridge formation. This resulting output response stability is also observed to be enhanced by application of input pulses, resembling a persistent increase in synaptic connection following higher-repetition stimulation by action potentials found in the biological nervous system 2 , as shown schematically in Fig. 1a . We investigated the stability of LTP as a function of input-pulse repetition rate, where we defined ‘LTP’ as a state maintaining a conductance higher than 77.5 μS after a time exceeding 20 s from the last electrical stimulus (see Supplementary Section S2 ). We also define ‘unstable potentiation’ as a temporary increase in conductance over 77.5 μS that is metastable and rapidly decays to a value smaller than 77.5 μS after an input pulse. After application of a total of 12 input pulses ( V =80 mV, W =0.5 s) with two different pulse intervals, it was observed that 97% and 78% of the recorded events corresponded to unstable potentiation with both shorter ( T =2 s) and longer ( T =20 s) intervals, respectively ( Fig. 2a ). The 2-s-interval pulses, however, caused 93% of the inorganic synapses to move to an LTP state whereas the 20-s-interval pulses gave only 19% of events resulting in LTP. The relatively high value of 19% of events resulting in LTP with longer interval was found to statistically relate to particular inorganic synapses that suddenly increased rather than gradually built up conductance as a function of input pulses. Each inorganic synapse operating in the STP state also showed a certain distribution of the threshold number of input pulses required for unstable potentiation (see Supplementary Fig. S5 ). Figure 2: LTP formation depending on input-pulse repetition time. a , The occurrence probabilities of inorganic synapses in which an ‘unstable potentiation’ state occurred during stimulation pulses and that formed an ‘LTP’ state at the end of the stimulation events. Input pulses ( V =80 mV, W =0.5 s) were applied 12 times at intervals of T =2 s or 20 s. b , The change in conductance after application of 12 input pulses with pulse intervals of T =20 and 2 s. The dashed line shows a conductance of 77.5 μS. Each circle represents an individual inorganic synapse. c , Typical change in the time constant of an inorganic synapse for the decay in the STP state. Full size image In Fig. 2b , we show that the conductance of an inorganic synapse after the 12th input pulse is smaller than 77.5 μS, and its subsequent decrease in conductance is characteristic of STP. Conversely, it was observed that its relatively easy to maintain an enhanced conductance higher than 77.5 μS, corresponding to the formation of a multi-atom metallic bridge. We found that the decay time of the STP state also maintains a history of the previous input pulses. This is shown in Fig. 2c , where the exponential decay time of STP is observed to increase with the number of input pulses. This history dependence indicates that a non-stoichiometry of Ag 2 S, supporting the Ag atomic bridge, rearranges 31 with subsequent pulses. The lack of Ag + ions in the Ag 2 S region, adjacent to the Ag cluster, can then recover with an increase in the number of input pulses. The measured time constant is of the order of seconds, indicating that the phenomenon is not purely electronic, but rather electroionic, because electrical capacitance effects have an estimated timescale of nanoseconds (see also Supplementary Section S4 ). The inorganic synapse was observed to go beyond purely mimicking synaptic-like biological behaviour, and may also be useful in psychology for implementing a model of human memory in the brain. It is believed that human memory is created by the dynamic change of neural circuits based on the synaptic connections, and it is accepted that some architecture for human memory exists in the brain, although its mechanism has not yet been fully elucidated. In 1968, Atkinson and Shiffrin proposed ‘the multistore model’ of human memory 32 , which is still the most accepted model in psychology. In this model, new information from the external environment is stored for a very short period of time in the sensory register as a sensory memory (SM), and then selected information is transferred from temporary short-term memory (STM) in the short-term store to a permanent long-term memory (LTM) in the long-term store, as illustrated in Fig. 3a . Importantly, Atkinson and Shiffrin assumed that STM can become LTM through a process of rehearsal, and that the probability of transfer to LTM increases with rehearsal repetition. It should be noted that STP and LTP are terms used in neuroscience whereas STM and LTM are terms used to describe psychological phenomena. In Fig. 3b , we present a memorization model inspired by the multistore model. The memorization level in SM mode increases slightly with initial rehearsals, and then, in STM mode, is temporarily enhanced before decaying. Subsequent frequent rehearsals are observed to result in LTM, leading to a long-term memorization. We propose that the conductance value of the inorganic synapse is analogous to the memorization level in this model of human memory, as shown in Fig. 1b,c , where the behaviour of the inorganic synapse corresponds to the simplified memorization model. Figure 3: The multistore model and the human-memory forgetting curve. a , The psychological model of human memory proposed by Atkinson and Shiffrin. This multistore model provides for three types of memory, that is, sensory memory (SM), short-term memory (STM) and long-term memory (LTM). b , Simplified memorization model in the inorganic synapse, which was inspired by the multistore model. After storing new information as SM, information is stored in STM for short periods of time, whereas repeated rehearsal events result in LTM. At a higher repetition rate, rehearsal before complete decay in memorization level forms LTM, as shown by the red line. Rehearsal at lower repetition rate cannot form LTM, as shown by the blue line. c , Typical change of memory retention in the inorganic synapse for the decay in STM mode. A power function, used to analyse psychological behaviour such as STM (ref. 35 ), y = b × t − m , was used to fit the conductance curves, where y is the memory retention, b is the fit constant for scaling, t is the time from the n th rehearsal and m is the power function rate. Memory retention was normalized using a conductance value of 77.5 μS. The conditions of the input pulse were V =80 mV, W =0.5 s and T =4 s. Full size image The decay phenomenon of conductance in STM mode also suggests a surprising similarity to ‘the forgetting curve’. Since Ebbinghaus developed the first approach to forgetting in 1885 (ref. 33 ), forgetting (or retention) curves have been derived 34 , 35 , and it is clear that the repetition rehearsal based on active recall is an appropriate method for increasing memory strength. Figure 3c shows the experimental decay curve of conductance of the inorganic synapse in STM mode. The decay rate, which corresponds to the rate of the power function ( m ), generally used in psychology, decreased and the ratio of memory retention increased with the increase in the number of rehearsals. This result correlates well to the forgetting curve in psychology and is well reproduced by the inorganic synapse. To demonstrate concrete psychological behaviour, the memorization of two images into a 7×7 inorganic synapse array was carried out, as shown in Fig. 4 and Supplementary Movie . An image of the numeral ‘2’ was stored using ten inputs with longer intervals ( T =20 s), and an image of the numeral ‘1’ was simultaneously stored using the same number of inputs, with equal amplitude and width, but with shorter intervals ( T =2 s). At first, the numeral ‘2’ (and ‘1’) emerged slightly on the first few inputs, corresponding to SM mode. Following this, the numerals ‘1’ and ‘2’ were temporarily stored in the inorganic synapse array while repeating the concurrent enhancement of conductance and spontaneous decay in conductance up to the last (tenth) input. This process corresponds to repeated rehearsal in STM mode. The numerals ‘1’ and ‘2’ appeared with higher conductance soon after the last input, which made it difficult to distinguish them from each other. However, the numeral ‘1’ was observed to persist after 20 s from the last input, owing to the forgetting of the numeral ‘2’, demonstrating that only the numeral ‘1’ was transferred to LTM mode. As the total numbers of inputs of images were the same for memorizing both images, the numerals ‘1’ and ‘2’ should have been stored at the same conductance level in a conventional switch array. However, our result indicates that a multistore model best described the observed behaviour. This experimental demonstration clearly shows implementation of a multistore model, which is a unique function of the inorganic synapse, and that it has three types of memory (SM, STM and LTM), including forgetting. The data indicate that we may apply a psychological memory model simultaneously with the emulation of biological synaptic-like behaviour. Figure 4: Image memorizing into an inorganic synapse array. a , Images of the numerals ‘1’ and ‘2’ were memorized simultaneously into a 7×7 inorganic synapse array by inputting the image ten times with intervals of T =2 and 20 s, respectively. The image of the numeral ‘2’ was stored into STM mode and that of the numeral ‘1’ into LTM mode, which caused the emergence of the numeral ‘1’ after 20 s from the last input. The numeral ‘2’ also emerged slightly by the first few inputs, which corresponds to SM mode (in this figure, the response by the first input was shown, and its scale bar is also different from the other two). The change in conductance at each pixel corresponds to the result for a single inorganic synapse, and the respective conductance profiles in the pixels are shown in b . b , Typical change in the conductance of an individual inorganic synapse, depending on input-pulse repetition time. The x axis is time and the two upper plots correspond to conductance (see the scale bar on the left). The bottom y axis shows the input pulse sequences. Full size image Here, we demonstrate that synaptic behaviours of an Ag 2 S inorganic system resemble some of the key features of a biological synapse, with clear experimental evidence of STP and LTP characteristics in a single device. A temporary increase in conductance and its spontaneous decay over time was observed using input stimuli at a lower repetition rate, and persistent enhancement was easily achieved by frequent input repetition. The results reported show that individual inorganic synapse elements may enable a new functional element suitable for the design of neural systems that can work without the need of the poorly scalable software and pre-programming currently employed in artificial neural network systems, with clear potential for hardware suited to artificially and physically intelligent systems. In addition, the wiring problem 6 , caused by the requirement for a much larger number of elements to imitate the human brain, may be solved by using inorganic synapses with previously proposed but experimentally unconfirmed architectures for hardware-implemented artificial neural networks 36 , 37 . Methods The inorganic synapse devices consisted of a nanoscale Ag 2 S-coated Ag electrode and a counter platinum electrode. Ag 2 S, which is a mixed ionic and electronic conductor 38 , 39 , was made by sulphurizing a Ag substrate in sulphur vapour at 150 °C for 70 min. A scanning tunnelling microscope, working at room temperature, was used to make the nanometre gap between the two electrodes. The initial conductance of each inorganic synapse was set to 1 μS, which corresponds to the initial conductance state. After formation of the nanogap, application of the voltage input pulse to the inorganic synapse was carried out. The conductance of the inorganic synapse was measured under an applied voltage of 10 mV using a series-connected reference resistance of 10 kΩ. The demonstration of the memorizing of two images into a 7×7 Ag 2 S inorganic synapse array was also carried out using a device structure with a nanogap formed by a scanning tunnelling microscope. The change in conductance at each pixel was obtained by the voltage pulse input–output measurement of each point on the Ag 2 S/Ag substrate. The row/column of the array was addressed by the moving of a counter platinum electrode (platinum tip) 49 times with an interval of 100 μm.

A joint research group of International Center for Materials Nanoarchitectonics, NIMS, and Department of Chemistry and Biochemistry, University of California, Los Angeles succeeded in developing a new inorganic device named "synapse device". National Institute of Materials Science (NIMS) and Japan Science and Technology Agency (JST) announced that a joint research group of International Center for Materials Nanoarchitectonics, NIMS, and Department of Chemistry and Biochemistry, University of California, Los Angeles succeeded in developing a new inorganic device named "synapse device", which automatically realizes two types of functions analogous to those of the human brain, i.e., memorizing and forgetting. Details are published online in Nature Materials. The device is made with the atomic switch which consists of an Ag2S-coated metal Ag electrode and a counter electrode of platinum Pt, having a nanometer gap between the two electrodes. The atomic switch works by the formation and annihilation of an Ag-atom bridge between the electrodes, which is realized by controlling the solid-state electrochemical reaction of a mixed ionic and electronic conductor Ag2S. The research group discovered that the device emulates two types of synaptic function, short-term plasticity and long-term potentiation by varying input pulse repetition time which controls the formation of the Ag-atom bridges. The published paper in Nature Materials remarks that the Ag2S device indicates a breakthrough in mimicking synaptic behavior essential for further creation of artificial neural systems that emulate human memories.

doi:10.1038/nmat3054

Physics

Using magnets to toggle nanolasers leads to better photonics

Nature Photonics, DOI: 10.1038/s41566-021-00922-8 Journal information: Nature Photonics

http://dx.doi.org/10.1038/s41566-021-00922-8

https://phys.org/news/2021-12-magnets-toggle-nanolasers-photonics.html

Abstract The nanoscale mode volumes of surface plasmon polaritons have enabled plasmonic lasers and condensates with ultrafast operation 1 , 2 , 3 , 4 . Most plasmonic lasers are based on noble metals, rendering the optical mode structure inert to external fields. Here we demonstrate active magnetic-field control over lasing in a periodic array of Co/Pt multilayer nanodots immersed in an IR-140 dye solution. We exploit the magnetic nature of the nanoparticles combined with mode tailoring to control the lasing action. Under circularly polarized excitation, angle-resolved photoluminescence measurements reveal a transition between the lasing action and non-lasing emission as the nanodot magnetization is reversed. Our results introduce magnetization as a means of externally controlling plasmonic nanolasers, complementary to modulation by excitation 5 , gain medium 6 , 7 or substrate 8 . Further, the results show how the effects of magnetization on light that are inherently weak can be observed in the lasing regime, inspiring studies of topological photonics 9 , 10 , 11 . Main Plasmonic nanostructures feature electromagnetic modes that confine light into sub-wavelength volumes. Consequently, the Purcell effect is strong, and when such structures are paired with emitters, enhancement of both spontaneous and stimulated emission can be achieved. Plasmonic lasers have been realized with architectures ranging from single nanoparticles and metal–insulator thin films to random, aperiodic, periodic and superperiodic arrays of nanoparticles 1 , 2 , 3 , 12 , 13 , 14 , 15 . Although the effects of the nanoparticle shape and arrangement have been extensively studied 5 , 8 , 14 , the material choice has been mostly limited to noble metals, except for some recently proposed alternatives 16 , 17 . The selection of noble metals, and in particular dielectric alternatives, is motivated by the minimization of ohmic losses typical for plasmonic materials. Lower damping leads to a higher mode quality ( Q ) factor and a stronger Purcell effect, which benefits lasing. Due to inherently high ohmic losses, nanostructures made of magnetic materials have been largely overlooked as a platform for plasmonic lasers, even when they, in principle, would offer the powerful possibility of modifying the optical modes by a magnetic field during device operation 17 . Here we demonstrate that this unique advantage can be experimentally realized. Contrary to noble metals, the permittivity tensor of magnetic metals contains non-zero off-diagonal components that depend on the direction of magnetization. Consequently, magnetic switching alters the optical response of magnetic materials, giving rise to magneto-optical phenomena such as the Faraday effect, magneto-optical Kerr effect and magnetic circular dichroism (MCD). Although these effects have been successfully exploited in the field of plasmonics for biosensing applications utilizing phase-sensitive detection 18 , the absolute modulation of light intensity through magneto-optics is weak. For instance, the polarization of light reflected from or transmitted through arrays of magnetic nanodots rotates by up to 1° 19 , which corresponds to an intensity modulation of the order 1%. Here we show that, although small, magneto-optical effects lead to subtle changes in the mode structure of the array; these, in turn, become prominently visible due to nonlinearities inherent in a lasing process, enabling full on–off switching of the lasing action by an external magnetic field. To demonstrate the magnetic-field control of plasmonic lasing, we fabricated square and rectangular arrays of Co/Pt multilayer nanodots on a Au/SiO 2 bilayer and immersed the structures in a gain medium consisting of a 12 mM IR-140 dye solution (Methods). Figure 1a shows the full Ta(2)/Pt(4)/[Co(1)/Pt(1)] 30 /Pt(2) nanodot multilayer stack, where the [Co(1)/Pt(1)] bilayer repeats 30 times and the numbers in the parentheses indicate the layer thickness in nanometres. Hereafter, we refer to these structures as Co/Pt nanodots. Co/Pt multilayers were selected as the nanodot material because of their abrupt magnetic switching in a perpendicular magnetic field and full remanence (Supplementary Fig. 1 ), enabling the non-volatile magnetic-field control of plasmonic lasing. Nanodots of diameter D = 220 nm were arranged in periodic lattices (Fig. 1a ) with periods p x = p y = 590 nm for square arrays and p y ranging from 520 to 540 nm in steps of 5 nm for rectangular arrays. Arranging metallic nanostructures into periodic lattices enables the excitation of surface lattice resonances (SLRs), which are hybridized modes of the lattice diffraction orders and localized surface plasmons (LSPs) of the nanodots 20 . The experimental reflectivity spectrum of the square array ( p x = p y = 590 nm) displays a pronounced minimum at the SLR wavelength (Fig. 1b ), in agreement with finite-element method (FEM) simulations (Supplementary Fig. 2 ). The SLR of the nanodot array overlaps with the emission spectrum of the gain medium (Supplementary Fig. 3 ), which is essential for optical feedback. Moreover, the Q factor of the SLR mode compares with that of noble-metal systems 19 , providing high rates of stimulated emission. Fig. 1: Magnetic-field control of plasmonic lasing in a square array of Co/Pt nanodots. a , Co/Pt nanodots (top) with diameter D = 220 nm, height of 68 nm and perpendicular magnetization arranged in periodic lattices ( p x = p y = 590 nm) on top of a Au/SiO 2 bilayer (bottom). b , Experimental MCD and reflectivity spectra. Both spectra show prominent features at the SLR wavelength determined by the lattice period. The MCD spectrum is obtained by subtracting the reflectivity spectra of σ + light for up and down magnetization and signal normalization. The dashed line marks the wavelength of the pump pulses in the lasing experiments. c , PL intensity spectra recorded normal to the sample plane using σ − and σ + excitation for three pump fluences. For clarity, the PL intensity spectra corresponding to 48.9 μJ cm –2 have been multiplied by a factor of 10 and each fluence pair is vertically offset by 2,000 counts. d , e , PL intensity as a function of the pump laser fluence for σ − ( d ) and σ + ( e ) excitation. In c – e , the blue and red data are recorded as a −0.5 and +0.5 T magnetic field saturates the perpendicular magnetization of the Co/Pt nanodots down and up, respectively. Source data Full size image Lasing is achieved when the square Co/Pt nanodot array is excited by linearly, left circularly ( σ − ) or right circularly ( σ + ) polarized 200 fs pulses at 800 nm, where the chirality is defined from the receiver’s point of view (Fig. 1a ). As an example, we show the photoluminescence (PL) intensity spectra for σ − and σ + pulses (Fig. 1c ). The strong nonlinear increase and narrowing of the emission peak signifies lasing above a critical pump fluence. The intensity of the lasing peak increases by about two orders of magnitude and its full-width at half-maximum (FWHM) is ~0.2 nm (Supplementary Fig. 4 ). More importantly, for circularly polarized pump pulses, the lasing threshold and lasing intensity depend on the direction of magnetization in the Co/Pt nanodots. For σ − excitation, the threshold fluence is the smallest when the magnetization points up (+0.5 T) and the largest when the magnetization points down (−0.5 T). Consequently, by selecting a pump fluence just above the lower threshold, we are able to drastically alter the lasing intensity through magnetic switching (75–90% intensity modulation at 52.7 μJ cm –2 ; Fig. 1c ). We note that the lasing threshold in our system is lower than that of other plasmonic array- and dye-based lasers, being more similar to the values reported for high-gain materials such as quantum dots 21 . The low threshold fluence could be due to an enhanced dye emission rate caused by the proximity of the molecules to the laser-irradiated nanodots and Au film 22 . Figure 1d summarizes the variation in PL intensity with pump fluence and the direction of magnetization for σ − pulses. The difference in threshold fluence for up and down magnetization is ~0.8 µJ cm –2 (4%). The magnetization state also affects the lasing intensity in saturation. Similar results are obtained under σ + excitation, but with reverse dependence on the direction of perpendicular magnetization (Fig. 1e ). Additionally, switching the magnetization state of the Co/Pt nanodots does not change the lasing intensity when using linearly polarized pulses (Supplementary Fig. 5 ). Next, we discuss angle-resolved PL measurements to identify the lasing mode. We consider the data recorded close to and above the lasing threshold (Fig. 2a ). The above-threshold data (Fig. 2b ) demonstrates highly directional lasing normal to the sample plane at 889.3 nm. The angle-resolved emission spectrum obtained near the lasing threshold (Fig. 2c ) reveals band dispersion of the plasmonic system. Calculations based on the empty-lattice approximation (Fig. 2d ) reproduce the main modes: SLRs supported by the square Co/Pt nanodot array and a surface plasmon polariton (SPP) excited at the Au/SiO 2 interface via grating coupling. For normal-incident light with linear polarization along the x axis, the modes with linear dispersion correspond to SLRs excited through the (0, ±1) diffracted orders, whereas the red parabolic band belongs to SLRs excited through the (±1, 0) diffracted orders (Methods). The broad parabolic band shown in Fig. 2c and the blue curve shown in Fig. 2d depict the SPP mode. From the analysis shown in Fig. 2 , we conclude that stimulated emission to modes that are located close to the diffracted orders at the Γ point (lattice momenta k x = k y = 0) produces lasing in our system. A more accurate determination of the lasing modes including their chirality is given below, after we consider the effect of magnetic field on optical absorption. Fig. 2: Analysis of the lasing mode for a square Co/Pt nanodot array. a , PL intensity spectra recorded normal to the sample plane for a pump fluence close to and above the lasing threshold. b , c , Angle-resolved emission spectra recorded at 55.0 μJ cm –2 ( b ) and 49.6 μJ cm –2 ( c ). The color bars are identical and use a natural logarithmic scale. d , Calculated dispersion bands for a sample with a square nanodot array. All the data are obtained for normal-incident light with linear polarization along the x axis. The angle corresponds to lattice momentum along the y direction. Source data Full size image Circularly polarized light excites the free electrons of a metal nanodot into spectrally degenerate rotational motion. When a magnetic field is applied or the nanodot exhibits a net magnetization, the Lorentz force on the electrons lifts the degeneracy 23 . Consequently, the optical response of a magnetic nanodot resides at slightly different frequencies for σ − and σ + excitation and the two frequencies interchange when an external magnetic field switches the magnetization. A similar MCD effect also appears in the SLR modes of a magnetic nanodot lattice 24 , as illustrated by the square array in Fig. 1b . The change in optical absorption due to the MCD effect could increase the excitation of dye molecules in the vicinity of the nanodot lattice. This effective increase in pumping would shift the lasing threshold curve, but only by the amount of increased absorption, that is, linearly. The MCD effect at the pump wavelength is below 1%, whereas the threshold fluence of the square array changes by about 4% (Fig. 1d,e ). Besides, a mere shift in the threshold curve does not produce a higher saturated PL intensity, in contradiction to our observation. Nonlinear changes in the saturated PL intensity and lasing threshold can, instead, be caused by changes in the lifetime of the lasing mode, as well as in the so-called β factor that quantifies which portion of the total emission of the gain medium goes to the lasing mode. To estimate such nonlinear effects, we now comprehensively analyse the properties of the lasing modes, as well as other modes overlapping with the emission spectrum of the molecules. The lasing mode identified in Fig. 2 turns out to be an out-of-plane SLR mode related to the (±1, 0) and (0, ±1) diffracted orders (Supplementary Fig. 6 ). In a square lattice, this mode is doubly degenerate due to x – y symmetry. FEM simulations of nanodot arrays with perpendicular magnetization show that the degeneracy is lifted, producing two modes of σ − and σ + character at wavelengths of 896.5 and 895.8 nm, respectively. The ~5 nm difference to the experimental lasing peaks is understandable as the refractive index of the dye solution may slightly change on pumping (note that n = 1.48 used in the simulations produces an excellent match with the experimental reflectivity spectra taken without pumping (Supplementary Fig. 2 )). The two modes and their field profile are shown in Fig. 3a,c (Supplementary Fig. 7 provides more details). In nanoparticle array lasers, the pump often not only excites the gain medium but also slightly excites the plasmonic modes, and this small excitation then stimulates emission, causing lasing polarization to be dependent on pump polarization 3 . It can, thus, be expected that σ − and σ + pump pulses trigger lasing in the corresponding modes. Indeed, this is observed in the experiments shown in Fig. 1c and Supplementary Fig. 8 : there are two lasing peaks, and the peaks interchange when either the magnetization direction or pump helicity switches, as expected for modes of opposite chirality. We find good agreement with the magnitude of splitting between the two modes given by simulations, namely, 0.7 nm, and the experimentally observed distance between the lasing peaks, that is, 0.46 ± 0.25 nm (averaged over six measurements). Further evidence of the chiral doublet is given by the fact that consistent with circularly polarized emission, we observe both x - and y -polarized components in the lasing emission (Supplementary Fig. 9 ). The observation of two chiral lasing modes is a remarkable result, since such splitting of degenerate modes by time-reversal symmetry breaking (magnetization direction) is needed in creating topologically non-trivial systems 9 . Here we show that the effects of magnetic time-reversal symmetry breaking in plasmonics, which are small in general due to the weakness of magneto-optical effects, become prominent through suitable mode design and visible in the lasing regime. Fig. 3: Chiral modes emerging in a square lattice of Co/Pt nanodots. a , FEM electric-field profile of the out-of-plane SLR lasing mode. In the presence of perpendicular magnetization, the degenerate SLR doublet splits into σ + and σ − modes (as shown in c ). b , FEM field profile of the hybrid mode (in-plane SLR with strong LSP contribution and hybridization to SPPs). The degeneracy of this mode is also lifted by perpendicular magnetization. The modes in a and b are σ + polarized, but the σ − modes look similar. c , Illustration of overlap between the emission spectrum of the dye solution (dashed line) and the lasing and hybrid modes. The hybrid mode at a lower wavelength has a bigger overlap with the emission spectrum of the dye, and consequently, the gain available for the lasing mode of the same chirality is lower. The schematic is not to scale, but it depicts the mode ordering as obtained from the simulations for magnetization pointing up. Full size image According to the FEM simulations, the σ − and σ + lasing modes have lifetimes that are too similar to explain the notable modulation of the lasing threshold. One should bear in mind, however, that the pump may provide a small seed excitation not only to the lasing mode but also to other modes. Stimulated emission to those modes can then deplete the gain available for the lasing mode. Using FEM simulations, we have identified a broad hybrid mode at higher wavelengths corresponding to an in-plane SLR with a strong LSP component. The SPPs of the Au film underneath are involved, too, as the mode does not appear at this wavelength when Au is absent (in that case, the in-plane SLR is close to the diffracted orders). Figure 3b shows how the field of the hybrid mode is strongly enhanced close to the nanodots. Just like the lasing mode, this mode is doubly degenerate in the non-magnetized case, and it splits into σ − and σ + modes at wavelengths of 1270 and 1257 nm, respectively, in the presence of perpendicular magnetization (Supplementary Figs. 6 and 7 ). Although these broad modes are spectrally far from the lasing wavelength, they still overlap with the emission spectrum of the molecules. Consequently, the pump can excite these modes in a chirality-dependent manner, which triggers stimulated emission either to the σ − or σ + mode. This emission reduces the gain available for the lasing mode, which can be described as an effective, chirality-dependent reduction in the β factor of the lasing mode. Based on the wavelengths and linewidths of the modes obtained from FEM simulations and the observed emission spectrum (Supplementary Fig. 7 ), we estimate a 1–2% change in the β factor of the lasing mode, which, in turn, changes the threshold and PL intensity in a nonlinear manner. For realistic parameters, 3–4% changes in the lasing threshold are simulated for this gain competition effect (Supplementary Figs. 10 and 11 ), which is comparable to our experimental observations. Direct experimental support for this scenario is given in Fig. 1c , always showing a lower threshold and higher PL for the lasing peak at higher wavelengths. This is consistent with the spectral ordering of the simulated modes (Fig. 3c ): the hybrid mode that is farther away from the lasing modes, and thus reduces the gain less, has the same chirality as the higher-wavelength lasing mode, explaining its lower threshold and higher PL. We note that σ − / σ + helicity is defined in the same way in both experiments and simulations, as confirmed by the similar behaviour of the simulated MCD (Supplementary Fig. 12 ) and measured data (Fig. 1b ). The proposed switching mechanism relies on stimulated emission, as spontaneous emission does not prefer the chirality of one mode above the other. To further confirm the mechanism and to exploit it in designing even stronger magnetic-field effects, one could maximize the amount of pump energy that is directly coupled to the plasmonic modes. This can be achieved by fabricating rectangular arrays in which one of the periods is selected to generate an SLR mode at the excitation wavelength (800 nm) and the other—similar to the square array—produces an SLR within the emission band of the dye molecules (~890 nm). We accomplished this by varying p y from 520 to 540 nm in steps of 5 nm as p x was fixed at 590 nm. Figure 4a depicts the MCD and reflectivity spectra of the rectangular array with p y = 530 nm. Compared with the square array (Fig. 1b ), the rectangular array absorbs about 60% more light at the pump wavelength (as shown in the reflectivity data in Fig. 4a ). As demonstrated by Fig. 4b , the lasing threshold approximately halves. Further, because the difference in the absorption of σ − and σ + light is larger in the rectangular array (the MCD effect is enhanced by about 50% compared with the square array), the chirality-dependent coupling to the broad hybrid modes changes accordingly. The lattice-designed increase in MCD enhances the change in lasing threshold when the magnetization is switched. Moreover, the hybrid modes in the rectangular array are pulled closer to the lasing mode according to our simulations (1266 nm for σ − light and 1253 nm for σ + light for p x = 590 nm and p y = 530 nm; Supplementary Figs. 7 and 13 ), which further increases the chirality-dependent gain depletion effect. Indeed, the experimentally observed changes in the lasing action are larger than the square array (Fig. 4b ). We estimate a 2.5–4.0% change in the effective β factor, which leads to a 6.0–26.0% change in the threshold for typical β factors (Supplementary Fig. 14 ). Data for other rectangular arrays demonstrating a direct correlation between the optical absorption of σ − and σ + light at the excitation wavelength and magnitude of the magnetic switching effect are shown in Supplementary Fig. 15 . In the rectangular lattice, the energy difference in the SLR modes related to the (±1, 0) and (0, ±1) diffracted orders breaks the degeneracy of the lasing mode. Consequently, a chiral doublet does not form in the presence of magnetization. FEM simulations indeed show that the lasing mode is y -polarized (Supplementary Fig. 13 ). This is consistent with the y -polarized lasing emission observed in our experiments (Supplementary Fig. 9 ), in clear contrast to the case of the square lattice. Lasing results for all the rectangular arrays are summarized in Supplementary Figs. 16 – 20 . Fig. 4: Magnetic-field control of plasmonic lasing in rectangular arrays of Co/Pt nanodots. a , MCD and reflectivity spectra of a rectangular array with p x = 590 nm and p y = 530 nm for σ + excitation. b , PL intensity as a function of the pump laser fluence for an array with p y = 530 nm and σ + excitation. The results recorded with the magnetization of the Co/Pt nanodots pointing up and down are shown in red and blue, respectively. c , PL intensity recorded at a σ + pump fluence of 25.1 μJ cm –2 with repeated switching of the magnetization of Co/Pt nanodots between down (blue) and up (red). Source data Full size image Tailoring of the SLR modes at the excitation wavelength and at a wavelength that overlaps with the emission spectrum of the gain medium provides full on–off switching of the lasing signal by an external magnetic field, as demonstrated in Fig. 4c . Here the modulation of PL intensity exceeds two orders of magnitude at a pump fluence of 25.1 μJ cm –2 . Although our proof-of-concept experiments are conducted by placing permanent magnets near the plasmonic sample, magnetic switching is inherently fast—a feature that is critical for non-volatile magnetic data storage technology 25 . Moreover, ultrafast all-optical helicity-dependent magnetic switching has been demonstrated for ferromagnetic metals, including Co/Pt 26 . The modulation of lasing intensity through magnetic switching as demonstrated here, therefore, provides a feasible pathway towards actively controlled coherent light sources for enhanced light–matter interactions on the nanoscale. Our results also hold promise for topological photonics 11 , 27 . Photonic analogues of topological insulators can, in theory, be realized with time-reversal symmetry breaking provided by material magnetization 9 . However, the effect is weak at visible frequencies and external magnetic fields are needed for topological lasing 10 . We have discovered that the interplay of magnetized plasmonic nanodots with the symmetry of the array can lead to a splitting of the chiral modes by about 0.5 nm or more, which is remarkable compared with the 42 pm topological bandgap 10 . The splits are below the natural linewidth of the modes, but they become visible in the lasing regime as shown here. Our results suggest magnetic nanodot arrays as an exciting platform for studies of topological photonics; with permanently magnetized nanodots, a topological system could be realized even without an external magnetic field. Methods Sample fabrication The plasmonic structures were fabricated on Si substrates with a native SiO 2 layer. First, a 2 nm Ti/150 nm Au film was grown by electron-beam evaporation. The metal film was then covered by 20 nm SiO 2 using atomic layer deposition. On top of the SiO 2 layer, Co/Pt nanodot arrays were patterned by electron-beam lithography using a Vistec EBPG5000pES system. After defining holes in a polymethyl methacrylate resist layer, a 2 nm Ta/4 nm Pt/[1 nm Co/1 nm Pt] 30 /2 nm Pt multilayer stack was grown by magnetron sputtering. The resist layer was lifted off using acetone. The array size was chosen to be 0.5 mm × 0.5 mm as this is approximately the minimum size required to measure their magneto-optical response in our home-built magneto-optical setup. Nonetheless, lasing from smaller arrays is expected 28 . Optical and magneto-optical characterization The optical and MCD response of the plasmonic structures was characterized in reflection using a supercontinuum laser (NKT SuperK EXW-12) and a photodetector (Hinds DET-200-002). The samples were placed between the poles of an electromagnet (GMW 3470) for the application of perpendicular magnetic fields. All the optical measurements were conducted near normal incidence through a hole in one of the pole pieces. The reflectivity spectra were recorded in zero magnetic field using linearly polarized light. A quarter-wave plate was used to circularly polarize ( σ − or σ + ) the incident laser beam in the MCD measurements. The MCD spectra were obtained by normalizing the difference in optical reflectivity for up and down magnetization to the reflectivity measured in zero external field. The magnetization of the Co/Pt nanodots was saturated up or down by a ±0.5 T magnetic field. The Co/Pt nanodots were immersed in an IR-140 dye solution using a cover glass to ensure an identical dielectric environment as in the lasing experiments. Angle-resolved PL measurements Angle-resolved PL measurements were performed using a 12 mM IR-140 dye solution. The IR-140 dye (Sigma-Aldrich) was dissolved in a 1:2 solution of dimethyl sulfoxide and benzyl alcohol to match the refractive index of the cover glass. The plasmonic structures were excited at 800 nm and normal incidence by 200 fs pulses from a Ti:sapphire laser (Coherent). The repetition rate was 1 kHz. We focused the laser beam to a diameter of 750 μm to fully irradiate the 500 × 500 μm 2 nanodot arrays. The PL of the arrays was collected using a CFI Plan Fluor 10X objective (Nikon) with a numerical aperture of 0.3. The back focal plane of the objective was imaged onto the entrance slit of a spectrometer (SP2500, Princeton Instruments) and projected onto a two-dimensional charge-coupled device detector (PIXIS: 400, Princeton Instruments), providing wavelength and angle-resolved images (Fig. 2b,c ). From these images, we extracted the single-pixel maximum intensity counts within the spectral range where lasing was observed to study the dependence of PL intensity on pump fluence. Three consecutive images were recorded and the extracted maximum PL intensity was averaged. The spectral resolution of the charge-coupled device camera was 0.14 nm. Lasing experiments were conducted for linearly polarized, σ − -polarized and σ + -polarized laser pulses. The magnetization of the Co/Pt nanodots was switched by rotating a permanent magnet behind the sample. The field at the sample location was ±0.5 T. Empty-lattice model We used an analytical model based on the empty-lattice approximation 29 , 30 to identify the modes present in the angle-resolved emission spectrum (Fig. 2c ). In this approach, the plasmon modes in the nanodot lattice are defined as 30 , 31 $$| {{{{\bf{k}}}}}_{| | }+{{{\bf{G}}}}| =\sqrt{\epsilon }\frac{\omega }{c},$$ (1) where k ∣ ∣ is the in-plane wave vector, \({{{\bf{G}}}}={n}_{x}{G}_{x}{{{{\hat{\mathbf{x}}}}}}+{n}_{y}{G}_{y}{{{{\hat{\mathbf{y}}}}}}\) , G x , y = 2π/ p x , y are the reciprocal vectors of the lattice, n x , y are integers (0, ±1,...) corresponding to the diffracted orders, ϵ is the dielectric constant of the surrounding environment, c is the speed of light and ω is the angular frequency. In our experiments, the entrance slit of the spectrometer was parallel to the y axis, k ∣ ∣ = k y and k x = 0. Hence, $$| {{{{\bf{k}}}}}_{y}+{{{\bf{G}}}}| =\sqrt{\epsilon }\frac{\omega }{c}.$$ (2) By definition, the Γ point is the frequency at which the diffracted orders n x , y = ±1 intersect for k ∣ ∣ = 0 in equation ( 1 ). Therefore, a fourfold degeneracy arises at that frequency. According to equation ( 2 ), the frequencies of the SLRs corresponding to the diffracted orders (0, ±1) linearly depend on the in-plane wave vector, resulting in two linear dispersion bands (Fig. 2d , black curves). Conversely, the SLRs corresponding to the diffracted orders (±1, 0) yield two degenerate parabolic dispersion bands (Fig. 2d , red curve). For the evanescent SPP modes propagating at the 150 nm Au/20 nm SiO 2 interface, the empty-lattice model yields the following relation 29 : $$| {{{{\bf{k}}}}}_{| | }+{{{\bf{G}}}}| ={k}_{\mathrm{SPP}}=\frac{\omega }{c}\sqrt{\frac{{\epsilon }_{\mathrm{m}}{\epsilon }_{\mathrm{d}}}{{\epsilon }_{\mathrm{m}}+{\epsilon }_{\mathrm{d}}},}$$ (3) where k SPP is the momentum of the SPP mode and ϵ m and ϵ d are the dielectric constants of Au and SiO 2 , respectively. Equation ( 3 ) also results in a parabolic variation in frequency with the in-plane wave vector. Consequently, the SPPs exhibit parabolic dispersion bands (Fig. 2d , blue curve) that are slightly redshifted compared with the parabolic SLR bands. Because of the transversal magnetic nature of the SPP modes, the counterpart of the linear SLR dispersion bands is not present in this case. A refractive index of n = 1.51 was used here to make the empty-lattice model match with the experiments shown in Fig. 2 ; the actual SLR modes are typically slightly away from the empty-lattice approximation. Rate equation analysis To model the interplay of the emission of the gain medium with the plasmonic modes supported by the magnetic nanodot lattice, we implement a standard rate equation approach, as shown in ref. 32 . We describe the molecules as optically pumped four-level quantum emitters with time-dependent populations ranging from N 0 to N 3 in increasing order of energy, as given by $$\frac{\mathrm{d}{N}_{3}}{\mathrm{d}t}=r(t)({N}_{0}-{N}_{3})-\frac{{N}_{3}}{{\tau }_{32}},$$ (4) $$\frac{\mathrm{d}{N}_{2}}{\mathrm{d}t}=-\beta {n}_{\mathrm{ph}}\frac{\left({N}_{2}-{N}_{1}\right)}{{\tau }_{21}}-\frac{{N}_{2}}{{\tau }_{21}}-\frac{{N}_{2}}{{\tau }_{20}}+\frac{{N}_{3}}{{\tau }_{32}},$$ (5) $$\frac{\mathrm{d}{N}_{1}}{\mathrm{d}t}=\beta {n}_{\mathrm{ph}}\frac{\left({N}_{2}-{N}_{1}\right)}{{\tau }_{21}}+\frac{{N}_{2}}{{\tau }_{21}}-\frac{{N}_{1}}{{\tau }_{10}},$$ (6) $$\frac{\mathrm{d}{N}_{0}}{\mathrm{d}t}=-r(t)({N}_{0}-{N}_{3})+\frac{{N}_{2}}{{\tau }_{20}}+\frac{{N}_{1}}{{\tau }_{10}},$$ (7) where r ( t ) is the time-dependent pumping rate corresponding to pulsed pump excitation. Also, r ( t ) = r f ( t ) and \(f(t)=\exp (-{(t-{t}_{0})}^{2}/{\tau }_{\mathrm{p}}^{2})\) , such that t 0 ≫ τ p and τ p = 200 fs (same as the pump-pulse length in the experiment). Lifetimes τ 32 = 500 fs, τ 20 = 240 ps and τ 10 = 4 ps account for non-radiative transitions, and τ 21 = 720 ps is the lifetime of the lasing transition. More details and definitions are provided elsewhere 32 . The number of photons n ph generated by both spontaneous and stimulated emission is determined by the expression $$\frac{\mathrm{d}{n}_{\mathrm{ph}}}{\mathrm{d}t}=\beta {n}_{\mathrm{ph}}\frac{\left({N}_{2}-{N}_{1}\right)}{{\tau }_{21}}+\beta \frac{{N}_{2}}{{\tau }_{21}}-\frac{{n}_{\mathrm{ph}}}{{\tau }_{\mathrm{cav}}},$$ (8) where τ cav is the cavity lifetime and the β factor accounts for the fraction of emitted photons that decay into the lasing mode. We simulated the PL intensity and threshold curve by solving the time evolution of the coupled equations (equations ( 4 )–( 8 )) using different values of pump intensity r and integrating n ph over a long time. The lifetimes of the lasing modes ( τ cav ) in our magnetic system were calculated from numerical FEM simulations of the electromagnetic fields. Further, we used the overlap of the molecule emission spectrum with lasing and hybrid plasmonic modes (Fig. 3c ) as an input for estimating the β factor. More details on these calculations are given in the Supplementary Information . Estimation of β factors We consider gain competition between the lasing and hybrid modes, and indicate the decay in emission to these modes by γ L (lasing mode) and γ −,hybrid and γ +,hybrid (hybrid modes); γ i denotes any of these, where i denotes the mode. We estimate γ i as $${\gamma }_{i}=\int\nolimits_{0}^{{E}_{\mathrm{cutoff}}}p(E){\alpha }^{2}{\sigma }_{i}(E)\mathrm{d}E,$$ (9) where E cutoff is the energy below which the lasing and hybrid modes are located (Supplementary Fig. 7 ), p ( E ) is the normalized emission spectrum, α 2 is projection of the mode to the σ − / σ + basis ( α 2 = 0.5 for the linearly polarized lasing mode of the rectangular case and α 2 = 1.0 for the other modes) and σ i ( E ) is the absorption probability of mode i . The emission spectrum is approximated with a Lorentzian fitted to the measured emission spectrum (Supplementary Figs. 3 and 7 ). The absorption spectra σ i ( E ) are estimated by Lorentzians with the parameters (peak energy and linewidth) obtained from FEM simulations (Supplementary Fig. 7 ). The β factor is defined as the proportional decay rate into the selected mode 33 , where the total decay ( γ tot ) includes other decay channels ( γ other ): $${\beta }_{i}=\frac{{\gamma }_{i}}{{\gamma }_{\mathrm{tot}}}=\frac{{\gamma }_{i}}{{\gamma }_{i}+{\gamma }_{\mathrm{other}}}.$$ (10) If γ i , γ j ≪ γ other , then β i / β j ≃ γ i / γ j . This is the case here as the β factors of the lasing and hybrid modes are very small (~10 −3 ). We assume that the lasing and hybrid modes compete about the same gain, and their β factors are of the same order of magnitude. Then, if the β factor of the hybrid mode changes, there is a corresponding change in the β factor of the lasing mode. That is, decreased/increased amount of emission to the hybrid mode is given to/taken away from the lasing mode. We estimate this change Δ in \(\frac{{\beta }_{-}}{{\beta }_{+}}=1+{{\varDelta }}\) as $${{\varDelta }}=\frac{{\gamma }_{-,\mathrm{hybrid}}}{{\gamma }_{\mathrm{L}}}\left(1-\frac{{\gamma }_{-,\mathrm{hybrid}}}{{\gamma }_{+,\mathrm{hybrid}}}\right).$$ (11) This corresponds to a change in the β factor of the hybrid mode, scaled by how important the decay in the hybrid mode is compared with the lasing mode ( \(\frac{{\gamma }_{-,\mathrm{hybrid}}}{{\gamma }_{\mathrm{L}}}\) , which is of the order one). FEM analysis To theoretically calculate the effect of nanodot magnetization, particularly to provide the wavelength and lifetime of the plasmonic modes in the system for the rate equations, we carried out numerical simulations of the electromagnetic fields using FEM. We modelled a three-dimensional realistic structure consisting of an infinite lattice of Co/Pt nanodots combined with the Au layer (Fig. 1a ) by simulating one unit cell and imposing periodic boundary conditions in the x and y directions. We used the geometrical parameters from the experiment and described the local optical response of the Co/Pt material through a non-diagonal permittivity tensor 34 : $$\epsilon =\left(\begin{array}{lll}{\epsilon }_{xx}&{\epsilon }_{xy}&0\\ {\epsilon }_{yx}&{\epsilon }_{yy}&0\\ 0&0&{\epsilon }_{zz}\end{array}\right),$$ (12) where ϵ x x = ϵ y y = ϵ z z = ϵ d ; ϵ x y = − ϵ y x = i ϵ 0 ; and ϵ d and ϵ o are complex-valued parameters coming from previously measured data 35 . Further, we model the metal response of the Au film using a complex frequency-dependent isotropic permittivity available in the experimental tabulated data 36 and assume that the spacer that separates the metallic substrate from the nanodots as well as the background medium have a refractive index of n = 1.48. The FEM analysis was performed by combining the eigenmode studies and frequency-domain scattering simulations. Eigenmode simulations allow us to identify the modes supported by the system without any external excitation, including dark modes, and provide their near fields together with a complex eigenfrequency: f = f ′ + i f ″. The energy, linewidth and lifetime of the mode are then given by E = hf ′, Γ = h f ″ and τ = 1/(2π f ″), respectively, where h is the Planck’s constant. The results of these simulations are shown in Fig. 3a,b and Supplementary Figs. 6 and 13 . In Fig. 3a,b , we show the results for one unit cell at planes intersecting the nanodot at its mid-height (Fig. 3a,b , left) and across its diameter (Fig. 3a,b , right). In addition, we calculated the optical reflectivity and MCD spectra of the system using frequency-domain scattering simulations with normal-incident plane-wave illumination. The simulated optical reflectivity and MCD spectra are shown in Supplementary Figs. 2 and 12 , respectively. Identifying chirality of eigenmodes Finite-element eigenmode simulations provide the energy, lifetime and electromagnetic fields of all the magnetized and non-magnetized modes in the system. However, identifying the polarization of the resulting modes is not straightforward as these simulations do not consider explicit coupling to external illumination. We develop a theoretical method to unveil the chirality of the in-plane and out-of-plane modes involved in our analysis. Considering that the magnetized modes emerge from a doublet of x - and y -polarized modes in the square array (Supplementary Fig. 6 b–e), we define our magnetized states m 1 and m2 according to the field distributions of the linearly polarized modes: $$\left|{m}_{1}\right\rangle ={C}_{x,{m}_{1}}\left|\leftrightarrow \right\rangle +{C}_{y,{m}_{1}}\left|\updownarrow \right\rangle,$$ (13) $$\left|{m}_{2}\right\rangle ={C}_{x,{m}_{2}}\left|\leftrightarrow \right\rangle +{C}_{y,{m}_{2}}\left|\updownarrow \right\rangle,$$ (14) with $${C}_{x,{m}_{1,2}}=\langle \leftrightarrow | {m}_{1,2}\rangle,$$ (15) $${C}_{y,{m}_{1,2}}=\langle \updownarrow | {m}_{1,2}\rangle ,$$ (16) where the scalar product is defined as the overlap integral $$\langle i| j\rangle ={{\int}_{{{{\rm{unit}} \, {\rm{cell}}}}} {\mathrm{d}}^{3}{{{\bf{r}}}}\,{{{{\bf{u}}}}}_{i}^{* }({{{\bf{r}}}})\cdot {{{{\bf{u}}}}}_{j}({{{\bf{r}}}}}),$$ (17) with the normalization $${{{{\bf{u}}}}}_{i}({{{\bf{r}}}})=\frac{{{{{\bf{E}}}}}_{i}({{{\bf{r}}}})}{\sqrt{{\int}_{{\rm{unit}} \, {\rm{cell}}}{\mathrm{d}}^{3}{{{\bf{r}}}}| {{{{\bf{E}}}}}_{i}({{{\bf{r}}}}){| }^{2}}}.$$ (18) Here E i is the electric field of the mode i . Moreover, we can project the x - and y -polarized basis onto the σ + - and σ − -polarized basis, obtaining $$\left|{m}_{1}\right\rangle ={C}_{{\sigma }^{-},{m}_{1}}\left|{\sigma }^{-}\right\rangle +{C}_{{\sigma }^{+},{m}_{1}}\left|{\sigma }^{+}\right\rangle,$$ (19) $$\left|{m}_{2}\right\rangle ={C}_{{\sigma }^{-},{m}_{2}}\left|{\sigma }^{-}\right\rangle +{C}_{{\sigma }^{+},{m}_{2}}\left|{\sigma }^{+}\right\rangle,$$ (20) with $${C}_{{\sigma }^{-},{m}_{1,2}}=\frac{1}{\sqrt{2}}({C}_{x,{m}_{1,2}}-\mathrm{i}{C}_{y,{m}_{1,2}}),$$ (21) $${C}_{{\sigma }^{+},{m}_{1,2}}=\frac{1}{\sqrt{2}}({C}_{x,{m}_{1,2}}+\mathrm{i}{C}_{y,{m}_{1,2}}).$$ (22) These amplitudes, and their moduli, reveal whether the mode is σ + or σ − polarized. Data availability Source data are provided with this paper. All other data from this work are available from the corresponding authors upon reasonable request.

A magnetic field can be used to switch nanolasers on and off, shows new research from Aalto University. The physics underlying this discovery paves the way for the development of optical signals that cannot be disturbed by external disruptions, leading to unprecedented robustness in signal processing. Lasers concentrate light into extremely bright beams that are useful in a variety of domains, such as broadband communication and medical diagnostics devices. About ten years ago, extremely small and fast lasers known as plasmonic nanolasers were developed. These nanolasers are potentially more power-efficient than traditional lasers, and they have been of great advantage in many fields—for example, nanolasers have increased the sensitivity of biosensors used in medical diagnostics. So far, switching nanolasers on and off has required manipulating them directly, either mechanically or with the use of heat or light. Now, researchers have found a way to remotely control nanolasers. "The novelty here is that we are able to control the lasing signal with an external magnetic field. By changing the magnetic field around our magnetic nanostructures, we can turn the lasing on and off," says Professor Sebastiaan van Dijken of Aalto University. The team accomplished this by making plasmonic nanolasers from different materials than normal. Instead of the usual noble metals, such as gold or silver, they used magnetic cobalt-platinum nanodots patterned on a continuous layer of gold and insulating silicon dioxide. Their analysis showed that both the material and the arrangement of the nanodots in periodic arrays were required for the effect. Photonics advances towards extremely robust signal processing The new control mechanism may prove useful in a range of devices that make use of optical signals, but its implications for the emerging field of topological photonics are even more exciting. Topological photonics aims to produce light signals that are not disturbed by external disruptions. This would have applications in many domains by providing very robust signal processing. "The idea is that you can create specific optical modes that are topological, that have certain characteristics which allow them to be transported and protected against any disturbance," explains van Dijken. "That means if there are defects in the device or because the material is rough, the light can just pass them by without being disturbed, because it is topologically protected." So far, creating topologically protected optical signals using magnetic materials has required strong magnetic fields. The new research shows that the effect of magnetism in this context can be unexpectedly amplified using a nanoparticle array of a particular symmetry. The researchers believe their findings could point the way to new, nanoscale, topologically protected signals. "Normally, magnetic materials can cause a very minor change in the absorption and polarization of light. In these experiments, we produced very significant changes in the optical response—up to 20 percent. This has never been seen before," says van Dijken. Academy Professor Päivi Törmä adds that 'these results hold great potential for the realization of topological photonic structures wherein magnetization effects are amplified by a suitable choice of the nanoparticle array geometry." The results are published in Nature Photonics. These findings are the result of a long-lasting collaboration between the Nanomagnetism and Spintronics group led by Professor van Dijken and the Quantum Dynamics group led by Professor Törmä, both in the Department of Applied Physics at Aalto University.

10.1038/s41566-021-00922-8

Medicine

A Braf kinase-inactive mutant induces lung adenocarcinoma

Patricia Nieto et al, A Braf kinase-inactive mutant induces lung adenocarcinoma, Nature (2017). DOI: 10.1038/nature23297 Journal information: Nature

http://dx.doi.org/10.1038/nature23297

https://medicalxpress.com/news/2017-08-braf-kinase-inactive-mutant-lung-adenocarcinoma.html

Abstract The initiating oncogenic event in almost half of human lung adenocarcinomas is still unknown, a fact that complicates the development of selective targeted therapies. Yet these tumours harbour a number of alterations without obvious oncogenic function including BRAF-inactivating mutations. Inactivating BRAF mutants in lung predominate over the activating V600E mutant that is frequently observed in other tumour types 1 . Here we demonstrate that the expression of an endogenous Braf(D631A) kinase-inactive isoform in mice (corresponding to the human BRAF(D594A) mutation) triggers lung adenocarcinoma in vivo , indicating that BRAF-inactivating mutations are initiating events in lung oncogenesis. Moreover, inactivating BRAF mutations have also been identified in a subset of KRAS-driven human lung tumours. Co-expression of Kras(G12V) and Braf(D631A) in mouse lung cells markedly enhances tumour initiation, a phenomenon mediated by Craf kinase activity 2 , 3 , and effectively accelerates tumour progression when activated in advanced lung adenocarcinomas. We also report a key role for the wild-type Braf kinase in sustaining Kras(G12V)/Braf(D631A)-driven tumours. Ablation of the wild-type Braf allele prevents the development of lung adenocarcinoma by inducing a further increase in MAPK signalling that results in oncogenic toxicity; this effect can be abolished by pharmacological inhibition of Mek to restore tumour growth. However, the loss of wild-type Braf also induces transdifferentiation of club cells, which leads to the rapid development of lethal intrabronchiolar lesions. These observations indicate that the signal intensity of the MAPK pathway is a critical determinant not only in tumour development, but also in dictating the nature of the cancer-initiating cell and ultimately the resulting tumour phenotype. Main The RAS–MAPK signalling cascade serves as a central node in transducing signals from membrane receptors to the nucleus. This pathway is aberrantly activated in a substantial fraction of human cancers 4 . Moreover, germline mutations resulting in limited activation of this signalling cascade cause developmental disorders known as RASopathies 5 . There is also abundant evidence that elevated RAS–MAPK signalling results in cellular toxicity that may serve as a natural barrier to cancer progression early in tumorigenesis 6 . Finally, genetic abrogation of this pathway in adult mice results in their rapid death 7 . These findings suggest that defined thresholds of RAS–MAPK activity are required for homeostasis as well as for malignant transformation, but compelling genetic evidence is missing. In order to augment MAPK signalling in controlled increments we have taken advantage of the expression of an endogenous Braf(D631A) kinase-dead isoform (corresponding to the human BRAF(D594A) mutant) that is known to induce Erk phosphorylation in a Craf-dependent manner 2 , 8 . This effect, known as the MAPK paradox, is due to enhanced heterodimerization and activation of the catalytically competent Craf protomer in Braf(D631A)–Craf complexes 2 , 3 . In agreement with these observations, lack of wild-type Braf expression in Kras G12V cell lines expressing Braf(D631A) increased the intensity and duration of MAPK signalling ( Extended Data Fig. 1 ), probably as a result of the exclusive formation of Braf(D631A)–Craf heterodimers. Thus, to generate controlled thresholds of MAPK intensity in vivo , we combined wild-type Braf , conditional knockout Braf lox and conditional knock-in Braf LSLD631A with an inducible Kras LSLG12Vgeo allele 9 (where LSL indicates a lox-STOP-lox motif). The resulting Kras +/ LSLG12Vgeo (hereto designated as K), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (designated as KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (designated as KBL) strains were intratracheally infected with adenovirus expressing Cre recombinase (Ad-Cre). Cre-mediated recombination of these alleles results in the induction of distinct levels of Ras–MAPK signalling, with Braf +/+ driving lower activity, Braf +/ D631A intermediate intensity and Braf − /D631A maximal activation. This strategy allowed us to investigate the effect of various MAPK activity thresholds on cell transformation, adenocarcinoma development and cellular toxicity in vivo . Lung cells expressing these conditional alleles were identified by co-expression of the Kras(G12V) oncoprotein with β-geo, a chimaeric bacterial protein with β-galactosidase activity 9 ( Extended Data Fig. 2 ). As illustrated in Fig. 1a , small X-gal + hyperplasias could be readily detected in the lungs of K mice 1 month after Ad-Cre infection. By constrast, Ad-Cre-infected KB mice displayed abundant hyperplastic areas together with adenomas, a lesion that is extremely infrequent in K mice at this early stage 10 . Remarkably, X-gal + alveolar hyperplasias and adenomas were nearly absent in Ad-Cre-infected KBL animals. These mice, however, displayed hypertrophic bronchi as their most prominent feature ( Fig. 1a ). The lack of Kras G12V -driven alveolar lesions in KBL mice is unlikely to be a consequence of the absence of an active Braf, since this kinase is dispensable for the development of Kras G12V -driven lung adenocarcinoma 7 , 11 . Thus, we explored the possibility that the absence of alveolar lesions was due to some sort of cellular toxicity such as that caused by the induction of senescence or DNA damage, two phenomena known to act as barriers to tumour development 6 , 12 , 13 . Indeed, examination of lung lysates and sections from KBL mice detected p19 Arf and p53 tumour suppressors, active caspase-3 and γ-H2AX, a marker for DNA damage, as early as 1 week following Ad-Cre infection ( Fig. 1b, c ). We did not observe differences in senescence-associated β-gal staining (data not shown). The concomitant increase in the phosphorylation of Erk1/2 and p90Rsk suggested that elevated MAPK signalling triggered a stress response that impaired tumour cell proliferation. Yet it has been recently described that mutant Kras -driven lung adenocarcinomas display a tonic activation of the DNA damage response to prevent the induction of genotoxic stress 14 . We hypothesize that the exclusive formation of Braf(D631A)–Craf complexes in KBL mice exceeded such a toxic signalling threshold and induced a stress response incompatible with tumour development. Thus, our results indicate that Ras-oncogenic toxicity is quantitatively dictated by MAPK function. In support of this hypothesis, whereas Mek inhibition prevented tumour growth in Ad-Cre-infected KB mice, it rescued the toxic phenotype in KBL animals in a dose-dependent manner and restored tumour progression at intermediate drug concentrations, most likely by curtailing MAPK activity to levels compatible with cell proliferation ( Fig. 1d ). This is in agreement with the observation that RAS-induced senescence in primary cells is bypassed by inhibition of MEK and ERK kinases 15 . These observations suggest that Kras G12V -driven lung tumour cells can only proliferate within a limited range of MAPK activity. Whereas excess MAPK activity induced DNA damage-mediated cellular toxicity, insufficient MAPK signalling cannot sustain tumour growth. Figure 1: Combinations of Kras G12V , Braf D631A and wild-type Braf alleles establish a MAPK activity window that determines cell transformation and oncogene toxicity. a , Whole-mount X-gal staining of representative lung sections ( n = 5 per genotype) from Kras +/ LSLG12Vgeo (K), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice 1 month after Ad-Cre infection. X-gal staining identifies β-galactosidase expression as a surrogate marker for Kras G12V -expressing cells. Scale bar, 1mm. Insets show high-magnification images. Scale bar, 100 μm. b , Western blot analysis of lung lysates from Kras +/ LSLG12Vgeo (K), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice 1 week after Ad-Cre infection. Migration of p19 ARF , p53, γ-H2AX, cleaved caspase-3 (C3A), p-Erk1/2, Erk1/2, p-p90Rsk and p90Rsk is indicated by arrowheads. Gapdh was used as loading control. Lysates from two independent animals per genotype are shown. c , Representative immunostaining of paraffin-embedded lung sections ( n = 5 per genotype) from Kras +/ LSLG12Vgeo (K), Kras +/ LSLG12Vgeo ; Braf +/LSLD631A (KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice 1 week after Ad-Cre infection using the indicated antibodies. Scale bar, 50 μm. d , Whole-mount X-gal staining of representative lung sections ( n = 3 per genotype) from Kras +/ LSLG12Vgeo (K), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice 1 month after infection with 10 8 Ad-Cre particles. During this period, mice were treated with the indicated doses of the Mek inhibitor PD-0325901. The percentage of Kras G12V -expressing cells (X-gal-positive area) per lung section is indicated in each panel (top left, shown as mean ± s.d.). Scale bar, 200 μm. Insets in each panel show high-magnification images (top right; scale bar, 50 μm) or representative images of p-Erk1/2 immunostaining (bottom right; scale bar, 25 μm). PowerPoint slide Full size image Next, cohorts of Ad-Cre infected K, KB and KBL mice were followed over time. In agreement with the earlier onset and the more rapid progression of the alveolar lesions, the KB animals displayed significantly shortened survival ( Fig. 2a ). Histopathological analysis of their lungs at 6 months post Ad-Cre infection revealed a 7.5-fold increase in tumour burden compared to K controls ( Fig. 2b ). Moreover, KB mice displayed advanced adenocarcinomas, a tumour stage that is extremely infrequent at this time in tumours driven by Kras G12V alone ( Fig. 2c ). Tumours present in KB mice displayed SPC + CC10 − immunostaining, which suggests an alveolar type II (AT2) origin as previously described for adenocarcinomas driven by oncogenic Kras alone 10 , 16 ( Fig. 2d ). Altogether, these observations suggest that MAPK hyperactivation by coexisting Kras(G12V) and Braf(D631A) mutations resulted in increased transformation of AT2 cells and accelerated tumour progression. The MAPK paradoxical activation model postulates that the observed tumour phenotype is mediated by Craf kinase activity 2 , 8 , 17 . To genetically validate this hypothesis in the lung tumours studied here, we added conditional knock-in Craf (also known as Raf1 ) kinase-dead alleles ( Craf LSLD468A ) to KB mice. The resulting strain, Kras +/LSLG12Vgeo ; Braf +/ LSLD631A ; Craf LSLD468A/LSLD468A (designated as KBC KD ) was used to determine whether genetic inhibition of the Craf kinase reverted the increased tumorigenic phenotype displayed by KB mice. Expression of the Craf(D468A) kinase-dead isoform led to a substantial decrease in the levels of phosphorylated (p-)Erk1/2 and overall tumour burden ( Fig. 2e, f and Extended Data Fig. 3 ), significantly extending the survival of Ad-Cre-infected KBC KD mice compared to KB animals. Similarly, Craf ablation (KBC L mice) also resulted in prolonged survival ( Fig. 2g ). Altogether, these results demonstrate that the increased tumour burden and faster adenocarcinoma progression in KB animals is due to Craf-mediated hyperactivation of MAPK signalling. Figure 2: Mice concomitantly expressing Kras(G12V) and Braf(D631A) show reduced survival and increased tumour development mediated by Craf kinase activity. a , Survival of Kras +/ LSLG12Vgeo (K, open circles, n = 14), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, solid circles, n = 22) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL, open triangles, n = 17) mice after Ad-Cre intratracheal infection. P < 0.0001, obtained using the log-rank test (Mantel–Cox). b , Quantification of average tumour (adenoma and adenocarcinoma) burden (per cent of total lung area) in lung sections from Kras +/ LSLG12Vgeo (K, n = 5), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, n = 7) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL, n = 7) mice 6 months after Ad-Cre intratracheal infection. ** P < 0.01, * P < 0.05. c , Representative haematoxylin and eosin staining of paraffin-embedded lung sections obtained from Kras +/ LSLG12Vgeo (K, n = 5), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, n = 7) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL, n = 7) mice 6 months after Ad-Cre infection. Scale bar, 5 mm. d , Representative immunostaining of paraffin-embedded sections showing tumours from Ad-Cre-infected Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, n = 7) mice killed at humane end point using antibodies against SPC and CC10. Scale bar, 1 mm. e , Quantification of average tumour (adenoma and adenocarcinoma) burden (per cent of total lung area) in lung sections from Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, n = 7) and Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A ; Craf LSLD468A/LSLD468A (KBC KD , n = 5) mice 6 months after Ad-Cre intratracheal infection. ** P < 0.01. f , Representative haematoxylin and eosin staining of paraffin-embedded lung sections ( n = 5 per genotype) from Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB), and Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A ; Craf LSLD468A/LSLD468A (KBC KD ) mice 6 months after Ad-Cre infection. Scale bar, 5 mm. g , Survival of Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB, solid circles, n = 22), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A ; Craf LSLD468A/LSLD468A (KBC KD , solid triangles, n = 12) and Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A ; Craf lox/lox (KBC L , empty squares, n = 7) mice after Ad-Cre intratracheal infection. P < 0.0001, obtained using the log-rank test (Mantel–Cox). PowerPoint slide Full size image As indicated above, the tumour burden of the KBL cohort was significantly decreased compared to KB mice, suggesting that excessive MAPK activity in the absence of wild-type Braf expression may be detrimental for lung adenocarcinoma development. Yet, in spite of the reduced tumour burden, KBL animals reached humane end point at the same time as KB mice ( Fig. 2a, b ). Detailed examination of the lungs of KBL mice revealed the presence of intrabronchiolar carcinomas (112 of 250 bronchi, 45%), a rare lesion in K (0 of 162) or KB (20 of 145, 14%) cohorts ( Fig. 3a ). Immunostaining with a collection of markers including Ttf1 and SPC, characteristic of lung adenocarcinoma and Sox2, p63 and CK5, diagnostic of squamous cell carcinoma, did not clarify the origin of these intrabronchiolar lesions since the resulting expression pattern was not in accordance with the expected profile of either of these tumour types 18 ( Fig. 3b ). Analysis of the initial tumour stages of these lesions revealed protruding papillary structures accompanied by loss of CC10 expression, a marker characteristic of the bronchiolar epithelium. Most of these lesions (95%) acquired expression of the AT2 marker SPC, thus suggesting a transdifferentiation process ( Fig. 3c and Extended Data Fig. 4 ). Figure 3: Hyperactive MAPK signalling triggers bronchiolar carcinoma by transdifferentiation of club cells. a , Haematoxylin and eosin staining of paraffin-embedded sections showing the presence of intrabronchiolar carcinoma in Ad-Cre-infected Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice killed at humane end point. Two different magnifications are shown. Scale bar, 200 μm (top) and 50 μm (bottom). b , Representative immunostaining of paraffin-embedded sections showing intrabronchiolar tumours from Ad-Cre-infected Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL, n = 5) mice using antibodies against, SPC, CC10, Ttf1, Sox2, p63 and Ck5. Scale bar, 50 μm. c , Immunostaining of paraffin-embedded consecutive sections showing early intrabronchiolar lesions (1 week after Ad-Cre infection) in Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice using antibodies against SPC and CC10. Arrowheads indicate protruding bronchiolar papillary growth undergoing transdifferentiation. Scale bar, 50 μm. d , Representative haematoxylin and eosin staining of paraffin-embedded lung sections ( n = 3 per genotype) from Kras +/ LSLG12Vgeo (K), Kras +/ LSLG12Vgeo ; Braf +/ LSLD631A (KB) and Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice killed 2 months after infection with cell-type-restricted Ad-Cre viruses that selectively induce Cre-mediated recombination in AT2 (Ad-SPC-Cre) (top) and club cells (Ad-CC10-Cre) (bottom). Scale bar, 500 μm. Insets show high-magnification images. Scale bar, 100 μm. e , SPC and CC10 immunostaining of paraffin-embedded consecutive sections showing intrabronchiolar tumours from Kras +/ LSLG12Vgeo ; Braf lox/LSLD631A (KBL) mice infected with lineage-specific Ad-CC10-Cre. Scale bar, 100 μm. PowerPoint slide Full size image To further clarify the origin of these intrabronchiolar lesions we performed in vivo tracing experiments by infecting K, KB and KBL mice with lineage-specific Ad-CC10-Cre or Ad-SPC-Cre viruses that restrict Cre-mediated recombination to club or AT2 cells respectively 19 . Analysis of lungs of K and KB mice after Ad-SPC-Cre infection revealed the presence of adenomas that, as expected, were larger and more abundant in KB animals. Likewise, KBL mice predominantly developed alveolar hyperplasias that failed to progress to advanced stages. Importantly, infection with Ad-CC10-Cre did not induce tumour growth in K or KB mice but caused abundant SPC + intrabronchiolar carcinomas in KBL animals ( Fig. 3d, e ). These results indicate that the threshold of MAPK activity toxic for AT2 cells resulted in the efficient transformation of cells present in bronchial epithelium that are usually refractory to Kras G12V -driven oncogenesis 10 , 20 , 21 . Altogether, these results suggest a quantitative impact of MAPK activity controlling various aspects of Kras oncogene-driven lung carcinogenesis. As such, oncogenic transformation may critically depend on cell- or tissue-specific programs that are hijacked to adjust MAPK activity and avoid senescence or other tumour-suppressive stress responses. This quantitative model is not exclusive to MAPK signalling as a similar threshold response level has been proposed for Wnt/β-catenin-driven 22 or PI3K/AKT-driven tumorigenesis 23 . Mutational analysis of different human cancers has recently uncovered that among the BRAF hot spots in lung adenocarcinoma, those resulting in inactivating mutations predominate over the V600E activating substitution 1 . However, the contribution of BRAF-inactive mutants to lung cancer progression is unclear. Interestingly, a percentage of BRAF-inactivating mutations (11%) coexist with upstream RAS alterations ( Table 1 ). Thus, we decided to investigate whether expression of the Braf(D631A) kinase-inactive mutant in pre-existing Kras G12V -driven tumours may enhance tumour progression. To this end, we added the Braf LSLD631A allele to a lung tumour model driven by the Flp recombinase. The resulting strain, Kras +/ FSFG12V ; Tg.hUb-cre-ERT2 +/ T ; Braf +/ LSLD631A (designated as K F B) allows the temporal separation of tumour initiation (Flp-mediated Kras(G12V) expression) from genetic events induced during tumour progression (Cre-mediated Braf(D631A) expression). Induction of the Braf(D631A) kinase-inactive isoform in pre-existing Kras G12V -driven tumours resulted in reduced survival owing to the accelerated progression of these lesions ( Fig. 4a and Extended Data Fig. 5 ). Thus, these results provide an experimental explanation for the concurrence of oncogenic KRAS and BRAF -inactivating mutations in certain human lung tumours. These observations also reinforce the current clinical indication that class I RAF inhibitors (vemurafenib and derivatives targeting the activated form of the BRAF kinase) should not be used in patients with KRAS -mutant lung adenocarcinoma. Table 1 BRAF hypoactive mutants with NF1 and RAS mutations Full size table Figure 4: Activation of endogenous Braf D631A results in lung adenocarcinoma development. a , Kras +/ FSFG12V ; Braf +/ LSLD631A ; Tg.hUb-cre-ERT2 +/ T (K F B) and control Kras +/ FSFG12V ; Tg.hUb-cre-ERT2 +/ T (K F ) mice were infected intratracheally with Ad-Flp and tumour formation was monitored by computed tomography (CT). Mice bearing tumours visible on CT (K F B, n = 29 tumours from 14 mice and K F , n = 38 tumours from 15 mice) were fed ad libitum a tamoxifen-containing diet resulting in expression of the Braf(D631A) kinase-dead isoform in K F B. Tumour volume increase (measured as fold change) was re-evaluated by CT after 8 weeks in continuous diet. ** P < 0.01. b , Haematoxylin and eosin (H&E) staining, as well as SPC and CC10 immunostaining, of paraffin-embedded lung sections from Ad-Cre-infected Braf +/ LSLD631A mice killed at humane end-point. Scale bar, 500 μm and 50 μm (top right). c , Representative images of p-Erk1/2 immunostaining of paraffin-embedded lung sections ( n = 3 per genotype) from Ad-Cre-infected Kras +/ LSLG12Vgeo , Kras +/ LSLG12Vgeo ; Trp53 lox/lox , Braf +/ LSLV637E (equivalent to human BRAF V600E ) or Braf +/ LSLD631A mice. Scale bar, 100 μm. PowerPoint slide Full size image Yet, the majority of BRAF-inactivating mutations present in human lung cancer (89%) do not coexist with RAS mutations ( Table 1 ). To assess whether Braf inactive mutants could induce lung adenocarcinoma formation in the absence of Kras mutations, we infected Braf +/ LSLD631A mice intratracheally with Ad-Cre. Analysis of their lungs 12 months after infection revealed the presence of tumours in 9 of 22 mice (41% incidence) compared to 14 of 18 (78%) in Kras +/ LSLG12Vgeo animals. All tumours studied displayed histology characteristic of lung adenocarcinoma with SPC + CC10 − immunostaining ( Fig. 4b and Extended Data Fig. 6a ). Notably, p-Erk1/2 levels in Braf D631A -driven lung adenocarcinomas were higher than those observed in Braf V637E (equivalent to human BRAF V600E ), Kras G12V or Kras G12V ; Trp 53 –/– tumours ( Fig. 4c ). As control, we confirmed that the Braf LSLD631A allele was efficiently recombined in tumour tissue. Moreover, we determined that there were no Ras mutations in these tumours (data not shown). These results suggest that Braf-inactivating mutations initiate lung adenocarcinoma ( Extended Data Fig. 6b ). Our results are in good agreement with data from the Rosen laboratory indicating that increased levels of wild-type Ras–GTP complexes can cooperate with Braf hypoactive mutants to trigger tumour development in epithelial cells 24 . Similarly, increasing the pool of Ras–GTP by a dominant active Sos is sufficient to induce MAPK hyperactivation and the formation of epithelial tumours in response to Raf inhibition 25 . Likewise, MAPK activation in primary keratinocytes carrying the Braf LSLD631A allele depended both on Craf and RTK signalling. Of note, elimination of the Braf wild-type allele induced a further increase in p-Erk1/2 ( Extended Data Fig. 6c ). In lung, the high levels of endogenous Ras–GTP present in adult AT2 cells 26 may explain the oncogenicity of the Braf(D631A) inactive mutant in the absence of Kras(G12V). By contrast, hypoactive BRAF mutants require the presence of RAS or NF1 mutations to trigger melanoma development, both in experimental GEM models 2 as well as in human tumours ( Table 1 ). In summary, we provide the first genetic evidence demonstrating that a kinase-inactivating Braf mutation induces lung adenocarcinoma development. Importantly, in lung adenocarcinoma patients kinase-inactivating BRAF mutations are more prevalent than the activating V600E allele 1 . Furthermore, a recent retrospective study identified BRAF(D594G) (affecting the same residue as the mouse mutation used in this study) as the most frequent BRAF alteration in these patients 27 . Our analysis of human lung adenocarcinomas with hypoactive BRAF revealed co-occurring alterations such as mutations in RTK signalling antagonists that might cooperate in sustaining MAPK activity ( Extended Data Fig. 7 ). The accompanying manuscript by Yao et al . 24 suggests that RTK profiling of tumours driven by kinase-impaired BRAF mutants can identify the dominant receptor for the design of specific treatment for individual patients. Alternatively, our results suggest that these patients could benefit from therapies based on selective CRAF inhibitors. The mouse model described here will be useful to evaluate effective drug combinations. Methods Mice Kras LSLG12Vgeo (ref. 9 ), Braf LSLD631A (ref. 2 , described therein as Braf LSLD594A ), Braf lox (ref. 28 ), Braf LSLV637E (ref. 29 , described therein as Braf CA ) and Trp53 lox/lox (ref. 30 ) strains have been previously described. The targeting vector for the Craf LSLD468A allele was generated by Gene Bridges GmBH. The D468A miscoding mutation (GAC (Asp) to GCC (Ala)) was engineered within the Craf locus present in the BAC RP23-37K24 by multi-site Red/ET (‘triple recombination’). A loxP-cDNA-STOP-PGK-Neo-loxP cassette containing a partial Craf cDNA sequence encompassing exons 13 to 17 was inserted 150 bp upstream of the 5′ end of the mutated exon13 by Red/ET recombination. The resulting targeting vector was linearized with NotI and SalI restriction enzymes and electroporated into B6129SF1/J ES cells. Clones having undergone proper homologous recombination were identified by Southern blot analysis. Two independent recombinant ES cell clones were microinjected into FVB donor blastocysts and implanted into pseudo-pregnant females. Chimaeric mice were backcrossed to C57BL/6J mice and germline transmission of the Craf LSLD468A allele was confirmed by Southern blot analysis. The targeting vector used to obtain the Kras FSFG12V allele was generated by Taconic Artemis. In brief, the homology arms including the first exon containing the oncogenic G12V mutation were amplified by PCR using as a template a targeting vector previously developed to generate a Kras LSLG12Vgeo allele 9 . A PGK-Neo-STOP cassette was generated by PCR amplification of the 1,377 bp STOP cassette derived from the Kras LSLG12Vgeo targeting vector with primers that incorporated NdeI restriction sites. The STOP cassette was subsequently cloned into the NdeI restriction site of pBASIC10 (Taconic Artemis) flanked by FRT sequences. The resulting targeting vector was linearized with NotI and electroporated into B6129SF1/J ES cells. Two independent recombinant ES cell clones were microinjected into C57BL/6J blastocysts and transplanted into pseudo-pregnant females. Chimaeric mice were backcrossed to C57BL/6J mice and germ line transmission of the targeted allele was confirmed by Southern blot analysis. All animal experiments were approved by the Ethical Committee of CNIO and performed in accordance with the guidelines stated in the International Guiding Principles for Biomedical Research Involving Animals, developed by the Council for International Organizations of Medical Sciences (CIOMS). All strains were genotyped by Transnetyx. Tumour induction and drug treatments Tumours were induced in 8- to 12-week-old mice (both male and female animals were used) by single intratracheal infection with 10 6 adenoviral particles (unless stated otherwise) after anaesthesia (i.p. injection of ketamine 75 mg kg −1 , xylazine 12 mg kg −1 ) as previously reported 7 . Ad-Cre, Ad-CC10-Cre, Ad-SPC-Cre and Ad-Flp were purchased from University of Iowa Vector Core Facility. The Mek inhibitor PD-0325901 (Tocris Bioscience) was dissolved in 0.5% hydroxypropyl methyl cellulose, 0.2% Tween-80 water solution and administered by daily gavage. Following the ethical committee guidelines all mice were euthanized when showing respiratory problems. Histopathology and immunohistochemistry For routine histological analysis, all lung lobes from each mouse were fixed in 10% buffered formalin (Sigma), embedded in paraffin and evaluated in serial sections by conventional haematoxylin and eosin (H&E) staining according to previously published criteria 31 . All whole mount X-gal-stained sections were counterstained with Nuclear Fast Red. Antibodies used for immunostaining included those raised against: SPC (Millipore, AB3786); CC10 (Santa Cruz Biotechnology, SC-9772); Ttf1 (Epitomics, 2044-1); Sox2 (Cell Signaling Technology, 3728); p63 (Thermo Scientific, MS-1081-P); Ck5 (Covance, PRB-160P) and p-Erk1/2 (Cell Signaling Technology; 9101). Cell culture Kras +/ G12V ; Araf –/– ; Braf lox/lox ; Craf lox/lox ; Trp53 –/– ; Tg.hUb-cre-ERT2 +/ T lung adenocarcinoma cell lines were generated from primary tumours. Cells were infected with lentiviruses expressing Braf, Braf(D631A) and Craf (pLVXpuro) using routine procedures and exposed to 50 MOI (multiplicity of infection) of Ad-Cre. 4-hydroxytamoxifen (4-OHT) (Sigma) was used at 600 nM. Cell lines were confirmed mycoplasma free at every freeze/thaw cycle. Primary keratinocytes were obtained from adult tail skin. The antibodies used for western blotting included those raised against: Araf (Cell Signaling Technology, 4432); Braf (Santa Cruz Biotechnology, SC-5284); Craf (BD Biosciences, 610151); p-Erk1/2 (Cell Signaling Technology, 9101); Erk1 (BD Biosciences, 554100); Erk2 (BD Biosciences, 610103); γ-H2AX (Millipore, 05-636); p-p90Rsk (Cell Signaling Technology, 9341); p90Rsk (Santa Cruz Biotechnology, SC-231); p53 (Cell Signaling Technology, 2524); p19 ARF (Abcam, ab80); cleaved caspase-3 (Cell Signaling Technology, 9661) and Gapdh (Sigma, G8795). Computed tomography Mice were anaesthetized with 1% to 3% flow of isoflurane/oxygen, and the chest area was visualized with the GE eXplore Locus micro PET-CT scanner (GE Healthcare). The resulting raw data were reconstructed to a final image volume of 875 × 875 × 465 slices at 93 μm 3 voxel dimensions. Reconstructed slices were output in the manufacturer’s raw format and corrected equal to Hounsfield units and analysed with MicroView analysis software (GE Healthcare). Digital image quantification Digital images of immunostained slides were obtained using a whole slide scanner (Dotslide Olympus) with resolution 0.32 μm per pixel (20×/NA 0.75). For automated image quantification the tumour areas on scanned H&E sections were manually delimited within the normal lung tissue according to histological criteria and quantified by Dotslide viewer software. Statistical and data analysis The product limit method of Kaplan and Meier was used for generating the survival curves, which were compared by using the log-rank (Mantel–Cox) test. A P value that was less than 0.05 was considered statistically significant for all datasets. All statistical analysis was performed using GraphPad Prism software. No statistical method was used to predetermine sample size in animal studies. Only mice carrying tumours >1 mm in diameter on CT were enrolled in longitudinal studies assessing tumour volume variation. For all animal experiments block randomization was used to ensure a balance in sample size across groups over time. The investigators were blinded during evaluation of tumour size variations following tamoxifen diet. Data availability All data generated or analysed during this study are included in this published article (and its Supplementary Information files ).

The initiating oncogenic event in almost half of human lung adenocarcinomas is still unknown, complicating the development of selective targeted therapies. Yet these tumours harbour a number of alterations without obvious oncogenic function, including BRAF-inactivating mutations. Researchers at the Spanish National Cancer Research Centre (CNIO) have demonstrated that the expression of an endogenous Braf (D631A) kinase-inactive isoform in mice, corresponding to the human BRAF(D594A) mutation, triggers lung adenocarcinoma in vivo, indicating that BRAF-inactivating mutations are initiating events in lung oncogenesis. The paper, published in Nature, indicates that the signal intensity of the MAPK pathway is a critical determinant not only in tumour development, but also in dictating the nature of the cancer-initiating cell and ultimately the resulting tumour phenotype. The RAS-MAPK signalling cascade serves as a central node in transducing signals from membrane receptors to the nucleus. This pathway is aberrantly activated in a substantial fraction of human cancers. There is also abundant evidence that elevated RAS-MAPK signaling results in cellular toxicity that may serve as a natural barrier to cancer progression early in tumorigenesis. These findings suggest that defined thresholds of RAS-MAPK activity are required for homeostasis as well as for malignant transformation, but compelling genetic evidence is missing. Mutational analysis of different human cancers has recently uncovered that among the BRAF hot spots in lung adenocarcinoma, which comprise a component of the RAS-MAP kinase pathway, those resulting in inactivating mutations predominate over the V600E activating substitution, the main oncogenic form in other tumours such as melanoma. However, the contribution of BRAF-inactive mutants to lung cancer progression is unclear. Using public databases, researchers have identified inactivating BRAF mutations in a subset of KRAS-driven human lung tumours. Subsequently, using mouse models, researchers have replicated these observations showing that the co-expression of oncogenic Kras and inactive Braf markedly enhances the onset of lung adenocarcinoma. Also, this combination accelerates tumour progression when the inactivating Braf mutation is genetically induced in advanced tumors. Surprisingly, in this same study, the researchers showed that individually, the inactivating mutations of Braf are also oncogenic events that induce the appearance of lung adenocarcinoma. The paper provides the first genetic evidence demonstrating that a kinase-inactivating Braf mutation induces lung adenocarcinoma development. Moreover, results suggest that lung adenocarcinoma patients with hypoactive BRAF could benefit from therapies based on selective CRAF inhibitors.

10.1038/nature23297

Earth

Researchers map symbiotic relationships between trees and microbes worldwide

Climatic controls of decomposition drive the global biogeography of forest-tree symbioses, Nature (2019). DOI: 10.1038/s41586-019-1128-0 , www.nature.com/articles/s41586-019-1128-0 Journal information: Nature

http://dx.doi.org/10.1038/s41586-019-1128-0

https://phys.org/news/2019-05-symbiotic-relationships-trees-microbes-worldwide.html

Abstract The identity of the dominant root-associated microbial symbionts in a forest determines the ability of trees to access limiting nutrients from atmospheric or soil pools 1 , 2 , sequester carbon 3 , 4 and withstand the effects of climate change 5 , 6 . Characterizing the global distribution of these symbioses and identifying the factors that control this distribution are thus integral to understanding the present and future functioning of forest ecosystems. Here we generate a spatially explicit global map of the symbiotic status of forests, using a database of over 1.1 million forest inventory plots that collectively contain over 28,000 tree species. Our analyses indicate that climate variables—in particular, climatically controlled variation in the rate of decomposition—are the primary drivers of the global distribution of major symbioses. We estimate that ectomycorrhizal trees, which represent only 2% of all plant species 7 , constitute approximately 60% of tree stems on Earth. Ectomycorrhizal symbiosis dominates forests in which seasonally cold and dry climates inhibit decomposition, and is the predominant form of symbiosis at high latitudes and elevation. By contrast, arbuscular mycorrhizal trees dominate in aseasonal, warm tropical forests, and occur with ectomycorrhizal trees in temperate biomes in which seasonally warm-and-wet climates enhance decomposition. Continental transitions between forests dominated by ectomycorrhizal or arbuscular mycorrhizal trees occur relatively abruptly along climate-driven decomposition gradients; these transitions are probably caused by positive feedback effects between plants and microorganisms. Symbiotic nitrogen fixers—which are insensitive to climatic controls on decomposition (compared with mycorrhizal fungi)—are most abundant in arid biomes with alkaline soils and high maximum temperatures. The climatically driven global symbiosis gradient that we document provides a spatially explicit quantitative understanding of microbial symbioses at the global scale, and demonstrates the critical role of microbial mutualisms in shaping the distribution of plant species. Main Microbial symbionts strongly influence the functioning of forest ecosystems. Root-associated microorganisms exploit inorganic, organic 2 and/or atmospheric forms of nutrients that enable plant growth 1 , determine how trees respond to increased concentrations 6 of CO 2 , regulate the respiratory activity of soil microorganisms 3 , 8 and affect plant species diversity by altering the strength of conspecific negative density dependence 9 . Despite the growing recognition of the importance of root symbioses for forest functioning 1 , 6 , 10 and the potential to integrate symbiotic status into Earth system models that predict functional changes to the terrestrial biosphere 10 , we lack spatially explicit quantitative maps of root symbioses at the global scale. Quantitative maps of tree symbiotic states would link the biogeography of functional traits of belowground microbial symbionts with their 3.1 trillion host trees 11 , which are spread across Earth’s forests, woodlands and savannahs. The dominant guilds of tree root symbionts—arbuscular mycorrhizal fungi, ectomycorrhizal fungi, ericoid mycorrhizal fungi and nitrogen-fixing bacteria (N-fixers)—are all based on the exchange of plant photosynthate for limiting macronutrients. Arbuscular mycorrhizal symbiosis evolved nearly 500 million years ago, and ectomycorrhizal, ericoid mycorrhizal and N-fixer plant taxa have evolved multiple times from an arbuscular-mycorrhizal basal state. Plants that are involved in arbuscular mycorrhizal symbiosis comprise nearly 80% of all terrestrial plant species; these plants principally rely on arbuscular mycorrhizal fungi for enhancing mineral phosphorus uptake 12 . In contrast to arbuscular mycorrhizal fungi, ectomycorrhizal fungi evolved from multiple lineages of saprotrophic ancestors and, as a result, some ectomycorrhizal fungi are capable of directly mobilizing organic sources of soil nutrients (particularly nitrogen) 2 . Associations with ectomycorrhizal fungi—but not arbuscular mycorrhizal fungi—have previously been shown to enable trees to accelerate photosynthesis in response to increased concentrations of atmospheric CO 2 when soil nitrogen is limiting 6 , and to inhibit soil respiration by decomposer microorganisms 3 , 8 . Because increased plant photosynthesis and decreased soil respiration both reduce atmospheric CO 2 concentrations, the ectomycorrhizal symbiosis is associated with buffering the Earth’s climate against anthropogenic change. In contrast to mycorrhizal fungi, which extract nutrients from the soil, symbiotic N-fixers (Rhizobia and Actinobacteria) convert atmospheric N 2 to plant-usable forms. Symbiotic N-fixers are responsible for a large fraction of biological soil-nitrogen inputs, which can increase nitrogen availability in forests in which N-fixers are locally abundant 13 . Symbioses with either N-fixers or ectomycorrhizal fungi often demand more plant photosynthate than does arbuscular mycorrhizal symbiosis 12 , 14 , 15 . Because tree growth and reproduction are limited by access to inorganic, organic and atmospheric sources of nitrogen, the distribution of root symbioses is likely to reflect environmental conditions that maximize the cost:benefit ratio of symbiotic exchange as well as physiological constraints on the different symbionts. One of the earliest efforts 16 to understand the functional biogeography of plant root symbioses categorically classified biomes by their perceived dominant mycorrhizal type, and hypothesized that seasonal climates favour hosts that associate with ectomycorrhizal fungi (owing to the ability of these hosts to compete directly for organic nitrogen). By contrast, it has more recently been proposed that sensitivity to low temperatures has prevented N-fixers from dominating outside of the tropics, despite the potential for nitrogen fixation to alleviate nitrogen limitation in boreal forests 15 , 17 . However, global-scale tests of these proposed biogeographical patterns and their climate drivers are lacking. To address this, we compiled a global ground-sourced survey database to reveal the numerical abundances of each type of symbiosis across the globe. Such a database is essential for identifying the potential mechanisms that underlie transitions in forest symbiotic state along climatic gradients 18 , 19 . We determined the abundance of tree symbioses using an extension of the plot-based Global Forest Biodiversity (GFB) database that we term the GFBi; this extended database contains over 1.1 million forest inventory plots of individual-based measurement records, from which we derive abundance information for entire tree communities (Fig. 1 ). Using published literature on the evolutionary histories of mycorrhizal and N-fixer symbioses, we assigned plant species from the GFBi to one of five root-associated symbiotic guilds: arbuscular mycorrhizal, ectomycorrhizal, ericoid mycorrhizal, N-fixer and weakly arbuscular or non-mycorrhizal. We then used the random-forest algorithm with K -fold cross-validation to determine the importance and influence of variables related to climate, soil chemistry, vegetation and topography on the relative abundance of each tree symbiotic guild (Fig. 2 ). Because decomposition is the dominant process by which soil nutrients become available to plants, we calculated annual and quarterly decomposition coefficients according to the Yasso07 model 20 , which describes how temperature and precipitation gradients influence mass-loss rates of different chemical pools of leaf litter (with parameters fit using a previous global study of leaf decomposition) (Fig. 3 , Supplementary Fig. 5 ). Finally, we projected our predictive models across the globe over the extent of global biomes that fell within the multivariate distribution of our model training data (Fig. 4 , Supplementary Figs. 14 , 15 ; see Methods for full description). Fig. 1: The global distribution of GFBi training data. The global map has n = 2,768 grid cells at a resolution of 1° × 1° latitude and longitude. Cells are coloured in the red, green and blue spectrum according to the percentage of total tree basal area occupied by N-fixer, arbuscular mycorrhizal (AM) and ectomycorrhizal (EM) tree symbiotic guilds, as indicated by the ternary plot. Full size image Fig. 2: A small number of environmental variables predict the majority of global turnover in forest symbiotic status. a – c , Partial feature contributions of different environmental variables to forest symbiotic state. Each row plots the shape of the contribution of the four most-important predictors of the proportion of tree basal area that belongs to the ectomycorrhizal ( a ), arbuscular mycorrhizal ( b ) and N-fixer ( c ) symbiotic guilds ( n = 2,768). Variables are listed in declining importance from left to right, as determined by the increase in node purity (inc. node purity), and with points coloured with a red to green to blue gradient according to their position on the x axis of the most-important variable (left-most panels for each guild), allowing cross-visualization between predictors. Each panel lists two measures of variable importance; inc. node purity (used for sorting) and percentage increase in mean square error (% inc. MSE) (see Supplementary Information). The abundance of each type of symbiont transitions sharply along climatic gradients, which suggests that sites near the threshold are particularly vulnerable to switching their dominant symbiont guild as climate changes. Warmest and wettest quarter, the warmest and wettest quarters of the year, respectively. Full size image Fig. 3: The distribution of forest symbiotic status across biomes is related to climatic controls over decomposition. a , Biome level summaries of the median ± 1 quartile of the predicted percentage of tree basal area per biome for ectomycorrhizal, arbuscular mycorrhizal and N-fixer symbiotic guilds ( n = 100 random samples per biome). b , The dependency of decomposition coefficients ( k , solid and dotted lines; in the region between the solid lines, the model transitions abruptly between dominant symbiotic status) on temperature and precipitation during the warmest quarter with respect to predicted dominance of mycorrhizal symbiosis. The transition from arbuscular mycorrhizal forests to ectomycorrhizal forests between k = 1 and k = 2 is abrupt, which is consistent with positive feedback between climatic and biological controls of decomposition. Full size image Fig. 4: Global maps of predicted forest-tree symbiotic state. a – c , Maps (left) and latitudinal gradients (right; solid line indicates median; coloured ribbon spans the range between the 5% and 95% quantiles) of the percentage of tree basal area for ectomycorrhizal ( a ), arbuscular mycorrhizal ( b ) and N-fixer ( c ) symbiotic guilds. All projections are displayed on a 0.5°-by-0.5° latitude and longitude scale. n = 28,454 grid cells. Full size image Our analysis shows that each one of the three most-numerically abundant guilds of tree symbiosis has a reliable environmental signature, with the four most-important predictors accounting for 81, 79 and 52% of the total variability in relative basal area for ectomycorrhizal, arbuscular mycorrhizal and N-fixer symbioses, respectively. Given the relative rarity of ericoid mycorrhizal and weakly arbuscular or non-mycorrhizal symbiotic states among trees, models for these symbioses lack strong predictive power—although the raw data do identify some local abundance hotspots for ericoid mycorrhizal symbiosis (Supplementary Fig. 1 ). As a result, we focus on the three major tree symbiotic states (ectomycorrhizal, arbuscular mycorrhizal and N-fixer). Despite the fact that data from North America and South America constitute 65% of the training data (at the 1°-by-1° grid scale), our models accurately predict the proportional abundances of the three major symbioses across all major geographical regions (Supplementary Fig. 10 ). The high performance of our models—which is robust to K -fold cross-validation and to rarefying samples such that all continents are represented with equal depth (Supplementary Figs. 11 , 12 )—suggests that regional variations in climate (including indirect effects on decomposition) and soil pH (for N-fixers) are the primary factors that influence the relative dominance of each guild at the global scale; geographical origin explained only approximately 2–5% of the variability in residual relative abundance (Supplementary Table 8 , Supplementary Fig. 10 ). Whereas a recent global analysis of root traits concluded that plant evolution has favoured a reduced dependence on mycorrhizal fungi 21 , we find that trees that associate with the relatively more carbon-demanding and recently derived ectomycorrhizal fungi 12 , 14 represent the dominant tree symbiosis. By taking the average proportion of ectomycorrhizal trees, weighted by spatially explicit global predictions for tree stem density 11 , we estimate that approximately 60% of tree stems on earth are ectomycorrhizal—despite the fact that only 2% of overall plant species associate with ectomycorrhizal fungi (versus nearly 80% that associate with arbuscular mycorrhizal fungi) 7 . Outside of the tropics, the estimate for the relative abundance of ectomycorrhizal symbiosis increases to approximately 80% of trees. Turnover among the major symbiotic guilds results in a tri-modal latitudinal abundance gradient, in which the proportion of ectomycorrhizal trees increases (and the proportion of arbuscular mycorrhizal trees decreases) with distance from the equator and the upper quantiles of nitrogen-fixing trees reach a peak in abundance in the arid zone at around 30° N or S (Figs. 3a and 4 ). These trends are driven by abrupt transitional regions along continental climatic gradients (Fig. 2 ), which skew the distribution of symbioses among biomes (Fig. 3a ) and drive strong patterns across geographical and topographic features that influence climate. Moving north or south from the equator, the first transitional zone separates warm (aseasonal) tropical broadleaf forests dominated by arbuscular mycorrhizal symbiosis (>75% median basal area versus 8% for ectomycorrhizal trees) from the rest of the world forest system, which is dominated by ectomycorrhizal symbiosis (Figs. 2a, b and 3a ). The transition zone occurs across the globe at around 25° N and S, just beyond the dry tropical broadleaf forests (which have 25% of their basal area consisting of ectomycorrhizal trees) (Fig. 3a ) in which average monthly temperature variation reaches 3–5 °C (temperature seasonality) (Fig. 2a, b ). Moving further north or south, the second transitional climate zone separates regions in which decomposition coefficients during the warmest quarter of the year are less than two (Fig. 3b gives the associated temperature and precipitation ranges). In North America and China, this transition zone occurs around 50° N, and separates the mixed arbuscular mycorrhizal and ectomycorrhizal temperate forests from their neighbouring ectomycorrhizal-dominated boreal forests (75 and 100% of their basal area, respectively, consisting of ectomycorrhizal trees) (Fig. 3a ). This transitional decomposition zone is not present in western Europe, which has a temperature seasonality of >5 °C but lacks sufficiently wet summers to accelerate decomposition coefficients beyond the values that are associated with mixed arbuscular mycorrhizal and ectomycorrhizal forests. The latitudinal transitions in symbiotic state observed among biomes are mirrored by within-biome transitions along elevation gradients. For example, in tropical Mexico decomposition coefficients of less than two during the warmest and wetter quarters of the year occur along the slopes of the Sierra Madre, where a mixture of arbuscular-mycorrhizal and N-fixer woodlands in arid climates transition to ectomycorrhizal-dominated tropical coniferous forests (75% basal area) (Figs. 3a and 4a–c , Supplementary Figs. 16 – 18 ). The Southern Hemisphere—which lacks the landmass to support extensive boreal forests—experiences a similar latitudinal transition in decomposition rates along the ecotone that separates its tropical and temperate biomes, at around 28° S. The abrupt transitions that we detected between forest symbiotic states along environmental gradients suggest that positive feedback effects may exist between climatic and biological controls of decomposition 10 , 20 . In contrast to arbuscular mycorrhizal fungi, some ectomycorrhizal fungi can use oxidative enzymes to mineralize organic nutrients from leaf litter and convert nutrients to plant-usable forms 2 , 5 . Relative to arbuscular mycorrhizal trees, the leaf litter of ectomycorrhizal trees is also chemically more resistant to decomposition, and has higher C:N ratios and higher concentrations of decomposition-inhibiting secondary compounds 10 . Thus, ectomycorrhizal leaf litter can exacerbate climatic barriers to decomposition and promote conditions in which ectomycorrhizal fungi have nutrient-acquiring abilities that are superior to those of arbuscular mycorrhizal fungi 5 , 10 . A recent game-theoretical model has shown that positive feedback effects between plants and soil nutrients can lead to local bistability in mycorrhizal symbiosis 22 . Such positive feedback effects are also known to cause abrupt ecosystem transitions along smooth environmental gradients between woodlands and grasses: trees suppress fires (which promotes seedling recruitment), whereas grass fuels fires that kill tree seedlings 23 . The existence of abrupt transitions also suggests that forests in transitional regions along decomposition gradients should be susceptible to marked turnover in symbiotic state with future environmental changes 23 . To illustrate the sensitivity of global patterns of tree symbiosis to climate change, we use the relationships that we observed for current climates to project potential changes in the symbiotic status of forests in the future. Relative to our global predictions that use the most-recent climate data, model predictions that use the projected climates for 2070 suggest that the abundance of ectomycorrhizal trees will decline by as much as 10% (using a relative concentration pathway of 8.5 W per m 2 ) (Supplementary Fig. 24 ). Our models predict that the largest declines in ectomycorrhizal abundance will occur along the boreal–temperate ecotone, where small increases in climatic decomposition coefficients cause abrupt transitions to arbuscular mycorrhizal forests (Fig. 2a, b ). Although our model does not estimate the time lag between climate change and forest community responses, the predicted decline in ectomycorrhizal trees corroborates the results of common garden transfer and simulated warming experiments, which have demonstrated that some important ectomycorrhizal hosts will decline at the boreal–temperate ecotone under altered climate conditions 24 . The change in dominant nutrient-exchange symbioses along climate gradients highlights the interconnection between atmospheric and soil compartments of the biosphere. The transition from arbuscular mycorrhizal to ectomycorrhizal dominance corresponds with a shift from phosphorus to nitrogen limitation of plant growth with increasing latitude 25 , 26 . Including published global projections of total soil nitrogen or phosphorus, microbial nitrogen or soil phosphorus fractions (labile, occluded, organic and apatite) did not increase the amount of variation explained by the model, or alter the variables identified as most important; we therefore dropped these projections from our analysis. However, our finding that climatic controls of decomposition are the best predictors of dominant mycorrhizal associations provides a mechanistic link between symbiont physiology and climatic controls on the release of soil nutrients from leaf litter. These findings are consistent with Read’s hypothesis 16 that slow decomposition at high latitudes favours ectomycorrhizal fungi owing to their increased capacity to liberate organic nutrients 2 . Thus, although more experiments are necessary to understand the specific mechanism by which nutrient competition favours the dominance of arbuscular mycorrhizal or ectomycorrhizal symbioses 18 , we propose that the latitudinal and elevational transitions from arbuscular-mycorrhizal-dominated to ectomycorrhizal-dominated forests be named ‘Read’s rule’. Our analyses focus on prediction at large spatial scales that are appropriate to the available data, but our findings with respect to Read’s rule also provide insight into how soil factors structure the fine-scale distributions of tree symbioses within our grid cells. For example, at a coarse scale, we find that ectomycorrhizal trees are relatively rare in many wet tropical forests; however, individual tropical sites in our raw data span the full range from 0 to 100% basal area dominated by ectomycorrhizal trees. In much of the wet tropics, these ectomycorrhizal-dominated sites exist as outliers within a matrix of predominantly arbuscular mycorrhizal trees. In an apparent exception that proves Read’s rule, in aseasonal, warm neotropical climates—which accelerate leaf decomposition and promote the regional dominance of arbuscular mycorrhizal symbiosis (Fig. 3 )—ectomycorrhizal-dominated tree stands can develop in sites in which poor soils and recalcitrant litter slow the rates of decomposition and nitrogen mineralization 18 , 27 . Landscape-scale variation in the relative abundance of symbiotic states also changes along climate gradients: variability is highest in xeric and temperate biomes (Supplementary Figs. 3 , 4 ), which suggests that the potential of local nutrient variability to favour particular symbioses is contingent on climate. Whereas ectomycorrhizal trees are associated with ecosystems in which plant growth is thought to be primarily nitrogen-limited, N-fixer trees are not. Our results highlight the global extent of the apparent ‘nitrogen cycling paradox’ in which some metrics suggest that nitrogen limitation is greater in the temperate zone 25 , 26 and yet nitrogen-fixing trees are relatively more common in the tropics 15 , 28 (Fig. 3a ). We find that N-fixers—which we estimate represent 7% of all trees—dominate forests with annual maximum temperatures >35 °C and alkaline soils, particularly in North America and Africa (Fig. 2c ). N-fixers have the highest relative abundance in xeric shrublands (24%), tropical savannahs (21%) and dry broadleaf forest biomes (20%), but are nearly absent from boreal forests (<1%) (Figs. 3a and 4 ). The decline in N-fixer tree abundance with increasing latitude that we observed is also associated with a previously documented latitudinal shift in the identity of nitrogen-fixing microorganisms, from facultative rhizobial N-fixers in tropical forests to obligate actinorhizal N-fixers in temperate forests 28 . Our data are not capable of fully disentangling the several hypotheses that have previously been proposed to reconcile the nitrogen cycling paradox 15 . However, our results are consistent with the model prediction 17 and regional empirical evidence 19 , 29 , 30 that nitrogen-fixing trees are particularly important in arid biomes. Based primarily on the observed positive nonlinear association of the relative abundance of N-fixers with the mean temperature of the hottest month (Fig. 2c ), our models predict a twofold increase in relative abundance of N-fixers when transitioning from humid to dry tropical forest biomes (Fig. 3a ). Although soil microorganisms are a dominant component of forests in terms of both diversity and ecosystem functioning 5 , 6 , 10 , identifying global-scale microbial biogeographical patterns remains an ongoing research priority. Our analyses confirm that Read’s rule—which is one of the first proposed biogeographical rules specific to microbial symbioses—successfully describes global transitions between mycorrhizal guilds. More generally, climate driven turnover among the major symbioses between plants and microorganisms represents a fundamental biological pattern in the Earth system, as forests transition from low-latitude arbuscular mycorrhizal through N-fixer to high-latitude ectomycorrhizal ecosystems. The predictions of our model (available in the Supplementary Data as global raster layers) can now be used to represent these critical ecosystem variations in global biogeochemical models that are used to predict climate–biogeochemical feedback effects within and between trees, soils and the atmosphere. Additionally, the raster layer that contains the proportion of nitrogen-fixing trees can be used to map potential symbiotic nitrogen fixation, which links atmospheric pools of carbon and nitrogen. Future work can extend our findings to incorporate multiple plant growth forms and non-forested biomes (in which similar patterns are likely to exist) to generate a complete global perspective. Our predictive maps leverage a comprehensive global forest dataset to generate a quantitative global map of forest tree symbioses, and demonstrate how nutritional mutualisms are coupled with the global distribution of plant communities. Methods We quantified the relative abundance of tree symbiotic guilds across >1.1 million forest census plots combined in the GFBi database, an extension of the plot-based GFB database 31 . The GFBi database consists of individual-based data that we compiled from all the regional and national GFBi forest-inventory datasets, including the French NFI (IGN—French National Forest Inventory, raw data, annual campaigns 2005 and following, , site accessed on 01 January 2015). The standardized GFBi data frame (that is, tree list) comprises tree identifier (ID) (a unique number assigned to each individual tree); plot ID (a unique string assigned to each plot); plot coordinates, in decimal degrees of the WGS84 datum; tree size, in diameter-at-breast-height; trees-per-hectare expansion factor; year of measurement; dataset name (a unique name assigned to each forest inventory dataset); and binomial species names of trees. We checked all species names from different forest inventory datasets for errors in three steps. First, we extracted scientific names from original datasets, and kept only the names of genus and species (authority names are removed). Next, we compiled all the species names into five general species lists (one for each continent). Finally, we verified individual species names against 23 online taxonomic databases using the ‘taxize’ package of the R programming language 32 . We assigned each morphospecies a unique name that comprised the genus, the string ‘spp’, followed by the dataset name and a unique number for that species. For example, ‘Picea sppCNi1’ and ‘Picea sppCNi2’ represent two different species under the genus Picea , observed in the first Chinese dataset (CNi). We derived plot-level abundance information in terms of species-abundance matrices. Each species-abundance matrix consisted of the number of individuals by species (column vectors) within individual sample plots (row vectors). In addition, key plot-level information was also added to the matrices, including plot ID, dataset name, plot coordinates, the year of measurement and basal area (that is, the total cross-sectional areas (in m 2 ) of living trees per one hectare of ground area). Tree genera were assigned to a plant family using a plant taxonomy lookup table generated by W. Cornwell (hosted on Github, ), which uses the accepted taxonomy from ‘The Plant List’ ( ). The majority (96.5%) of genera of the species in the GFBi were successfully matched to family; for those that could not be assigned, we manually checked the genus and species in the GFBi against synonyms from The Plant List. Of the 1,038 mismatches that remained after automated assignment to families, an additional 440 genera were assigned to family either by updating older genera and species names with their more-recent synonyms or by correcting obvious misspellings. The remaining 598 entries that could not be matched to family were excluded from further analysis. We used a taxonomically informed approach to assign symbiotic states to plant species from the GFBi. Plant species were assigned to one of five symbiotic guilds; ectomycorrhizal, arbuscular mycorrhizal, ericoid mycorrhizal, weakly arbuscular mycorrhizal or non-mycorrhizal (AMNM) or N-fixer (Supplementary Table 1 ). Although we did not model the relative abundance of ericoid mycorrhizal trees (owing to their rarity), we have included a map of their relative abundance from our grid (Supplementary Fig. 1 ). We also include the full species list as Supplementary Data; this list includes the columns used to assign species to guilds. We also include a list of families and genera assigned to all guilds except the arbuscular mycorrhizal guild (Supplementary Tables 2 – 5 ), with notes for cases of species of individual genera that were assigned to two guilds simultaneously (for example, Alnus is an N-fixer and ectomycorrhizal) or for cases in which species from individual genera were split between two different guilds (for example, some Pisonia sp. are AMNM and some are ectomycorrhizal). An arbuscular mycorrhizal summary table is excluded from the Supplementary Tables for length considerations; this information is available as Supplementary Data (file name ‘SymbioticGuildAssignment.csv’). The taxonomy of species in our inventory was compared with recently published literature on the evolutionary history of mycorrhizal symbiosis 7 , 33 and nitrogen fixation 34 , 35 , 36 , 37 . For most species, symbiotic status could be reliably assigned at the genus (for example, Dicymbe ) or family level (for example, Pinaceae). For the few groups for which status was unreliable or variable within a genus (for example, Pisonia ), we conducted additional literature searches. We assigned species to the ectomycorrhizal category in three stages: first, at the family level (for example, Pinaceae); then, at the genus level (for example, Dicymbe ); and, finally, by using literature searches for genera for which the status was unclear (for example, in the genus Pisonia some species are arbuscular mycorrhizal and others are ectomycorrhizal). We used a published list 38 to sort species into the appropriate guild. For the genus Acacia , we followed previous work 7 by assuming that only endemic Australian species associate with ectomycorrhizal fungi (we sorted Acacia species according to provenance using ). The AMNM category grouped all genera of terrestrial, non-epiphytic plants that either lack arbuscular mycorrhizal fungi or have low or inconsistent records of arbuscular mycorrhizal fungi colonization of roots. For example, although there are some published records of arbuscular mycorrhizal fungi colonization in the roots of plants of the Proteaceae family, these records are inconsistent and colonization is generally low. Further, as Proteaceae are associated with a non-mycorrhizal root morphology (the cluster or proteoid root system) that allows them to access otherwise unavailable forms of soil nutrients 39 , we placed the entire family within AMNM. The family Urticaceae (which we also characterized as AMNM) was problematic—early successional species from tropical forests, such as those in the genus Cecropia , have records of both low and absent arbuscular mycorrhizal fungi colonization 40 . Our approach was to use the most broadly inclusive categorization for AMNM plants. N-fixer status was assigned at the genus level, using previously compiled databases of global symbiotic N 2 fixation 34 , 35 , 36 , 37 . Given that symbiotic N 2 fixation with rhizobial or Frankia bacteria has evolved only in four orders (Rosales, Cucurbitales, Fabales and Fagales) 41 , all species outside of this nitrogen-fixing clade were assigned non-fixing status. Some species could not be assigned an N-fixer status because they were typed to a higher taxonomic level (for example, family) that is ambiguous from the perspective of N-fixer status. We recorded when our assignment of N-fixer status was based on phylogenetic criteria, but where symbiotic nitrogen fixation is evolutionarily labile. Because these cases are more likely to be mis-assigned, we excluded them from the nitrogen-fixation category. The N-fixer group contains species that are colonized by arbuscular mycorrhizal fungi (for example, most genera from Leguminosae) and others that are colonized by ectomycorrhizal fungi (for example, Alnus sp.). Most plant species form arbuscular mycorrhizal symbioses, the basal symbiotic state relative to the later-derived ectomycorrhizal and nitrogen-fixing symbioses. Furthermore, many ectomycorrhizal and nitrogen-fixing plants maintain the ability to form arbuscular mycorrhizal symbioses. Thus, a tree species is most likely to be arbuscular mycorrhizal if it does not form associations with another symbiotic guild (or forgoes root symbiosis entirely), as evidenced by its inclusion in exhaustive databases of plant symbiotic state 7 , 33 , 34 , 35 , 36 , 37 , 40 . In keeping with other large-scale studies in the field 33 , we assigned tree species from the GFBi database an arbuscular-mycorrhizal-exclusive state if they belonged to taxa that were not matched to ectomycorrhizal, ericoid mycorrhizal, AMNM or N-fixer symbioses. Thus, the arbuscular mycorrhizal and N-fixer groups in our dataset are non-overlapping, despite the fact that most N-fixers also associate with arbuscular mycorrhizal fungi. The proportions of tree basal area and tree individuals were aggregated to a 1°-by-1° grid by taking the weighted average of the plot-level proportions (Supplementary Table 6 ). This resulted in a total of 2,768 grid cells, each with a score for the proportional abundance of ectomycorrhizal, arbuscular mycorrhizal, N-fixer, ericoid mycorrhizal and AMNM trees. We calculated two measures of relative abundance for each symbiotic guild: the proportion of tree stems and the proportion of tree basal area. Because the measurements are highly correlated with one another (Supplementary Fig. 2 ), we chose to model only the proportion of total tree basal area, which should scale more closely to proportion of tree biomass as it accounts for differences in size among individual stems. Additionally, we quantified variability among plots within each grid cell by calculating the weighted standard deviation across the grid (Supplementary Information, Supplementary Figs. 3 , 4 ). To identify the key factors that structure symbiotic distributions, we assembled 70 global predictor layers: 19 climatic indices (relating to annual, monthly and quarterly temperature and precipitation variables), 14 soil chemical indices (relating to total soil nitrogen density, microbial nitrogen, C:N ratios and soil phosphorus fractions, pH and cation exchange capacity), 5 soil physical indices (relating to soil texture and bulk density), 26 vegetative indices (relating to leaf area index, total stem density, enhanced vegetation index means and variances) and 5 topographic variables (relating to elevation and hillshade) (Supplementary Table 7 ). Because decomposition is the dominant process by which soil nutrients become available to plants, we generated five additional layers that estimate climatic control of decomposition. We parameterized decomposition coefficients according to the Yasso07 model 20 , 42 , using the following equation: k = exp(0.095 T − 0.00014 × T 2 ) × (1 − exp[−1.21 × P ]), in which P and T are precipitation and mean temperature (either quarterly or annually) of a grid cell, and the constants 0.095, 0.00014 and −1.21 are parameters that were fit using a previous global study of leaf litter mass loss 20 . Although local decomposition rates can vary considerably based on litter quality or microbial community composition 43 , climate is the primary control at the global scale 20 . Decomposition coefficients describe how fast different chemical pools of leaf litter lose mass over time, relative to a parameter ( α ) that accounts for leaf chemistry. Decomposition coefficients ( k ) with values of 0.5 and 2 indicate a halving and doubling of decomposition rates, respectively, relative to α (Supplementary Information, Supplementary Fig. 5 ). We implemented the random-forest algorithm using the ‘randomForest’ package in R. Random-forest models average over multiple regression trees, each of which uses a random subset of all the model variables to predict a response. We first determined the influence and relationship of all 75 predictor layers on forest symbiotic state, and then optimized our models using a stepwise reduction in variables from least to most important. Variable importance was measured in two ways: increase in node purity and percentage increase in MSE (with values reported in Fig. 2 ). The increase in node purity of variable x considers the decrease in the residual sum of squares that results from splitting regression trees using variable x . The percentage increase in MSE quantifies the increase in model error as a result of randomly shuffling the order of values in the vector x . We chose to rank variables according to the increase in node purity because we found that higher increases in node purities were associated with larger effect sizes, whereas larger percentage increases in MSE were associated with more-linear responses with smaller effect sizes. Whereas our inspection of partial feature contributions is derived from univariate random-forest models, we additionally ran multivariate random forests that predict the proportional abundance of ectomycorrhizal, arbuscular mycorrhizal and N-fixer trees for each pixel. The multivariate models were run using 50 regression trees each, with the unique set of the best 4 predictor variables for each symbiotic guild in the univariate models (Fig. 2 , Supplementary Table 7 ). Despite strong negative correlations between the proportions of ectomycorrhizal and arbuscular mycorrhizal basal area (Supplementary Fig. 22 ), the results from multivariate and univariate random forests are strongly correlated with one another (Supplementary Fig. 23 ). Using model selection based on eliminating variables with a low increase in node purity, we removed most soil nutrient, vegetative and topographic variables from our models (Supplementary Figs. 6 , 7 ). Our final models include the remaining 34 predictor layers with climate, decomposition and some soil physical and chemical information (Supplementary Fig. 8 ). To determine the parsimony of our models, we compared the coefficient of determination in models run with a stepwise reduction in the number of variables (starting with those with the lowest increase in node purity). Based on performance of the ratio of coefficient of determination in models with 4 versus 34 variables, we determined that the 4 most-important variables accounted for >85% of the explained variability (Supplementary Fig. 9 ). We also compared model performance visually with plots of actual versus predicted proportions of each tree symbiotic guild among continents and geographical subregions (Supplementary Fig. 10 ). We used the ‘forestFloor’ package in R to plot the partial variable response of tree symbiotic guilds to each predictor variable (Fig. 2a–c , see Supplementary Figs. 19 – 21 for partial plots of the partial feature contributions of all 34 variables). To test the sensitivity of model performance and predictions, we performed cross-validation in R using the ‘rfUtilities’ package 44 . K -fold cross-validation tests the sensitivity of model predictions to losing random subsets from the training data. For ectomycorrhizal, arbuscular mycorrhizal and N-fixer models, we ran 99 iterations that withheld 10% of the model training data. We assessed the decrease in model performance in the 99 iterations by manually calculating the coefficient of determination, which uses the following formula: 1 − Σ(actual percentage basal area – predicted percentage basal area) 2 /Σ(actual percentage basal area − mean actual percentage basal area) 2 . For all symbiotic guilds, withholding 10% of the training data resulted in a mean loss in variance explained of less than 1% (Supplementary Fig. 11 ). This shows that our training data have sufficient redundancy to ensure that our model conclusions are robust. Similarly, to determine whether our random-forest models would make similar predictions if data were equally distributed among continents, we rarefied our aggregated grid of symbiotic states and predictor layers to an even depth. Specifically, we sub-sampled all continents—North America (including Central America and the Caribbean), South America, Europe, Asia and Oceania—to match the number of grid pixels from Africa ( n = 50). This is a much more aggressive reduction of training data than is typically used in K -fold cross-validations, as it involves dropping ~90% of training data rather than retaining the same amount. We performed 99 iterations of rarefaction each for the three symbiotic guilds. On average, models run with the rarefied data explained about 10% less variance over the full training data (the entire predictor/response grid) than did models run with all of the training data (Supplementary Figs. 12 , 13 ). To avoid projecting our random-forest models outside the ranges of their training data (for example, grid cells with higher mean annual temperatures than the maximum used to fit the models), we subset a global grid of predictor layers depending on whether (1) the grid cell fell within the top 60% of land surface with respect to tree stem density 11 and either (2) fell within the univariate distribution of all the predictor layers from our training data and/or (3) fell within an 8-dimensional hypervolume defined by the unique set of the 4 best predictors of the relative abundance of each guild (Supplementary Fig. 14 ). We then projected our models across only those grid cells that met these criteria, which constitutes 46% of the global land surface and 88% of global tree stems (Fig. 1 , Supplementary Fig. 15 ). Model projections were made at two resolutions: 1°-by-1° and 0.5°-by-0.5° (Fig. 4 ). Although model validation indicates that our projections are robust, additional studies to ground-truth these predictions and identify any discrepancies would be valuable. If such discrepancies exist, they can help to fine-tune climate–symbiosis models, or identify areas in which climate might favour invasion by symbioses that have not yet evolved in or dispersed to a particular biogeographical region. We used the following equation to estimate the percentage of global tree stems that belong to each tree symbiotic guild: Σ i ((predicted proportion of trees of guild g in pixel i ) × (total number of tree stems in pixel i ))/Σ i (total number of tree stems in pixel i ). The proportion of tree stems and the proportion of tree basal area in each guild are highly correlated throughout the training data (Supplementary Fig. 4 ). The figures cited in the main text for each guild were calculated using model projections across all pixels, even those that did not meet the criteria for model projection because they fell outside the multivariate distribution of the predictor layers or had insufficient stem density. However, our estimates for the global percentage of trees occupied by each tree symbiotic guild change by <1% when using only those pixels that met our criteria for model projection. In the main text, we state that sharp transitions between dominant symbiotic states with climate variables could lead to declines in ectomycorrhizal trees, particularly in the southern range limit of the northern boreal forests. To determine this, we projected our random-forest models for each symbiotic guild using climate-change projections over our 19 bioclimatic variables (Supplementary Table 7 ), including the decomposition coefficients that use temperature and precipitation values. Specifically, we considered the 2070 scenario with a relative concentration pathway of 8.5 W per m 2 , which predicts an increase of greenhouse gas emissions throughout the twenty-first century 45 . We plot the difference in the proportion of forest basal area between the projections for 2070 and projections that use current climate data (Supplementary Table 7 , Supplementary Fig. 24 ). We qualify this prediction with the note that vegetative changes to forests are constrained by rates of mortality, recruitment and growth. After training and cross-validating our models with GFBi data exclusively, we additionally tested whether our models accurately predicted the previously published 46 symbiotic state of Eurasian forests. We assigned symbiotic status to all of the trees in this previous publication, and aggregated plot-level data to a 1°-by-1° grid using the same methods as with the GFBi dataset (Supplementary Fig. 25 ). We found that—on average—our models predicted the symbiotic state in the regional dataset within 13.6% of the value of this previously published dataset (Supplementary Fig. 26 ). For projected maps in Fig. 4a–c , we included the previously published 46 data with the GFBi training data to increase geographical coverage throughout Eurasia. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper. Data availability Information regarding symbiotic guild assignments, model selection (including global rasters of our model projections for ectomycorrhizal, arbuscular mycorrhizal and N-fixer proportion of tree basal area) and analyses is available as Supplementary Data. The GFBi database is available upon written request at . Any other relevant data are available from the corresponding authors upon reasonable request. Change history 28 June 2019 In this Letter, a middle initial and additional affiliation have been added for author G. J. Nabuurs; two statements have been added to the Supplementary Acknowledgements; and a citation to the French National Institute has been added to the Methods; see accompanying Author Correction for further details.

In and around the tangled roots of the forest floor, fungi and bacteria grow with trees, exchanging nutrients for carbon in a vast, global marketplace. A new effort to map the most abundant of these symbiotic relationships—involving more than 1.1 million forest sites and 28,000 tree species—has revealed factors that determine where different types of symbionts will flourish. The work could help scientists understand how symbiotic partnerships structure the world's forests and how they could be affected by a warming climate. Stanford University researchers worked alongside a team of over 200 scientists to generate these maps, published May 16 in Nature. From the work, they revealed a new biological rule, which the team named Read's Rule after pioneer in symbiosis research Sir David Read. In one example of how they could apply this research, the group used their map to predict how symbioses might change by 2070 if carbon emissions continue unabated. This scenario resulted in a 10 percent reduction in the biomass of tree species that associate with a type of fungi found primarily in cooler regions. The researchers cautioned that such a loss could lead to more carbon in the atmosphere because these fungi tend to increase the amount of carbon stored in soil. "There's only so many different symbiotic types and we're showing that they obey clear rules," said Brian Steidinger, a postdoctoral researcher at Stanford and lead author of the paper. "Our models predict massive changes to the symbiotic state of the world's forests—changes that could affect the kind of climate your grandchildren are going to live in." One of three maps showing the distribution of trees likely to associate with the three major types of symbiotic bacteria or fungi. Credit: Brian Steidinger Three symbioses Hidden to most observers, these inter-kingdom collaborations between microbes and trees are highly diverse. The researchers focused on mapping three of the most common types of symbioses: arbuscular mycorrhizal fungi, ectomycorrhizal fungi and nitrogen-fixing bacteria. Each of these types encompasses thousands of species of fungi or bacteria that form unique partnerships with different tree species. Thirty years ago, Read drew maps by hand of where he thought different symbiotic fungi might reside, based on the nutrients they provide. Ectomycorrhizal fungi feed trees nitrogen directly from organic matter—like decaying leaves—so, he proposed, they would be more successful in cooler places where decomposition is slow and leaf litter is abundant. In contrast, he thought arbuscular mycorrhizal fungi would dominate in the tropics where tree growth is limited by soil phosphorous. Research by others has added that nitrogen-fixing bacteria seem to grow poorly in cool temperatures. Testing Read's ideas had to wait, however, because proof required gathering data from large numbers of trees in diverse parts of the globe. That information became available with the Global Forest Biodiversity Initiative (GFBI), which surveyed forests, woodlands and savannas from every continent (except Antarctica) and ecosystem on Earth. The team fed the location of 31 million trees from that database along with information about what symbiotic fungi or bacteria most often associates with those species into a learning algorithm that determined how different variables such as climate, soil chemistry, vegetation and topography seem to influence the prevalence of each symbiosis. From this, they found that nitrogen-fixing bacteria are probably limited by temperature and soil acidity, whereas the two types of fungal symbioses are heavily influenced by variables that affect decomposition rates—the rate at which organic matter breaks down in the environment—such as temperature and moisture. One of three maps showing the distribution of trees likely to associate with the three major types of symbiotic bacteria or fungi. Credit: Brian Steidinger "These are incredibly strong global patterns, as striking as other fundamental global biodiversity patterns out there," said Kabir Peay, assistant professor of biology in the School of Humanities and Sciences and senior author of the study. "But before this hard data, knowledge of these patterns was limited to experts in mycorrhizal or nitrogen-fixer ecology, even though it is important to a wide range of ecologists, evolutionary biologists and earth scientists." Although the research supported Read's hypothesis—finding arbuscular mycorrhizal fungi in warmer forests and ectomycorrhizal fungi in colder forests—the transitions across biomes from one symbiotic type to another were much more abrupt than expected, based on the gradual changes in variables that affect decomposition. This supports another hypothesis, the researchers thought: that ectomycorrhizal fungi change their local environment to further reduce decomposition rates. This feedback loop may help explain why the researchers saw the 10 percent reduction in ectomycorrhizal fungi when they simulated what would happen if carbon emissions continued unabated to 2070. Warming temperatures could force ectomycorrhizal fungi over a climatic tipping point, beyond the range of environments they can alter to their liking. Mapping collaboration The data behind this map represents real trees from more than 70 countries and collaboration, led by Jingjing Liang of Purdue University and Tom Crowther of ETH Zürich, between hundreds of researchers who speak different languages, study different ecosystems and confront different challenges. "There are more than 1.1 million forest plots in the dataset and every one of those was measured by a person on the ground. In many cases, as part of these measurements, they essentially gave the tree a hug," said Steidinger. "So much effort—hikes, sweat, ticks, long days—is in that map." The maps from this study will be made freely available, in hopes of helping other scientists include tree symbionts in their work. In the future, the researchers intend to expand their work beyond forests and to continue trying to understand how climate change affects ecosystems.

10.1038/s41586-019-1128-0

Chemistry

Researchers devise cheaper, faster way to continuously produce amines

Recyclable Cooperative Catalyst for Accelerated Hydroaminomethylation of Hindered Amines in a Continuous Segmented Flow Reactor, Nature Communications (2022). DOI: 10.1038/s41467-022-30175-0 Journal information: Nature Communications

https://dx.doi.org/10.1038/s41467-022-30175-0

https://phys.org/news/2022-05-cheaper-faster-amines.html

Abstract Synthesis of hindered amines using the atom-efficient hydroaminomethylation (HAM) route remains a challenge. Here, we report a general and accelerated HAM in segmented flow, achieved via a cooperative effect between rhodium (Rh)/N-Xantphos and a co-catalyst (2-Fluoro-4-methylbenzoic acid) to increase the reactivity by 70 fold when compared to Rh/Xantphos in batch reactors. The cooperation between Rh and the co-catalyst facilitates the cleavage of the H–H bond and drives the equilibrium-limited condensation step forward. Online reaction optimization expands the scope to include alkyl, aryl, and primary amines. In-flow solvent tuning enables selectivity switching from amine to enamine without the need for changing the ligand. Furthermore, leveraging the ionic nature of the catalyst, we present a robust Rh recovery strategy up to 4 recycles without loss of activity. Introduction Amines are essential building blocks in agrochemicals, detergents, and fine chemicals 1 , 2 , 3 . For example, alkylated anilines and carbazoles are used as chromophores in synthesis of organic light emitting diodes (OLEDs) 4 , 5 , alkylated phenylenediamines are applied as antioxidants in rubber, and long chain fatty amines are components in personal care and lubrication formulations. Moreover, the amine moiety is common in several drugs, including Ibutilide, Melperone, Terfenadine, Aripiprazole, Fluspirilene, and Difenidol 1 . Carbon–nitrogen (C–N) bonds are widely common in bioactive molecules and thus, rapid C–N bond formation reactions are essential tools for the generation of large molecular libraries for structure-activity-relationship mapping (SAR), and the on-demand synthesis of isotopically labeled tracers with relatively short half-life such as the 11 C labeled PET tracers 6 , 7 , 8 . The emerging trend towards decentralized manufacturing of pharmaceuticals and fine chemicals promotes continuous flow production as the preferred method of choice 9 , 10 , 11 . Four conventional processes are typically used in the large scale synthesis of aliphatic amines: (1) amination of alcohols, (2) amination of alkyl halides, (3) reduction of nitrile compounds, and (4) reductive amination of carbonyl compounds 3 , which all require multistep processing and purification. HydroAminoMethylation (HAM) is an alternative one-pot synthetic route for sustainable manufacturing of amines from economic starting materials (olefins) without the need for the multistep rection-separation as needed in routes (1), (3), and (4) or the generation of metal halide solid waste that requires disposal as in route (2). HAM is an atom-efficient route as it produces no waste other than water as the sole byproduct. Substitution on the amine α carbon is often introduced in active pharmaceutical ingredients to control their lipophilicity and rate of metabolism 12 . However, hindered amines are usually synthesized by less economic routes such as the amination of alkyl halides. The HAM reaction proceeds through three main steps in an auto-tandem fashion (Fig. 1a ): (i) hydroformylation, (ii) condensation, and (iii) enamine hydrogenation. Steps (ii) and (iii) are often referred to as the reductive amination of aldehydes. Steps (i) and (iii) are gas–liquid reactions and thus provide an exciting opportunity for intensification through flow chemistry (segmented gas–liquid flow format). Step (ii) is a liquid phase condensation reaction that can be equilibrium-limited with sterically hindered amines and thus the undesired aldehyde self-condensation becomes the major reaction. This is a significant drawback that has limited the scope of HAM reactions to non-hindered amines 13 , 14 . Recent studies have demonstrated advances in the rate acceleration and substrate scope expansion of HAM reactions 15 , 16 . However, examples of HAM with sterically hindered amines are rare, and alternative complex routes are developed specifically for their alkylation 17 . The cost associated with catalyst is another drawback that limits the use of HAM in large scale production. In addition to the increasing demand for expanding the substrate scope of HAM reactions to sterically hindered amines, there is a need for continuous operation of HAM with recycling the expensive rhodium (Rh) catalyst to enhance the process sustainability and promote decentralized manufacturing of pharmaceuticals and fine chemicals. Fig. 1: HAM reaction scheme and a summary of prior HAM work vs. this work. a Amine formation by tandem hydroformylation-condensation-hydrogenation reactions. M: metal, L: ligand. b – e Different approaches to amine syntheses via HAM reaction. Full size image The application of the bidentate, commercially available phosphine ligand, Xantphos, in the Rh-catalyzed HAM reactions to maintain high selectivity towards the linear amine was pioneered by Beller and co-workers 18 . However, the HAM reaction is still plagued by the use of expensive cationic Rh(cod) + BF 4 − catalyst and the slow kinetics (batch reaction time > 24 h at 115 °C, TOF < 200 mol/mol Rh/h) which makes it unsuitable for continuous synthesis (Fig. 1b ). The reaction kinetics were enhanced by Hartwig and co-workers. through the addition of the iridium (Ir)-based Xiao’s hydrogenation catalyst and sodium formate to accelerate the enamine hydrogenation step (Fig. 1a , step iii), resulting in a complete conversion in 20 h at 80 °C (Fig. 1c ) 19 . Despite the addition of 1 mol% of Xiao’s Ir catalyst, the hydroformylation step (Fig. 1a , step i) remained slow. To overcome the slow rate of the hydroformylation step, the Rh catalyst loading was increased by 5 times relative to the Rh loading in the study by Beller and co-workers, which resulted in TOF values below 50 mol/mol Rh/h. A more economic Rh source, Rh(acac)(CO) 2 , was used by Hartwig and co-workers. and the bidentate phosphene ligand, BISBI, showed better yields when compared to Xantphos under similar reaction conditions. Despite the significant advancements of the HAM synthetic route in batch reactions, no α-branched amines have been demonstrated. This major limitation arises from the equilibrium-limited condensation and slow hydrogenation of the branched enamine in batch reactors. Wasserscheid and co-workers demonstrated that Rh/Xantphos catalyst for continuous gas phase HAM reaction (TOF values up to 500 mol/mol Rh/h) could be recycled by immobilizing the catalyst in supported ionic liquid phase on activated carbon 20 . Recently, continuous HAM of 1-decene with the non-hindered diethylamine was demonstrated by Seidensticker and co-workers with catalyst recycling (0.1 mol% Rh loading and SulfoXantphos ligand) in a pressurized continuously stirred tank reactor (Fig. 1d ) 21 . TOF values up to 200 mol/mol Rh/h was achieved and the catalyst recycle was accomplished by carrying out the reaction in a thermomorphic multiphase solvent system which allowed for product/catalyst separation upon reaction cooling. In this work, we present a synthetic route for accelerated HAM reactions with hindered amines by leveraging a cooperative effect between Rh/N-Xantphos and fluorinated benzoic acid catalyst. Compared to the traditional Rh/Xantphos system, the developed cooperative catalyst increases the kinetics of HAM reactions by up to 70 fold when conducted in flow, while achieving high linear/branched selectivity ( l / b ). Furthermore, we demonstrate HAM turnover frequencies (TOF) up to 10,000 mol/mol Rh/h with the cooperative catalyst when the reaction is conducted continuously in a gas–liquid segmented flow reactor at moderate temperature and pressure (115 °C and 28 bar). The segmented flow refers to a regular train of alternating gas–liquid segments continuously moving along the flow reactor, which is commonly referred to plug flow format (i.e., gas plugs surrounded by a continuous liquid phase) 22 . The synergistic effect of the developed cooperative catalyst system with the heat and mass transfer advantages of the segmented flow reactor allows for the HAM reaction with sterically hindered amines at good to excellent yields (71–92%) in less than 3 h with 0.1 mol% Rh(acac)(CO) 2 . Our mechanistic investigations reveal that the HAM rate acceleration results from the cooperation between the Rh/N-Xantphos and the benzoic acid to lower the energy barrier for the H–H bond cleavage to form the Rh hydride needed to drive the enamine hydrogenation. Rapid enamine hydrogenation drives the equilibrium-limited condensation with α-branched amines in the forward direction and avails Rh to catalyze the olefin hydroformylation. Gas–liquid segmented flow format enhances the hydrogen transfer rate into the liquid phase and maximizes the concentration of the active Rh–H species. We demonstrate that the gas–liquid segmented flow reactor can enable selective acceleration of the slow elementary step in the HAM reaction, which is the enamine hydrogenation, but not the undesired aldol condensation. Through reaction condition optimization, we extended the substrate scope of HAM to alkyl and aryl amines as well as primary amines and styrenes. Another unique aspect of the developed cooperative catalyst is its recyclability. Leveraging the ionic nature of the cooperative catalyst system, we demonstrate catalyst precipitation and recycle for up to 4 cycles. Additionally, through on-the-fly solvent switching from methanol/toluene mixture to toluene we present an on-demand selectivity switching from HAM to hydroaminovinylation without the need for ligand replacement. The developed cooperative catalyst in combination with the scalable segmented flow reactor can facilitate decentralized continuous manufacturing of a wide range of APIs and fine chemicals through the atom-efficient HAM reactions with a high degree of flexibility and catalyst recyclability. Results Reaction optimization in flow We began our investigations of the HAM synthetic route by studying the effect of ligand (L) and co-catalyst on the HAM of 1-octene with piperidine to form N-nonylpiperidine ( 1a ) using a more economic catalyst precursor Rh(acac)(CO) 2 instead of Rh(cod)BF 4 . We developed a gas–liquid segmented flow reactor, shown in Fig. 2 , for in-flow studies of the HAM reactions (see “Methods”, General Procedure 1). Fig. 2: Schematic illustration of the flow chemistry platform utilized for the accelerated HAM reactions. The sample collection chamber allows for liquid sampling at different reaction (residence) times by varying the total feed flow rate. MFC: Mass Flow Controller, BPR: Back Pressure Regulator, DPR: Digital Pressure Regulator, 2F: 2-fluoro-4methylbenzoic acid. Full size image Seeking to address the aforementioned challenges of BISBI and Xantphos ligands in HAM reactions and inspired by the cooperative hydrogenation work of de Bruin and co-workers 23 , we wondered if a benzoic acid co-catalyst in combination with the enhanced mass transport rate of the gas–liquid segmented flow reactor could provide a synergistic effect to accelerate the kinetics of HAM reactions. The co-catalyst screening was carried out with bidentate phosphine ligands because of the ligand previous success in HAM of olefins 18 , 19 . Table 1 presents a summary of the ligand and co-catalyst screening experiments of HAM of 1-octene. Xantphos resulted in an amine yield of 6.6% after 15 min residence (reaction) time (Table 1 , entry 1). Higher yield and l / b were obtained with BISBI ligand. However, the yield of the unreacted aldehyde and enamine remained significant (Table 1 , entry 2). Changing the substitution on the phosphorous from phenyl to alkyl or amine did not result in an enhanced activity (Table 1 , entries 3 and 4). Switching to N-Xantphos ligand increased the amine yield by 4 times to reach 28.8%. However, a significant portion of the substrate were accumulated as the enamine intermediate, indicating that the overall HAM transformation is limited by the enamine hydrogenation step (Fig. 1a , step iii). N-Xantphos, in comparison with Xantphos, increased the amine l / b ratio from 50 to 98 and the enamine to aldehyde ratio from ca . 2 to 5 (Table 1 , entry 4), indicating the superior performance of N-Xantphos in terms of hydroformylation linear selectivity and enamine hydrogenation activity. Without any co-catalyst addition, the only side product observed were cis- and trans- 2 octene. Table 1 Ligand and co-catalyst screening of HAM reaction in segmented flow. Full size table Interestingly, addition of 0.5 mol% of the co-catalyst, 4-trifluoromethylbenzoic acid, to the Rh/N-Xantphos catalyst increased the amine yield to 71.8% and decreased the yield of the aldehyde and enamine intermediates down to zero and 4.5%, respectively (Table 1 , entry 6). This result indicates a catalytic effect of the benzoic acid additive on either the condensation or the hydrogenation step (Fig. 1a , steps ii and iii). However, introduction of the co-catalyst decreased the amine l / b selectivity from 97 to 55. Building on the superior performance of the Rh/N-Xantphos system in the presence of the co-catalyst, in the next set of experiments we explored the effect of the co-catalyst structure on the HAM reaction (Supplementary Table 1 ). High-throughput flow screening of a rationally selected library of 10 different benzoic acids unveiled the fluorine substitution at the ortho position of the benzoic acids results in the highest amine l / b . This result could be attributed to the formation of the less sterically hindered Rh carboxylate species that contribute to the pro-branched hydroformylation without fluorine substitution at the ortho position. In particular, the highest amine yield was achieved with 2-fluoro-4-methylbenzoic acid 2F (Table 1 , entry 7). To demonstrate the rate acceleration by the segmented flow reactor, in a separate experiment, we conducted the HAM reaction in a batch reactor using the co-catalyst condition of entry 7 in Table 1 . The batch HAM reaction (Table 1 , entry 8) resulted in a significantly lower 1-octene conversion, amine yield, and amine l / b selectivity compared to the segmented flow reactor. The higher 1-octene conversion and amine/enamine ratio in the flow reactor supports the accelerating effect of the gas–liquid segmented flow format on the tandem hydroformylation and hydrogenation steps (Fig. 1a , steps i and iii). After identifying N-Xantphos and 2F as the optimal ligand and co-catalyst for the proposed accelerated HAM synthetic route, we studied the effect of the co-catalyst loading as well as the solvent and gas composition on the amine yield and selectivity. This systematic study (Fig. 3 ) allowed us to investigate and elucidate the effect of each reaction parameter on the three elementary steps of the HAM reaction; hydroformylation, condensation, and hydrogenation. The amine yield monotonically increased with increasing 2 F loading from zero to 0.6 mol% to reach a maximum of 86%, while the amine l / b ratio decreased from 117 to 59. The increase in the amine yield was accompanied by a decrease in the aldehyde yield and enamine intermediates (0% at 0.6 mol% and 0.9 mol% of 2F ), shown in Fig. 3a . No evidence of benzoic acid esterification or hydrogenation was observed at the highest 2F loading. The solvent composition also has a significant impact on the HAM reaction when 2F is added at 2 mol%. In the absence of methanol (i.e., toluene only), both hydroformylation and enamine hydrogenation (Fig. 1a , steps i and iii) were slow as indicated by the incomplete conversion of 1-octene (79%), and the high yield of the enamine (55%), shown in Fig. 3b . Gradual increase in the methanol volumetric ratio results in suppression of the unreacted 1-octene and enamine (0% unreacted 1-octene and enamine at the methanol:toluene volumetric ratio of 3.77). The aldehyde yield increases with increasing methanol content and reaches a maximum at methanol:toluene volumetric ratio of 0.83, before it decreases to 0% at methanol:toluene volumetric ratio of 3. The increase in the aldehyde yield can be attributed to the acceleration in the rate of hydroformylation at low methanol content and is reversed by the acceleration in the rate of condensation at the high methanol content. Interestingly, no branched product was observed at the low methanol content up to methanol:toluene volumetric ratio of 2.3. Although further increasing of the volumetric ratio of methanol increases the amine yield, it results in a lower amine l / b ratio. Increasing the methanol:toluene volumetric ratio beyond 3.7 does not result in an increase in the amine yield but reduces the amine l / b . Furthermore, varying the reaction solvent composition did not affect the yield of n-octane and 2-octene isomers. Fig. 3: In-flow reaction optimization. Optimization of the HAM reactions for a 2F loading at MeOH:MePh volume ratio of 5 and H 2 :CO ratio of 3.5, b MeOH:MePh volumetric ratio at 2 mol% 2F and H 2 :CO ratio of 3.5, and c H 2 :CO ratio at 2 mol% 2F and MeOH:MePh volume ratio of 4. General HAM reaction conditions: 125 °C, 15 min residence time, 24 barg total pressure, 3.5 gas:liquid volumetric ratio, 0.45 M 1-octene, 0.1 mol% Rh(acac)(CO) 2 , and 0.4 mol% N-Xantphos. Full size image The results shown in Table 1 suggest that the HAM reaction is limited by the enamine hydrogenation. Since the required stoichiometric ratio of H 2 to CO for a complete HAM transformation is 2, in the next set of experiments we studied the effect of H 2 :CO ratio from 2 to 5, while maintaining the total pressure at 24 barg. At H 2 :CO ratio of 2, an incomplete conversion of 1-octene was achieved (Fig. 3c ). Increasing the partial pressure of H 2 (H 2 :CO = 3) resulted in an increase in both 1-octene conversion and amine yield from 70 to 85% at 2 mol% loading of 2F . Further increase of the H 2 :CO ratio resulted in an increase of the 2-octenes yield, while decreased the amine l / b to 50. The results of the gas composition screening, shown in Fig. 3c , shows the H 2 :CO ratio of 3–4 achieves the best combination of high amine yield and reasonably high amine l/b (>20). In order to maximize the flow reactor throughput when continuously conducting the HAM reaction, we increased the concentration of the olefin substrate to 1 M, the pressure to 28 barg, and the gas:liquid volumetric ratio to 5; The gas:liquid volumetric ratio was increased to 5 to provide excess CO gas relative to the olefin and suppress n -octane formation. In order to suppress the amine yield loss due to olefin isomerization to 2-octene and hydrogenation to n -octane, we decreased the reaction temperature from 125 to 115 °C and achieved an amine yield of 92% (Figs. 4 , 1a ). Fig. 4: In-flow HAM substrate scope screening with the cooperative catalyst. Rh and N-Xantphos loading were 0.1 and 0.4 mol%, respectively. For each in-flow HAM reaction, 4 ml of the effluent liquid was collected, and the product was isolated and purified for analysis. The reported yields are the isolated yields, and the l/b values are measured by gas chromatography–mass spectrometry (GC-MS). Full size image Substrate scope With the optimized conditions in hand, we next investigated the scope of the accelerated HAM reaction with the cooperative catalyst in flow (Fig. 4 ). Specifically, in the substrate scope study, we focused on more challenging a-branched amines. Using the optimized cooperative catalyst, continuous synthesis of N-nonylpiperidine (1a) at 0.93 mmol/h was achieved with in only 18 min residence time (92% yield). Morpholine was also alkylated in 18 min residence time to produce 1b with a yield of 86.6%. The residence time needed to alkylate the acyclic dihexylamine with the cooperative catalyst to obtain 1c at 83.6% yield had to be increased to 30 min to account for the slower hydroformylation observed with this amine. For comparison, the batch alkylation of dihexylamine under similar reaction conditions to the in-flow HAM with the cooperative catalyst resulted in only 48% yield and amine l / b of 13 after 4 h. This result further supports the superior performance of the segmented flow reactor in promoting both HAM reactivity and selectivity. HAM of anilines results in products and intermediates commonly applied in pharmaceuticals, plastics, and rubber additives, as well as OLEDs 1 , 2 , 3 , 4 , 5 . In-flow HAM of N-methylaniline with the cooperative catalyst resulted in 86% yield of 1d and l / b of 159. The more sterically hindered 2-methylpiperidine was successfully alkylated in the segmented flow to give 79% yield of 1e and l / b of 213. The 2F loading was increased from 2 to 5 mol% to accelerate the relatively slow reaction with the hindered enamine. Under these conditions, no aldol products were detected, which further supports the effectiveness of 2F in selectively catalyzing the reductive amination with the hindered amine over the aldehyde self-condensation. For comparison, batch alkylation of 2-methylpiperidine under similar conditions to the flow reactor resulted in 74% yield of 1e with l / b of 11 after 24 h reaction time. This result of batch vs. flow HAM of the hindered amines illustrates the effectiveness of the segmented flow reactor in promoting both the HAM reactivity and selectivity for the sterically hindered amines. The substrate scope was then successfully extended to 2-phenylpiperidine with a 89% yield of 1f and l / b of 87 in only 18 min, using 5 mol% of the co-catalyst 2F . Encouraged by the results obtained with the slightly hindered 2-substituted piperidines, we moved to the more hindered dicyclohexylamine. To our delight, dicyclohexylamine was successfully alkylated at 71% yield of 1 g with 10 mol% of the co-catalyst 2F at the residence time of 130 min in the segmented flow reactor. To the best of our knowledge, this is the only HAM report of hindered amines at this high rate and selectivity. Attempting to extend the scope to 2,2,4,4-tetremethylpiperidine failed (no 1 h product was formed) and resulted in the accumulation of the unreacted aldehyde. Phenylaniline was also alkylated in flow with the cooperative catalyst (72% yield of 1i with l / b of 160). For this relatively hindered, less basic substrate, 2F loading had to be increased to 10 mol % and the residence time to 130 min. Next, HAM of piperidine with styrene was performed at 81% yield ( 1j ) and l / b of 2.1. The relatively low l / b selectivity for product 1j is attributed to the stabilization of the benzylic Rh species induced by the η2 electron donation from the arene ring 24 . Continuous HAM of the disubstituted olefin, α-methylstyrene, was accomplished with the cooperative catalyst at an increased temperature (135 °C) and residence time (120 min), resulting in a 74% yield of 1k and l / b of 66. Additional in-flow HAM optimization of the primary amine, aniline, revealed that a higher yield can be obtained when 2F loading is lowered to 1 mol % to suppress the double alkylation of the amine. At this lowered loading of the co-catalyst, the residence time in the segmented flow reactor was increased to 30 min to obtain a 77% yield of the monoalkylated product 1l and l / b of 35. Next, we investigated in-flow HAM with cyclohexylamine. Our in-flow screening experiments revealed that the optimal HAM of 1-octene with cyclohexylamine can be achieved when the reaction is carried out in two tandem steps with the in-line addition of the co-catalyst 2F for the second step. The H 2 :CO ratio was maintained at 2 for the first step to selectively form the imine N-nonylcyclohexylimine, while the second step was performed under 26 barg of H 2 for 14 min to obtain the desired monoalkylated product 1m with a 79% yield and l / b of 62. In-flow HAM of carbazole could not be achieved because of the ultra-low solubility of carbazole in the reaction solvent used in this study at room temperature. Moreover, ammonia could not be alkylated under the developed HAM synthetic route because of the slow imine hydrogenation relative to condensation side reactions. The intramolecular HAM of o -isopropenylaniline resulted in low yield of the desired product, 4-methyl-1,2,3,4-tetrahydroquinoline, because of the competing intermolecular reaction. Catalyst recycling Owing to the high cost of Rh catalysts, it is necessary to maximize the productivity of the catalyst before sending for precious metal reclaiming 25 , 26 . During the product isolations of the in-flow HAM reactions, we observed that an orange solid precipitates upon solvent removal from the crude product mixture under vacuum. In most cases, the precipitated solids were insoluble in the product amine. Upon addition of pentane to the residue after solvent removal, the powder precipitates and can be separated by settling or centrifugation. Additional washing of the powder with more pentane followed by pentane removal allows for complete separation of the product amine from the powder, as shown in Fig. 5a . To assess the HAM reactivity of the isolated powder, the solid precipitate was dried under vacuum and then dissolved in the reaction solvent (MeOH:MePh volumetric ratio of 4:1) at room temperature under inert atmosphere. Fresh reactants were then added to the powder, and the HAM reaction was performed under the same reaction conditions. A rapid decay in activity was observed over the first three recycles (Fig. 5b ). Both yields and l / b selectivity of amine 1a decreased after catalyst recycling (from 93% yield to less than 10% and l / b of 60 to 5), while the yield of the enamine 2a and the aldol side products increased. Incomplete consumption of 1-octene was also observed after the third catalyst recycling. The loss in activity was first observed in the hydrogenation step, followed by the hydroformylation step. We postulated that the observed decrease in the HAM activity after the catalyst recycling was due to the loss of the co-catalyst 2F during the recycling process. To validate this hypothesis, we replenished the reaction mixture with fresh 1 mol% of 2F following each recycle round. When catalyst recycle was performed with the addition of 1 mol % of 2F , an excellent recyclability and activity of the catalyst system was achieved, Fig. 5c . The amine yield and l / b using the developed HAM synthetic route and catalyst recycling process were maintained higher than 85% and 55, respectively, after three rounds of catalyst recycles. We attribute the enhanced recyclability of the Rh/N-Xantphos catalyst upon the addition of the co-catalyst 2F to stabilization of the ionic Rh carboxylate complex (Fig. 5 ), that is insoluble in long-chain, low polarity amines. Furthermore, the fluorine substitution on 2F could also contribute to the catalyst recyclability by minimizing the solubility of the Rh complex in the product amines. The Rh loss to the amine product was found to be less than 5% of the initial mass of Rh (measured by ICP-MS). This concept has previously been utilized to synthesize fluorinated ligands that can facilitate Rh extraction from organic media in hydroformylation reactions 27 . The presence of the co-catalyst 2F , in addition to reducing the catalyst solubility, enhances the stability of the ionic species when exposed to air, as evident by the persistence of the orange color of the powder when fresh 2F was added after each round of the catalyst recycle, that is contrary to the change in the powder color to pale green without fresh 2F addition. Recycling of Rh from amination solutions are less explored in the literature 20 , 21 , 28 . This study illustrates the effectiveness of a commercially available co-catalyst in accelerating the HAM reaction and allowing for catalyst recycling at the same time. Fig. 5: Catalyst recycling scheme. a Schematic of the developed catalyst recycling process. HAM performance of the Rh/N-Xantphos catalyst recycled without ( b ) and with c the addition of 2F. Full size image On-demand selectivity switching to enamine Enamines are nucleophilic reactants that can undergo alkylation or acylation with alkyl halides and acyl halides, respectively, and thus are widely used in organic synthesis 29 . Selective formation of enamines by condensation of an aldehyde with an amine often requires the slow addition of the aldehyde or the addition of the amine in large excess to suppress the aldehyde self-condensation. The hydroaminovinylation of the olefin with the amine is an alternative strategy that allows for the one-pot, auto-tandem hydroformylation-condensation route that suppresses the formation of the undesired aldol. The main limitation in this strategy is that the formed enamine often undergoes hydrogenation under the hydroformylation conditions. The ligand NAPHOS was shown to be more selective to enamine formation than Xantphos in toluene at 65 °C with enamine l / b selectivity exceeding 90 30 . However, the reaction required 16 h to reach completion. More recently, hydroaminovinylation was shown to be driven to completion in 1 h at 130 °C in solvent-free conditions with a diphosphite ligand, but with enamine l / b values lower than 25 31 . To address these limitations, we explored the performance of N-Xantphos ligand in the hydroaminovinylation of 1-octene with piperidine in flow under the optimized HAM conditions developed in this work. Without the co-catalyst 2F , the yield of enamine 2a increased from 0 to 32%. However, the yield of amine 1a remained high at 42% (Supplementary Fig. 1 ). Lowering the reaction temperature to 95 °C and the H 2 :CO ratio to 1 did not completely suppress 2a hydrogenation to 1a , but drastically reduced the hydroformylation reactivity (less than 15% yield of 2a + 1a ). Through our solvent screening experiments, we discovered that on-the-fly solvent switching to 100% toluene could suppress 2a hydrogenation and maintain 2a yield at 55% with the remaining yield being unreacted aldehyde and olefins. Increasing the co-catalyst loading to 2 mol% and reaction temperature to 125 °C, further increased the yield of 2a to 74% with an enamine l / b of 90 (Supplementary Fig. 1 ). Inspired by this finding, we demonstrated the unique ability of the flow reactor to switch on-demand from HAM synthetic route to hydroaminovinylation (Fig. 6 ) with the same cooperative catalyst (N-Xantphos ligand and 2F ). Fig. 6: Online reaction product switching. On-the-fly selective switching from the amine to enamine formation with Rh/N-Xantphos in the presence of the co-catalyst 2F in the segmented flow reactor. G:L is gas to liquid volumetric feed ratio, t : residence time. Full size image Mechanistic study Next, we investigated the mechanism by which the cooperative catalyst (Rh/N-Xantphos + 2F ) rapidly catalyzes the HAM. The Rh-catalyzed hydrogenation of enamine 2a was performed with N-Xantphos and Xantphos ligands to compare the hydrogenation activity of the two ligands under CO/H 2 atmosphere (Table 2 ). The reaction was performed at 105 °C and 20 barg to slow down the hydrogenation and allow for accurate reactivity comparison in batch and flow reactors. The yield of the amine product with N-Xantphos was 39% which is 3 times more than the yield obtained with Xantphos after 30 min in the batch reactor (Table 2 , entries 1 and 2). The higher hydrogenation activity with N-Xantphos could be attributed to the electron donating character of the –NH moiety 32 , that can lower the barrier to the oxidative addition of the H–H bond. Addition of 2 mol% of the co-catalyst 2F resulted in an enhanced hydrogenation activity with both ligands (Table 2 , entries 3, and 4). This result suggested that the accelerating effect realized by 2F is not due to protonation of the –NH moiety on the N-Xantphos by the acid. Indeed, the 1 H and 31 P NMR shifts of the ligand N-Xantphos in deuterated methanol/toluene solvent are essentially unchanged when the acid 2F or the amine dicyclohexylamine are added (see Supplementary Table 2 ). It is worth noting here that deprotonation of the ligand N-Xantphos by a strong base resulted in an improved activity towards oxidative addition to aryl chlorides 33 . However, it is unlikely that this effect occurs under the less basic HAM conditions as supported by the NMR data in Supplementary Table 2 . The amine yield by addition of the co-catalyst 2F was increased to 99% in only 10 min of the reaction time, when the reaction was performed in the segmented flow reactor (Table 2 , entry 5). Table 2 Enamine hydrogenation activity of N-Xantphos vs. Xantphos with and without 2F. Full size table Contrary to hydrogenation, there is no catalytic effect of 2F on piperidine condensation with n-nonanal in the absence of Rh catalyst; the yield of enamine 2a slightly decreased from 64 to 58% upon 2F addition to a 60 min batch condensation reaction (Supplementary Table 3 ). The decrease in the enamine yield was accompanied by an increase in the C 18 aldol products (from ca . 12 to 22%). The boost in the desired reaction was only realized when 2F was added in the presence of Rh catalyst which resulted in an increase in the combined yield of 1a and 2a to 92% and decreased the aldol yield to less than 5% (Supplementary Table 3 ). The same observation applies to the condensation/reductive amination of the sterically-hindered dicyclohexylamine with n-nonanal. Without Rh catalyst, the combined yield of the amine and enamine remained at 3% with and without the co-catalyst 2F (Supplementary Table 4 , entries 1 and 2). Addition of the Rh/N-Xantphos catalyst increased the amine yield to 20% (Supplementary Table 4 , entry 3). The increase in the amine yield is because the condensation step is equilibrium-limited, and the Rh addition helps drive the condensation equilibrium in the forward direction by hydrogenating the enamine product. The combined yield almost doubled to 41% when both 2F and Rh/N-Xantphos catalyst were added (Supplementary Table 4 , entry 4). This result further supports the cooperative effect of the co-catalyst and Rh/N-Xantphos on catalyzing the enamine hydrogenation and thus driving the condensation equilibrium forward. Following the identification of the cooperative effect of 2F with Rh/N-Xantphos on catalyzing the reductive amination, we proceeded with hydrogen-deuterium (H/D) scrambling experiments. The H/D ratio of the hydrogenation step represents the kinetic isotope effect on the reductive amination. Changing the gas from H 2 to D 2 decreased the % hydrogenation from 50.2 to 39.1% (Table 3 , entries 1 and 2). The H/D ratio was found to be 1.28, indicating that the hydrogenation reaction could be limited by a hydride formation or hydride delivery step, but not the protonation of the Rh-aminyl complex or the enamine coordination (see Fig. 7 for the proposed mechanism). The decrease in the H/D ratio to 1.18 in the presence of the co-catalyst indicates the lower hydridic nature of the rate-limiting step upon the addition of 2F . The same trend was observed in the reductive amination with the less hindered piperidine, where the measured H/D ratio of the dehydrogenation decreased from 1.13 to 1.03 upon the addition of the co-catalyst 2F (Supplementary Table 5 ). Table 3 H/D effect on the reductive amination of dicyclohexylamine with 2F + Rh/N-Xantphos. Full size table Fig. 7: Proposed reaction mechanism. Proposed mechanism of the HAM reaction with the cooperative catalyst developed in this work. Full size image The olefin hydroformylation reactivity was not affected by 2F addition or switching between H 2 and D 2 (Supplementary Table 6 ). This result suggests that the hydroformylation with Rh/N-Xantphos ligand is not limited by hydride formation or delivery as indicated by the H/D ratio of 1.03. Earlier studies with N-Xantphos showed that the hydroformylation is limited by olefin coordination or CO dissociation 32 , which is in agreement with the results shown in Supplementary Table 6 . The hydroformylation experiment was performed without amine addition to eliminate the effect of enamine hydrogenation on the hydroformylation reactivity. It should be noted that the N-Xantphos ligand has been shown to be slower than Xantphos in the hydroformylation of 1-octene under typical hydroformylation conditions 32 . However, the N-Xantphos is faster in 1-octene hydroformylation under the HAM synthetic route developed in this study. This dichotomy is attributed to the faster enamine hydrogenation by N-Xantphos, and thus promoting the Rh catalyst to catalyze olefin hydroformylation under this HAM synthetic route. Same explanation applies to the indirect accelerating effect of 2F on the hydroformylation rate under HAM conditions where the co-catalyst accelerates the enamine hydrogenation, resulting in enhanced availability of Rh for the hydroformylation step of the HAM synthetic route. From our results of the H/D scrambling and control experiments, we propose that the HAM reaction, developed in this work, proceeds via the formation of Rh/N-Xantphos- 2F complex i formed form the loss of CO and the deprotonation of 2F by the amine reactant, as shown in Fig. 7 . The heterolytic cleavage of the H-H bond is believed to be more facile than the homolytic oxidative addition that results in Rh dihydride. Similar effect has been proposed for Co 23 , Ni 34 , and Ru 35 catalysts. Ligand exchange between 2F and the enamine results in the formation of the Rh hydride amine complex iii that can undergo hydride transfer to the unsaturated C=C bond. P -methyl carboxylic acids are less acidic than their unsubstituted analogs, and thus should undergo a more facile transition from species i to ii in the proposed reaction mechanism. This is also supported by the higher amine/enamine ratio by 2F in Supplementary Table 1 . Protonation of the formed Rh amine complex v by the ammonium cation results in the formation of the product amine and the Rh- 2F complex i . From the H/D scrambling experiments, it is reasonable to propose that the rate-limiting step is either the cleavage of the H–H (formation of ii ) bond or the hydride delivery (formation of iv ) in the enamine hydrogenation. The need for high H 2 pressure to drive the enamine hydrogenation in combination with the comparable hydrogenation obtained for the less sterically hindered amine 1a and the more sterically hindered amine 1g suggest that the oxidative addition of the H–H bond is the hydrogenation rate-limiting step 36 . The hydroformylation cycle proceeds via the formation of the rhodium carbonyl hydride vi followed by the olefin coordination, and the hydride addition to form complexes vii and viii, respectively. CO migratory insertion results in the formation of the acyl complex ix that can perform the oxidative addition of H–H bond followed by the reductive elimination to produce the aldehyde and the Rh hydride vi. The aldehyde condensation with the amine is not often regarded as a metal-catalyzed reaction, and thus the ligand structure does not play a role in this step. Instead, solvent polarity and medium acidity are often increased to catalyze this step 37 , 38 . However, the condensation of the aldehyde with sterically hindered amines is equilibrium-limited 39 , and thus, fast enamine hydrogenation results in faster condensation relative to the aldehyde aldol condensation, as demonstrated by the cooperative catalyst system developed in this study. Under HAM conditions, both hydroformylation and enamine hydrogenation occur simultaneously and share the available Rh catalyst, and thus their rates are expected to be lower than the rates obtained when each of the two reactions is performed separately. To measure the TOFs of both reactions that can be achieved in flow under typical HAM conditions, the catalyst loading was reduced from 0.1 to 0.01 mol% and the 1-octene reaction with piperidine was performed under variable residence times (ligand to Rh ratio at 4 and co-catalyst 2F loading at 2 mol %, at 115 °C and gas to liquid of 5). The residence (reaction) time was varied by tuning the total volumetric flow rate. The initial hydroformylation TOF was 9000 mol/mol Rh/h, while the HAM TOF was 4000 mol/mol Rh/h (Supplementary Fig. 2a ). To the best of our knowledge, this is the highest HAM TOF reported up to date. The HAM TOF could be further boosted in flow by increasing the gas to liquid volumetric ratio. High gas to liquid volumetric ratio results in smaller liquid segments that exhibit shorter H 2 diffusion length. The enhanced diffusion results in maximization of the Rh-hydride active species ii that catalyze enamine hydrogenation which promotes the Rh catalyst for catalyzing the hydroformylation step. When the gas to liquid volumetric ratio is 10, the HAM TOF approaches 10,000 mol amine/mol Rh/h (Supplementary Fig. 2B ). While an ultra-high TOF can be achieved at high gas to liquid volumetric ratios, the product throughput decreases. The choice of the optimum gas to liquid volumetric ratio and Rh loading in HAM segmented flow reactors depends on process economics. Discussion In summary, we introduced the fastest Rh-catalyzed hydroaminomethylation process known to date by leveraging a synergistic effect of the cooperative Rh/N-Xantphos and the co-catalyst (fluorinated benzoic acid 2F ), and the enhanced gas–liquid mass transfer in the segmented flow reactor as an enabling factor. The process furnishes cyclic and acyclic alkylamines, anilines, and morpholines in high yield and regioselectivity, 70 times faster than the state-of-the-art Xantphos-ligated Rh catalyst in a batch reaction. The HAM scope was expanded to include primary amines without the need for Ir hydrogenation catalyst and included hindered amines, such as alkylated dicyclohexylamine. Rapid enamine hydrogenation is the key feature in the developed HAM synthetic route and is enabled by the cooperative effect of the co-catalyst and Rh/N-Xantphos that facilitates the oxidative addition of the H–H bond to Rh. The rapid enamine hydrogenation drives the condensation equilibrium forward and enables the HAM of hindered amines over the undesired aldol products. Leveraging the ionic nature of the catalyst system (Rh/N-Xantphos and 2F ), we developed a robust recycle strategy of the increasingly expensive Rh metal and achieved TOF values exceeding 10,000 mol amine/mol Rh/h. On-the-fly tuning of the flow reactor conditions, enabled on-demand switching of the product selectivity from the amine to the enamine. The unique online reaction product switching from amine to enamine combined with the recyclability of the developed cooperative catalyst illustrates unique capabilities of this catalyst system for continuous manufacturing of hindered amines through the atom-efficient HAM route. The accelerated in-flow production of hindered amines through the scalable hydroaminomethylation synthetic route, developed in this work, will facilitate on-demand/on-site synthesis of a wide range of specialty compounds in chemical and pharmaceutical industries. Furthermore, the proposed catalyst cooperativity and recycling protocol of the N-Xantphos ligated precious metals with 2F is amenable to automation and will find applications in other challenging homogeneous catalytic reactions. Methods General Procedure 1: HAM in the segmented flow reactor The catalyst, dicarbonyl 2,4-pentanedionato rhodium(I), and the ligand, 4,6-Bis(diphenylphosphino)-10 H -phenoxazine (N-Xantphos), were dissolved in toluene under inert atmosphere inside a glovebox. Methanol was then added to dilute the catalyst/ligand solution to the target solvent composition. An 8 mL-stainless steel syringe connected to a Teflon tubing (fluorinated ethylene propylene (FEP), inner diameter (ID): 0.01″, outer diameter (OD): 1/16″) was filled with the catalyst and ligand solution under inert atmosphere. Another 8 mL-stainless steel syringe connected to a 0.01″ FEP tubing was filled with a mixture of the olefin, the amine, and the co-catalyst dissolved in methanol/toluene solvent under inert atmosphere. The syringe outlets were capped under inert atmosphere with Teflon screw caps (IDEX Health & Science) before being transferred outside of the glovebox and connected to a PEEK cross connection (IDEX Health & Science), where the solutions from the syringes were mixed at a pre-selected volumetric ratio before contacting the gas. During the ligand and co-catalyst screening experiments (results shown in Table 1 and Supplementary Table 1 ), an additional 8 ml stainless steel syringe was utilized to separately load the ligand or co-catalyst, thereby providing additional flexibility for online tuning of the co-catalyst and ligand loading. Gas–liquid segmentation was achieved by contacting the combined liquid stream with the gas flow in a stainless steel T-junction (1/8″ OD, Swagelok) before entering the stainless-steel flow reactor (1/8″ OD, 1/16″ ID, and 94 cm length). Liquid flowrates were controlled by syringe pumps (Harvard PHD ULTRA) and gas flowrates were controlled by individual mass flow controllers (EL-Flow ® , Bronkhorst). The residence time is calculated on cold basis using the flow reactor volume and the feed flow rates. Time-to-distance transformation was used to vary the residence (reaction) time by tuning the total volumetric flow rate at the same gas:liquid feed ratio. The flow reactor temperature was controlled through a hotplate and oil bath with a temperature probe immersed in the oil bath. The flow reactor pressure was controlled with a nitrogen (N 2 ) pressure connected to the control port of a backpressure regulator (Equilibar) integrated at the outlet of the flow reactor coil. The flow reactor effluent was passed through a 10-way selector valve (VICI, EUHB) and directed to a custom-designed waste collection chamber equipped with an exhaust line for unreacted carbon monoxide (CO) and hydrogen (H 2 ) removal. Prior to each in-flow hydroaminomethylation (HAM) reaction, the fluidic path, including the feed lines, and discharge lines were rinsed with 8 ml toluene and 16 ml methanol, then dried with N 2 flow. After changing reaction conditions, the flow reactor was allowed to stabilize for two residence times before a sample was collected by directing the selector valve towards a collection vial. Following the sample collection, the flow reactor effluent was directed to the waste collection vial during the transient period of the next reaction condition. Product analysis was performed by GC-MS and NMR. General Procedure 2: HAM in the batch reactor The catalyst, dicarbonyl 2,4-pentanedionato rhodium(I) catalyst and the ligand, 4,6-Bis(diphenylphosphino)-10 H -phenoxazine (N-xathphos), were dissolved in a stock solution of toluene and methanol at the target solvent composition under inert atmosphere inside a glovebox. The co-catalyst was weighed in an 8 ml glass vial and dissolved in a measured amount of the catalyst stock solution. The olefin and amine were subsequently added to the vial and the solution was further diluted with toluene/methanol to the desired concentration. The glass vial containing the liquid mixture and a magnetic stir bar was placed in an autoclave (Buchiglass Tinyclave, 35 mL under inert atmosphere). Supplementary Figure 3 shows a picture of the autoclave and the glass vial insert. The autoclave was connected to the gas supply manifold. Prior to each experiment, the lines and the autoclave were purged three times with N 2 . CO and H 2 were then sequentially introduced into the autoclave at their target pressures. The autoclave was disconnected and placed in an oil bath heated with a temperature-controlled stir plate. The stirring rate was set at 800 rpm. A separate batch experiment performed at a stirring rate of 1100 rpm resulted in the same yield and selectivity obtained with the 800 rpm stirring rate reaction. Upon reaction completion, the autoclave was removed from the oil bath and cooled in a water bath until it reached room temperature. The autoclave was then vented through the manifold and purged three times with N 2 before opening. Aliquots were taken from the reaction mixture. Product analysis was performed by GC-MS and NMR. Catalyst recycling procedure (Fig. 5 ) Following General Procedure 2, 9.24 mg of the co-catalyst 2F were weighed in an 8 ml glass vial. 1.5 ml of the toluene/methanol (1:4 volumetric raio) stock solution, containing 0.77 mg of the Rh catalyst and 6.6 mg of the ligand was transferred to the glass vial. 0.47 ml of 1-octene and 0.296 ml of piperidine were added and the solution was diluted to the desired concentration (1 M 1-octene) with toluene/methanol solvent. The reaction was performed at 115 °C for 60 min residence time after reaching thermal equilibrium. The cold pressure was 24 barg (total pressure) and the initial H 2 /CO ratio was 3.5. 50 μl aliquot was taken for analysis following the reaction completion and then the solvent was removed under high vacuum at 75 °C. The formed dispersion was redissolved in 2 ml pentane to precipitate the solid and the pentane was removed under high vacuum. The pentane wash was repeated three times. After the third time, the amine solution in pentane was separated from the solid by settling and transferred to another vial. The vial that contained the solid residue was transferred to a glovebox and redissolved in 1.5 ml toluene/methanol solvent. Next, fresh reactants were added as following: 0.47 ml of 1-octene, and 0.296 ml of piperidine, followed by dilution to the desired concentration (1 M 1-octene) by addition of toluene/methanol solvent. The vial was transferred to the autoclave for a new round of the HAM reaction with the recycled catalyst. In the experiment where fresh 2F was added in every recycle, 4.6 mg of 2F was weighed in the vial that contained the solid before adding the fresh solvent. On-the-fly switching from HAM to hydroaminovinylation (Fig. 6 ) Following General Procedure 1, The solution for loading in the catalyst syringe was prepared by dissolving 5.16 mg (0.1 mol %) of the catalyst, dicarbonyl 2,4-pentanedionato rhodium(I), and 44.1 mg of the ligand, N-Xantphos, in 2 ml toluene (4 to 1 ligand to Rh ratio). The solution was diluted with 8 ml methanol solvent and loaded into an 8-ml stainless steel syringe. 1-octene (3.13 ml, 1 M) and piperidine (1.97 ml, 0.99 equiv.) were added to 61.6 mg of co-catalyst 2F (2.0 mol %), diluted with toluene/methanol solvent to 10 ml total volume, and loaded into an 8 ml stainless steel syringe. The flow reactor pressure was set at 28 barg and the flow rate from the syringes was set at 8.48 μl/min each. The H 2 and CO flow rates were set at 1.995 mln/min and 0.57 mln/min, respectively. The flow reactor temperature was set at 115 °C. The reaction was run for 40 min before collecting samples. The collected crude mixture was analyzed by GC-MS and the amine l / b was 84. To switch to condition (2), zero 2F concentration, a third reactant syringe that contained 1-octene (3.13 ml, 1 M) and piperidine (1.97 ml, 0.99 equiv.) diluted with methanol solvent to 10 ml total was connected to the flow reactor and used as the reactant feed syringe. To switch to condition (3), the flow reactor temperature was lowered to 95 °C and the H 2 and CO flow rates were set at 1.28 mln/min and 1.28 mln/min, respectively. To switch to condition (4), fresh catalyst and reactant syringes were prepared the same way as described for condition (1), except that methanol was replaced by toluene in both syringes and no co-catalyst 2F was added. The flow reactor temperature was set at 115 °C and the H 2 and CO flow rates were set at 1.71 mln/min and 0.86 mln/min, respectively. To switch to condition (5), a fresh reactant syringe was added that contained 61.6 mg of co-catalyst 2F in toluene. To switch to condition (6), the flow reactor temperature was set at 125 °C. After switching to each new condition, the flow reactor was allowed to stabilize for 40 min before collecting samples for the shown length of time on Supplementary Fig. 1 (TOS). Samples were collected and analyzed at each condition and the amine and enamine yield are reported in Supplementary Fig. 1 . The reaction was run under condition (6) for 4 h and the product was collected and solvent was removed under high vacuum. The residue was washed three times with pentane and the enamine product yield was measured at 72% with an l / b of 90, measured by GC-MS. Data availability The authors declare that all data supporting the findings of this study are available within the main text and Supplementary Information .

Researchers at North Carolina State University have developed a faster, less expensive technique for producing hindered amines—a class of chemicals used as building blocks in products ranging from pharmaceuticals and agrochemicals to detergents and organic light emitting diodes. "Hindered amines are used in a tremendous variety of products, but all of the existing techniques for producing these amines are complicated and expensive," says Milad Abolhasani, corresponding author of a paper on the new technique and an associate professor of chemical and biomolecular engineering at NC State. "We set out to develop a better method for synthesizing these hindered amines, and we were successful." One of the less expensive techniques for producing hindered amines is hydroaminomethylation, or HAM. However, the chemical industry has largely avoided using HAM, because there are too many ways things can go wrong—leaving producers with undesirable chemicals instead of the functionalized amines they were trying to make. Researchers have improved the HAM process over the years. But all of the techniques for avoiding undesirable byproducts have meant extending the timeframe of the HAM process, so that it takes hours to perform all of the necessary reactions. Until now. "We've developed a HAM technique that makes use of continuous flow reactor technologies to produce hindered amines more efficiently," Abolhasani says. "Our HAM process takes less than 30 minutes in most cases. The only products are hindered amines and water. And we are able to recycle the primary catalyst, rhodium/N-Xantphos, which further drives down costs." The success of the new technique is made possible by two things. First, by using a continuous flow reactor that allows for continuous flow of both gases and liquids in a segmented flow format, the researchers were able to make the kinetics of the reaction far more efficient. Second, the new technique makes use of a co-catalyst—fluorinated benzoic acid—which reduces the amount of energy needed to perform some of the necessary reactions in the HAM process. Ultimately, this technique drives down the cost of producing hindered amines using inexpensive feedstock, allowing users to produce them more quickly and with no toxic byproducts. "By designing a cooperative catalyst system, we've demonstrated that the rate of the HAM reactions in our system can be 70 times higher than the existing state-of-the-art processes," says Malek Ibrahim, first author of the paper and a former postdoctoral researcher at NC State. "This process is also a good example for how flow chemistry platforms can improve catalyst turnover frequency, which is increasingly important as the price of rhodium catalysts goes up." The new technique is particularly attractive for decentralized manufacturing operations, since the small footprint of the necessary equipment and its scalability allows users to efficiently produce hindered amines on site and on demand. "What's more, the same technique can also be used to produce enamines—which are other chemical building blocks—on demand, simply by tuning the solvents we use in the flow reactor," Ibrahim says. "You can literally switch back and forth between producing amines and enamines without having to stop the production process, since the only thing you're changing is the solvent mixture." The researchers have filed a provisional patent on the new technique and are now looking for industrial partners to put the technique into widespread use. The paper, "Recyclable Cooperative Catalyst for Accelerated Hydroaminomethylation of Hindered Amines in a Continuous Segmented Flow Reactor," will be published May 4 in the journal Nature Communications.

10.1038/s41467-022-30175-0

Medicine

'Nano-sensing' drives melanoma cells' invasion

Directed migration of cancer cells by the graded texture of the underlying matrix, Nature Materials, DOI: 10.1038/nmat4586 Journal information: Nature Materials

http://dx.doi.org/10.1038/nmat4586

https://medicalxpress.com/news/2016-03-nano-sensing-melanoma-cells-invasion.html

Abstract Living cells and the extracellular matrix (ECM) can exhibit complex interactions that define key developmental, physiological and pathological processes. Here, we report a new type of directed migration—which we term ‘topotaxis’—guided by the gradient of the nanoscale topographic features in the cells’ ECM environment. We show that the direction of topotaxis is reflective of the effective cell stiffness, and that it depends on the balance of the ECM-triggered signalling pathways PI(3)K–Akt and ROCK–MLCK. In melanoma cancer cells, this balance can be altered by different ECM inputs, pharmacological perturbations or genetic alterations, particularly a loss of PTEN in aggressive melanoma cells. We conclude that topotaxis is a product of the material properties of cells and the surrounding ECM, and propose that the invasive capacity of many cancers may depend broadly on topotactic responses, providing a potentially attractive mechanism for controlling invasive and metastatic behaviour. Main Living cells have evolved a range of mechanisms to recognize a diverse set of environmental cues, including those present in spatially graded doses. For instance, in addition to being sensitive to spatial gradients of various dissolved chemical factors (chemotaxis) 1 , many eukaryotic cell types can detect gradients in the chemical or physical properties of the cell adhesion substratum, such as the graded density of the surface-bound ECM proteins (haptotaxis) 2 , 3 or graded substratum rigidity (durotaxis) 4 , 5 . Within these gradients, individual cells can migrate towards higher ECM densities or stiffer areas of the substratum. Our understanding of the mechano-chemical guidance cues associated with adhesion substrata comes mostly from studies in which the cell substratum is defined to be flat and featureless. However, the more native, in vivo cell adhesion surfaces are topographically more complex, primarily owing to a large diversity of ECM features spanning multiple scales of size and organization. For example, collagen fibrils and fibres interlinked within complex matrices are exemplary of this three-dimensional (3D) topographic complexity 6 . A convenient way to mimic and study the effects of complex ECM topographies, while retaining the advantages of essentially 2D experimentation, is to use quasi-3D, nanopatterned surfaces, capturing the in vivo geometry and size ranges of large ECM fibres. In our previous analysis, we found that many types of mammalian cell have the ability not only to anisotropically orient their migration and polarization in contact with ridged nano-topographic structures 7 , 8 , 9 , 10 , but also to detect and respond to gradients of these nanoscale features by biasing their directional migration 11 , 12 . This phenomenon of single-cell sensitivity to the topography gradient, also reported on the microscale 13 , which we will term here ‘topotaxis’, is still poorly understood. In particular, it is not clear whether it is a version of the more accepted haptotaxis and durotaxis processes, or if it is distinct from them in some essential way. Furthermore, the molecular basis of topotaxis is still not explored. Finally, it has not been addressed whether there is a potential for topotaxis to affect the invasive behaviour of cancer cells interacting with the surrounding ECM. We thus set out to examine topotaxis in the context of one of the most invasive cancers, melanoma. Melanoma, an aggressive cancer mostly affecting the skin, results in the highest percentage of skin-cancer-related deaths 14 . Melanoma tumours can transition from more benign radial growth patterns to more invasive, vertical growth 14 . In the latter case, cell invasion takes place through the dermis, a collagen-rich and cell-poor layer of connective tissue. Within dermis, collagen fibres are highly organized and frequently aligned, presenting an organized ECM-based adhesion substratum. As cancer cells migrate through the collagen matrix, they frequently express proteins, such as matrix metalloproteinases (MMPs), that can break down collagen fibres, which can cause the inhomogeneous density in the matrix and generate arrays of severed fibre bundles 15 , 16 . Melanoma cells and resident fibroblasts can also deposit matrix components, for example, fibronectin (FN), which is essential for invasive cell migration 17 , 18 . Increasing melanoma invasiveness is frequently associated with a range of genetic changes, one of which is a loss of functional PTEN, which can lead to over-activation of the PI(3)K–Akt signalling pathway 14 . Although this pathway has been associated with controlling cell migration, how it can influence melanoma invasiveness is unknown. Here, we provide evidence for topotaxis of melanoma cells, and show that it depends on the material properties of both the model matrix environment (density and structure) and the cell itself (stiffness). Genetic changes, such as loss of PTEN, and the local density of ECM can determine the directionality of topotaxis-driven cell migration. In particular, we show that there exist conditions under which more aggressive melanoma cells can shift in the direction of sparser ECM-mimicking quasi-3D matrix, whereas, strikingly, non-invasive melanoma cells shift in the opposite direction. We suggest a model accounting for this behaviour and show that, in agreement with model predictions, the manipulation of the PI(3)K–Akt- or ROCK–MLCK-dependent signalling pathways can determine the direction of the topotactic movement. The results have important implications for understanding the interplay between genetic mutations associated with cancer aggressiveness and the invasive properties of the cells. The results can also account for basic mechanisms underlying topotaxis in different cell types, and underscore the importance of accounting for materials properties of cells and their microenvironment in understanding physiological and pathological processes. Melanoma cells undergo topotaxis To examine topotactic properties of melanoma cells, we used capillary force lithography to fabricate arrays of nanoscale posts with graded post density. We refer to these arrays as graded post density arrays (GPDAs). The GPDAs had the gradient of post spacing in one of the orthogonal directions ( x direction in Fig. 1a ), with the density varying between 0.3 and 4.2 μm, and constant spacing (600 nm) in the other direction ( y direction in Fig. 1a ). Each post was 600 nm in diameter, within the range of natural collagen fibre (from 50 nm to 20 μm; refs 6 , 19 , 20 ), and was then coated with FN to focus on the matrix component essential for invasive migration, and melanoma cells were plated on the surface ( Fig. 1a and Supplementary Fig. 1 ). As the scale of a single cell is one order of magnitude larger than that of the nano-posts, such anisotropic nano-topography can present a directional physical cue to a cell attached onto the substrata. Invasive 1205Lu melanoma cells adhering to this substratum exhibited spreading and active migration, similar to their overall behaviour on flat adhesion substrata. However, individual cells also frequently exhibited distinct organization of membrane protrusions, depending on the local post density: long/parallel filopodia on the sparser post density side and short/randomly oriented, thicker protrusions on the denser post density side ( Fig. 1b ). The bundles of long and parallel filopodia observed on the sparser density side are similar to those found at the leading edge of cells undergoing directional migration 21 . This result was consistent with the possibility of topotactic cell behaviour, which we directly explored next. Figure 1: Topotactic migration of melanoma cells guided by the gradient of the post density and pre-coated ECM density. a , Schematic of a cell culture on the graded post substratum. b , Scanning electron micrographs of a 1205Lu melanoma cell cultured on the graded post substratum. Magnified areas focus on different formations of filopodia in the denser (blue rectangle) versus sparser post region (red rectangle). c , Changes in the number of cells as a function of time and the post density for 1205Lu cells and SBcl2 cells on GPDA, pre-coated with 10 μg ml −1 and 50 μg ml −1 FN. d , Trajectories of 1205Lu and SBcl2 cell migration on the GPDA pre-coated with 10 μg ml −1 and 50 μg ml −1 FN. e , Biased migration of invasive (1205Lu and WM983B) and non-invasive (SBcl2 and WM1552) melanoma cells evaluated as the average shift from the initial position on 10 μg ml −1 and 50 μg ml −1 FN in the direction parallel to the gradient of post density (displacement x ) over 5 h. Positive values indicate the migration from denser to sparser post regions. Asterisks indicate the statistical significance of biased directional migration on GPDA compared with migration on flat substrata. ∗ P < 0.05 and ∗ ∗ ∗ P < 0.005; all paired two-sample Student’s t -test. All error bars are s.e.m. Full size image We tracked the biased shift in the location of invasive (1205Lu) and non-invasive (SBcl2) melanoma cells lines on GPDAs coated with FN at two distinct densities, 10 and 50 μg ml −1 , for 3 days. In particular, we explored whether cells would accumulate in the sparser or denser areas of the substratum. Interfacing with denser arrays of posts can present cells with both a higher overall amount of ECM and a less deformable, thus, more effectively rigid surface. Therefore, both haptotactic and durotactic mechanisms of cell migration would predict that cells might accumulate over time in denser parts of the substratum 12 , 13 . Surprisingly, in contrast to this prediction, we observed that both cell lines accumulated in sparser areas, if the surface was pre-coated with 10 μg ml −1 FN ( Fig. 1c and Supplementary Fig. 2 ). At 50 μg ml −1 FN, the results were divergent, with 1205Lu cells accumulating in the sparser array areas, whereas SBcl2 cells accumulated in the denser array zones. These results could only be accounted for by directed cell migration. Cell proliferation for individual cell lines was independent of the post density, although invasive cells proliferated faster than non-invasive cells, particularly on lower FN density ( Supplementary Fig. 3 ). To investigate whether this biased accumulation results from a directional migration mechanism and to explore its generality in other cell types, we then tracked live cells for a much shorter time of 5 h, which allowed us to perform continuous live cell imaging. Consistent with the results above ( Fig. 1c ), we found for two invasive melanoma cell lines examined, 1205Lu and WM983B, that cells migrated towards sparser post array zones of GPDA pre-coated with 10 and 50 μg ml −1 FN, again suggesting that this topotactic cellular migration was guided by a mechanism distinct from haptotaxis or durotaxis ( Fig. 1d, e ). The direction of topotactic migration of two non-invasive melanoma cell lines, SBcl2 and WM1552, was again FN-density dependent. These cells exhibited directional cell migration opposite to that of invasive cells (that is, towards dense post array zones on GPDA) on surfaces pre-coated with 50 μg ml −1 FN; however, on GPDA pre-coated 10 μg ml −1 FN, they switched the migration directionality and shifted from dense to sparse array zones (the same direction as their invasive counterparts; Fig. 1d, e ). These results suggest that the melanoma cells are highly sensitive to the gradient of the density of topographic features, and that this topotactic behaviour is not equivalent to either haptotactic- or durotactic-directed cell guidance. Invasive melanoma cells migrate to sparser posts We confirmed that 1205Lu melanoma cells showed significant biased topographic shifts, with a broad distribution of single-cell trajectories on GPDAs pre-coated with 10 and 50 μg ml −1 FN ( Fig. 2a ). We also found that the migration bias was at least partially accounted for by a bias in motility persistence ( Supplementary Fig. 4 ). We then used high-resolution cell imaging to address putative mechanisms of the topotactic migration. In particular, we examined the possibility that topotaxis results from the differential ability of individual cells to conform to the topography of the underlying substratum. Cells whose plasma-membrane-associated cortical cytoskeleton is relatively stiff can, as a consequence, have less deformable membranes 22 , 23 , 24 , not capable of full penetration into the areas separating the nano-posts, thus being unable to fully conform to the shapes of the gaps between topographic features. Hence, this limited membrane deformability could lead to a decreased overall contact between the plasma membrane and the substratum-bound ECM. In the extreme, it can prevent contact with the lateral post surfaces, that is, essentially localizing spreading cells only to the tops of the posts. On the other hand, cells with less stiff cortical cytoskeleton and thus more deformable membranes might successfully envelop the posts and spaces between them, thus increasing their exposure to ECM. Of course, this differential ability to conform to the complex surface topography depends on the features of the topography itself, for example, the local spacing between the posts. To explore this conceptual model, we first directly addressed the possibility of differential cell penetration into the inter-post spaces by high-resolution 3D imaging the focal adhesion (FA) complexes. We found that FA complexes (indicating the location of cell–ECM interfaces) indeed exhibited differential depth of penetration into the inter-post spaces in the sparser versus denser areas of GPDA ( Fig. 2b, c ). Furthermore, scanning electron microscopy imaging suggested the presence of pseudopods between posts in sparser but not denser areas ( Supplementary Fig. 5a ). As a result of differential penetration, cells migrating towards areas maximizing their ECM contact would be expected to migrate to lower post densities for softer cells, consistent with the experimental observations and thus supporting the conceptual model of topotaxis. Figure 2: Correlation of the topotactic migration direction and cell stiffness as a function of the local post density. a , Fractions of 1205Lu cell displacements in the x direction for 5 h on GPDAs pre-coated with 10 μg ml −1 and 50 μg ml −1 FN. Positive values indicate the migration from denser to sparser post regions. b , The fractions of 1205Lu cells that have relatively deeper FA penetration into the substratum in the sparser versus denser post array zones of GPDAs coated with 10 μg ml −1 and 50 μg ml −1 FN (see c for examples of the image analysis). Asterisks indicate the statistical significance of deviation from the zero mean that indicates vertically even distribution of FAs in individual cells. ∗ ∗ ∗ P < 0.005. Wilcoxon signed rank test for zero mean. c , Confocal images of vinculin-stained cells showing differential vertical protrusions of 1205Lu cells, as a function of the post density, on indicated [FN]. For each [FN], we analysed the depths of FAs, as represented by vinculin staining intensity, located in the denser region, 1–4 (blue), and in the sparser region, 5–8 (red). In the top right panel, we show z -series images and analysis of signal intensity of corresponding FAs in the z direction; FAs in denser regions show peak intensity (dashed lines) at higher z distance measured from the glass bottom (the z position of 0) than those in sparser regions. In the bottom panels, x – z and y – z projections of FAs are shown with respect to the glass bottom (with the maximum vinculin staining positions determined in the top panels marked with dashed lines and arrows). d , Dependence of cell stiffness on the local post density on GPDAs coated with 10 μg ml −1 and 50 μg ml −1 FN, with the gradient range divided into three equally sized zones of the different local post densities. Asterisks indicate the statistical significance between the values on the dense versus sparse post density zones. ∗ P < 0.05, and ∗ ∗ ∗ P < 0.005. All paired two-sample Student’s t -test. All error bars are s.e.m. e , ECM-stimulated ROCK and PI(3)K activity in 1205Lu cells evaluated by immunoblotting of their substrates: Akt and MLC. pAkt, phosphorylated Akt; pMLC, phosphorylated MLC; GADPH, glyceraldehyde-3-phosphate dehydrogenase. All error bars are s.e.m. ( n = 3). f , A schematic depiction of ECM-triggered signalling pathway activities in 1205Lu cells shown in e and their effect on membrane protrusion and cell stiffness, assuming relative dominance of the PI(3)K pathway, which leads to a decrease in cell stiffness (and thus increasing substratum penetration) with increasing ECM contact. Full size image Although cell stiffness may be an inherent property of a specific cell type, it may also depend on the input from the cell adhesion substratum, and in particular the local density of the topographic features. To explore this possibility, we directly measured effective stiffness of the cortical cytoskeleton and the associated plasma membrane 25 , 26 . Specifically, we examined micro-rheological properties of 1205Lu cells by magnetic twisting cytometry, using RGD-coated beads bound to the integrin receptors on the cell surfaces. We analysed displacements of these beads in response to magnetic forces, which allowed us to determine the effective resistance of these beads to forces displacing them and thus to evaluate cell stiffness and its dependence on the local post density. Interestingly, the effective stiffness values of 1205Lu cells were significantly lower in the sparser post zones, compared with the values found in denser zones ( Fig. 2d ). This result suggested that, owing to both a lower cell stiffness and increased inter-post distances, 1205Lu cell–ECM contact can be relatively more extensive in the lower versus higher post density zones, consistent with the model connecting the direction of the topotactic motility with the ability to conform to the topographic features of the cell adhesion substratum. PI(3)K versus ROCK balance defines topotaxis direction What can account for the dependence of the cell stiffness on the post density? It is generally accepted that the cortical cytoskeleton can increase its apparent stiffness as a function of enhanced crosslinking of actin polymers due to augmented myosin activity 27 . Myosin can in turn be regulated by phosphorylation of the myosin light chain (MLC) by the MLC kinase (MLCK). MLCK activity is regulated by elevated activation of the small GTPase RhoA causing upregulation of the activity of the RhoA-dependent kinase 28 , 29 , 30 (ROCK). RhoA activity can be triggered by an increased ECM engagement by integrin receptors, and thus be enhanced by more extensive cell–ECM contact. We indeed found that MLC phosphorylation was correlated with the density of FN in 1205Lu cells ( Fig. 2e ). It thus seemed paradoxical that the stiffness of 1205Lu cells would decrease rather than increase on sparser post arrays, where the cell–substratum contact is expected to be higher. However, this paradox can be resolved if another signalling pathway triggered by ECM acts to counteract the effects of the ECM-activated RhoA–ROCK–MLCK pathway. One such pathway operates through the activation of another small GTPase, Rac1, frequently accompanied by elevated activity of PI(3)K (refs 31 , 32 ). We indeed found that, in 1205Lu cells, an increasing engagement of FN triggered progressively higher PI(3)K activation in an ECM-dose-dependent fashion, as evaluated by phosphorylation of the downstream kinase Akt ( Fig. 2e ). These results thus suggested that both ROCK and PI(3)K signalling can be enhanced by ECM, with the potential for opposing effects on the membrane protrusion and effective stiffness. Furthermore, we found evidence for negative cross-regulation between ROCK and PI(3)K that could enhance this effect ( Supplementary Fig. 6 ). For 1205Lu cells, the effect of PI(3)K could dominate that of ROCK, thus leading to decreasing rather than increasing effective membrane stiffness with decreasing post density ( Fig. 2f ). On the basis of these observations, we tested this model in the next set of experiments. If the directionality of topotaxis is defined by differential control of cell stiffness, which in turn depends on the ECM-regulated interplay between PI(3)K- and ROCK-activated pathways, pharmacological perturbations of these pathways are expected to affect topotaxis properties ( Fig. 3a ). In particular, a sufficiently strong inhibition of the PI(3)K signalling pathway could allow the ROCK-activated pathway to dominate the outcome ( Fig. 3a ). This could lead to a reversal of the stiffness dependence on the post density, and thus could in turn reverse the direction of the topotactic cell displacement. Indeed, we observed that PI(3)K inhibition reversed the directionality of topotactic motility of 1205Lu cells, at both 10 and 50 μg ml −1 FN coating density ( Fig. 3b–e and Supplementary Fig. 4 ). In agreement with the model above, the cell stiffness profile was also reversed by PI(3)K inhibition, with the higher cell stiffness values now observed at the sparser density zones of the post arrays ( Fig. 3f, g ). However, inhibition of ROCK did not change either the direction of topotaxis or stiffness profile, regardless of FN coating density ( Fig. 3b–g ), also in agreement with the model ( Fig. 3a ). Figure 3: Modulation of topotaxis direction of 1205Lu cells by perturbations of PI(3)K and ROCK signalling. a , Predicted changes of activity of the signalling network regulating cell stiffness of 1205Lu cells by pharmacological inhibition of PI(3)K (LY294002, red arrow) and ROCK (Y27632, blue arrow). PI(3)K inhibition is predicted to generate a ‘reversal zone’: the ECM density range where ROCK activity becomes dominant over PI(3)K activity, which results in a reversal of the direction of the topotactic cell migration. b , c , Cell displacements in the direction parallel to post density gradient on 10-μg ml −1 -FN-pre-coated GPDA ( b ) and 50-μg ml −1 -FN-pre-coated GPDA ( c ) over 5 h. Asterisks indicate the statistical significance of the biased deviation from the zero mean in the x direction. ∗ ∗ P < 0.01 and ∗ ∗ ∗ P < 0.005. Wilcoxon signed rank test for zero mean. All error bars are s.e.m. d , e , The distributions of topotactic cell displacements on 10-μg ml −1 -FN-pre-coated GPDA ( d ) and 50-μg ml −1 -FN-pre-coated GPDA ( e ) following treatment with a ROCK inhibitor (5 μM of Y27632) and a PI(3)K inhibitor (10 μM of LY294002). f , g , Modulation of cell stiffness dependence on the post density of 10-μg ml −1 -FN- ( f ) and 50-μg ml −1 -FN-coated GPDA ( g ) following PI(3)K and ROCK signalling perturbations. Asterisks indicate the statistical significance of the difference between the values on the dense versus sparse regions. ∗ P < 0.05, and ∗ ∗ ∗ P < 0.005. All paired two-sample Student’s t -test. All error bars are s.e.m. Full size image Our topotaxis model also implied that, owing to differential contact with ECM across the length of a single cell, PI(3)K activity is expected to be polarized. Notably, spatially localized PI(3)K activation has been associated with directional migration in different cell types 1 , 33 , 34 . To directly assess local PI(3)K activity, we transfected cells with 3-phosphoinotitide-specific Akt pleckstrin-homology domain fused with fluorescent protein (mRFP in 1205Lu or GFP in SBcl2), thus directly tracking PI(3)K activity in individual cells. We indeed observed that spatial localization of PI(3)K signalling in individual 1205Lu cells was polarized towards sparser post arrays, where effective membrane stiffness decreases ( Supplementary Fig. 7 ). PTEN reverses topotaxis of non-invasive melanoma The dominance of PI(3)K activity over RhoA–ROCK–MLCK signalling may be a specific characteristic of invasive melanoma cell lines, such as 1205Lu, known to have a loss of functional PTEN, a PI(3)K antagonist 14 . We confirmed that these cell lines had elevated basal levels of Akt phosphorylation compared with non-invasive SBcl2 and WM1552 cells ( Fig. 4a ). Conversely, the activity of ROCK was relatively higher in the non-invasive cells, suggesting a change in the relative balance of both PI(3)K and ROCK pathways, affecting the cell stiffness and protrusive activity. In spite of this pathway ‘rebalancing’, we found that both PI(3)K and ROCK activities again correlated with the surface FN density ( Fig. 4b ). Using these data, we predicted that PI(3)K dominance over ROCK signalling would occur in SBcl2 cells at low but not high ECM density ( Fig. 4c ), explaining the reversal of topotactic direction in non-invasive compared with invasive cells at 50 μg ml −1 , but not at 10 μg ml −1 (Figs 1d, e and 4d ). This explanation was further supported by the reversal of the dependency of the cell stiffness on the post density and of the spatial bias of PI(3)K signalling towards denser post regions in SBcl2 versus 1205Lu cell lines, observed at high but not low FN coating density ( Fig. 4e and Supplementary Fig. 7c ). These results suggested that the stiffness of non-invasive cells at higher ECM density values can be such that they would exhibit essentially no penetration in the inter-post spaces. Furthermore, these results imply that increasing inter-post distance, which might have gradually become more permissive for inter-post membrane protrusion, is compensated in these cells at high ECM density by increasing cell stiffness, ensuring that cells essentially remain at the top of the posts at any post density. We indeed found that, whereas at 10 μg ml −1 FN the penetration of the plasma membrane (as evaluated by FA confocal microscopy imaging) was similar to that found in 1205Lu cells, at 50 μg ml −1 FN the average penetration was close to zero ( Fig. 4f, g ). In addition, in contrast to non-invasive cells cultured on GPDA pre-coated with 10 μg ml −1 FN and invasive cells on both examined FN densities, which showed differential vertical penetration of pseudopods, pseudopods of SBcl2 on 50-μg ml −1 -FN-pre-coated GPDA were exclusively found on the tops of posts, independent of the post density, suggesting very limited penetration between posts ( Supplementary Fig. 5b ). Figure 4: Functional PTEN in non-invasive SBcl2 modulates its topotactic properties accounting for the responses shown in Fig. 1 . a , Non-invasive SBcl2 melanoma cells but not invasive 1205Lu cells express PTEN, a PI(3)K antagonist. Non-invasive melanoma cell lines, SBcl2 and WM1552, exhibit lower PI(3)K activity and higher ROCK activity versus invasive 1205Lu and WM983B cell lines on GPDA coated with 1 μg ml −1 FN. b , Dependence of the PI(3)K and ROCK activity in SBcl2 cells on the ECM density measured by assaying their substrates, Akt and MLC correspondingly. The data were quantified and normalized using the analysis in a to show the relative PI(3)K and ROCK activities in SBcl2 (solid lines) and 1205Lu (dashed lines) cell lines. All error bars are s.e.m. c , PI(3)K and ROCK signalling pathway activities in SBcl2 (solid lines) and 1205Lu (dashed lines) cell lines, inferred from b , showing the predicted change in the balance of the relative PI(3)K and ROCK signalling activities and the resulting switch in the topotaxis directionality. This change in the activity balance is also schematically depicted in the diagram on the right. d , The distributions of topotactic displacements of SBcl2 cells on 10-μg ml −1 -FN- and 50-μg ml −1 -FN-coated GPDA over the course of 5 h. e , Reversal of the SBcl2 cell stiffness dependence on the post density in cells cultured on 50 μg ml −1 versus 10 μg ml −1 FN coating density on GPDA. Asterisks indicate the statistical significance between the values on dense versus sparse regions. ∗ P < 0.05; all paired two-sample Student’s t -test. All error bars are s.e.m. f , The fractions of SBcl2 cells that have relatively deeper FA penetration into the substratum in the sparser versus denser post array zones on 10-μg ml −1 -FN- and 50-μg ml −1 -FN-coated GPDA. Asterisks indicate the statistical significance of the deviation from the zero mean. NS, not significant and ∗ ∗ ∗ P < 0.005. Wilcoxon signed rank test for zero mean. g , Confocal images showing differential vertical protrusions of SBcl2 cells depending on post density and ECM density. The analysis is equivalent to that shown in Fig. 2c . Full size image To further confirm the role of differential balancing between PI(3)K and ROCK pathways in defining topotactic migration in the non-invasive SBcl2 cell line, we explored the effect of pharmacological perturbations of these pathways. In this case, PI(3)K inhibition was predicted to enhance the reversal of topotactic migration of SBcl2 cells versus 1205Lu cells, so that the direction would be now reversed at both high and low FN coating densities, permitting consistent cell displacement from lower to higher post density areas. Inhibition of ROCK, on the other hand, was predicted to abolish the relative migration reversal and to induce directed migration from denser to sparser zones at all FN coating densities ( Fig. 5a ). These effects were indeed observed ( Fig. 5b–g ). Furthermore, the directionality of topotactic migration was again consistent with the effect of pharmacological perturbations on how cell stiffness or persistence of cell migration depended on post and ECM density ( Fig. 5f, g and Supplementary Fig. 4 ). Finally, to mimic genetic loss of PTEN, expected to enhance PI(3)K activity and potentially suppress ROCK activity, we pharmacologically inhibited this phosphatase. As expected in accordance with our model, non-invasive SBcl2 and WM1552 cells, at 50 μg ml −1 FN, showed topotactic migration from dense to sparse areas of the post array, thus now showing topotaxis directionality consistent with that of invasive cells ( Fig. 5h, i ). Figure 5: Modulation of topotaxis direction of SBcl2 cells by perturbations of PI(3)K and ROCK signalling. a , Predicted changes of activity of the signalling network regulating cell stiffness of SBcl2 cells by pharmacological inhibition of PI(3)K (LY294002, red arrow) and ROCK (Y27632, blue arrow). PI(3)K inhibition is predicted to extend the reversal of the balance between PI(3)K and ROCK activities to the whole ECM density range in contrast to ROCK inhibition removing the reversal. b , c , SBcl2 cell displacements in the direction parallel to post density gradient on 10-μg ml −1 -FN-pre-coated GPDA ( b ) and 50-μg ml −1 -FN-pre-coated GPDA examined for 5 h ( c ). Asterisks indicate the statistical significance of the biased deviation from the zero mean in the x direction. ∗ ∗ P < 0.01 and ∗ ∗ ∗ P < 0.005. Wilcoxon signed rank test for the zero mean. All error bars are s.e.m. d , e , The distributions of topotactic SBcl2 cell displacements on 10-μg ml −1 -FN-pre-coated GPDA ( d ) and 50-μg ml −1 -FN-pre-coated GPDA ( e ) following treatment with a ROCK inhibitor (5 μM of Y27632) and a PI(3)K inhibitor (10 μM LY294002). f , g , Modulation of cell stiffness dependence on the post density of 10-μg ml −1 -FN- ( f ) and 50-μg ml −1 -FN-coated GPDA ( g ) following PI(3)K and ROCK signalling perturbations. Asterisks indicate the statistical significance between the values on the dense versus sparse regions. ∗ P < 0.05; all paired two-sample Student’s t -test. All error bars are s.e.m. h , Modulation of topotaxis direction of non-invasive melanoma cells cultured on GPDA pre-coated with a high FN density, following inhibition of PTEN. SBcl2 and WM1552 cell displacements in the direction parallel to post density gradient on GPDA pre-coated with 50 μg ml −1 FN and examined for 5 h following treatment with 100 nM of bpV(HOpic), a PTEN inhibitor. Asterisks indicate the statistical significance of difference in displacement in the x direction between with and without drug. ∗ ∗ ∗ P < 0.005. All paired two-sample Student’s t -tests. All error bars are s.e.m. i , The distributions of topotactic SBcl2 and WM1552 cell displacements measured in h . Asterisks indicate the statistical significance of the biased deviation from the zero mean in the x direction. ∗ ∗ ∗ P < 0.005. Wilcoxon signed rank test for the zero mean. Full size image Outlook Oriented cell motility in gradients of topographic features, termed here ‘topotaxis’, has remained largely unexplained. The data presented in this study suggest a mechanism that strongly depends on the ability of the plasma membrane and the underlying cortical cytoskeleton to conform to the topographic complexity of the cell adhesion substratum. The complex substratum topographies can be a reflection of the ECM structure in the cell vicinity, composed of both intact and potentially severed fibres, interlinked around a cell in a complex meshwork. Our analysis suggests that a gradient of density of such topographic features can be differentially interpreted by a cell as either attractive or repulsive, depending on whether the cell forms ECM contacts only owing to exploration of the immediate surfaces of the topographic features (for example, the tops of the posts in this study), or it enhances the ECM contact by penetrating into spaces separating these features (for example, exploring inter-post spaces in this study). We found that this differential ability to conform to the topographic complexity was a function of the genetic differences naturally present in melanoma cells of different degrees of invasiveness, as well as the signalling activity triggered by engagement of ECM. Importantly, both genetic and signalling inputs affected the same signalling circuit, involving PI(3)K- and ROCK-dependent pathways, which in turn regulated the effective stiffness of the cortical cytoskeleton and the plasma membrane and thus directionality of topotaxis. In spite of its apparent complexity, the mechanism of topotaxis is easily generalized to three possible classes of cell interactions with topographically complex ECM environment. If a cell is either fully compliant with the ECM topography (can fully envelop ECM fibres) or is fully non-compliant (is localized on the tops of ECM fibres, being incapable of enveloping them), the guidance effect of a graded ECM topography is similar to the haptotactic guidance, in which cells are expected to migrate up the gradient of ECM density, and thus higher density of topographic features. However, if the cell has an intermediate compliance, it may have higher ECM penetration and thus ECM contact in sparser rather than denser matrix zones, which can guide its migration from denser to sparser feature zones. Cell compliance to the matrix structure is a function of the cell stiffness, which in turn can be a function of ECM-triggered signalling activity, resulting in a feedback control of topotactic migration. Interestingly, this conceptual model allows for the possibility that even the same topography gradient can specify opposite migration directions. This has been observed to occur when fibroblasts undergo topotaxis in two opposite directions on the same ECM topography gradient, migrating towards an optimum ECM fibre density (that is, converging away from very sparse or very dense areas) 11 . In this case the fibroblasts are predicted to be fully compliant in the sparser ECM zones and only partially compliant in denser ECM areas. Membrane stiffness and thus cell compliance to the topographic features of the surrounding matrix and, ultimately, the topotaxis direction, are found here to depend on a specific signalling network architecture, in which two parallel signalling pathways, both triggered by the ECM–integrin engagement, have opposite effects on the cortical cytoskeleton and the effective membrane stiffness. Thus, the stiffness value or, conversely, protrusive activity of the plasma membrane, is a result of a balance of activities in these pathways. With increasing dominance of PI(3)K–Akt activity, the stiffness value decreases and thus membrane compliance to the matrix topography can increase ( Fig. 6a ). Conversely, if ROCK–MLCK pathway activity dominates, the membrane stiffness can increase, limiting compliance to the topographic features ( Fig. 6b ). The relative abundance of signalling activities in these pathways, such as PI(3)K and its antagonist, PTEN, can shift across the length of a single cell, which is expected to orient directional cell migration 35 , 36 . Thus, as observed in this study, locally polarized PI(3)K activity biased by inhomogeneous cell–ECM contact could help explain topotactic migration on a single-cell level. As demonstrated in this study, both genetic and pharmacological perturbations of PI(3)K and ROCK–MLCK signalling pathways can change their relative dominance, and switch the directionality of topotactic migration. Interestingly, this signalling network architecture is structurally similar to the one proposed to account for chemotactic cell migration (the so-called LEGI network) 1 , 37 , suggesting that similar principles underlie the control of distinct tactic mechanisms. Figure 6: A schematic graphic model of the topotaxis in invasive and non-invasive cells. The directionality of topotactic migration depends on differential conformity of cellular membrane to local topographic structure. The degree of conformity depends on the cell rigidity, which in turn is controlled by the interplay between two ECM-activated signalling pathways: PI(3)K–Akt increasing the degree of conformity and matrix penetration and ROCK–MLCK decreasing the degree of conformity. The balance between the activities of these pathways can be determined by the local ECM signalling input or genetic changes affecting these pathways, and can be modulated by pharmacological perturbations. a , The relatively higher activity of the PI(3)K–Akt pathway in the invasive melanoma cells can lead to differential penetration of the matrix and lead to a higher matrix contact in the sparser versus denser matrix zones, generating bias for migration from the denser to sparser matrix density. b , On the other hand, at high enough ECM signalling inputs, the relatively higher activity of the ROCK–MLCK pathway in non-invasive melanoma cells limits cell penetration into the matrix, leading to haptotactic migration up the density gradient. Full size image The mechanism of topotaxis proposed here emphasizes intimate interdependence of the matrix structure and the activity of the signalling networks controlling cell mechanics. This framework can provide a unifying view of the effects of different genetic mutations implicated in progression from non-invasive to invasive cell behaviour. For instance, the loss of PTEN is frequently observed in transition of cancer cells to aggressive and invasive growth, but the exact reason for this switch in cell phenotype is not clear. Furthermore, it has been shown in mouse models that a gradual decrease in PTEN gene dosage progressively increases the ability of melanoma cells and normal melanocytes to penetrate the ECM (ref. 38 ). The topotaxis model proposed here strongly suggests that the increased dominance of the PI(3)K–Akt signalling pathway precipitated by the loss of PTEN leads to an increased cell compliance and topotaxis from dense to sparse matrix areas. As cancer invasion is also frequently accompanied by partial degradation of the ECM due to an increased expression of MMPs, migration from denser to sparser matrix areas can enhance the ability of the cells to shift to the areas of sparser and, possibly more aligned matrix fibres, enabling more directed and efficient aggressive spread. Loss of PTEN may enhance MMP secretion thus additionally contributing to this process 39 . On the other hand, less aggressive cells, expressing functional PTEN, would be less compliant, which can direct their topotactic migration into denser matrix areas, complicating migratory behaviour, even if the matrix is partially degraded. Many other genetic alterations could potentially affect cell stiffness, either through the ECM-regulated PI(3)K–Akt and ROCK–MLCK signalling pathways, or through other mechanisms, accounting for increased invasive behaviour. Compliance with the complex substratum topography can also influence formation of focal adhesions of different size and maturity, which might also contribute to the topotaxis control 40 . Conveniently, this network-based view can also suggest multiple potential interventions that may help shift signalling activity and alter cell stiffness, thus reversing aggressive phenotypes. Our analysis illustrates the importance of changing the prevailing emphasis on cell analysis on flat adhesion substrata, where various tactic phenomena have been described so far, to more complex but also more realistic environments, in which the haptotactic, durotactic and topotactic guidance effects can be intertwined. For instance, studies of cell migrations on nanostructured surfaces also suggested that cells may actively sense the rigidity of the local microenvironment 41 , 42 . The initial analysis undertaken here, with the assistance of tools that are more complex than the 2D environment of the Petri dish, but less daunting than the 3D environments found in many tissues, can help establish the framework for quantitative description of the mechanical and chemical properties of both ECM and the live cells actively interacting with it. Methods Fabrication of topographical pattern arrays. To construct a graded post density array (GPDA), poly(urethane acrylate) (PUA) was used as a mould material from the silicon master as previously described 43 ( Supplementary Fig. 1 ). Briefly, the ultraviolet-curable PUA was drop-dispensed onto a silicon master and then a flexible and transparent polyethylene terephthalate (PET) film was brought into contact with the dropped PUA liquid. Subsequently, it was exposed to ultraviolet light ( λ = 200–400 nm) for 30 s through the transparent backplane (dose = 100 mJ cm −2 ). After ultraviolet curing, the mould was peeled off from the master and additionally cured overnight to terminate the remaining active acrylate groups on the surface before use as a first replica. The resulting PUA mould used in the experiment was a thin sheet with a thickness of ∼ 50 μm. The topographic patterns with variable local density and anisotropy were fabricated on glass coverslips, using ultraviolet-assisted capillary moulding techniques 44 . Before application of the PUA mould, the glass substratum was cleaned with isopropyl alcohol, thoroughly rinsed in distilled ionized water, and then dried in a stream of nitrogen. Subsequently, an adhesive agent (phosphoric acrylate/propylene glycol monomethyl ether acetate, 1:10, volume ratio) was spin-coated to form a thin layer ( ∼ 100 nm) for 30 s at 3,000 r.p.m. A small amount of the same PUA precursor was drop-dispensed on the substrate and a PUA mould was directly placed onto the surface. The PUA precursor spontaneously filled the cavity of the mould by means of capillary action and was cured by exposure to ultraviolet light ( λ = 250–400 nm) for ∼ 30 s through the transparent backplane (dose = 100 mJ cm −2 ). After curing, the mould was peeled off the substrate using sharp tweezers. Patterns were viewed and photographed with a LEO FESEM 1530 operating at 1 kV. The nano-post array used in this study, GPDA, has the post spacing gradient along the x direction and post spacing to the y direction, resulting in an anisotropic density gradient. Thus, the nano-posts used had a constant diameter of 600 nm with the varying post-to-post spacing from 300 nm (spacing ratio (spacing/diameter) of 0.5) to 4.2 μm (spacing ratio of 7) in the x direction, but constant post-to-post spacing of 600 nm (spacing ratio of 1) to the y direction. Cell culture. We used several cell lines identified by the Wistar Institute (Pennsylvania, USA) 1205Lu, WM983B, SBcl2 and WM1552, a gift from R. Alani (Boston University, USA). We also used 1205Lu and SBcl2 transfected with plasmids coding for 3-phosphoinotitide-specific Akt pleckstrin-homology domain fused with mRFP in 1205Lu or GFP in SBcl2. We transfected mRFP–Akt PH plasmid in 1205Lu and GFP–Akt PH plasmid in SBcl2 using Lipofectamine 2000 transfection reagents (Thermo Fisher, 11608027) as described in the product’s manual. mRFP and GFP–Akt PH plasmids were gifts from J. Zhang’s laboratory (UCSD; ref. 45 ). After cloning the Akt PH domain from GFP–Akt PH plasmids, we ligated the mRFP plasmid to it. The cells were cultured in Dulbecco’s modified Eagle’s medium (Gibco) supplemented with 10% fetal bovine serum (Gibco), 50 U ml −1 penicillin, and 50 μg ml −1 streptomycin (Invitrogen) at 37 °C, 5% CO 2 and 90% humidity. These were split 1:4 after trypsinization every 2–3 days. A glass coverslip with the topographical pattern substratum was previously glued onto the bottom surface of the custom-made MatTek dish (P35G-20-C). Then, we replated cells on the pattern pre-coated by 10 or 50 μg ml −1 FN (Sigma-Aldrich, F1141) for 3 h and incubated up to 12 h for the attachment. For the perturbation of the signalling pathway related to cell migration, 5 μM Y27632 (Sigma-Aldrich, Y0503), 10 μM LY294002 (Sigma-Aldrich, L9908) and 100 nM bpV(HOpic) (Sigma-Aldrich, SML0884) were added. Mycoplasma contamination was tested with Hoechst (Invitrogen, 33342). Western blot analysis. Cells were homogenized in RIPA buffer (Thermo Scientific) with protease inhibitor cocktail (Thermo Scientific), and then centrifuged at 12,000 g at 4 °C for 30 min. The supernatant was collected and protein was quantified using Bradford assay kit (Pierce). Protein samples were subsequently diluted with sample buffer, and heated at 70 °C for 10 min. These samples were separated on a 4–20% w/v SDS–PAGE gel (BioRad), and transferred to a nitrocellulose membrane (BioRad). This membrane was blocked for 1 h in AquaBlock/EIA/WB (EastCoast Bio) and incubated in a 1:1,000 diluted solution of PTEN (Cell Signaling, 9552), a 1:1,000 diluted solution of myosin light chain (Abcam, ab11082), a 1:500 diluted solution of phosphorylated myosin light chain (Cell Signaling, 3674), a 1:1,000 diluted solution of Akt (Cell Signaling, 9272), a 1:1,000 diluted solution of phosphorylated Akt (Cell Signaling, 9271) and a 1:1,000 diluted solution of GAPDH (Abcam, ab9485) in blocking solution at 4 °C overnight. After washing (three times for 10 min) in TBST (10 mM Tris, pH 8.0 and 0.1% v/v Tween 20), goat anti-rabbit and anti-mouse secondary antibodies for Odyssey western blotting (Li-cor) were added. Finally, after washing with TBST, the Odyssey CLx infrared imaging system (Li-cor) was applied to blotting. For quantification, each blot was repeated at least three times. Time-lapse microscopy. Images were acquired after cells were fully attached. For long-term observation, the custom-made MatTek dish integrated with topographically patterned substratum was mounted onto the stage of a motorized inverted microscope (Zeiss Axiovert 200M) equipped with a Cascade 512B II CCD (charge-coupled device) camera containing an environmental chamber. Phase-contrast and epi-fluorescent images of cells were automatically recorded using Slidebook 4.1 (Intelligent Imaging Innovations) for 5 h at 10 min intervals. Tracking cells. To investigate the migratory behaviour on the pattern, we tracked cells manually in every image taken by time lapse by outlining the boundary of each cell. Then, using a custom MATLAB code (The MathWorks), we obtained the positions of cell centroids as a function of time. The code is accessible in the Supplementary Information . D / T was defined as the ratio of the shortest distance between the initial location of an individual cell and it‘s final location to the length of the whole itinerary of the cell during live cell imaging. Quantification of spatial PI(3)K translocation. Segmentation of epi-fluorescent images was performed as follows. MATLAB Image Processing Toolbox allows reading the pixel values that describe the intensity in a manually isolated single cell. Hotspots, the regions that have a higher intensity value on the cell membrane than the average intensity of the total pixels in a single cell, were segmented. Then, we grouped each hotspot after cutting off the area below the threshold area for ensuring continuous activation of signal. On the basis of these groups of hotspots, we acquire the area ( A ), average intensity ( F ), and centroid coordinates of each hotspot region to calculate the overall signalling vector within a cell. The position of the hotspot relative to the cell centroid, x i = ( x i , y i ), was defined by subtracting the coordinates of the cell’s centroid from those of its hotspots, i . Its vector, s i , is described by the magnitude equal to the total fluorescence intensity of the spot, ( A i F i ); and the overall signalling vector, S , describing spatial PI(3)K translocation is the sum of s i . These vectors indicated the direction of spatial PI(3)K activation in an individual cell 46 . Then, we averaged all of the signalling vectors of each time point and calculated the direction of the averaged signalling vector of each cell for 5 h ( Supplementary Fig. 8a ). Magnetic twisting cytometry. To quantify the stiffness of the living adherent cell, we used magnetic twisting cytometry as described previously 25 , 26 . In brief, a functionalized ferrimagnetic microbead bound to the cytoskeleton through cell surface integrin receptors was magnetized horizontally and then twisted in a vertically aligned homogeneous magnetic field that varied sinusoidally in time. The sinusoidal twisting field causes both a rotation and a pivoting displacement of the bead. As the bead moves, the cell develops internal stresses that in turn resist bead motions. Lateral bead displacements in response to the resulting oscillatory torque were detected using a CCD camera (Orca II-ER, Hamamatsu) attached to an inverted optical microscope (Leica Microsystems), and with an accuracy of 5 nm using an intensity-weighted centre-of-mass algorithm. We defined the ratio of specific applied torque to lateral bead displacements as the complex elastic modulus of the cell, g ∗ ( f ) = g ′( f ) + ig ′′( f ), where g ′ is the storage modulus (cell stiffness), g ′′ is the loss modulus (cell friction), and i 2 = −1. Cell stiffness and friction are expressed in units of pascal per nanometre (Pa nm −1 ). Statistical analyses were performed using unpaired two-tailed Student’s t -tests. To satisfy the normal distribution assumptions, cell stiffness data were square root transformed and confirmed by the Jarque–Bera test. Immunofluorescence staining. Cultures were conducted on the nanofabricated coverslip as described above after coating with FN. We fixed with ice-cold 4% paraformaldehyde for 20 min, washed two times with phosphate-buffered saline (PBS) and permeabilized with 0.1% Triton X-100 in PBS for 5 min. After washing with PBS, cultures were blocked by 10% goat serum for 1 h, and then incubated with primary antibody against vinculin (1:200, Sigma-Aldrich, V9131) for 3 h at room temperature. After washing, the cultures were incubated with secondary antibody for vinculin (1:500), Alexa Fluor 594-conjugated phalloidin (1:40, Molecular Probes, A12381) and Hoechst (Invitrogen, 33342) for 1 h at room temperature. The slides were mounted with anti-fade reagent (SlowFade gold, Invitrogen) and examined using a confocal microscope (Zeiss 510 Meta confocal) with a ×63 oil immersion objective (Zeiss, 1.6 NA). Proliferation assay. 1205Lu and SBcl2 cells were plated on the nanofabricated coverslips pre-coated with 10 or 50 μg ml −1 FN. After 1 day of culture, cells were immunofluorescently stained with Ki67 (Abcam, ab15580) and Hoechst (Invitrogen, 33342) as described above. Next, we counted all cells and the cells positive for Ki67, a proliferation marker, in sparse and dense post regions on GPDAs. Then, the ratios of the number of Ki67-positive cells to that of total cells in each region analysed were calculated to assess the proliferation rate. Measurement of relative depths of vinculin-stained focal adhesions. Cells with immunofluorescence-stained vinculin were imaged using a confocal microscope (Leica SP8) with a ×63 oil immersion objective (Zeiss, 1.6 NA), and the z direction re-construction was performed with customized MATLAB code (The MathWorks). The code is accessible in the Supplementary Information . We found the centroid coordinates of the re-constructed vinculin-stained focal adhesions and measured their depths relative to the zero-plane level (at the top of the posts) based on the z coordinates. Then, we divided the vinculin-marked focal adhesions in individual cells into two groups, depending on their locations corresponding to the local post density, that is, those in the sparser versus denser post regions, and averaged their depths within each group. Relative depths were calculated by subtracting the mean depth of the focal adhesions on denser relative to the sparser region in an individual cell; positive relative depth indicates deeper average FA penetration on the sparser post region than on the denser post region. Scanning electron microscopy (SEM). For scanning electron microscopy, cultured melanoma cells on the graded nano-post density adhesion substrata were washed with PBS (pH 7.4, Gibco Invitrogen) and fixed in 3% glutaraldehyde (Sigma-Aldrich) in PBS for 1 h. After fixation, we rinsed samples in 0.1 M sodium cacodylate for 30 min at 4 °C and then post-fixed in 2% osmium tetroxide for 1 h at the same temperature. After a brief rinse with distilled H 2 O, samples were en bloc stained in 2% aqueous uranyl acetate (0.22 μm filtered) for 1 h at room temperature in the dark. Following a graded ethanol dehydration cells were critical point dried with liquid CO 2 , mounted onto SEM stubs with double-stick carbon tape, and sputter-coated with 10 nm gold palladium. Samples were viewed and photographed with a LEO FESEM 1530 operating at 1 kV. Change history 22 March 2016 In the version of the Article originally published, the original title omitted a word and should have read 'Directed migration of cancer cells guided by the graded texture of the underlying matrix'. This has been corrected in all versions of the Article.

A new study sheds light on how melanoma cells change from benign to malignant, and how the complex interaction between the cells and their surrounding environment affects outcomes of the cancer. The transition from the radial growth patterns of benign melanoma cells to the vertical patterns of malignant cells has long been a mystery to researchers. But a study from the lab of Andre Levchenko, the John C. Malone Professor of Biomedical Engineering and director of the Yale Systems Biology Institute, has found that the cells' stiffness—determined by the specific balance of two signaling pathways—plays a major role in directing cell migration through the complex environment. Cells essentially sense and follow the nanoscale topography, a phenomenon the researchers have coined "topotaxis." The findings, published in the online edition of the journal Nature Materials on March 14, could lead to new treatments and diagnostic tests, the researchers said. Levchenko and his colleagues found that melanoma cells in the extracellular matrix (ECM)—the body's living tissue—migrate toward either dense or sparse areas of the matrix. Benign cells veer toward the dense area, where they have little room to move and stay on the surface. Malignant cells move toward sparse areas that allow them to grab onto the fibers of the matrix and more effectively spread out. This directed migration of the cells is partly dependent on the cells' material properties: Because they can negotiate the topography better, soft invasive cells are more capable of spreading out than stiff cells. How aggressively the cells move is determined by the chemical cues they receive. Contained in the matrix itself, these cues are processed by a network of pathways that breaks off into two branches. The direction of the cells' migration depends on which branch is dominant. The balance of these pathways is determined by the genetic state of the cells. A mutation that causes the loss of the gene PTEN—a common genetic mutation in aggressive forms of melanoma—will alter the balance. "We know now that we can interpret genetic changes in the context of this signaling network, which enables us to understand better why genetic changes may lead to metastatic spread," said Levchenko, who is a member of the Yale Cancer Center. These findings also suggest avenues for reversing that invasive transition in cell behavior. It doesn't necessarily require restoring the PTEN gene, but restoring the balance of the pathways to their original state through genetic manipulations or using a set of drugs. "In our experiments, we can take a cell that's aggressive and make it revert to a benign form and vice versa," he said. "As soon as you change the balance, the cells start moving in the opposition direction." To study the cell behavior, Levchenko's lab is using models of the matrix made from precisely nanofabricated quasi-3D environments. More complex than the 2D environments of Petri dishes, these environments feature the complex topography and textures of ECM, but are more readily analyzed than the actual tissue. "Here it's more controlled—we can measure cell stiffness, we can measure activity of these pathways and networks, we can very precisely control the activity of the cells." Levchenko also sees a potential in the findings for a less invasive diagnostic test. "We can take cells and drop them on these nanofabricated surfaces to mimic the matrix, and depending on where they move, we can actually tell whether they're benign or malignant," he said. "That gives us an interesting possibility that didn't exist before—to characterize the degree of invasiveness based on how the cells behave."

10.1038/nmat4586

Physics

Researchers use 3-D imaging to improve diagnosis of muscle diseases

Dominik Schneidereit et al, Optical prediction of single muscle fiber force production using a combined biomechatronics and second harmonic generation imaging approach, Light: Science & Applications (2018). DOI: 10.1038/s41377-018-0080-3 Journal information: Light: Science & Applications

http://dx.doi.org/10.1038/s41377-018-0080-3

https://phys.org/news/2018-11-d-imaging-diagnosis-muscle-diseases.html

Abstract Skeletal muscle is an archetypal organ whose structure is tuned to match function. The magnitude of order in muscle fibers and myofibrils containing motor protein polymers determines the directed force output of the summed force vectors and, therefore, the muscle’s power performance on the structural level. Structure and function can change dramatically during disease states involving chronic remodeling. Cellular remodeling of the cytoarchitecture has been pursued using noninvasive and label-free multiphoton second harmonic generation (SHG) microscopy. Hereby, structure parameters can be extracted as a measure of myofibrillar order and thus are suggestive of the force output that a remodeled structure can still achieve. However, to date, the parameters have only been an indirect measure, and a precise calibration of optical SHG assessment for an exerted force has been elusive as no technology in existence correlates these factors. We engineered a novel, automated, high-precision biomechatronics system into a multiphoton microscope allows simultaneous isometric Ca 2+ -graded force or passive viscoelasticity measurements and SHG recordings. Using this MechaMorph system, we studied force and SHG in single EDL muscle fibers from wt and mdx mice; the latter serves as a model for compromised force and abnormal myofibrillar structure. We present Ca 2+ -graded isometric force, pCa-force curves, passive viscoelastic parameters and 3D structure in the same fiber for the first time. Furthermore, we provide a direct calibration of isometric force to morphology, which allows noninvasive prediction of the force output of single fibers from only multiphoton images, suggesting a potential application in the diagnosis of myopathies. Introduction Structure and function are inevitably related to each other. A desired function requires a tailored structure, while from a given structure, deductions regarding its function may be derived. This particularly applies to the concept of organ and tissue structures and functions. Skeletal muscle, for instance, is highly ordered and hierarchically structured by parallel and serial polymeric motor proteins, i.e., actomyosin filaments, and series elastic elements, to perform as a linear bioactuator to enable movement and to give in passively to external forces 1 . Within each muscle fiber, the concept of highly ordered elements, i.e., myofibrils aligned in parallel by sarcomeric and extrasarcomeric proteins at the z-disks 2 , ensures the well-known striation-pattern 3 and directs force production along the main fiber axis 4 . Thus, apart from fast signaling-related activation processes that determine muscle performance, e.g., excitation contraction coupling and Ca 2+ homeostasis 5 , a long-term predictor of muscle function is found in structural changes. This is particularly important as skeletal muscle has a high plasticity to respond to exercise with hypertrophy, to disuse with atrophy 6 , and to injury with complete regeneration and repair 7 . Similarly, muscle structure and cytoarchitecture are also major targets in muscle diseases, such as chronic inflammatory or degenerative diseases associated with ongoing degeneration-regeneration cycles, imperfect repair, and tissue and cellular remodeling. Examples include (poly/dermato-)myositis 8 , cancer cachexia 9 , muscular dystrophies 10 , and even aging 11 . Muscle structure remodeling involves not only the extracellular matrix (e.g., fibrosis) but also the remodeling of the sarcomere and the myofibrillar cytoarchitecture. For instance, a hallmark of structural cellular changes as a result of regeneration has been found in single fiber branching and splitting, for example, in Duchenne muscular dystrophy 12 , 13 , 14 , following weight-lift exercise 15 or toxin-induced myonecrosis 16 . Thus, it is not surprising that chronic conditions with ongoing remodeling are associated with progressive muscle weakness 12 , 17 . Therefore, the three-dimensional sterical arrangement of myofibrillar and cytoarchitecture may be considered an anatomical correlate to predict muscle force outcomes. It is tempting to speculate that imaging of the 3D muscle fiber architecture can be a modality for extrapolating mechanical performance instead of executing strenuous force recordings using force transducer technologies, in particular in cases involving single fibers. However, the assessment of skeletal muscle fiber architecture is not straightforward. Clinically, the diagnosis of myopathy in patients presenting muscle weakness is usually based on histology section analysis from muscle biopsies. Although cross-cut sections allow visualization of multiple cells at once, this approach is limited to one plane without any 3D aspect. Alternatively, confocal laser scanning microscopy has frequently been used to determine muscle fiber structure 13 , 18 , 19 . However, it always requires external labels, which may be harmful and prone to bleaching. More importantly, single muscle fibers with diameters exceeding 50 µm must already be considered thick samples where photon scattering and absorption can become substantial within the illuminated z-cone. An elegant alternative is multiphoton microscopy where infrared light ensures deeper penetration depths and less scattering while laser pulse excitation ensures excitation only within a focal volume of ∼ 1 µm 3 . Using a special nonlinear mode of multiphoton excitation through second harmonic generation (SHG), intrinsic signals can be visualized label-free. Biomolecules susceptible for SHG comprise collagen-I and myosin-II, enabling label-free, detailed structural analysis of subcellular cytoarchitecture and myofibrillar geometry in 3D 3 , 20 , 21 . Furthermore, several mathematical analysis strategies have been developed to describe the degree of myofibrillar disarray using quantitative morphometry in 2D images 14 , 20 , 22 and in 3D volumes of muscle or single fiber SHG stacks 23 , 24 , 25 . Our group introduced boundary tensor orientation analysis of myofibrillar striation patterns to extract two very sensitive parameters of ‘myofibrillar disorder,’ namely, cosine angle sums (CAS) and vernier densities (VDs). This approach has successfully been employed to describe alterations in 3D cytoarchitecture associated with various disease models 14 , 24 , 26 , 27 . In particular, such studies enabled the identification of the correlation between structure and morphology for the chronic progression of muscle weakness due to age or inherited myopathies. However, despite all the progress made in SHG imaging and pattern recognition analysis defining ultrastructural alterations, to date there is still no direct proof of how an altered sterical myofibrillar cytoarchitecture is correlated to an impaired muscle force. To obtain such a calibration, simultaneous measurements of SHG signals and isometric force are required in the same single muscle fiber. Therefore, we combined biomechatronics and optical engineering to engineer a miniaturized biomechatronics system, MechaMorph , which allocates an optically based force transducer sensor and length/force-feedback controlled voice coil actuator technology in a multiphoton microscopy-adapted stage. To employ a large range of morphological parameters of myofibrillar alignment and isometric force amplitudes, we used single EDL muscle fibers from wild-type and mdx mice. We present, for the first time, detailed correlations of the biomechanical parameters of (i) active isometric force and myofibrillar Ca 2+ sensitivity, (ii) passive viscoelastic parameters, and (iii) optically SHG-derived morphometry (SHG, VD) in a calibration of structure to force. Using this approach, we propose a quantitative prediction of muscle function exclusively from an optical assessment of the structure, without the need for sophisticated biomechanical recordings. This may be very valuable for diagnostics in the scenario of myopathies in future clinical settings involving volumetric SHG imaging of muscle samples/biopsies. Results Ultrastructural myofibrillar architecture assessed using second harmonic generation (SHG) is a predictor of Ca 2+ - activated specific force and myofibrillar Ca 2+ sensitivity In previous studies, the mdx model presented a vastly altered myofibrillar cytoarchitecture, fiber branching and enhanced muscle weakness 12 , 14 , 17 . To directly link structure and mechanical function, we sought to calibrate both parameters, SHG morphometry and isometric force amplitudes, within the same single fiber. To exploit a wide range of SHG parameters of CAS, VD, and isometric force, the mdx model was chosen to provide values of decreased CAS and increased VD, which are usually not seen in wt single fibers, and to assess the graded isometric force in conjunction with SHG imaging in the same single fiber using our MechaMorph system (Fig. 1 ). Figure 2 shows a direct comparison of isometric force recordings in a single wt and mdx EDL muscle fiber that is gradually exposed to solutions with increasing free Ca 2+ concentrations (decreasing pCa). To complete a pCa step, the solution was exchanged at least three times, as reflected by the force artifacts during the manual exchange. At the end of each pCa step, an SHG stack was recorded, from which the CAS and VD (#/100 µm 2 ) were calculated. Figure 2 demonstrates lower isometric force amplitudes in mdx mouse samples versus the wt mouse samples and lower CAS and higher VD values throughout the pCa steps. An interesting and somewhat unexpected finding was the increase in VD and decrease in CAS at a higher Ca 2+ activation indicative of less ordered myofibrillar alignment, regardless of the underlying genotype. Figure 3 shows analyses of those forces and SHG-derived structural parameters in more detail. First, Fig. 3a shows the SHG images of a wt single fiber and an mdx single fiber under the relaxed and pCa 6.03-activated conditions, revealing that the contraction is not purely isometric and that the sarcomere lengths decrease with Ca 2+ activation due to the finite stiffness of the transducer pin needle. This needle allows deflections at given forces that are detected using SHG morphometry. As seen in Fig. 3b , those deflections are larger in the wt fibers than in the mdx fibers due to the larger force production range in the wt fibers. When correlating specific force values from single wt and mdx fibers with the CAS and VD over a large pCa range, the Ca 2+ -graded increase in myofibrillar disorder (CAS↓, VD↑), shown in Fig. 2 , was fully confirmed with a more pronounced range in the otherwise highly ordered wt fibers under the resting pCa 9 condition compared to the already disordered mdx fibers. The mdx fibers in particular, started with very large VD values at pCa 9 that were already approximating maximum values and remained fairly constant during the activation with decreasing pCa ( Fig. 3d ) . An advantage of the graded pCa activation was the possibility of extracting Ca 2+ -force biosensor curves alongside the SL, diameter, CAS and VD values from the SHG analysis and the corresponding pCa-force curves, as exemplarily shown for wt and mdx fibers in Fig. 4a . The sensor curve was right-shifted for the mdx fiber, resulting in a lower half-activation pCa, pCa 50 , of the contractile apparatus, which is indicative of a reduced myofibrillar Ca 2+ sensitivity. This was also confirmed in the group analysis of several single fibers reaching statistical significance for smaller pCa 50 (Fig. 4d ), alongside with significantly reduced maximum force and specific force (Fig. 4a, b ), and aberrant ultrastructure (CAS↓, VD↑) in dystrophic fibers (Fig. 4f, g ). When combining the wt and mdx data to correlate the normalized forces and contractile Ca 2+ -sensitivity with ultrastructural parameters of myofibrillar order (VD, CAS), significant correlations of structure with biomechanical performance were found (Fig. 5 ). Thus, based solely on the SHG assessment of the ultrastructure of a single fiber, a prediction of the active force is possible. Fig. 1: A novel biomechatronics system (MechaMorph) for the simultaneous assessment of isometric, Ca 2+ -activated force and SHG multiphoton imaging in single muscle fibers. a 3D CAD sketch of the miniaturized biomechatronics device containing an interchangeable muscle fiber chamber to fit onto a stage of a multiphoton microscope. A force transducer and voice coil actuator pin are connected to a horizontal trough, each fabricated from cannula needles to take up a single muscle fiber ( d ). b photograph of the engineered device which is inserted in between two objective lenses to record forward scattered SHG, while the custom-made biomechatronics software runs on a parallel computer ( c , d ) Full size image Fig. 2: Ca 2+ -activated force and SHG multiphoton imaging simultaneously performed in single wt and mdx EDL muscle fibers. Representative example recordings of force (top) and myosin SHG signals (bottom) from a single EDL fiber from a wt mouse (left) and an mdx mouse (right) during successive solution exchange for increasing Ca 2+ concentrations (decreasing pCa). Brief positive force spikes represent the time points of manual solution exchange performed several times per pCa step. Scale bar: 20 µm. CAS: cosine angle sum (a.u.). VD: vernier density (#/100 µm 2 ) Full size image Fig. 3: Distortion of sarcomeric and myofibrillar structures in single EDL muscle fibers from both wt and mdx mice during active Ca 2+ -induced isometric force generation. a representative example SHG images of a single EDL fiber from a wt mouse (17 weeks) and an mdx mouse (27 weeks) in the relaxed pCa 9 and pCa 6.03 activated states. During mechanical activation, although isometric, the sarcomere lengths visibly shorten and fiber diameters increase. b analysis of the reduction in single fiber sarcomere length during Ca 2+ -dependent force generation (logarithmic plot of specific force) in wt mice (filled circles) and mdx mice (open squares); pCa values color-coded. c , d , changes in myofibrillar/sarcomeric ultrastructural parameters of the cosine angle sum, CAS ( c ), and vernier density, VD ( d ), indicate a Ca 2+ -graded increase in the myofibrillar disorder (CAS↓, VD↑) that is more pronounced in the ordered wt EDL fibers, while single fibers from mdx mice are already highly disordered under relaxed conditions Full size image Fig. 4: Myofibrillar ultrastructural disorganization in relaxed mdx EDL fibers is a predictor of reduced contractile performance and reduced Ca 2+ sensitivity of the contractile apparatus. a SHG images of single EDL fibers from a wt mouse (14 weeks) and an mdx mouse (79 weeks) in the relaxed state (pCa 9) with indicated sarcomere lengths (SL), maximum diameters, VD and CAS. Steady-state force at a given pCa normalized to the maximum force with indicated pCa 50 and Hill parameters for the shown fibers. Single mdx EDL fibers showed vast biomechanical deficits in active force production: significantly reduced max. absolute ( b ) and specific force ( c ) at a pCa of 4.92–5.67 and significantly reduced pCa 50 values indicative of myofibrillar Ca 2+ desensitization ( d ) with otherwise similar Hill parameters ( e ). This correlates well with myofibrillar structural deficits, as shown by the markedly increased VD ( f ) and decreased CAS ( g ) in mdx mice over wt mice. At similar sarcomere lengths ( h ), fiber diameters were larger in mdx EDL fibers compared than in wt fibers ( i ). Box plots with box (25 to 75 percentiles), median (line), whiskers (5–95 percentiles), minimum and maximum (x), mean (rectangle) and significances from one-way ANOVA with post hoc Bonferroni test (equal variance) or post hoc Tukey test (no equal variance) indicated as p < 0.05 (*) and p < 0.01 (**) Full size image Fig. 5: Pearson correlations show significant correlations between SHG-derived structural data and active isometric force parameters in single EDL fibers. Both structural parameters, the CAS and VD, obtained in relaxed (pCa 9) wt (gray) and mdx (red) single fibers show significant correlations with pCa 50 and the maximum force per diameter in the activated state. In particular, the more ordered the myofibrillar arrangement (CAS↑, VD↓), the higher the probability for a higher Ca 2+ sensitivity (larger pCa 50 ) ( a , b ), and thus, a higher predicted force production upon Ca 2+ activation ( c , d ). Thus, the CAS positively correlates with both pCa 50 and the max. force ( a , c ), while VD negatively correlates with both ( b , d ) Full size image Passive stretch impairs myofibrillar alignment by increased axial stress to similar extents in wt and mdx fibers Apart from assessing active myofibrillar Ca 2+ -activated forces, the design of our MechaMorph system also allows assessment of SHG ultrastructural changes with passive strains in single fibers. Figure 6a shows a sequence of recorded passive restoration forces in response to very fast 50 µm step extensions in a single mdx EDL fiber and the corresponding SHG image taken at the end of the viscous relaxation phase. That particular mdx fiber showed substantial ultrastructural abnormality, e.g., fiber branching. The restoration force pattern always followed an instantaneous elastic restoration force to F max , followed by exponential viscous relaxation to a steady-state level F eq , as indicated in the force trace from another mdx fiber in the inset of Fig. 6a . The mdx fibers were more fragile during step length changes (Fig. 6b ), which was probably the result of a significantly reduced viscous relaxation, as indicated by the reduced ( F max − F eq )/ F max amplitudes in the dystrophic fibers (Fig. 6k ). The sarcomere lengths (Fig. 6d ), fiber diameter (Fig. 6e ), rupture stress (Fig. 6h ), maximum stress (stress corresponding to F max ) (Fig. 6i ), and equilibrium stress (stress corresponding to F eq ) (Fig. 6j ) were not significantly different between the two genotypes. Intriguingly, when looking at the SHG ultrastructure, ongoing fiber stretch not only resulted in spreading of the sarcomere lengths but, most importantly, in a marked myofibrillar disorder, as reflected by a marked decline in the CAS (Fig. 6c ). In particular, there was a prominent bend of the myofibrillar lattice spacing towards the periphery of the fibers (Fig. 6c ). Lumping together the CAS and VD values for all stretches did not reveal any difference between the wt and mdx fibers (Fig. 6f, g ). When stretching fibers, however, one has to keep in mind that with an increasing SL, fewer verniers will remain in the respective field of view. This phenomenon can be compensated for by using a stretch-corrected VD, as described in the Methods. Thus, when correlating CAS and the stretch-corrected VD to the maximum stress values and the SL in all stretched single fibers, Pearson correlations confirmed significant relationships between the SL and the degree of myofibrillar disorder, i.e., a massively declining CAS and an increasing corrected VD (Fig. 7a, b ). Since increasing the SL also means increasing restoration forces and thus increasing maximum stress, the resulting correlations for the maximum stress values corresponded to the SL-CAS/-VD corr relationships (Fig. 7c, d ). Thus, these data confirm a marked myofibrillar disorder with stretch, which was similar in the wt and mdx fibers and thus was not related to the presence or absence of dystrophin in single EDL fibers. Fig. 6: Simultaneous assessment of passive single fiber viscoelasticity biomechanics and SHG myofibrillar ultrastructure in wt and mdx EDL muscles. a sequence of SHG images (left) taken during a protocol stretching a single EDL fiber from an mdx mouse in 50 µm steps and recording restoration force (right). Images were taken at the time points indicated. The inset shows force recording of another mdx fiber: instantaneous maximum restoration force F max at each stretch was followed by a double-exponential viscous relaxation to a new steady-state elastic force level F eq . b mdx single EDL fibers already broke at lower strains compared to wt fibers, compatible with larger stiffness, respective lower viscous relaxation. c sequence of SHG images taken from a wt single EDL fiber stretched to the indicated sarcomere lengths (SL) and analyzing CAS values. Note that with the increase in stretch, a marked decline in CAS can be detected as visualized by A-band bending across the fiber cross-section. The SL value ranges were similar in wt and mdx fibers ( d ), as were the fiber diameters ( e ), overall CAS values ( f ) and VDs ( g ). Biomechanical passive parameters of the maximum rupture stress ( h ), maximum stress values from all stretches ( i ) and equilibrium strains ( j ) were similar among the wt and mdx fibers while the viscous stress relief ΔF/F max values ( k ) were significantly smaller in the mdx fibers than the wt fibers, indicating a lower viscosity in the dystrophic genotype. # P < 0.025. n = (a/b/c) depicts the data from a images on b fibers from c animals Full size image Fig. 7: Pearson correlations show significant correlations between the SHG-derived structural data and the passive elastic parameters, as well as a higher degree of myofibrillar disorder with the sarcomere length in stretched single EDL fibers. Both structural parameters, the CAS ( a ) and the stretch-corrected VD ( b ), obtained in relaxed (pCa 9) wt (gray) and mdx (red) single fibers show significant correlations with the sarcomere lengths. In particular, the higher the stretch was (SL↑), the less ordered the myofibrillar arrangement became (CAS↓, VD↑). Similarly, the myofibrillar disarray, i.e., the low CAS ( c ) and large VD values ( d ), significantly correlated with increased maximum stress values that occurred during stretching Full size image Discussion Structure and function studies in skeletal muscle Apart from organ-tissue-cellular function being regulated on short to intermediate time scales by a plethora of signaling pathways, e.g., involving fast Ca 2+ signaling or slower G-protein related second messenger cascades, a long-term predictor of cell function is also encoded in the architecture, either within the cells (cytoarchitecture) or affecting the extracellular matrix, ECM (e.g., in fibrosis). In skeletal muscle, genetic, degenerative, or chronic inflammatory diseases have regularly been associated with mechanisms of weakness related to aberrant signaling and structural long-term remodeling, but to date the direct interaction between these two processes has not been determined. This is mostly due to a lack of appropriate technologies that would allow assessment of the structure–function relationship on the single fiber level. While force production in single muscle fibers or even in myofibrils has been studied over the recent decades using various force transducer technologies (e.g., refs 28 , 29 , 30 ), many more studies assessing active forces in muscles are available than those assessing passive viscoelastic behavior. This fact is mostly due to the necessity to manually operate many of the custom-made and commercial systems used to date. The systems usually lack a degree of automation and µm-level precision required for quasistatic steady-state resting-length tension curves or ultrafast and precise stretch experiments to study viscoelastic behavior in single fibers, unlike the more coarse systems used for whole muscle 31 . Therefore, we have very recently introduced a novel automated voice coil actuator-driven system that allows full control of fiber lengths at very slow or fast speeds to record active and passive forces in single muscle fibers 32 . To obtain detailed structural information on the same fiber in situ, we used this concept to integrate our so-called MechaMorph into a multiphoton microscopy environment. This unique system allows simultaneous assessment of structure and force, as shown in the present study. SHG has increasingly been used in recent years to study structural changes in skeletal muscle disease models. In healthy muscle, the regular sarcomere pattern reflected by the aligned myofibrillar architecture can be elegantly exploited by the myosin-origin of skeletal muscle SHG to determine the myofibrillar cytoarchitecture in 3D 3 , 33 or even to dissect contractile states of the motor protein interaction 34 , 35 . In lysosomal storage disease, SHG in conjunction with 2-photon excited fluorescence was used in muscle biopsies from patients with Pompe’s disease or acid α-glucosidase knockout mice to identify intramyoplasmic areas void of SHG signals, representing autophagic debris accumulation 36 , 37 . In fibers from these models, wavy and pitted myofibrillar patterns were predominant. Since similar patterns, i.e., wavy and disorganized sarcomeres, were also observed in dystrophic muscle, both in sections 20 and in single fibers 14 , quantitative morphometry approaches were subsequently developed by various groups 20 , 38 . In addition to Fourier-transform based analyses of sarcomere patterns 38 or single-frequency wavelet-based Gabor filtering to quantify structural disorder 22 , our group developed pattern recognition algorithms based on boundary tensor analysis into fully automated pattern extraction routines to define two major structural parameters of myofibrillar order, reflected by the CAS (a measure for myofibrillar parallelism) and the density of so-called verniers (a measure for out-of-register appearances of myofibrils) 14 , 23 , 24 , 25 . In various subsequent age-related studies applying this set of SHG morphometry tools to dystrophic mdx or R349P mutant desmin muscle, we established these morphometric parameters for structural diagnosis of ‘myopathy’ and for monitoring disease progression with age 24 , 25 , 26 , 27 . Our results using the MechaMorph fully confirm the structural CAS and VD parameters in mdx single fibers under resting pCa 9 conditions from our previous studies. Moreover, with the advent of combined force-SHG recordings, we can now directly observe the myofibrillar ultrastructure during defined Ca 2+ activation and during passive stretch. In conjunction with the multiphoton endoscopy system that is currently being developed in our institute, we hope to apply our SHG diagnostics and analysis of muscle tissue in vivo through cannula-based endoscopy in the near future. Even though the backscattered SHG signal will be of lower intensity as compared to transmitted forward-detected signal, we estimate, based on published work, that the signal-to-noise ratio is high enough to conduct morphology analysis 39 , 40 . Fibrillar molecules such as collagen-I emit coherent SHG in backward direction 41 , as opposed to myofibrillar-based SHG that mainly appears in backward detection due to scattering effects. Polarization-selective analysis of the signal 42 , 43 in backward direction might help to further decrease noise in a non-transmission setup. Combined SHG morphometry and active biomechanics in single fibers Although the isometric force levels during maximum Ca 2+ activation in dystrophic mdx fibers in our setting confirmed the findings from the literature of significantly reduced specific forces (Fig. 4 ) 17 , 44 , 45 , a very novel and somewhat unexpected finding was that with increasing myofibrillar tension, myofibrillar disorder increased as indicated by decreasing CAS values and increasing VDs (Fig. 3 ). This increase in the myofibrillar angular variability was similar to that of mdx fibers even in the wt fibers and seems to represent a feature that is unrelated to muscular dystrophy. There is not much information about changes in sarcomere geometries under isometric contractile conditions with graded Ca 2+ activation. Early snap-freezing and subsequent EM studies in glycerinated fiber bundles from rabbit psoas muscle showed no major changes in the filament lengths of actin or myosin within sarcomeres under rigor, isometric contraction or relaxing conditions 46 . Using an E-shaped tendon force transducer clamped to a tibialis anterior muscle of living mice and applying SHG imaging to an electrically stimulated whole muscle in vivo, a vast increase in the sarcomere length variability was seen as a broadened SL distribution, in particular at short muscle lengths (90° ankle) over long muscle lengths (180° ankle) 47 . That very recent study focused on planar images from the top 100 µm and collecting backscattered SHG signals from whole muscle without further details on the 3D aspect of myofibrillar structure under contractile activation; however, it may provide a basis to explain our observed increase in angular disorder of myofibrils if one assumes that lateral forces resulting from different sarcomere lengths of adjacent myofibrils connected by extrasarcomeric proteins, for example desmin, would induce tilts and twists to myofibrils. A study by Nucciotti et al. 35 also used SHG-imaging of single fibers connected to a force transducer to perform SHG polarization anisotropy (SPA) recordings at different sarcomere lengths under relaxed versus rigor conditions, but not under defined Ca 2+ activations. Although myosin motor head states could be derived from those SPA analyses, the 3D ultrastructure was not obtained. Thus, our approach represents a more intuitive assessment of myofibrillar alignment, both angular as well as in-register alignment in full 3D with the simultaneous biomechanical assessment of the active and passive force. While VDs increased with the specific force (increasing Ca 2+ ) in wt fibers, they remained stationary in mdx fibers, probably already reflecting a maxed out upper limit for axial misalignment in this model. The graded Ca 2+ activation clearly provided a wealth of novel information on not only the correlation between maximum isometric force and corresponding stress to myofibrillar ultrastructure but also on the myofibrillar Ca 2+ sensitivity which, in our results, was significantly reduced. This may initially appear to conflict with unchanged myofibrillar Ca 2+ sensitivity in previous mdx studies 48 , 49 , 50 . However, in particular in such mdx fibers presenting with a high degree of angular variability, the effective contractile Ca 2+ sensitivity, which is the contractile readout variable in the axial direction, may still be reduced since mechanical fiber activation by Ca 2+ will inevitably activate force contributions deviating from the main fiber axis 14 , thus requiring higher Ca 2+ levels to compensate for this angular disorder of the myofibrillar pull. The maximum specific force values assessed in our EDL single fibers of ~10 N/cm 2 in the wt fibers are well comparable to other studies using single skinned mouse muscle fibers 49 or fibers from humans 51 . The reduced specific force in dystrophic muscle is, to varying extents, well established in mdx mice 17 , 49 , while in DMD patient quadriceps or biceps single muscle fibers, the maximum isometric force was not compromised 51 . However, our study was not intended to further elaborate on biomechanics of single mdx muscle fibers. Rather, the unique opportunity to assess Ca 2+ -graded force and SHG morphometry in the same fiber motivated us to use the mdx model as a tool to obtain SHG-force data sets that we would normally not obtain from healthy muscle fibers, vastly compromised myofibrillar 3D structures and reduced isometric forces. Thus, combining the data sets from single wt and mdx EDL fibers enabled us to calibrate the SHG parameters to the isometric force over a large range ( Fig. 5 ) . These significant correlations now prove what has only been suggested previously: a linear decline in force and Ca 2+ -sensitivity of the contractile apparatus with myofibrillar disorder (indicated by a decreasing CAS and an increasing VD). This is a very new finding, here experimentally verified for the first time. It will open new methods of projecting the force output of muscle fibers, either isolated or within tissue samples, by simply investigating them using SHG imaging. Combined SHG morphometry and passive biomechanics in single fibers Implementing our previous voice coil strategy 32 , also in the MechaMorph system, allowed us to perform detailed viscoelasticity assessments on the single fiber level, which is very rare in the literature (for rabbit single psoas and soleus muscle fibers; 51 for mouse single tibialis ant . muscle fibers 52 ). As previously described, sudden stretches are answered by an instantaneous increase in the passive restoration force, followed by a decline in the force due to viscous relaxation of titin filaments 32 , 51 , 52 . Our passive stresses are in the range of tens of kPa and increase with stretching, which is in agreement with other studies on single muscle fibers 52 . In our stretch experiments, mdx EDL fibers showed similar morphometric parameters to wt fibers, e.g., sarcomere lengths, SHG-derived CAS and vernier densities (corrected for the stretch amount to account for sarcomeres leaving the field of view during stretching), but significantly reduced force relaxation at an unaltered maximum stress, compatible with a preferential impairment of the viscoelasticity over the steady-state stiffness. Additionally, mdx fibers broke at lower strains, supporting a compromised passive relaxation. Within the literature, differential results were reported on dystrophic mdx EDL muscles either showing no compromise in passive mechanical properties during maturation in mice up to 35 days of age 53 or in 21 months old mdx mice 54 , while another detailed age-related study showed a consistently increased passive stress over a large range of strains in animals aged between 2 and 20 months, compatible with a higher stiffness and compromised viscosity 55 . One reason for the disparate results may be that all those mentioned studies exclusively used whole muscle. Specifically, connective tissue may have a vast impact on passive muscle biomechanics 55 , although one study found that the extracellular collagen content did not alter passive mechanical properties in whole muscles of mdx mice 56 . We are not aware of any study on single fiber passive biomechanics in the mdx model, and thus our results truly reflect the myofiber viscoelasticity without the influence of ECM components. Similar to the active force data, the mdx model was merely included to extend the biomechanical stress range. However, despite being similar to the wt range, the mdx dataset actually expanded the wt dataset toward a higher number of structure-biomechanics pairs for the Pearson correlations, demonstrating, for the first time, that increasing the stretch and thus the passive stress resulted in a vast increase in myofibrillar disorder, as seen by the decrease in CAS and increase in VD to levels beyond those observed in resting fibers, even in strongly abnormal mdx fibers 14 , 24 . The significant linear correlations confirm a direct influence of the mechanical stress on the myofibrillar structure. The prominent drop in the CAS is particularly explained by the observed bending of A-band signals with stretching ( Fig. 6c ) that increases the angular distribution of the local force vector orientations and therefore, diminishes the CAS. In summary, our new combined SHG-biomechatronics approach introduced as the MechaMorph system with high precision voice coil technology has enabled us, for the first time, to obtain direct structure–function data pairs in terms of the isometric force, passive viscoelasticity and SHG quantitative morphometry and to establish significant linear correlations between the structure and function at the single fiber level. Thus, we propose that our MechaMorph approach adds a further dimension to SHG-derived morphometry resulting in a new noninvasive methodology to predict forces and biomechanical performances in skeletal muscle exclusively using optical assessments, suggesting a translational potential for monitoring disease progression or remission in patients. Material and methods Animals and single muscle fiber preparation Adult mdx mice carrying a missense mutation in exon 23 of the dystrophin gene (strain: C57BL/10ScSn- Dmd mdx /J) were compared to wild type (wt) mice (C57BL/6N, obtained from Charles River Company). The wt mice were between 13 and 21 weeks of age, and the mdx mice were between 27 and 91 weeks of age. We chose that particular age difference because the primary focus was not to compare the differences between the wt and mdx phenotypes but to collect a wide range of myofibrillar disorganization and associated impaired biomechanics. In particular, single fibers from older mdx mice show reduced CAS levels and increased VD levels unparalleled by adult or aged wt animals. Therefore, the obtained range of optical parameters of the myofibrillar architectures and isometric forces allowed us to correlate both sets of data for a force-SHG calibration. Animal handling was in accordance with the German Animal Welfare Act (Tierschutzgesetz) as well as the German Regulation for the protection of animals used for experimental purposes or other scientific purposes (Tierschutz-Versuchstierverordnung). The investigations were approved by the governmental Office for Animal Care and Use (Regierung von Mittelfranken, Ansbach, Germany; reference number TS-14/2015). All applicable international, national, and institutional guidelines for the care and use of animals were followed. Animals were anesthetized using Isofluorane inhalation. After verifying deep sedation by the absence of a pain reflex when pinching the skin, animals were sacrificed by neck dislocation and both hindlimbs were cut off. The extensor digitorum longus (EDL) muscle was manually dissected in Ringer’s solution, pinned to an elastomer-coated dish and solution exchanged to a ‘high K + ‘-relaxing solution (HKS, in mM: K-glutamate 140, Hepes 10, glucose 10, MgCl 2 10, EGTA 1, pH 7.0) under isometric conditions. Single muscle fiber segments of 2–3 mm length were then dissected in HKS solution through manual tethering of muscle fascicles using fine forceps 26 , 27 . MechaMorph biomechatronics system Our system (Fig. 1 ) consists of a force transducer element (FT) using in-built optical metrology to measure pin deflection in response to external forces (TR5 S, resonance frequency 550 Hz, range 1.5 µN–0.5 N, compliance 0.7 µm/mN, Scientific Instruments, Heidelberg, Germany), and a software-controlled voice coil actuator, VCA, (SMAC CAL12-010-51BSA, Ispringen, Germany) that allows force- and length-controlled precision feedback positioning 32 . Both the sensor and actuator are aligned in the same plane and are screwed to a steel frame containing one microscrew-driven sledge to lower or lift the apparatus from a single fiber bath chamber clipped into a drilled groove holder on top of the microscope stage plate (Fig. 1a ). A coarse micrometer screw allows preadjustment of the x-coordinate along the fiber axis, and the VCA then performs controlled fine-tuned length adjustments in the µm domain (x-position). Fig. 1b shows a photograph of the MechaMorph prototype, and Fig. 1c shows a schematic of the system incorporation within the two-objective multiphoton microscope. Both the transducer and voice coil pin were glued to a horizontal trough made from cut-grinding a half-opened polyethylene (PE) tubing (PORTEX SX05, 0.58 × 0.96 mm, A. Hartenstein GmbH, Würzburg, Germany) on each side (Fig. 1d ). Multiphoton SHG microscopy Single fibers were imaged using a multiphoton microscope, MPM (TriMScope II, LaVision BioTec, Bielefeld, Germany). A mode-locked ps-pulsed Ti:Sa laser (Chameleon Vision II, Coherent, Santa Clara, CA, USA) was used to excite the SHG signal of myosin-II. The average laser power on the sample was ∼ 16 mW, the pulse duration was ∼ 150 fs and the repetition rate was 80 MHz. A symmetric transmitted light configuration of two water immersion objectives was used for detection. On the excitation side (backscattered, descanned), an LD C-Apochromat lens (40 × /1.1/UV-VIS-IR/WD 0.62, Carl Zeiss, Jena, Germany) was used, and on the transmission side (forward scattered, non-descanned), a W Plan-Apochromat lens (20 × /1.0/(UV)VIS-IR/WD 1.88/DIC M27 75 mm, Carl Zeiss) was used. In the active force experiments, the recorded images had a size of 200 × 200 µ m, consisting of 1024 × 1024 pixels, and two to three scans of each pixel were performed at a line frequency of 600 Hz or 1000 Hz and were averaged. The averaging was performed line by line, where the arithmetic mean was taken of the intensity of each pixel. The frame time of a double line averaged image is approximately 4.2 s. With a pixel dwell time of 0.8 µs and a sweep time of approximately 4 µs, the energy that is deposited in each pixel of the sample is 64 nJ, which is approximately three orders of magnitude below the reported damage thresholds 57 , 58 , 59 . SHG signals were excited at 810 nm and detected using a 405/20 nm single bandpass filter (Chroma Technology group, Acal BFi Germany GmbH, Gro¨benzell, Germany) and an ultrasensitive, non-descanned transmission photomultiplier tube (PMT) (H 7422-40 LV 5 M, Hamamatsu Photonics). The linearly polarized excitation light is aligned to 50° or 130° of the fiber orientation 34 by a rotating half-wave plate, which is located just before the excitation side objective, to maximize the SHG signal intensity. The z-stacks had a step-size of 0.5 µ m for recordings in the relaxed state (pCa 9) up to 6.0 µm for recordings at different pCa steps. For the stretch experiments, the z-stack step sizes were 1.0 µ m. Image processing and morphometric analysis of the SHG images were conducted as previously described 24 , 26 , 27 to detect verniers , i.e., Y-shaped deviations from the sarcomere pattern in a z-stack of the SHG images and to measure the CAS, reflecting the degree of local angular deviation of myofibrillar bundles from the main fiber axis. The normalized VD and the CAS were calculated over fiber z-stacks in the relaxed state before Ca 2 + activation (i.e., the relaxed state, pCa 9) and in steady-state of each indicated Ca 2+ activation (pCa) step. For processing of the SHG image stacks that included vast changes in the SHG intensity with sarcomere lengths 3 during the stretch experiments (see below), a wavelet-based filter procedure was applied to increase the signal-to-noise ratio 60 . The fiber diameter was obtained by plotting the width of the optical fiber section of each image slice within a z-stack and fitting a Gaussian profile to the data. From the mean fiber diameter, the circular cross-sectional area (CSA) was calculated. The sarcomere lengths (SL) were obtained from the SHG images by thresholding and particle analysis. Specifically, a linear plot in the Fourier domain along the direction of the fiber was created and its peaks were analyzed. The peak with the second highest amplitude was considered to represent the main spatial signal frequency in the main fiber direction, thereby representing the mean SL of the image slice. The mean SL of the entire stack was determined by averaging the SLs over each slice. Slices in which less than 10% of the pixel entries accounted for intensities of at least 10% of the maximum intensity in the stack were not included in the evaluation. To automate the evaluation of the CSA and SL, a macro was written in ImageJ (NIH software, ). Active, Ca 2+ -activated force recordings Single EDL fiber segments were mounted between the FT and the VCA pin with the actuator-sensor block lowered into a chamber filled with HKS. The ends of the cut fiber were placed into the troughs and clamp-fixed by clipping in a polyethylene tubing clip. The solution in the chamber was first exchanged for 0.1% (w/v) saponin (Sigma-Aldrich Chemie GmbH, Steinheim, Germany) in a high relaxing solution (HR, mM: Hepes 30, Mg(OH) 2 6.25, EGTA 30, Na 2 ATP 8, Na 2 -creatine phosphate 10, pH 7.2) for 20 s to chemically permeabilize the fiber. The fiber was then washed with HR, and the device was mounted onto the adapted MPM stage. By moving the static pin under microscopic control, the sarcomere length (SL) of a fiber was adjusted to 2.2 – 3.1 µm and a z-stack of the fiber was imaged to capture the structure in the relaxed state (pCa 9). The Ca 2+ -activated force was assessed by successively bathing the fiber in solutions with successively decreasing pCa and recording the force until steady-state levels were reached. The solution exchange at each pCa was manually performed using Eppendorf pipettes as that version of the MechaMorph system did not yet contain a microfluidics control system (engineering in progress). The maximum activation was measured at a pCa of 4.92 in an undiluted highly activating solution (HA, mM: Hepes 30, Mg(OH) 2 6.05, EGTA 30, CaCO 3 29, Na 2 ATP 8, Na 2 CP 10, pH 7.2). Specific force was calculated using the fiber diameter in the z-stack taken at the respective pCa. Conducting a sigmoidal Hill fit to the data yielded the Ca 2+ -sensitivity of the contractile apparatus in each single fiber (pCa 50 ) and the slope of the sensor curve. Passive viscoelasticity recordings during step-stretch experiments Single EDL fiber segments were clamp-fixed between FT and VCA pins using a PE tubing clip while submerged in an HKS solution. After recording the initial clamping length as a baseline, the fiber was stretched stepwise by moving the VCA using custom-written LabView software. This involved initial stretches of 50–100 µm, eventually up to 400 µm, if the fiber was initially slightly slack and thus, the restoration forces were still low. After each step of length change, the restoration force F R was continuously recorded until, after its initial instantaneous increase to a maximum value F max , the following relaxation phase reached a steady-state of force decline ( F eq , at least 5 min to ensure steady-state). Then, an SHG 3D image stack of the fiber was recorded before proceeding to the next stretch step. Axial fiber strain was obtained from the current VCA position. F max was automatically determined from the min–max analysis of the force trace during each step in which F eq was extracted from an exponential fit to the force decline during holding of the fiber at the given stretch. The force values were converted to stress using the CSA extracted from the SHG image stacks. Statistical analysis For data comparison, a one-way ANOVA analysis (Sigma Plot, Systat Software) was applied on the two genotypes, wt and mdx , with a post hoc Bonferroni test (equal variance) or post hoc Tukey test (no equal variance) where indicated. p < 0.05 was considered significant (*), and p < 0.01 was considered highly significant (**). The normality of data was tested using the Shapiro–Wilk test. Data are presented as box plots (median value: line, quartiles: whiskers 5–95 percentiles, minimum/maximum values:x, mean: rectangle). Pearson correlations were calculated using the online tool , or SigmaPlot.

Biotechnologists at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) have developed a system to accurately measure muscle weakness caused by structural changes in muscle tissue. The new method allows muscle function to be assessed using imaging without the need for sophisticated biomechanical recordings, and could in future even make taking tissue samples for diagnosing myopathy superfluous. The results have been published in the renowned journal Light: Science & Applications. The muscle is a highly ordered and hierarchically structured organ. This is reflected not only in the parallel bundling of muscle fibres, but also in the structure of individual cells. The myofibrils responsible for contraction consist of hundreds of identically structured units connected one after another. This orderly structure determines the force which is exerted and the strength of the muscle. Inflammatory or degenerative diseases or cancer can lead to a chronic restructuring of this architecture, causing scarring, stiffening or branching of muscle fibres and resulting in a dramatic reduction in muscular function. Although such changes in muscular morphology can already be tracked using non-invasive multiphoton microscopy, it has not yet been possible to assess muscle strength accurately on the basis of imaging alone. New system correlates structure and strength Researchers from the Chair of Medical Biotechnology have now developed a system that allows muscular weakness caused by structural changes to be measured at the same time as optically assessing muscular architecture. "We engineered a miniaturized biomechatronics system and integrated it into a multiphoton microscope, allowing us to directly assess the strength and elasticity of individual muscle fibres at the same time as recording structural anomalies," explains Prof. Dr. Oliver Friedrich. In order to prove the muscle's ability to contract, the researchers dipped the muscle cells into solutions with increasing concentrations of free calcium ions. Calcium is also responsible for triggering muscle contractions in humans and animals. The viscoelasticity of the fibres was also measured, by stretching them little by little. A highly-sensitive detector recorded mechanical resistance exercised by the muscle fibres clamped on the device. Data pool for simplified diagnosis The technology developed by researchers at FAU is, however, merely the first step towards being able to diagnose muscle disorders much more easily in future: "Being able to measure isometric strength and passive viscoelasticity at the same time as visually showing the morphometry of muscle cells has enabled us, for the first time, to obtain direct structure-function data pairs," Oliver Friedrich says. "This allows us to establish significant linear correlations between the structure and function of muscles at the single fibre level." The data pool will be used in future to reliably predict forces and biomechanical performances in skeletal muscle exclusively using optical assessments based on SHG images (the initials stand for Second Harmonic Generation and refer to images created using lasers at second harmonic frequency), without the need for complex strength measurements. At present, muscle cells still have to be removed from the body before they can be examined using a multiphoton microscope. However, it is plausible that this may become superfluous in future if the necessary technology can continue to be miniaturized, making it possible for muscle function to be examined, for example, using a micro-endoscope.

10.1038/s41377-018-0080-3

Other

Study finds nationality is not a good indicator of work-related cultural values

Vas Taras et al, Does Country Equate with Culture? Beyond Geography in the Search for Cultural Boundaries, Management International Review (2016). DOI: 10.1007/s11575-016-0283-x

http://dx.doi.org/10.1007/s11575-016-0283-x

https://phys.org/news/2016-05-nationality-good-indicator-work-related-cultural.html

Abstract Traditionally, cultures have been treated as though they reside exclusively within, or perfectly overlap with countries. Indeed, the terms “country” and “culture” are often used interchangeably. As evidence mounts for substantial within-country cultural variation, and often between-country similarities, the problem with equating country and culture becomes more apparent. To help resolve the country-culture conundrum, we evaluate the extent to which political boundaries are suitable for clustering cultures based on a meta-analysis of 558 studies that used Hofstede’s (Culture’s consequences: international differences in work-related values. Sage Publications, Beverly Hills, 1980 ) cultural values framework. The results reveal that approximately 80 % of variation in cultural values resides within countries, confirming that country is often a poor proxy for culture. We also evaluate the relative suitability of other demographic and environmental characteristics, such as occupation, socio-economic status, wealth, freedom, globalization, and instability. Our results suggest that it may be more appropriate to talk about cultures of professions, socio-economic classes, and free versus oppressed societies, than about cultures of countries. Access provided by Universität des es, -und Working on a manuscript? Avoid the common mistakes 1 Introduction Traditionally, cultures have been assumed to reside within countries. From the earliest studies of cultural differences dating back over two centuries (Darton 1790 , see Fig. 1 ) to more recent research by Hofstede ( 1980 ) and the GLOBE team (House et al. 2004 ), the unit of analysis in cross-cultural studies has typically been country. The focus on national cultures has appropriate applications, but its dominance in research has led to a lack of attention to other plausible organizing units of culture. Equating cultures with countries and using country of origin and individual culture interchangeably became a common practice (Brewer and Venaik 2012 ). However, the appropriateness of this trend depends on answers to two questions, including: (1) are countries good proxies for cultures, and (2) could other factors be superior for describing boundaries of cultural regions or groups of people who display similar cultural values? Fig. 1 Page from one of the earliest studies into cultural differences by William Darton ( 1790 ) Full size image Culture is a multi-faceted construct. First, the word culture has many meanings, from a collective of people who share a common history, language and traditions, to characteristics of such a collective in terms of its artifacts, practices, and value systems. Illustrating the pervasiveness and possibly ambiguity of the term “culture,” a Google search for “culture” and its derivatives returns 1.5 billion hits, making it one of the most popular words on the Web. More than 60 years ago, Kroeber and Kluckhohn ( 1952 ) found 164 distinct definitions of culture. Despite the variety of definitions, several elements are common across most of them, principally that culture is: (1) a relatively stable, (2) multi-level construct comprised of values, beliefs, norms, traditions, and artifacts that (3) are shared in a given population (cf. Taras et al. 2009 ). Early empirical research on cross-cultural differences has been largely qualitative and focused on describing artifacts, rituals, and social institutions. However, following the publication of Hofstede’s ( 1980 ) seminal book, Culture’s Consequences, the focus shifted to cultural values. As noted by Taras et al. ( 2009 ), “culture is values” has become one of the commandments of cross-cultural management research. The almost exclusive focus on cultural values is often justified, but also problematic and limiting in many ways, particularly if measured using Hofstede’s ( 1980 ) framework, one that was developed based on a survey not actually originally intended for cultural analysis (cf. Baskerville 2003 ; McSweeney 2002 ; Taras and Steel 2009 for reviews of problems with Hofstede’s framework in general and for studying “culture” in particular). For the purpose of the present study, however, the question of whether or not the Hofstede’s framework is suitable for studying culture due to its limitations is secondary because most empirical research on culture over the past 35 years has relied on Hofstede’s framework. As Taras and Steel ( 2009 ) concluded, since the publication of Culture’s Consequences (Hofstede 1980 ), research on culture effectively became research on values. Indeed, reviews of cross-cultural research published in management, psychology and related disciplines confirm that empirical measurement of what is called “culture” has almost exclusively focused on assessing cultural values, usually by the means of self-response questionnaires (Caprar et al. 2015 ). Subsequent models following Hofstede’s work primarily refined his framework (e.g., House et al. 2004 (GLOBE); Schwartz 1994 ) rather than substantively altering it. Even though this work changed the wording of items and the list of the dimensions, the underlying practices remained the same and are subject to the same limitations (McSweeney 2013 ). Accordingly, the outcome of just about every major cross-cultural comparative study has been a set of national cultural means and country rankings along dimensions of cultural values (Taras et al. 2009 ). Thus, to settle the culture-versus-country debate, we rely on the approach to culture that is at the foundation of most of the empirical literature on the topic. Interpreting culture as inherently inseparable from country has become popular enough that the two terms often are used synonymously. For example, the word “culture” has been routinely included in titles of publications that provided cross-country (but not culture) comparisons, as in “ The Perception of Distributive Justice in Two Cultures ” (Marin 1982 ), “ Rules for Social Relationships in Four Cultures ” (Argyle 1986 ), “ The Effect of Culture on the Curvilinear Relationship between Performance and Turnover ” (Sturman et al. 2012 ), “A Cross - Cultural Examination of Self - Leadership” (Houghton et al. 2014 ), and many others (e.g., Bagozzi et al. 2003 ; Cialdini et al. 1999 ; Goodwin and Plaza 2000 ). Similarly, there are numerous examples when nationality or country of residence are used as proxies for cultural values, as illustrated by such quotes as “ cultural background was measured by the current citizenship (passport status) of each of the managers ” (Offermann and Hellmann 1997 , p. 346), “ Individualism - collectivism was operationalized by the respondent’s native culture (country of origin) ” (Trubisky et al. 1991 , p. 73), or “ participants were divided into high and low Power Distance groups by county - of - origin ” (Eylon and Au 1999 , p. 378), and “ across two cultures (the U.S. and Korea) ” (Lee et al. 2014 , p. 692). This is not to say that the interest of cross-cultural management and psychology researchers in values is misplaced. Indeed, it is primarily the core values and beliefs, not the external cultural artifacts, that affect organizational behaviors and attitudes, and the effect of cultural values on work-related outcomes appears to be significantly stronger than that of other commonly used predictors, such as demographics or personality (for meta-analytic reviews see Fischer and Smith 2003 ; Stahl et al. 2010 ; Taras et al. 2010 ). Measuring tacit values and beliefs, however, is no easy task. Self-response questionnaire has been the method of choice, but the efficacy of this approach has inherent limitations (Riordan and Vandenberg 1994 ; Taras et al. 2009 ; Taras and Steel 2009 ). As a result, it would be very convenient and advantageous if one’s country of origin was actually a good proxy of cultural values. But, the question remains: Is it? This question has an extended history. The problem of equating country and culture has been recognized and sharply criticized for at least three decades, though attempts to address the problem have been predominantly theoretical, and much of the criticism has gone unheeded in subsequent research. As for empirical investigations, the evidence has typically targeted assessing within-country variation in cultural values, demonstrating that nations are imperfect indicators of its individual citizen’s values (e.g., Coon and Kemmelmeier 2001 ; Kaasa et al. 2014 ; Lenartowicz et al. 2003 ). The present study moves beyond confirming whether country is a good proxy for culture. By directly comparing within- and between-country variance in cultural values in a large global sample, we provide a new level of precision by assessing the exact extent to which national borders are suitable as boundaries for cultures. In addition to potentially pointing out that, yet again, cultures do not neatly compartmentalize between countries, we theorize what could be better ways to think about cultural clusters and empirically test the comparative worth of a number of alternatives, providing a foundation for moving forward on this long debated issue. 2 Theoretical Background 2.1 A Brief Review of the Culture vs. Country Discussion Thinking about cultures and countries as overlapping concepts appears to date as far back as the construct of country itself. Consistent with the fundamental cognitive bias of group stereotypes, here based on country of origin (Verlegh and Steenkamp 1999 ), we have an innate readiness to equate country with culture and have done so for centuries (e.g., Darton 1790 ). Given this inclination to conflate, Hofstede’s work providing country-averages for cultural values only made it more convenient to equate the two. His readily-available sets of national cultural indices provided a practical, low-cost and consequently attractive option for empirical research on culture. As Hofstede’s national cultural indices gained in popularity, the debate around the assumption that cultures are contained within countries was becoming increasingly pervasive. Even though the problem was identified early on, the discussion revolved around pointing out that cultures are not homogeneous within countries, and thus national averages may not adequately represent the distribution of the values in the population. Still, the solutions continued to adopt geography as the sorting mechanism, generating distinct sets of indices for different geographic regions within countries, such as assessing Switzerland’s German, Italian and French regions or the Anglophone versus Francophone provinces in Canada (e.g., House et al. 2004 ; Punnett 1991 ). Ironically, despite trying to rectify the issue of excessive within-group heterogeneity, these more granular reports do not often include within-region variance statistics, effectively treating these geographic regions themselves as culturally homogenous. Thus, even though reporting regional averages is a refinement from the practice of relying on political borders, dividing countries into regions implicitly assumes geography as the appropriate clustering dimension and does not necessarily solve the underlying problem. Hofstede responded to this criticism by stressing that his model is only suitable for the nation level of analysis and highlighting the importance of matched sampling to make the comparison of national averages meaningful (Hofstede 2002a , b , 2006 ). As a result, the discussion focused on the issue of ecological fallacy (i.e., acting as if the average represents the instance), which might have had the unintended consequence of doing more harm than good for the development of the field of cross-cultural research. Specifically, instead of trying to find more suitable dimensions for clustering cultures, researchers were preoccupied with the question of whether or not Hofstede’s indices can be generalized to the individual level of analysis (Spector et al. 2001 ), developed instruments for measuring individual cultural values (Maznevski et al. 2002 ), and researched cultural regions within countries (Huo and Randall 1991 ). Country as a proxy for culture may have been discredited, but in the absence of a better alternative, the old framework held its ground. Even the more recent large-scale studies on cultural regions remained stuck in the “country’s culture” paradigm (e.g., GLOBE country indices, House et al. 2004 ). Clearly recognizing the problem without offering a viable alternative resulted in a curious phenomenon. In the past decade or so, researchers would still use Hofstede’s (or GLOBE’s) country indices, but add a paragraph on the limitations of this approach, thereby showing their understanding of the problem and attempting to rebuff potential criticism. It also became common to add the word “national” (as in “national culture”, see Fig. 2 ) to signal the understanding of the controversy surrounding the use of national averages proxies for cultures and warn that the findings apply only at the national level, thereby making criticism redundant (e.g., Han et al. 2010 ; Kanagaretnam et al. 2011 ). Unfortunately, acknowledging limitations does not resolve them. Fig. 2 The frequency of use of the term “National Culture”, Google Ngram Full size image As a partial solution, many researchers moved on to directly measuring values in their samples by the means of surveys similar to Hofstede’s value survey module (Kirkman et al. 2009 ; Yoo and Donthu 2002 ). Even though this approach mitigates the problem of ecological fallacy, it does not address the issue of cultural boundaries. Culture is an inherently group level phenomenon, and values must be shared in a group to form a culture. Otherwise, we are likely dealing with personality, or the study of individual differences, even if we aggregate these personality profiles to the level of groups, such as nations (Hofstede and McCrae 2004 ). Unless we know the boundaries of the population in which those values are shared, direct measurement of participant values does not solve the problem of generalizability. If cultures do not cluster within countries, then just as national averages should not be taken to accurately represent individuals, so value effects found in a sample cannot be appropriately generalized to a country. If we do not know the boundaries of the population in which the given values are shared, we haphazardly generalize findings from the sample to that population. Even though numerous Editorial Letters have called for banning the “passport” approach in cross-cultural research (e.g., Caprar 2015 ; Jackson and Aycan 2006 ; Lenartowicz et al. 2003 ; Lonner and van de Vijver 2004 ; Pudelko et al. 2006 ; Tung and Verbeke 2010 ; Zander 2004 ), there has been limited response to these recommendations in empirical studies. Studies that use country of residence as a proxy for culture remain commonplace, even in leading journals (Han et al. 2010 ; Kanagaretnam et al. 2011 ; Ng et al. 2009 ). Remarkably, the effect of the country-equals-culture paradigm is so strong that cross-cultural comparative studies do not often report standard deviations or other variance statistics for the national average (e.g., Hofstede 1980 ; House et al. 2004 ); nor are the standard deviations reported (or taken into account) when calculating national averages in meta-analytic reviews of the literature on the topic, given that most of the studies never provided these statistics (e.g., Oyserman et al. 2002 ; Steel and Taras 2010 ; Taras et al. 2012 ). 2.2 Arguments for Using Country as a Proxy for Culture Before moving to analyzing the suitability of country as a proxy for culture, it is important to note that the use of country as a proxy for culture and the use of national cultural indices is sometimes justified. However, two conditions must be met in order to effectively equate the two, including: (1) within-country variance must be small; and (2) between-country variance must be large (cf. Gerhart and Fang 2005 ; Matsumoto et al. 2001 ; Taras and Steel 2009 ). That is, first, residents of a given country must have similar values. Second, values of residents of one country must be different from values of residents of another country. Expecting values to vary little within countries and greatly across countries may not be completely unreasonable. Political, economic, and societal institutions tend to reside neatly within national borders. Education systems, cost and quality of life, and media and other entertainment options tend to be relatively homogenous within nations. Geographic distance as well as national borders have, to various degrees, limited interpersonal exchanges, flows of information, labor and products. Although globalization forces may be now attenuating these effects, for most of human history the speed and cost of international transportation has been exorbitant (Hummels 2007 ), exacerbated in the nineteenth century by need to obtain passports for border crossings (Torpey 2000 ). Since factors that profoundly affect values were relatively localized, it would make sense to expect values to cluster within country borders. As Peterson and Smith ( 1997 , p. 934) note, “the link between nation and culture tends to occur because people prefer to interact with other people and be guided and politically government by institutions consistent with their values.” Nation-states can thus be a result of shared values and national institutions that, in turn, further perpetuate shared values. Furthermore, many economic and societal development indicators are inherently national or measured and tracked at the country level, such as GNP, GDP, FDI, international trade, spending on education and military, and corruption levels. Thus, it may be feasible to use nation-level cultural indices to study relationship between culture and these indicators. 2.3 Arguments Against Using Country as a Proxy for Culture Opponents of the culture-overlaps-with-country argument point out that even though these factors indeed historically align with national borders, it is increasingly no longer the case. The cultural convergence and modernization theories, which received ample empirical support, explain how cross-border personal and organizational exchanges blur the cultural differences among countries (Inglehart and Welzel 2005 ; Ralston et al. 1997 ). First, cross-border travel and long-term migration is rapidly increasing. The cost of instant communication internationally (e.g., voice, chat, video) is becoming effectively zero. Media and entertainment are more international than ever. Major music labels and movie studios have a truly global outreach. Institutions are morphing and converging in response to regional integration forces. Education is becoming increasingly international, with extended-time study abroad becoming the norm in many countries. Second, many national borders were not aligned with ethic and tribal boundaries in the first place, particularly in regions that experienced colonization where borders between what became sovereign states were often arbitrarily drawn. For example, modern Nigeria was created in 1914 as a matter of administrative convenience by the British, who merged two contiguous colonies. As with Nigeria, the effects of tribal histories and traditions often outweigh those of national institutions. It is not uncommon in some regions to observe a virtually identical culture on both sides of a national border and different cultures within. Supporting these concerns, numerous studies have revealed significant within-country variance in cultural values (e.g., Au and Cheung 2004 ; Coon and Kemmelmeier 2001 ; Dolan et al. 2004 ; Lenartowicz and Roth 2001 ). Indeed, a recent meta-analysis of research using Hofstede’s framework revealed that as much as 70–90 % of variance in cultural values may reside within countries, with only 10–30 % of variance residing between (Steel and Taras 2010 ). The results reported by Gerhart and Fang ( 2005 ) are even more striking. Based on a re-analysis of Hofstede’s data, they found that country of residence explains only 1–7 % of variance in responses to survey questions that were used to operationalize cultural values. The figures are similar to those reported by Hofstede ( 1980 ) himself, who acknowledged that country of residence accounted for just 4.2 % of the variance in survey responses. In Anderson’s ( 2006 ) terms and the title of his book on the topic, nations are largely “Imagined Communities,” with far greater variation within than most acknowledge. 2.4 In Search of More Relevant Boundaries of Culture Even though there are legitimate reasons for assessing culture at the nation level, the concerns previously listed suggest considerable skepticism regarding the degree that country is a good proxy for culture. Certainly, national cultural averages may still yield useful information, particularly if they were derived based on matched representative samples. However, a lack of precision in defining the boundaries of cultural communities makes the cultural scores less than ideally representative of these communities. For example, Weber ( 1930 ) argued that children’s stories were a key mechanism for instilling protestant work values. If the stories had strong images and narratives promoting hard work, the children and subsequent adults would have higher levels of achievement motivation and economic success. One could then study the relationship geographically by assessing the percentage of schools in regions of a country that adopted schoolbooks with achieving imagery. If the motivational climate in regions is higher where a greater percentage of the schools had textbooks with strong achieving imagery, this would support Weber’s thesis. Engeser et al. ( 2009 ) used this precise geographic methodology to compare the effects of imagery contained in the textbooks from two German federal states (i.e., Baden-Württemberg and Bremen). Finding their expected results supports the notion that regardless of whether region is randomly or weakly connected to adoption of particular reading books, some analyses can be done. On the other hand, consider how much more precise regarding the effects our schoolbook imagery if, instead of using region averages, we identified what textbooks school districts, individual schools, and specific classrooms actually adopted. In this way, national averages may provide some useful information, but the estimates would be much more precise and informative if the averages represented properly delineated cultural communities, rather than the countries whose borders may not represent the true borders of the cultural ethos or a population segment or segments that display similar cultural values. The commonly used solution of using averages for smaller geographic areas can provide only a partial solution. Cutting regions small enough can help achieve reasonable homogeneity within a region and thus satisfy the first condition of low within-region variation. However, it will not address the problem of low inter-regional variation, which is the second necessary condition for using country (or region) as a proxy for country. Needless to say, cultural values, at least as conceptualized by Hofstede, may systemically vary even within the same corporate building, which can have considerable vertical variation; that is, culture in the basement mailroom may be very different from culture in the executive suite on the top floor, so geographic boundaries would have to be very thinly sliced indeed to ensure within-region cultural homogeneity. An equally important concern is that even though slicing populations into smaller groups and geographic regions may reduce within-group variance, it will inevitably result in separate areas that score similarly on cultural value tests, and thus also smaller between-group differences. A number of researchers have tried to address this problem by looking for supra-national geographic cultural regions. Hofstede’s ( 1980 ) original Culture’s Consequences provided a series of maps that grouped various countries into cultural regions based on closeness of their cultural scores. Likewise, Ronen and Shenkar ( 1985 , 2013 ) started inquiry into this issue over thirty years ago, recently publishing an updated version of their study. However, on these maps, countries from different continents often would appear in the same cultural clusters. Due to the absence of a logical explanation for why some of these countries form a homogeneous cultural region, these maps often only add to the confusion. The solution, then, is not more segmentation, but rather a search for a smaller number of culturally homogeneous populations that are distinct from one another, irrespective of their geography. Unfortunately, inquiries into whether other characteristics are more suitable for defining cultural boundaries have been limited. The field appears to agree with Hofstede’s ( 2002a ) belief that “[nations] are usually the only kind of units available for comparison.” Essentially, we have settled upon an organizing schema (e.g., countries, regions within countries) without determining first the extent to which this is appropriate and whether other dimensions may prove superior. Ironically, the direction for searching for more relevant dimensions for clustering cultures may have been in plain sight all along. In Culture’s Consequences, Hofstede ( 1980 , 2001 ) reported substantive correlations between his cultural indices and various national and personal characteristics, such as wealth, economic development, freedom, as well as age, gender, and occupational rank. Perhaps these could provide a better way to delineate cultural clusters? The present study assesses this possibility. As Steel and Taras ( 2010 , p. 211) note, very little causal work has been done with culture, in that “Cultural values at both the national and individual levels have typically been assumed to be extremely stable and thus studying factors shaping culture have been deemed unjustifiable.” However, this dearth does not preclude correlational work, so our search for alternatives to national borders for clustering cultural values was guided by prior research into the correlates of cultural values. As noted earlier, even though Hofstede was preoccupied with culture’s consequences, he did explore the relationship between cultural values and demographics, social position, national development indicators, and the like. Regardless of the causality, the fact that cultural values appear to correlate with basic personal and environmental characteristics provides a prior foundation for identifying populations with shared cultural values. As a brief overview of the theories that guided our choice of possible culture-boundary factors, we explore the effects of such personal and environmental characteristics as wealth (i.e., socio-economic status as the individual-level characteristic and GDP/capita as the national level characteristic), safety (i.e., age and the sense of security that comes with it as the individual characteristic and corruption or political freedom at the environment-level indicator), and modernization and progress (i.e., level of education at the personal level and urbanization or HDI as an environment-level indicator). As noted by Hofstede in Culture’s Consequences and reviewed elsewhere in more detail (Basabe et al. 2002 ; Inglehart and Welzel 2005 ; Johnson and Lenartowicz 1998 ; Steel and Taras 2010 ; Waldron 2003 ), individual and societal wealth reduces one’s dependence on the group and, therefore, is expected to make people more individualistic and less power distance-oriented, more tolerant of uncertainty and promoting such masculine values as achievement and competition. Likewise, the sense of safety and stability, be it in personal life or society in general, also reduces one’s need to depend on a group and induces individualism, tolerance of uncertainty, and the ability to be less concerned about pleasing figures of power. Finally, modernization and progress have been shown to bring about an increase in individualism and masculinizing and a drop on power distance and uncertainty avoidance orientations. The nature of our study, however, is exploratory. As a first step, it appears most sensible at this stage to compare the suitability of country of residence as a cultural boundary versus the more basic, easily observable and commonly used personal and environmental characteristics. The goal is not to definitively establish the best way to think about cultural regions, but to determine how justifiable the current country-equals-culture approach is and whether or not there could be more worthwhile ways to approach the problem. Even though correlations and associations do not confirm causation or explanatory mechanisms, they are a requisite property for the latter and provide promising lines of inquiry. Using meta-analytic data, we test whether or not country is a reasonable “container” for culture by comparing within- and between-country variance in cultural values. Our approach is similar to Minkov and Hofstede’s ( 2012 ) regional level analysis, except we go a step further and consider groups based on factors other than geography. Then, using latent class analysis, we explore how cultural values cluster along other theoretically relevant factors besides geography. Although the underlying theories and rationales linking cultural values and demographic- and national-level characteristics have been reviewed at length in many earlier publications (e.g., Hofstede 2001 ; Steel and Taras 2010 ), for purposes of parsimony and the extensive number of alternative clustering dimensions we consider, we do not reiterate the theories in our analysis. Instead, we pose the following research question: Research Question: To what extent is country a proxy for culture, and are there more accurate ways to define boundaries of cultures than along country borders? We must stress again that the present study is concerned with culture in the Hofstedean sense. That is, the present study focuses on “culture” as a set of cultural values, including individualism, power distance, masculinity, or uncertainty avoidance. These values have been shown to explain and predict workplace behaviors (Taras et al. 2010 ). So, our search for dimensions for clustering “cultures” is an attempt to identify populations with similar values, such as the values described in models that have dominated the field of international business research since Hofstede’s ( 1980 ) Culture’s Consequences, and thereby aid with interpretation and generalizing the results of research into the effects of such values. For other attributes of culture—such as language, customs, or artifacts—alternate clustering dimensions may be more meaningful, and indeed country may be the optimal proxy for distinct groups. Also, other cultural values not included in Hofstede’s model may indeed operate differently, though a case would have to be made for any hypothesized radical difference. 3 Method 3.1 Literature Search The present study is based on a meta-analytic dataset, and so each data point represents a group or sample. Similar to previous meta-analyses of cultural values (e.g., Coon and Kemmelmeier 2001 ; Oyserman et al. 2002 ; Steel and Taras 2010 ; Taras et al. 2012 ), data from studies that involved assessment of cultural values of survey respondents were consolidated into a large multi-country longitudinal dataset. Studies that used Hofstede’s ( 1980 ) four-dimensional cultural model, the Value Survey Module (VSM), or related instruments, were included. The overriding benefit of using such meta-analytic data is that it enables this research in the first place, which is unusual. As Peterson and Søndergaard ( 2011 , p. 1549) note, “One of the main constraints on testing culture theory based on alternatives to national boundaries has been the limited availability of cultural data about within-nation regions and the limited availability of criterion data.” In contrast to our meta-analytic data, consider the matched samples design, a dominant choice among large-scale cross-cultural comparative studies. For example, we have cross-cultural studies based solely on information technology employees (Hofstede 1980 ), teachers (Schwartz 1994 ), mid-level managers (House et al. 2004 ), or students (Maznevski et al. 2002 ), not to mention further refinement by industry and organization. The matched sample design does have its advantages, as it potentially allows for isolating country-specific differences in cultures by minimizing data contamination due to sample background differences. On the other hand, matched samples designs have fixed or largely removed differences in terms of age, education level, occupation, and other relevant characteristics, a practice that simultaneously recognizes that these are related to culture while making it impossible to analyze the effects of these factors on cultural values of the respondents. Meta-analytic data, which is drawn from wide variety of samples and sources, possess sufficient diversity to allow clustering on these factors. We conducted a comprehensive literature search to locate relevant empirical studies for the meta-analysis. First, we conducted a computer search using electronic databases of scholarly publications. Second, we reviewed over two-dozen relevant journals for studies that used Hofstede’s VSM or similar instruments for assessment of cultural values. Third, for each article being coded, we checked the reference section for links to publications potentially containing data for the meta-analysis. Fourth, we used the “cited by” function of the Web of Science and Google Scholar databases to identify publications citing articles coded for our meta-analysis and, if relevant, included in our dataset. Finally, as a part of a larger meta-analytic project, we sent out a call via the Academy of International Business and Academy of Management list serves for studies that used Hofstede’s ( 1980 ) framework to measure culture. We received over two dozen responses and included all relevant papers in our meta-analytic database. 3.2 Inclusion Criteria A common challenge in meta-analysis is that the summarized studies rarely use identical research design and methodology (Rosenthal and DiMatteo 2001 ). Scale length modification (e.g., 1–5 modified to 1–7), change in the sequence of the survey items, and other minor differences are not likely to lead to a substantial alteration of the construct. However, if the studies are substantively different, aggregation becomes questionable, leading to the so-called problem of “apples and oranges” (Sharpe 1997 ). The tradeoff is that, on the one hand, the more relaxed are the inclusion criteria, the more studies are included in the meta-analytic dataset. Indeed, the larger sample size improves estimates and strengthens the validity of the findings. On the other hand, relaxing inclusion criteria lowers consistency across the studies and increases error, thereby lowering reliability and validity of the findings. To deal with the issue of commensurability, we relied on content validation in which multiple coders determined if instruments were similar by conducting a thorough item analysis and ensuring the instrument is consistent with Hofstede’s model of culture and is compatible with the different versions of Hofstede’s VSM (Hofstede 1982 ). This established meta-analytic methodology has been successfully used in earlier meta-analyses (e.g., Steel et al. 2008 ; Steel and Taras 2010 ; Taras et al. 2010 ). To minimize inconsistencies, we attempted to be as conservative as possible when making our inclusion decisions. That is, when in doubt, we excluded a measure, making errors of omission rather than errors of commission. Only studies that defined and operationalized cultural values consistently with Hofstede’s ( 1980 ) framework qualified for inclusion, such as those that used various versions of Hofstede’s original VSM. Studies that used other instruments to quantify cultural values posed a greater challenge. We conducted a thorough item evaluation and content analysis of individual survey instruments considered for inclusion in our sample. Drawing on the catalog of over 100 cultural measures collected by Taras ( 2011 ), eight measures and their variations satisfied our inclusion criteria. Table 1 provides a summary of the number of data points derived using each of the eight instruments. Further easing concerns of commensurability, the vast majority (up to 79.1 %, depending on the dimension) of the data points in our meta-analytic dataset were derived using Hofstede’s original VSM and its variations. Table 1 Data points in the meta-analytic dataset, by instrument Full size table 3.3 Data Coding Procedures Hofstede’s ( 1980 ) model of culture is comprised of the following value dimensions: (1) Power Distance the extent to which the less powerful persons in a society accept inequality in power and consider it as normal; (2) Individualism the degree to which people prefer to act as individuals rather than as members of groups. In individualist cultures, people look primarily after their own interest, while in collectivist cultures people are assumed to belong to tight in-groups that protect interest of its members in return for their loyalty; (3) Masculinity the degree to which values like assertiveness, performance, success, and competition prevail over values like the quality of life, maintaining warm personal relationships, service, care for the weak, and solidarity; (4) Uncertainty Avoidance the extent to which people are made nervous by situations that they perceive as unstructured, unclear, or unpredictable. The Long- versus Short-Term Orientation dimension (i.e., Confucian Dynamism) was later added to the model (Hofstede and Bond 1988 ), but we could not include it in our meta-analysis due to its rare use. Aside from the previous four cultural dimension sample means, key information extracted from studies in our meta-analytic database included: sample size, survey year, survey country, and sample demographics (i.e., percent male, average age, education level, and occupation). Using sample descriptions, we also coded respondents’ socio-economic status and calculated the year in which the respondents were born, which we coded as generation. Further, using publicly available data published by the World Bank, Heritage Foundation, and Transparency International, we linked these sample descriptions with measures of wealth, freedom, and economic and societal conditions for each country, as well for each matching time period (e.g., the 1980s). Specifically, the following 20 variable groups are included in the analysis: 3.3.1 Cultural Values Power distance, individualism-collectivism, masculinity-femininity, and uncertainty avoidance, as measured by different versions of the Hofstede VSM and compatible instruments, were used as the dependent variable. To provide a foundation for the analysis, we converted all culture data into a common metric. First, the scales were transformed to a 0-1 range to resolve the issue of range differences (e.g., 1–5, 1–7). Then, we standardized the scores (mean 0, SD 1) within data subsets corresponding to each instrument type in our pool to address the differences in item functioning and scoring schemes of different instruments. These removed any remaining data inconsistencies stemming from instrument differences. 3.3.2 Year Recording year, when available, enabled us to match country characteristics with the respective period. For example, if country X ’s data were collected in the 1990s, national indicators, such as civil freedom or HDI, would also be from the 1990s. In addition, this chronological data allowed us to calculate when respondents were born and to test the effect of the historic period on cultural values of the respondents. 3.3.3 Country We coded the country of residence of the respondents as a nominal categorical variable. Data from a total of 32 countries and regions were available for the analysis. Due to limited data availability for some countries, they were grouped into regions. For consistency, when possible, the grouping was done following the schema used by Hofstede ( 1980 ). Specifically, Arab countries were grouped together and coded as one region, as were the countries of Africa (South Africa was coded separately), Caribbean region, former USSR republics, Scandinavian countries, as well as smaller Latin American countries (e.g., Brazil, Argentina and Mexico were coded as separate regions). 3.3.4 Gender Since we only had sample-level data available for analysis, gender was coded as percent male in a given sample. For the Intra Class Correlational (ICC) analysis, described in the following Data Analysis section, the continuous variable was split into three groups: mostly female (<35 % male), mixed (35–65 %) and mostly male (>65 %). 3.3.5 Age We coded age as the average age in a given sample. For ICC analysis, we split the continuous variable into approximately nine equal groups, starting with young teenager (<15) and up to senior citizen (60+), with intervals of about 5 years in the younger age, increasing to about 10 year intervals for older people. 3.3.6 Generation We coded generation as the year in which the respondents in a particular sample were born by subtracting their average age from the year in which the data were collected. We split the data into generations by decade, resulting in six generations, starting with 1925–1935 through 1975–1985, in 10-year intervals. 3.3.7 Education Information about education of the respondents was typically presented as the average years of schooling or as the mode of the highest degree attained in the sample. Both pieces of information were coded and, if one of them was missing in a given publication, it was inferred from the other. The continuous version of the variable (i.e., years of schooling) was used in the Latent Class Modeling (LCM) analysis, while the ordinal categorical variable (i.e., highest degree) was used in ICC tests. 3.3.8 Occupation The list of occupations, as described in each study, was fairly lengthy with a total of about 30 categories. They all were collapsed into seven general categories including: worker, clerk, professional, middle-level manager, top manager, student, and graduate student. 3.3.9 Socio-Economic Status (SES) We coded SES as an ordinal variable with five categories: lower, lower middle, middle, upper middle, and upper class. Information about the respondent’s income level (lowest to highest income quintile in the corresponding country), occupation (by skill requirement and supervisory role; from unskilled to highly skilled with supervisory functions), or prestige of university for students (lowest to highest quintile of the university ranking) were used to estimate the social class of the respondents. 3.3.10 Civil and Political Freedom We used data provided by the Freedom House to code this variable. A 7-point scale is used by this organization. The data were recoded so that seven represents the highest degree of freedom. 3.3.11 Economic Freedom We used the Heritage Foundation data to operationalize this variable. Originally, it was a continuous variable designed to range from 0 to 100. For ICC analysis, we split the sample into six roughly equal groups. 3.3.12 GDP/Capita at PPP We used World Bank data, with values adjusted for inflection and expressed in year 2000 US dollar equivalents. For ICC analysis, the countries were split into ten roughly equal groups with GDP/capita at PPP ranging from less than $5000 to over $30,000, with the step of about $3000 dollars in the lower range and up to $7000 in the upper range. 3.3.13 Human Development Index (HDI) We used United Nations Development Program (UNDP) data to operationalize human development level in a given country in a given time period. The originally continuous variable was later split into five approximately equal classes for ICC analysis. 3.3.14 Globalization Index We used Kearney/Foreign Policy Magazine data to describe the extent of globalization (i.e., the degree of connectivity, integration, and interdependence). We split the originally continuous variable with a possible range of 0–100 into eight approximately equal classes for ICC analysis. 3.3.15 Long-Term Unemployment We used World Bank data to capture unemployment. Rather than using simple unemployment indicators (i.e., percent of people currently out of job and actively seeking one), long-term unemployment (i.e., percent of people unemployed for 6+ months) was used as the former index is too volatile and can drastically change in a matter of weeks. We split this variable into ten approximately equal groups, with the class range of about two percent in the lower range and about 5 percent in the upper range. 3.3.16 Urbanization We used UNDP data to record organization rate in a given society at a given time. Urbanization was operationalized as percent of people living in cities, split into nine classes ranging from less than 20 % to over 90 %, with the step of about 10 percentage points. 3.3.17 Income Inequality We used the World Bank’s estimates of the Gini index to operationalize income inequality. The continuous data, ranging from about 25 to about 75, were split into eight approximately equal-range classes for the purpose of ICC analysis. 3.3.18 Corruption We used the Corruption Perception index provided by Transparency International to operationalize this variable. The estimates are done using a 10-point scale, with greater values representing a lower level of corruption. 3.3.19 Crime Rate We used Euromonitor International data, collected in partnership with Interpol, on the number of criminal offences per 100,000 inhabitants to quantify criminal situation in a given country. The data were split into ten approximately equal classes for the purpose of ICC analysis. 3.3.20 Employment in Agriculture Finally, to capture the structure of the economy, we used UNDP data to code percent of people employed in agriculture. The continuous variable was later split into seven groups, with the class range of about 2 percentage points in the lower range to about 10 percentage points in the upper range. 3.4 Data Analysis Consistent with the meta-analytic procedures of Hunter and Schmidt ( 2004 ), all estimates in the tests described below were weighted by the sample size, allowing more emphasis on data points with less error. The estimates, however, were not corrected for unreliability, because internal consistency indices (along with other estimates of heterogeneity) have not been reported in most of the studies. The present study relies on, first, ICC analysis to assess suitability of the different dimensions for clustering cultural values. Then, we use LCA to explore optimal grouping of cases into distinct cultural entities. 3.4.1 Intra Class Correlation (ICC) First, we wanted to test how well cultural values cluster along various dimensions, such as geography (i.e., countries), demographic, and environmental characteristics. Cluster analysis is not suitable for this kind of test as it does not involve a dependent variable, which in our case is culture. Instead, we performed a series of one-way ANOVA tests to estimate ICC and calculate within- and between-group variance in cultural values for countries and other clustering dimensions to see what percentage of differences between subjects reside within and between groups, or how similar subjects in groups are to one another, and how distinct groups are from one another. A simple correlation analysis was also performed to explore more closely the relationship between the cultural value predictors tested using one-way ANOVA and ICC analysis. The disadvantage of ICC analysis is that it does not allow for identifying clusters that maximize between-group and minimize within-group variance; rather, it simply estimates the within- versus between-group variance for existing groups. For nominal variables, such as country of residence or occupation, such a split into groups is natural, and ICC analysis would be sufficient. For continuous variables, such as level of education or age, the respondents could be divided into groups in many ways and depending on grouping, the results may be very different. Accordingly, we grouped values along our continuous variables by commonly used categories. For example, in the case of years of education, such categories as “less than high-school”, “high school”, “some college”, “college”, “master’s” and “doctorate”, corresponding to up to 10, 12, 14, 16, 18, and 20+ years of education, seemed most natural. However, even though these simple tests were very informative, a more sophisticated analysis was required to explore the issue in more depth. 3.4.2 Latent Class Analysis (LCA) To go beyond uni-dimensional testing of pre-existing groups, we employed LCA, which allows for multi-dimensional search for latent classes. First, like factor analysis and cluster analysis, LCA allows for classifying cases according to their maximum likelihood class membership, but unlike factor or cluster analysis, LCA does so with respect to a dependent variable, which in our case is culture. Second, the tests can be performed with multiple predictors simultaneously and the solution, similar to that in structural equation modeling or multivariate regression, provides coefficients that indicate how well each criterion predicts membership in the latent classes. Most importantly, LCA does not require a pre-set number of classes, as would be the case for ICC analysis. Instead, in a very computation-intensive procedure, it tests every possible combination and identifies a solution that produces the best latent-class fit based on a set list of predictors. So rather than testing how well cultural values cluster within, for example, 32 countries or eight socio-economic classes, LCA identifies the number of distinct classes (in our case, cultural clusters) that results in the greatest between- to within-class variance ratio. 4 Results 4.1 Sample The final pool contained 558 empirical publications (see Appendix 1 for the complete list). Due to space restrictions, the complete list of studies included in our meta-analytic dataset could not be provided here, but is obtainable from the first author upon request. Of the 558 studies, 419 were published in peer-reviewed journals, 10 in book chapters, 120 were doctoral dissertations, and nine were Master’s theses. On average, the respondents were 28.9 years old, had 14.1 years of education and 49.1 % of them were male. 4.2 ICC Table 2 provides a summary of our main findings. As can be seen, “country” as a clustering function had an unfair advantage. A total of 32 countries were included in the analysis, as compared to only 5–10 categories along the other dimensions. Generally, as the number of categories increases, more small groups are created, the between-group variance increases, and the within-group variance decreases. Therefore, we also provide F-statistics, which are an indicator of group variance overlap, adjusted for the sample size and number of groups in the sample. A larger F-statistic means that the groups are more distinct from one another and shows how meaningful our grouping method is in terms of producing homogeneous separate entities. Table 2 Results of correlational and ANOVA tests Full size table Somewhat consistent across all four cultural dimensions, only about 16–21 % of the variance in cultural values resides between countries. The results indicate that although country of residence is not an irrelevant dimension in terms of grouping cultural values, the between-country variance numbers are certainly too low to treat terms “country” and “culture” as synonymous or interchangeable; indeed, 79–84 % of variance in cultural values resides within countries. Clearly, our data confirm that national averages are poor estimates of cultural values of individuals or small groups. Table 2 also provides correlations between cultural values and various personal- and environment-level indicators. For demographics, occupation stands out as a comparatively good dimension for clustering cultural values. With only seven categories, it closely aligns with cultural values, accounting for approximately 50 % of between-occupation variance in power distance. The results are less impressive along the other dimensions of culture, although for individualism the between-group variance is greater than that for country. For masculinity, the ICC analysis results favor country over occupation (i.e., between group variance at 12.8 versus 18.5 %), but accounting for the number of categories in each factor, occupation still emerges as a more meaningful predictor of cultural clusters (i.e., F-statistics of 18.3 versus 6.2). We observed similar results for socio-economic status, but this could be expected given that this variable was largely coded based on the profession and occupation of the respondents. For uncertainty avoidance and power distance, education level and generation (i.e., decade in which the respondent was born) provide comparable results to those obtained for country, though the between-group variance statistics are somewhat lower for these two factors compared to country. However, taking into account the much lower number of groups created along these two predictors, the F-statistics actually suggest education and generation are more meaningful than country for clustering into cultural entities based on their uncertainty avoidance, power distance, and individualism values. We found that age and gender were generally not good for clustering cultures. Differences among individuals within the same age groups and sexes were far greater than differences between different age groups and sexes. That is, on average, cultural values of different age cohorts and men versus women were found to be negligible, with great overlap among the groups and almost all variation residing within groups. With respect to environmental characteristics, importantly, they were almost all superior to geography in terms of model significance. Most notably, F-statistics consistently favor economic freedom, globalization extent, long-term unemployment, wealth distribution inequality, corruption, crime rate, and the share of employment in agriculture over country as a clustering function. Finally, the historic time period—year and decade when the data were collected—provided mixed results. For power distance and uncertainty avoidance, which have been steadily falling worldwide as indicated by negative correlations, the time period when the data were collected provides a meaningful cultural cluster, at least as good as geography. For individualism and masculinity, which are more stable longitudinally, the timing of the data collection seems largely irrelevant. 4.3 LCA ICC analysis showed that country is a poor proxy for culture. With the exception of HDI and urbanization rate, every criterion we considered outperformed geography (i.e., country) as a criterion for setting boundaries for cultural entities. At the same time, none of the tested criteria by themselves seemed to be a perfect “container” for cultural values. In all cases, the between-group variance for cultural values remains less than 50 percent, although the within-between variance split was close to 50–50 for economic freedom and long-term unemployment along power distance as well as for corruption perception along uncertainty avoidance. To explore more complex multi-dimensional solutions we turned to LCA. Table 3 provides a summary of the findings. The results of LCA provide further support for the conclusions suggested by our ICC tests. Table 3 Results of latent class analysis Full size table First, country is only a marginal predictor for culture. Even though we had 32 countries in our dataset (i.e., the greater the number of categories, the more likely the variable would be significant), country was a statistically significant predictor only for masculinity and uncertainty avoidance, and then only at p < 0.05. Second, using study level data, there appear to be fewer cultural regions in the world than there are countries. The greatest number of latent classes (i.e., groups with relatively homogeneous cultural values) suggested by LCA was eight. Solutions with a greater number of classes were clearly inferior to those more parsimonious; and, hence, our report includes goodness-of-fit indices for up to eight latent classes. In fact, a solution with even fewer latent classes seemed preferred in most cases. For instance, let us consider individualism, where a four-class solution seemed optimal. The classes were found to be unbounded by geography and, instead, were largely driven by wealth at the personal (SES) and environment levels (GDP/capita, PPP). Put simply, there appears to be only four cultural regions with respect to individualism, and they seem to be split along the individual and societal wealth lines; that is, poorer people are more collectivistic, richer people are more individualistic, and there are two intermediate individualism classes in between. Occupation and crime rate may slightly modify this picture, but this is fundamentally the conclusion suggested by our results. The solutions for the other three cultural dimensions are more complex and suggest that seven or even eight distinct cultural classes may exist in the world. Either way, the number is still substantively fewer than the lengthy lists of countries or geographic regions normally assumed to be culturally distinct. Depending on the dimension, different criteria seem most relevant to identifying cultural classes. Consistent with the results of the ICC and bivariate correlation tests, wealth at both demographic (SES, occupation) and environmental (GPD/capita, PPP) levels are almost universally significant factors. Also, the amount of freedom and globalization tend to matter across all cultural dimensions. Unsurprisingly, indicators of society traditionalism (e.g., percent of population in agriculture) and indicators of economic and societal turbulence (e.g., unemployment, inflation, and crime rate) are also significant, particularly with respect to uncertainty avoidance and power distance. 5 Discussion As Tov and Diener ( 2009 , p. 33) warn, “By equating entire nations with single cultures, we risk overlooking important differences within nations, as well as similarities that extend beyond national borders.” The present study addresses this concern by assessing the suitability of national borders for delineating groups of individuals with similar cultural values and by exploring what other relevant personal and environmental group characteristics might be more suitable. Using a meta-analysis of data from 558 studies, we conducted tests for each of the four dimensions that comprise Hofstede’s ( 1980 ) original cultural value framework. Performing a series of one-way ANOVA tests (estimates weighted by sample size), we obtained ICC statistics and calculated within- and between-group variances in cultural values. Given the roughly equal number of observations, but the different number of categories along different variables (i.e., over 30 countries versus only five to ten groups along the other predictors), we also compared F-statistics to assess the overall significance of the different models. We draw the following five conclusions based on our study’s results. The first three reconfirm the limitations of using geography for clustering, and the final two suggests superior units of culture. First, if culture is a collection of four sets of values as conceptualized by Hofstede, then country does not equate to culture . Even though country of residence may be of some value in predicting cultural values of respondents, there is far more variation in cultures within countries than between countries. Consistent with McSweeney’s ( 2013 ) recent review, we empirically confirm that the practice of using the two terms interchangeably is unwarranted, despite its popularity: country of residence certainly does not equal culture. Even when cultural dimensions were found to have a statistically significant relationship with country (i.e., masculinity, uncertainty avoidance), the link was rather weak overall, and much weaker than that with demographic and environmental factors, such as gender, age, SES, occupation, economic and political environment. Second, statistically significant differences in national averages do not mean that people in the countries are actually different. Essentially, this is the classical and pervasive statistical error of comparing means without considering within group variance (Lubinski and Humphreys 1996 ). Given the high within-country, and low between-country, variation in cultural values, even if the country averages are significantly different, there are likely to be more residents in one country whose values are identical to those in another country than otherwise (cf. Matsumoto et al. 2001 ). Alternatively, the concept of cultural overlap has been recently introduced by Maseland ( 2011 ), and it may prove more suitable to operationalize cultural differences among countries than the traditional use of national averages. Third, regional indices do not solve the problem of within - country variations in culture. In fact, geography is largely limited when defining cultural entities. Even though offering separate averages for within-country geographic regions in countries such as Switzerland and Canada is a refinement upon country-based cross-cultural research (Kaasa et al. 2013 ), it is still not an adequate or reliable answer. If the goal is to produce homogenous cultural groups, geography is unlikely to produce them, with rare exceptions being perhaps the thinly sliced neighborhood or gated community (Pow 2011 ). For example, as Haidt ( 2012 , p. 25) noted when researching moral values, which have strong overlap with cultural values, “I had flown five thousand miles south to search for moral variation when in fact there was more to be found a few blocks west of campus, in the poor neighborhood surrounding my university.” Furthermore, when such micro-cultures do emerge, the underlying reason is probably due to the demographics of those residing there or because of the regions’ environmental characteristics. People in a village in China are not more collectivist than people in lower Manhattan just because the two regions are on different continents, but likely because the residents in these two locations differ in terms of their socio-demographics and the politico-economic environments, which in turn lead to different needs, views, and values. Fourth, if we are seeking to cluster at the group level, there are far fewer distinct cultural entities than there are countries (or otherwise geographically distinct regions). Eight or even as few as four cultural clusters seem sufficient to classify people into distinct cultural groups. This suggests that having separate cultural profiles for each nation is an unnecessary complication. There are simpler organizational schemas available, which brings us to the last, and perhaps the most important, point. Fifth, if not geography, what then should be used to delineate cultural regions? Demographic and environment characteristics appear more relevant. Even though nationality still has relevance in cross-cultural studies, where the focus can be explicitly the country, our results show that there are better “containers” for culture than country. Based on an assessment of demographic and environmental characteristics, even though none of them can be considered proxies, we conclude that a number of them appear more suitable than countries for clustering cultures. Even though our results here are based on associations, these point the way towards understanding the basic mechanisms by which cultural groups are created or coalesced. While our study could not provide definitive answers, it offered initial empirical support for the notion that we should not be focusing exclusively on cultures of countries, but rather exploring and comparing cultures of socio-economic classes, professions, age cohorts, historic time periods, geographic or social environments characterized by certain level of wealth, freedom, equality, instability, and globalization. The story told by our data has far-reaching implications. Our innate readiness to form group stereotypes based on country of origin needs to be consistently challenged. For example, Huntington’s ( 1996 ) overwhelmingly influential book, “The Clash of Civilizations,” is fundamentally based on cultural and religious entities forming along rigid national geographic lines, which our findings indicate as extremely unlikely. Essentially, work-related values of a lawyer from Manhattan may in fact have more similar to those of a lawyer from Shanghai, than to those of a construction worker working on a site just a few blocks away, just as a construction worker from Shanghai may have more in common with a construction worker in Manhattan than with a lawyer from a Shanghai office. This commonality would not extend to external cultural attributes such as language, food tastes, and traditions. But as far as cultural values as defined by Hofstede, geography may be inferior to demographics and environment characteristics as a clustering factor. In summary, our results indicate that we have been overemphasizing a single organizational frame when discussing culture. Instead of just focusing on what country is being dealt with when discussing the cultural implications of managerial, interpersonal, and strategic advice, it would be better to also consider the group’s level of freedom, globalization, and above all, its wealth. Consistent with F. Scott Fitzgerald’s famous quote from The Great Gatsby (i.e., “Let me tell you about the very rich. They are different from you and me.”) and what we found here, there is substantial support that socio-economic bracket predicts and determines a wide range of values and perceptions (cf. Kraus et al. 2012 ). Highlighting the limitations of geography to identify the wealthy, an extensive review by Hay and Muller ( 2012 ) discussed how those who are in the top 1 % (or perhaps the top 0.1 %) have increasingly become transnational or de-territorialized. 5.1 Limitations and Directions for Future Research As with all research, our study is not without limitations. We review several of them here and provide four suggested directions for building upon our research and further advancing the field based on these limitations. First, our study relied on Hofstede’s ( 1980 ) framework, which makes it subject to the same drawbacks as his original study. Although very popular, which is a necessity for using a meta-analytic methodology, his approach has been criticized in a number of subsequent publications (e.g., Baskerville 2003 ; McSweeney 2002 ; Taras and Steel 2009 ; Williamson 2002 ). It remains an open question as to the degree to which our findings generalize to values assessed using other models of culture (e.g., House et al. 2004 ; Schwartz 1994 ), such as the 27 dimension suggested by Taras et al. ( 2009 ) in addition to the six moral values put forth by Haidt ( 2012 ). Unfortunately, the number of studies that used models of culture other than that offered by Hofstede has been limited. However, as more studies become available, a meta-analysis in these areas may become possible in the near future. Second, the present study is a meta-analysis. As is true for any meta-analysis, one of the threats to validity of the findings presented here is the issue of data commensurability. Each study included in our meta-analytic database was based on different sampling and methodology. Although we were very selective to avoid the “apples and oranges” problem, the only way to resolve the issue is to conduct a study using an original individual-level dataset. Unfortunately the earlier large-scale cultural comparison studies may not be suitable for the task (such as those by Hofstede 1980 ; House et al. 2004 ; Maznevski et al. 2002 ; Schwartz 1994 ; Trompenaars 1993 ), as all of them were based on matched samples. That is, the respondents were selected so that the demographics of the samples from each country were as close as possible (which implicitly recognizes that these indeed influence cultural values). Even though this approach minimized between-country sample differences, lack of variation in sample characteristics precludes an investigation of alternative dimensions for clustering cultures. Consequently, this meta-analysis can be considered an intermediate step, highlighting the need for an original multi-level dataset that spans across multiple countries and demographic groups. Such projects are extremely ambitious, but the resulting dataset would be based on a more comprehensive sample, collect more characteristics, and provide information at a participant level, rather than at a study or group level as done here. Spurred by our results and their implications, we hope future researchers will be motivated to collect such data and better determine where culture coalesces. Third, a number of seemingly relevant factors have not been included in the analysis. Due to limited data availability, we could not cluster along factors such as race or language. In particular, organizational membership has long been offered as an alternative or competing cultural frame (e.g., Schneider 1988 ), with Gerhart and Fang ( 2005 , p. 982) concluding that “organization differences are larger than country differences in cultural values.” The concept of person-organization fit is well established (Kristof-Brown et al. 2005 ), in which people are attracted to, and are less likely to leave, organizations sharing similar values to themselves. Research into which cultural values best represent organizations, how many value clusters exist at an organizational level, the advantages and disadvantages of organizational homogeneity, and the determinants of organizational values are all valuable areas for investigation. We could, for example, consider the degree that organizations inculcate values versus attract those consistent with them. Finally, the present study was conducted using four independent datasets, one for each cultural dimension in Hofstede’s model, which did not allow us to search for multi-dimensional cultural regions or test how the four different dimensions interact. To conduct such tests, a dataset with all cultural dimensions in it (i.e., each respondent assessed along cultural dimensions) would be needed. As noted earlier, existing cultural databases, such as those compiled by Hofstede and the GLOBE team, will not work due to their matched sampling designs. Although its dimensionality does not exactly replicate those of popular models of culture, the World Value Survey is more demographically diverse and could potentially serve as the next stop on the route of exploring cultural regions unconstrained to geography. Early investigations along these lines are promising (e.g., Kaasa et al. 2013 ). 6 Conclusion In the field of personality, twin research uncovered the provocative finding that relatively little variance in personality can be attributed to parental upbringing or shared family environment (Bouchard and Loehlin 2001 ). At a cultural value level, our results speak similarly to our shared country environment, supporting Anderson’s conclusion that people’s belief in community far exceeds reality. In Anderson’s ( 2006 , p. 7) words, “regardless of the actual inequality and exploitation that may prevail in each, the nation is always conceived as a deep, horizontal comradeship. Ultimately, it is this fraternity that makes it possible, over the past two centuries, for so many millions of people, not so much to kill, as willingly to die for such limited imaginings.” This illusion of national homogeneity may be a useful fiction though, allowing a broad level of cooperation and sacrifice not easily achieved otherwise. What it does not allow for are country averages to act as proxies for cultural values of individuals or small groups from these countries. National cultural indices are appropriate if the study itself is at a national level, such as examining GDP per capita (McSweeney 2013 ). Here and only here are national averages, like those offered by Hofstede ( 1980 ) or the GLOBE team (House et al. 2004 ), truly suitable. On the other hand, if we want to determine the drivers of culture or what identifies a homogenous culture, then we need to look beyond geography. Our findings not only corroborate the earlier warnings that national borders may not be a good way to delineate cultural regions, but also provide solid initial evidence that other dimensions may be more suitable for clustering cultures. Cultural clusters derived based on profession, socio-economic class, political, economic, and societal characteristics of the environment may be more homogeneous within, and more heterogeneous between, compared to cultures of countries. It is this exact movement beyond country that we wish to inspire with the present research.

Researchers and businesses have often operated under the idea that work-related cultural values are defined by country - just think of stereotypes about countries that are known to have hard workers or are team-oriented. A new study finds that nationality is actually a bad proxy for work-related cultural values, and points to other groupings - such as occupation - as more reliable indicators. "I study work-related values - how culture informs our beliefs and behaviors related to work," says Bradley Kirkman, co-author of a paper on the work. "This field has long defined and measured culture based on national borders. We wanted to know if nationality is really the best way to delineate cultural values and boundaries. And we learned that it's not a very good marker." Kirkman is the General (Ret.) H. Hugh Shelton Distinguished Professor of Leadership and head of the Department of Management, Innovation and Entrepreneurship at North Carolina State University. The paper was co-authored by researchers at the University of North Carolina at Greensboro and the University of Calgary. To examine the issue, the researchers looked at data from 558 studies on work-related values. The studies covered 32 countries from around the world, including the United States, Brazil, France, South Africa and China. Specifically, the researchers evaluated variation, both within each country and between countries, on four work-related cultural values: Individualism, which measures the extent to which a society places emphasis on individuals as opposed to groups;Power distance, which measures the importance of status and hierarchy in work settings;Uncertainty avoidance, which measures the extent to which cultures are willing to accept ambiguity or the unknown; andQuantity versus quality of life, which measures emphasis on competition and material wealth versus emphasis on societal welfare and well-being. The researchers found that approximately 80 percent of variation in these values was within countries. For example, at the low end, only 16.6 percent of the variation on individualism was between countries - 83.4 percent of the variability was within countries. At the high end, 20.8 percent of variation on power distance was between countries - which still left 79.2 percent of the variability within countries. "This told us that country does not equal culture on work-related values," Kirkman says. Researchers then evaluated other demographic and economic indicators to see if they could find a better proxy for work-related values - and they did. For example, occupation and socioeconomic status both significantly outperformed country as indicators on power distance. For example, only 20.8 percent of variation was between countries. But differences between occupations accounted for 50.1 percent, while differences between socioeconomic status accounted for 32.2 percent. Occupation and socioeconomic status also significantly outperformed country on individualism, but fared worse than nationality as indicators on uncertainty avoidance and quantity versus quality of life. "This work highlights the illusion of national homogeneity and shows that there is a real danger in equating country and culture in a work context," Kirkman says. "Making generalizations based on country can lead people to draw very inaccurate conclusions that may influence both individual and organizational business and management relationships. "For example, if a U.S. manager is transferred to a foreign office and makes decisions based on national stereotypes about workplace culture, it could blow up in his or her face," Kirkman says. The paper, "Does Country Equate with Culture? Beyond Geography in the Search for Cultural Boundaries," is published online in the journal Management International Review. Lead author of the paper is Vas Taras of UNC-Greensboro. The paper was co-authored by Piers Steel of the University of Calgary.

10.1007/s11575-016-0283-x

Medicine

Consumers of commercial genetic tests understand more than many believe

"How Well Do Customers of Direct-to-Consumer Personal Genomic Testing Services Comprehend Genetic Test Results? Findings from the Impact of Personal Genomics Study." Public Health Genomics. DOI: 10.1159/000431250

http://dx.doi.org/10.1159/000431250

https://medicalxpress.com/news/2015-06-consumers-commercial-genetic.html

Abstract. Aim: To assess customer comprehension of health-related personal genomic testing (PGT) results. Methods: We presented sample reports of genetic results and examined responses to comprehension questions in 1,030 PGT customers (mean age: 46.7 years; 59.9% female; 79.0% college graduates; 14.9% non-White; 4.7% of Hispanic/Latino ethnicity). Sample reports presented a genetic risk for Alzheimer's disease and type 2 diabetes, carrier screening summary results for &gt;30 conditions, results for phenylketonuria and cystic fibrosis, and drug response results for a statin drug. Logistic regression was used to identify correlates of participant comprehension. Results: Participants exhibited high overall comprehension (mean score: 79.1% correct). The highest comprehension (range: 81.1-97.4% correct) was observed in the statin drug response and carrier screening summary results, and lower comprehension (range: 63.6-74.8% correct) on specific carrier screening results. Higher levels of numeracy, genetic knowledge, and education were significantly associated with greater comprehension. Older age (≥60 years) was associated with lower comprehension scores. Conclusions: Most customers accurately interpreted the health implications of PGT results; however, comprehension varied by demographic characteristics, numeracy and genetic knowledge, and types and format of the genetic information presented. Results suggest a need to tailor the presentation of PGT results by test type and customer characteristics.","pageStart":"216","pageEnd":"224","siteName":"Karger Publishers","thumbnailURL":" Well Do Customers of Direct-to-Consumer Personal Genomic Testing Services Comprehend Genetic Test Results? Findings from the Impact of Personal Genomics Study","image":" Cover"} var SCM = SCM || {}; SCM.pubGradeAdsEnabled = true; SCM.pubGradeJSLibrary = ' var googletag = googletag || {}; googletag.cmd = googletag.cmd || []; googletag.cmd.push(function () { googletag.pubads().disableInitialLoad(); googletag.pubads().setTargeting("Profession Group", "N/A"); googletag.pubads().setTargeting("Profession", "N/A"); googletag.pubads().setTargeting("Specialization", "N/A"); googletag.pubads().setTargeting("Country", "DE"); googletag.pubads().setTargeting("Returning Visitor", "No"); googletag.pubads().setTargeting("url", " googletag.pubads().setTargeting("Page_Type", "Article"); googletag.pubads().setTargeting("ProductCode", "phg"); googletag.pubads().setTargeting("IsLicensedPhysician", "no"); googletag.pubads().setTargeting("Subjects", ); }); .MathJax_Hover_Frame {border-radius: .25em; -webkit-border-radius: .25em; -moz-border-radius: .25em; -khtml-border-radius: .25em; box-shadow: 0px 0px 15px #83A; -webkit-box-shadow: 0px 0px 15px #83A; -moz-box-shadow: 0px 0px 15px #83A; -khtml-box-shadow: 0px 0px 15px #83A; border: 1px solid #A6D ! important; display: inline-block; position: absolute} .MathJax_Menu_Button .MathJax_Hover_Arrow {position: absolute; cursor: pointer; display: inline-block; border: 2px solid #AAA; border-radius: 4px; -webkit-border-radius: 4px; -moz-border-radius: 4px; -khtml-border-radius: 4px; font-family: 'Courier New',Courier; font-size: 9px; color: #F0F0F0} .MathJax_Menu_Button .MathJax_Hover_Arrow span {display: block; background-color: #AAA; border: 1px solid; border-radius: 3px; line-height: 0; padding: 4px} .MathJax_Hover_Arrow:hover {color: white!important; border: 2px solid #CCC!important} .MathJax_Hover_Arrow:hover span {background-color: #CCC!important} #MathJax_About {position: fixed; left: 50%; width: auto; text-align: center; border: 3px outset; padding: 1em 2em; background-color: #DDDDDD; color: black; cursor: default; font-family: message-box; font-size: 120%; font-style: normal; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 15px; -webkit-border-radius: 15px; -moz-border-radius: 15px; -khtml-border-radius: 15px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} #MathJax_About.MathJax_MousePost {outline: none} .MathJax_Menu {position: absolute; background-color: white; color: black; width: auto; padding: 5px 0px; border: 1px solid #CCCCCC; margin: 0; cursor: default; font: menu; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 5px; -webkit-border-radius: 5px; -moz-border-radius: 5px; -khtml-border-radius: 5px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} .MathJax_MenuItem {padding: 1px 2em; background: transparent} .MathJax_MenuArrow {position: absolute; right: .5em; padding-top: .25em; color: #666666; font-size: .75em} .MathJax_MenuActive .MathJax_MenuArrow {color: white} .MathJax_MenuArrow.RTL {left: .5em; right: auto} .MathJax_MenuCheck {position: absolute; left: .7em} .MathJax_MenuCheck.RTL {right: .7em; left: auto} .MathJax_MenuRadioCheck {position: absolute; left: .7em} .MathJax_MenuRadioCheck.RTL {right: .7em; left: auto} .MathJax_MenuLabel {padding: 1px 2em 3px 1.33em; font-style: italic} .MathJax_MenuRule {border-top: 1px solid #DDDDDD; margin: 4px 3px} .MathJax_MenuDisabled {color: GrayText} .MathJax_MenuActive {background-color: #606872; color: white} .MathJax_MenuDisabled:focus, .MathJax_MenuLabel:focus {background-color: #E8E8E8} .MathJax_ContextMenu:focus {outline: none} .MathJax_ContextMenu .MathJax_MenuItem:focus {outline: none} #MathJax_AboutClose {top: .2em; right: .2em} .MathJax_Menu .MathJax_MenuClose {top: -10px; left: -10px} .MathJax_MenuClose {position: absolute; cursor: pointer; display: inline-block; border: 2px solid #AAA; border-radius: 18px; -webkit-border-radius: 18px; -moz-border-radius: 18px; -khtml-border-radius: 18px; font-family: 'Courier New',Courier; font-size: 24px; color: #F0F0F0} .MathJax_MenuClose span {display: block; background-color: #AAA; border: 1.5px solid; border-radius: 18px; -webkit-border-radius: 18px; -moz-border-radius: 18px; -khtml-border-radius: 18px; line-height: 0; padding: 8px 0 6px} .MathJax_MenuClose:hover {color: white!important; border: 2px solid #CCC!important} .MathJax_MenuClose:hover span {background-color: #CCC!important} .MathJax_MenuClose:hover:focus {outline: none} .MathJax_Preview .MJXf-math {color: inherit!important} .MJX_Assistive_MathML {position: absolute!important; top: 0; left: 0; clip: rect(1px, 1px, 1px, 1px); padding: 1px 0 0 0!important; border: 0!important; height: 1px!important; width: 1px!important; overflow: hidden!important; display: block!important; -webkit-touch-callout: none; -webkit-user-select: none; -khtml-user-select: none; -moz-user-select: none; -ms-user-select: none; user-select: none} .MJX_Assistive_MathML.MJX_Assistive_MathML_Block {width: 100%!important} #MathJax_Zoom {position: absolute; background-color: #F0F0F0; overflow: auto; display: block; z-index: 301; padding: .5em; border: 1px solid black; margin: 0; font-weight: normal; font-style: normal; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; -webkit-box-sizing: content-box; -moz-box-sizing: content-box; box-sizing: content-box; box-shadow: 5px 5px 15px #AAAAAA; -webkit-box-shadow: 5px 5px 15px #AAAAAA; -moz-box-shadow: 5px 5px 15px #AAAAAA; -khtml-box-shadow: 5px 5px 15px #AAAAAA; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} #MathJax_ZoomOverlay {position: absolute; left: 0; top: 0; z-index: 300; display: inline-block; width: 100%; height: 100%; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)} #MathJax_ZoomFrame {position: relative; display: inline-block; height: 0; width: 0} #MathJax_ZoomEventTrap {position: absolute; left: 0; top: 0; z-index: 302; display: inline-block; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)} .MathJax_Preview {color: #888} #MathJax_Message {position: fixed; left: 1px; bottom: 2px; background-color: #E6E6E6; border: 1px solid #959595; margin: 0px; padding: 2px 8px; z-index: 102; color: black; font-size: 80%; width: auto; white-space: nowrap} #MathJax_MSIE_Frame {position: absolute; top: 0; left: 0; width: 0px; z-index: 101; border: 0px; margin: 0px; padding: 0px} .MathJax_Error {color: #CC0000; font-style: italic} .MJXp-script {font-size: .8em} .MJXp-right {-webkit-transform-origin: right; -moz-transform-origin: right; -ms-transform-origin: right; -o-transform-origin: right; transform-origin: right} .MJXp-bold {font-weight: bold} .MJXp-italic {font-style: italic} .MJXp-scr {font-family: MathJax_Script,'Times New Roman',Times,STIXGeneral,serif} .MJXp-frak {font-family: MathJax_Fraktur,'Times New Roman',Times,STIXGeneral,serif} .MJXp-sf {font-family: MathJax_SansSerif,'Times New Roman',Times,STIXGeneral,serif} .MJXp-cal {font-family: MathJax_Caligraphic,'Times New Roman',Times,STIXGeneral,serif} .MJXp-mono {font-family: MathJax_Typewriter,'Times New Roman',Times,STIXGeneral,serif} .MJXp-largeop {font-size: 150%} .MJXp-largeop.MJXp-int {vertical-align: -.2em} .MJXp-math {display: inline-block; line-height: 1.2; text-indent: 0; font-family: 'Times New Roman',Times,STIXGeneral,serif; white-space: nowrap; border-collapse: collapse} .MJXp-display {display: block; text-align: center; margin: 1em 0} .MJXp-math span {display: inline-block} .MJXp-box {display: block!important; text-align: center} .MJXp-box:after {content: " "} .MJXp-rule {display: block!important; margin-top: .1em} .MJXp-char {display: block!important} .MJXp-mo {margin: 0 .15em} .MJXp-mfrac {margin: 0 .125em; vertical-align: .25em} .MJXp-denom {display: inline-table!important; width: 100%} .MJXp-denom > * {display: table-row!important} .MJXp-surd {vertical-align: top} .MJXp-surd > * {display: block!important} .MJXp-script-box > * {display: table!important; height: 50%} .MJXp-script-box > * > * {display: table-cell!important; vertical-align: top} .MJXp-script-box > *:last-child > * {vertical-align: bottom} .MJXp-script-box > * > * > * {display: block!important} .MJXp-mphantom {visibility: hidden} .MJXp-munderover {display: inline-table!important} .MJXp-over {display: inline-block!important; text-align: center} .MJXp-over > * {display: block!important} .MJXp-munderover > * {display: table-row!important} .MJXp-mtable {vertical-align: .25em; margin: 0 .125em} .MJXp-mtable > * {display: inline-table!important; vertical-align: middle} .MJXp-mtr {display: table-row!important} .MJXp-mtd {display: table-cell!important; text-align: center; padding: .5em 0 0 .5em} .MJXp-mtr > .MJXp-mtd:first-child {padding-left: 0} .MJXp-mtr:first-child > .MJXp-mtd {padding-top: 0} .MJXp-mlabeledtr {display: table-row!important} .MJXp-mlabeledtr > .MJXp-mtd:first-child {padding-left: 0} .MJXp-mlabeledtr:first-child > .MJXp-mtd {padding-top: 0} .MJXp-merror {background-color: #FFFF88; color: #CC0000; border: 1px solid #CC0000; padding: 1px 3px; font-style: normal; font-size: 90%} .MJXp-scale0 {-webkit-transform: scaleX(.0); -moz-transform: scaleX(.0); -ms-transform: scaleX(.0); -o-transform: scaleX(.0); transform: scaleX(.0)} .MJXp-scale1 {-webkit-transform: scaleX(.1); -moz-transform: scaleX(.1); -ms-transform: scaleX(.1); -o-transform: scaleX(.1); transform: scaleX(.1)} .MJXp-scale2 {-webkit-transform: scaleX(.2); -moz-transform: scaleX(.2); -ms-transform: scaleX(.2); -o-transform: scaleX(.2); transform: scaleX(.2)} .MJXp-scale3 {-webkit-transform: scaleX(.3); -moz-transform: scaleX(.3); -ms-transform: scaleX(.3); -o-transform: scaleX(.3); transform: scaleX(.3)} .MJXp-scale4 {-webkit-transform: scaleX(.4); -moz-transform: scaleX(.4); -ms-transform: scaleX(.4); -o-transform: scaleX(.4); transform: scaleX(.4)} .MJXp-scale5 {-webkit-transform: scaleX(.5); -moz-transform: scaleX(.5); -ms-transform: scaleX(.5); -o-transform: scaleX(.5); transform: scaleX(.5)} .MJXp-scale6 {-webkit-transform: scaleX(.6); -moz-transform: scaleX(.6); -ms-transform: scaleX(.6); -o-transform: scaleX(.6); transform: scaleX(.6)} .MJXp-scale7 {-webkit-transform: scaleX(.7); -moz-transform: scaleX(.7); -ms-transform: scaleX(.7); -o-transform: scaleX(.7); transform: scaleX(.7)} .MJXp-scale8 {-webkit-transform: scaleX(.8); -moz-transform: scaleX(.8); -ms-transform: scaleX(.8); -o-transform: scaleX(.8); transform: scaleX(.8)} .MJXp-scale9 {-webkit-transform: scaleX(.9); -moz-transform: scaleX(.9); -ms-transform: scaleX(.9); -o-transform: scaleX(.9); transform: scaleX(.9)} .MathJax_PHTML .noError {vertical-align: ; font-size: 90%; text-align: left; color: black; padding: 1px 3px; border: 1px solid} .frontend-filesViewer-components-fileDescription-module__descriptionContainer--5IiG5 { line-height: 21px; } .frontend-filesViewer-components-fileDescription-module__title--Zri4r { margin-bottom: 7px; font-weight: bold; } .frontend-filesViewer-components-fileDescription-module__description--rwuvv { margin-bottom: 7px; } .frontend-filesViewer-components-skipButton-index-module__fsButton--lRWNp, .frontend-filesViewer-components-skipButton-index-module__button--wpP\+- { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-components-skipButton-index-module__fsButton--lRWNp:focus, .frontend-filesViewer-components-skipButton-index-module__button--wpP\+-:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-skipButton-index-module__fsButton--lRWNp::before, .frontend-filesViewer-components-skipButton-index-module__fsButton--lRWNp::after, .frontend-filesViewer-components-skipButton-index-module__button--wpP\+-::before, .frontend-filesViewer-components-skipButton-index-module__button--wpP\+-::after { display: inline; font-size: 10px; } .frontend-filesViewer-components-skipButton-index-module__button--wpP\+-.frontend-filesViewer-components-skipButton-index-module__hidden--TAOdS { position: absolute; left: -10000px; top: auto; width: 1px; height: 1px; overflow: hidden; } .frontend-filesViewer-components-skipButton-index-module__button--wpP\+-.frontend-filesViewer-components-skipButton-index-module__hidden--TAOdS:focus { width: auto; height: auto; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__fs-icon-base--qOzXx, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg { height: 32px; color: #556471; text-decoration: none; display: flex; align-items: center; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg .figshare-logo { width: 32px; height: 32px; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg .figshare-logo .style0 { fill: #58585a; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg .figshare-logo .style1 { fill: #d1d2d4; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg .figshare-logo .style2 { fill: #818286; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg .figshare-logo .style3 { fill: #a8a9ad; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg::after { content: "\F144"; font-size: 10px; line-height: 32px; color: transparent; margin-left: 3px; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus::after, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:hover::after { color: #556471; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus .figshare-logo .style0, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:hover .figshare-logo .style0 { fill: #566471; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus .figshare-logo .style1, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:hover .figshare-logo .style1 { fill: #A2CD3C; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus .figshare-logo .style2, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:hover .figshare-logo .style2 { fill: #C54C59; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:focus .figshare-logo .style3, .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareButton--jh7bg:hover .figshare-logo .style3 { fill: #5BC4BD; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareText--yx0hv { font-size: 13px; margin-left: 7px; line-height: 32px; } .frontend-filesViewer-inlineMode-footer-figshareButton-module__mobileMode--yZZCk::after, .frontend-filesViewer-inlineMode-footer-figshareButton-module__mobileMode--yZZCk .frontend-filesViewer-inlineMode-footer-figshareButton-module__figshareText--yx0hv { display: none; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__fs-icon-base--OVZ9y, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__fsButton--\+W\+Ap, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__fsButton--\+W\+Ap:focus, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__fsButton--\+W\+Ap::before, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__fsButton--\+W\+Ap::after, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t::before, .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t { padding: 0; margin-left: 7px; } .frontend-filesViewer-inlineMode-fileMenu-trigger-module__listButton--Rn67t::after { content: "\F151"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-shared-components-arrowContainer-module__arrow--xh8Ew { position: absolute; } .frontend-shared-components-arrowContainer-module__arrow--xh8Ew::before, .frontend-shared-components-arrowContainer-module__arrow--xh8Ew::after { position: absolute; display: inline-block; content: ""; } .frontend-shared-components-arrowContainer-module__topArrow--Qtj6U::before { bottom: -1px; left: -11px; border-left: 11px solid transparent; border-right: 11px solid transparent; border-bottom: 12px solid #ddd; } .frontend-shared-components-arrowContainer-module__topArrow--Qtj6U::after { bottom: -1px; left: -10px; border-left: 10px solid transparent; border-right: 10px solid transparent; border-bottom: 10px solid #fff; } .frontend-shared-components-arrowContainer-module__rightArrow--Py2q4::before { top: -11px; left: -1px; border-top: 6px solid transparent; border-bottom: 6px solid transparent; border-left: 22px solid #ddd; } .frontend-shared-components-arrowContainer-module__rightArrow--Py2q4::after { top: -10px; left: -1px; border-top: 5px solid transparent; border-bottom: 5px solid transparent; border-left: 20px solid #fff; } .frontend-shared-components-arrowContainer-module__bottomArrow--VlRJp::before { top: -1px; left: -11px; border-left: 11px solid transparent; border-right: 11px solid transparent; border-top: 12px solid #ddd; } .frontend-shared-components-arrowContainer-module__bottomArrow--VlRJp::after { top: -1px; left: -10px; border-left: 10px solid transparent; border-right: 10px solid transparent; border-top: 10px solid #fff; } .frontend-shared-components-arrowContainer-module__leftArrow--nJYMJ::before { top: -11px; right: -1px; border-top: 6px solid transparent; border-bottom: 6px solid transparent; border-right: 22px solid #ddd; } .frontend-shared-components-arrowContainer-module__leftArrow--nJYMJ::after { top: -10px; right: -1px; border-top: 5px solid transparent; border-bottom: 5px solid transparent; border-right: 20px solid #fff; } /* required styles */ .leaflet-map-pane, .leaflet-tile, .leaflet-marker-icon, .leaflet-marker-shadow, .leaflet-tile-pane, .leaflet-tile-container, .leaflet-overlay-pane, .leaflet-shadow-pane, .leaflet-marker-pane, .leaflet-popup-pane, .leaflet-overlay-pane svg, .leaflet-zoom-box, .leaflet-image-layer, .leaflet-layer { position: absolute; left: 0; top: 0; } .leaflet-container { overflow: hidden; touch-action: none; } .leaflet-tile, .leaflet-marker-icon, .leaflet-marker-shadow { -webkit-user-select: none; -moz-user-select: none; -ms-user-select: none; user-select: none; -webkit-user-drag: none; } .leaflet-marker-icon, .leaflet-marker-shadow { display: block; } /* map is broken in FF if you have max-width: 100% on tiles */ .leaflet-container img { max-width: none !important; } /* stupid Android 2 doesn't understand "max-width: none" properly */ .leaflet-container img.leaflet-image-layer { max-width: 15000px !important; } .leaflet-tile { filter: inherit; visibility: hidden; } .leaflet-tile-loaded { visibility: inherit; } .leaflet-zoom-box { width: 0; height: 0; } /* workaround for */ .leaflet-overlay-pane svg { -moz-user-select: none; } .leaflet-tile-pane { z-index: 2; } .leaflet-objects-pane { z-index: 3; } .leaflet-overlay-pane { z-index: 4; } .leaflet-shadow-pane { z-index: 5; } .leaflet-marker-pane { z-index: 6; } .leaflet-popup-pane { z-index: 7; } .leaflet-vml-shape { width: 1px; height: 1px; } .lvml { behavior: url(#default#VML); display: inline-block; position: absolute; } /* control positioning */ .leaflet-control { position: relative; z-index: 7; pointer-events: auto; } .leaflet-top, .leaflet-bottom { position: absolute; z-index: 1000; pointer-events: none; } .leaflet-top { top: 0; } .leaflet-right { right: 0; } .leaflet-bottom { bottom: 0; } .leaflet-left { left: 0; } .leaflet-control { float: left; clear: both; } .leaflet-right .leaflet-control { float: right; } .leaflet-top .leaflet-control { margin-top: 10px; } .leaflet-bottom .leaflet-control { margin-bottom: 10px; } .leaflet-left .leaflet-control { margin-left: 10px; } .leaflet-right .leaflet-control { margin-right: 10px; } /* zoom and fade animations */ .leaflet-fade-anim .leaflet-tile, .leaflet-fade-anim .leaflet-popup { opacity: 0; transition: opacity 0.2s linear; } .leaflet-fade-anim .leaflet-tile-loaded, .leaflet-fade-anim .leaflet-map-pane .leaflet-popup { opacity: 1; } .leaflet-zoom-anim .leaflet-zoom-animated { transition: transform 0.25s cubic-bezier(0, 0, 0.25, 1); } .leaflet-zoom-anim .leaflet-tile, .leaflet-pan-anim .leaflet-tile, .leaflet-touching .leaflet-zoom-animated { transition: none; } .leaflet-zoom-anim .leaflet-zoom-hide { visibility: hidden; } /* cursors */ .leaflet-clickable { cursor: pointer; } .leaflet-container { cursor: -webkit-grab; cursor: -moz-grab; } .leaflet-popup-pane, .leaflet-control { cursor: auto; } .leaflet-dragging .leaflet-container, .leaflet-dragging .leaflet-clickable { cursor: move; cursor: -webkit-grabbing; cursor: -moz-grabbing; } /* visual tweaks */ .leaflet-container { background: #ddd; outline: 0; } .leaflet-container a { color: #0078A8; } .leaflet-container a.leaflet-active { outline: 2px solid orange; } .leaflet-zoom-box { border: 2px dotted #38f; background: rgba(255, 255, 255, 0.5); } /* general typography */ .leaflet-container { font: 12px/1.5 "Helvetica Neue", Arial, Helvetica, sans-serif; } /* general toolbar styles */ .leaflet-bar { box-shadow: 0 1px 5px rgba(0, 0, 0, 0.65); border-radius: 4px; } .leaflet-bar a, .leaflet-bar a:hover { background-color: #fff; border-bottom: 1px solid #ccc; width: 26px; height: 26px; line-height: 26px; display: block; text-align: center; text-decoration: none; color: black; } .leaflet-bar a, .leaflet-control-layers-toggle { background-position: 50% 50%; background-repeat: no-repeat; display: block; } .leaflet-bar a:hover { background-color: #f4f4f4; } .leaflet-bar a:first-child { border-top-left-radius: 4px; border-top-right-radius: 4px; } .leaflet-bar a:last-child { border-bottom-left-radius: 4px; border-bottom-right-radius: 4px; border-bottom: none; } .leaflet-bar a.leaflet-disabled { cursor: default; background-color: #f4f4f4; color: #bbb; } .leaflet-touch .leaflet-bar a { width: 30px; height: 30px; line-height: 30px; } /* zoom control */ .leaflet-control-zoom-in, .leaflet-control-zoom-out { font: bold 18px 'Lucida Console', Monaco, monospace; text-indent: 1px; } .leaflet-control-zoom-out { font-size: 20px; } .leaflet-touch .leaflet-control-zoom-in { font-size: 22px; } .leaflet-touch .leaflet-control-zoom-out { font-size: 24px; } /* layers control */ .leaflet-control-layers { box-shadow: 0 1px 5px rgba(0, 0, 0, 0.4); background: #fff; border-radius: 5px; } .leaflet-control-layers-toggle { background-image: url(data:image/png;base64,ZXhwb3J0IGRlZmF1bHQgX193ZWJwYWNrX3B1YmxpY19wYXRoX18gKyAiOTI4OWQ2OTRlN2E4MzJlODE1NDlhMTEzNzY1NjA3MGQucG5nIjs=); width: 36px; height: 36px; } .leaflet-retina .leaflet-control-layers-toggle { background-image: url(data:image/png;base64,ZXhwb3J0IGRlZmF1bHQgX193ZWJwYWNrX3B1YmxpY19wYXRoX18gKyAiNjljMDVlM2I2N2UyMThkNTZlNjcwZWY1YmFhODM4NjcucG5nIjs=); background-size: 26px 26px; } .leaflet-touch .leaflet-control-layers-toggle { width: 44px; height: 44px; } .leaflet-control-layers .leaflet-control-layers-list, .leaflet-control-layers-expanded .leaflet-control-layers-toggle { display: none; } .leaflet-control-layers-expanded .leaflet-control-layers-list { display: block; position: relative; } .leaflet-control-layers-expanded { padding: 6px 10px 6px 6px; color: #333; background: #fff; } .leaflet-control-layers-selector { margin-top: 2px; position: relative; top: 1px; } .leaflet-control-layers label { display: block; } .leaflet-control-layers-separator { height: 0; border-top: 1px solid #ddd; margin: 5px -10px 5px -6px; } /* attribution and scale controls */ .leaflet-container .leaflet-control-attribution { background: #fff; background: rgba(255, 255, 255, 0.7); margin: 0; } .leaflet-control-attribution, .leaflet-control-scale-line { padding: 0 5px; color: #333; } .leaflet-control-attribution a { text-decoration: none; } .leaflet-control-attribution a:hover { text-decoration: underline; } .leaflet-container .leaflet-control-attribution, .leaflet-container .leaflet-control-scale { font-size: 11px; } .leaflet-left .leaflet-control-scale { margin-left: 5px; } .leaflet-bottom .leaflet-control-scale { margin-bottom: 5px; } .leaflet-control-scale-line { border: 2px solid #777; border-top: none; line-height: 1.1; padding: 2px 5px 1px; font-size: 11px; white-space: nowrap; overflow: hidden; box-sizing: content-box; background: #fff; background: rgba(255, 255, 255, 0.5); } .leaflet-control-scale-line:not(:first-child) { border-top: 2px solid #777; border-bottom: none; margin-top: -2px; } .leaflet-control-scale-line:not(:first-child):not(:last-child) { border-bottom: 2px solid #777; } .leaflet-touch .leaflet-control-attribution, .leaflet-touch .leaflet-control-layers, .leaflet-touch .leaflet-bar { box-shadow: none; } .leaflet-touch .leaflet-control-layers, .leaflet-touch .leaflet-bar { border: 2px solid rgba(0, 0, 0, 0.2); background-clip: padding-box; } /* popup */ .leaflet-popup { position: absolute; text-align: center; } .leaflet-popup-content-wrapper { padding: 1px; text-align: left; border-radius: 12px; } .leaflet-popup-content { margin: 13px 19px; line-height: 1.4; } .leaflet-popup-content p { margin: 18px 0; } .leaflet-popup-tip-container { margin: 0 auto; width: 40px; height: 20px; position: relative; overflow: hidden; } .leaflet-popup-tip { width: 17px; height: 17px; padding: 1px; margin: -10px auto 0; transform: rotate(45deg); } .leaflet-popup-content-wrapper, .leaflet-popup-tip { background: white; box-shadow: 0 3px 14px rgba(0, 0, 0, 0.4); } .leaflet-container a.leaflet-popup-close-button { position: absolute; top: 0; right: 0; padding: 4px 4px 0 0; text-align: center; width: 18px; height: 14px; font: 16px/14px Tahoma, Verdana, sans-serif; color: #c3c3c3; text-decoration: none; font-weight: bold; background: transparent; } .leaflet-container a.leaflet-popup-close-button:hover { color: #999; } .leaflet-popup-scrolled { overflow: auto; border-bottom: 1px solid #ddd; border-top: 1px solid #ddd; } .leaflet-oldie .leaflet-popup-content-wrapper { zoom: 1; } .leaflet-oldie .leaflet-popup-tip { width: 24px; margin: 0 auto; -ms-filter: "progid:DXImageTransform.Microsoft.Matrix(M11=0.70710678, M12=0.70710678, M21=-0.70710678, M22=0.70710678)"; filter: progid:DXImageTransform.Microsoft.Matrix(M11=0.70710678, M12=0.70710678, M21=-0.70710678, M22=0.70710678); } .leaflet-oldie .leaflet-popup-tip-container { margin-top: -1px; } .leaflet-oldie .leaflet-control-zoom, .leaflet-oldie .leaflet-control-layers, .leaflet-oldie .leaflet-popup-content-wrapper, .leaflet-oldie .leaflet-popup-tip { border: 1px solid #999; } /* div icon */ .leaflet-div-icon { background: #fff; border: 1px solid #666; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fs-icon-base--2oV\+5, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__mobileMode--XTVaW.frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O::before, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__mobileMode--XTVaW.frontend-filesViewer-inlineMode-fileMenu-fileItem-module__downloadButton--7F7Y4::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileDetails--tczV1 { display: flex; flex-direction: row; flex-wrap: nowrap; align-items: center; border-bottom: 1px solid #ddd; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileDetails--tczV1:hover { background-color: #f5f5f5; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileDetails--tczV1:last-child { border-bottom: 0px none transparent; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileName--TOIPE, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__downloadButton--7F7Y4 { padding: 14px 7px; margin: 0 7px; line-height: 16px; font-size: 13px; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileName--TOIPE:focus, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O:focus, .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__downloadButton--7F7Y4:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; outline: none; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileName--TOIPE { display: flex; flex-direction: row; flex-grow: 1; text-align: left; width: 50%; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileBase--GHrON { white-space: nowrap; text-overflow: ellipsis; overflow: hidden; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__fileExt--lAwtx { white-space: nowrap; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O { margin-right: 0; text-decoration: none; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O[disabled] { visibility: hidden; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__mobileMode--XTVaW.frontend-filesViewer-inlineMode-fileMenu-fileItem-module__viewButton--XDE5O::before { content: "\F19C"; margin: 0 7px; font-size: 9px; vertical-align: middle; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__downloadButton--7F7Y4 { margin-left: 0; text-decoration: none; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__mobileMode--XTVaW.frontend-filesViewer-inlineMode-fileMenu-fileItem-module__downloadButton--7F7Y4::before { content: "\F12D"; margin: 0 7px; font-size: 10px; vertical-align: middle; } .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__mobileMode--XTVaW .frontend-filesViewer-inlineMode-fileMenu-fileItem-module__buttonText--WsCw4 { display: none; } .frontend-filesViewer-inlineMode-fileMenu-menu-module__arrowContainer--UrZHs { position: absolute; top: 0; left: 0; width: 100%; height: 100%; max-height: 100%; background: rgba(255, 255, 255, 0.75); } .frontend-filesViewer-inlineMode-fileMenu-menu-module__listContainer--a2JRc { position: absolute; bottom: 0; left: 0; right: 0; max-height: 80%; background: #fff; border: 1px solid #ddd; box-sizing: border-box; box-shadow: 0 -4px 8px rgba(221, 221, 221, 0.88); z-index: 0; overflow: auto; } .frontend-filesViewer-inlineMode-footer-fileControls-module__fs-icon-base--DuTWO, .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7::after, .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-footer-fileControls-module__fsButton--mgzir, .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7, .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-footer-fileControls-module__fsButton--mgzir:focus, .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7:focus, .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-footer-fileControls-module__fsButton--mgzir::before, .frontend-filesViewer-inlineMode-footer-fileControls-module__fsButton--mgzir::after, .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7::before, .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7::after, .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz::before, .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-footer-fileControls-module__container--uW8u2 { display: flex; align-items: center; } .frontend-filesViewer-inlineMode-footer-fileControls-module__info--rnSm4 { font-size: 13px; font-weight: bold; color: #464646; margin-right: 14px; } .frontend-filesViewer-inlineMode-footer-fileControls-module__mobileMode--iswi4 .frontend-filesViewer-inlineMode-footer-fileControls-module__info--rnSm4 { display: none; } .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7 { padding: 0; margin-right: 3px; } .frontend-filesViewer-inlineMode-footer-fileControls-module__prevButton--ducF7::after { content: "\F179"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz { padding: 0; } .frontend-filesViewer-inlineMode-footer-fileControls-module__nextButton--yR8Qz::after { content: "\F17B"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__fs-icon-base--OwY4q, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq::before, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__mobileMode--J5z8\+::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__fsButton--Ih8rm, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__fsButton--Ih8rm:focus, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__fsButton--Ih8rm::before, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__fsButton--Ih8rm::after, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq::before, .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq { margin-left: 7px; display: flex; white-space: nowrap; align-items: center; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__downloadButton--UsBvq::before { content: "\F12C"; margin-right: 6px; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__mobileMode--J5z8\+ { padding: 0; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__mobileMode--J5z8\+::after { content: "\F12C"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__mobileMode--J5z8\+::before { display: none; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__buttonText--GEnvG { display: inherit; font-size: 13px; } .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__mobileMode--J5z8\+ .frontend-filesViewer-inlineMode-downloadMenu-trigger-module__buttonText--GEnvG { display: none; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__wrapper--pWKJY { position: absolute; top: 0; left: 0; width: 100%; height: 100%; max-height: 100%; background: rgba(255, 255, 255, 0.75); } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__container--7gpjl { position: absolute; bottom: 0; left: 0; right: 0; max-height: 80%; background: #fff; border: 1px solid #ddd; box-sizing: border-box; box-shadow: 0 -4px 8px rgba(221, 221, 221, 0.88); display: flex; flex-direction: column; z-index: 0; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__downloadItem--4hbaF { padding: 14px; border-bottom: 1px solid #ddd; font-size: 13px; font-weight: bold; text-align: left; text-decoration: none; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__downloadItem--4hbaF:last-child { border: 0; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__downloadItem--4hbaF:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; outline: none; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__downloadItem--4hbaF[disabled] { color: #444; font-weight: normal; } .frontend-filesViewer-inlineMode-downloadMenu-menu-module__downloadItem--4hbaF[disabled]:hover { color: #444; } .frontend-filesViewer-inlineMode-footer-genericControls-module__fs-icon-base--a-WBW, .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC::after, .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::before, .frontend-filesViewer-inlineMode-footer-genericControls-module__mobileMode--cGFEe .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-footer-genericControls-module__fsButton--IvGIW, .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC, .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-footer-genericControls-module__fsButton--IvGIW:focus, .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC:focus, .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-footer-genericControls-module__fsButton--IvGIW::before, .frontend-filesViewer-inlineMode-footer-genericControls-module__fsButton--IvGIW::after, .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC::before, .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC::after, .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::before, .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-footer-genericControls-module__container--Ok8Uj { display: flex; align-items: center; } .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC { padding: 0; } .frontend-filesViewer-inlineMode-footer-genericControls-module__enlargeButton--qJnBC::after { content: "\F13F"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD { margin-left: 7px; display: flex; white-space: nowrap; align-items: center; } .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::before { content: "\F175"; margin-right: 6px; } .frontend-filesViewer-inlineMode-footer-genericControls-module__mobileMode--cGFEe .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD { padding: 0; } .frontend-filesViewer-inlineMode-footer-genericControls-module__mobileMode--cGFEe .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::after { content: "\F175"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-inlineMode-footer-genericControls-module__mobileMode--cGFEe .frontend-filesViewer-inlineMode-footer-genericControls-module__shareButton--RoEPD::before { display: none; } .frontend-filesViewer-inlineMode-footer-genericControls-module__hide--xlLS- { display: none; } .frontend-filesViewer-inlineMode-footer-genericControls-module__buttonText--eWe\+K { display: inherit; font-size: 13px; } .frontend-filesViewer-inlineMode-footer-genericControls-module__mobileMode--cGFEe .frontend-filesViewer-inlineMode-footer-genericControls-module__buttonText--eWe\+K { display: none; } .frontend-filesViewer-inlineMode-footer-index-module__footer--9uUmL { padding: 14px; box-sizing: border-box; background: #f5f5f5; border-top: 1px solid #ddd; } .frontend-filesViewer-inlineMode-footer-index-module__navigation--\+GhAf { display: flex; align-items: center; justify-content: space-between; width: 100%; } .frontend-filesViewer-inlineMode-footer-index-module__figshareButton--nJvGg { order: 0; } .frontend-filesViewer-inlineMode-footer-index-module__fileControls--KCjEL { order: 1; } .frontend-filesViewer-inlineMode-footer-index-module__genericControls--q\+AJj { order: 2; } .frontend-filesViewer-inlineMode-header-module__header--L5zPq { display: flex; max-width: 100%; flex-direction: row; flex-wrap: nowrap; border-bottom: 1px solid #ddd; overflow: hidden; background: #f5f5f5; } .frontend-filesViewer-inlineMode-header-module__titleSection--cXZcY { flex-grow: 1; width: 50%; padding: 14px; } .frontend-filesViewer-inlineMode-header-module__title--WcAab { margin: 0; font-size: 18px; line-height: 25px; font-weight: bold; color: #464646; } .frontend-filesViewer-inlineMode-header-module__fileInfoSection--ory0g { font-size: 13px; line-height: 18px; overflow: hidden; white-space: nowrap; text-overflow: ellipsis; color: #464646; } .frontend-filesViewer-inlineMode-header-module__fileName--l1HGx { white-space: nowrap; color: #111; } .frontend-filesViewer-inlineMode-header-module__statsContainer--K8Fp7 { display: flex; flex-direction: row; } .frontend-filesViewer-inlineMode-header-module__statsSection--IicLP { display: flex; align-items: center; justify-content: center; flex-direction: column; padding: 0 14px; border-left: 1px solid #ddd; } .frontend-filesViewer-inlineMode-header-module__statsCount--gpeZo, .frontend-filesViewer-inlineMode-header-module__statsType--eE7Fw { width: 100%; text-align: center; } .frontend-filesViewer-inlineMode-header-module__statsCount--gpeZo { font-size: 14px; } .frontend-filesViewer-inlineMode-header-module__statsType--eE7Fw { font-size: 11px; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt { flex-direction: column; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__titleSection--cXZcY { width: auto; padding: 7px 14px; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsContainer--K8Fp7 { border-top: 1px solid #ddd; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsSection--IicLP { width: 100%; flex-direction: row; align-items: baseline; padding: 12px 7px 7px 7px; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsSection--IicLP:first-child { border-left: 0 none transparent; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsCount--gpeZo, .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsType--eE7Fw { width: auto; } .frontend-filesViewer-inlineMode-header-module__mobileMode--r3Hjt .frontend-filesViewer-inlineMode-header-module__statsType--eE7Fw { margin-left: 5px; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fs-icon-base--\+d8bC, .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-expand::before, .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-collapse::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq { color: #464646; line-height: 21px; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-collapsed .fs-content-wrapper { max-height: 55px; overflow: hidden; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-toggle { position: relative; margin-top: -10px; padding: 10px 0 0 0; background: linear-gradient(to bottom, transparent 0px, #eee 10px); text-align: center; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-toggle button::before { display: inline-block; font-size: 10px; transform: scale(0.5); margin-right: 5px; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-expanded { height: 86px; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-expanded .fs-toggle { top: 55px; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-expanded .fs-content-wrapper { position: absolute; left: 0; bottom: 30px; width: 100%; max-height: 50%; padding: 12px; overflow-y: auto; background: #eee; border-top: 1px solid #ddd; box-sizing: border-box; box-shadow: 0 0 10px rgba(255, 255, 255, 0.5); } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-expand::before { content: "\F133"; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__fileDescription--l1BZq .fs-collapse::before { content: "\F11D"; } .frontend-filesViewer-enlargedMode-fileDetails-description-module__mobileDescription--U5tqf .fs-expanded .fs-content-wrapper { max-height: 90%; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fs-icon-base--UECC3, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__mobileDownloadButton--VSsFt::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fsButton--Ajee3, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fsButton--Ajee3:focus, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fsButton--Ajee3::before, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fsButton--Ajee3::after, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i::before, .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i::after { display: inline; font-size: 10px; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__mobileDownloadButton--VSsFt { padding: 0; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__mobileDownloadButton--VSsFt::after { content: "\F12D"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i { font-size: 11px; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__desktopDownloadButton--rDV8i strong { margin-right: 6px; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__hideContent---vFyd { display: none; } .frontend-filesViewer-enlargedMode-fileDetails-downloadButton-module__fileSize--Y5L2N { white-space: nowrap; } .frontend-filesViewer-enlargedMode-fileDetails-index-module__fs-icon-base--RO0BQ { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-enlargedMode-fileDetails-index-module__container--3tezS { width: 100%; background: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileDetails-index-module__section--75P43 { padding: 10px; border-top: 1px solid #ddd; font-size: 11px; } .frontend-filesViewer-enlargedMode-fileDetails-index-module__titleSection--CKY3U { margin-right: 10px; display: flex; justify-content: space-between; align-items: center; } .frontend-filesViewer-enlargedMode-fileDetails-index-module__title--Xs\+MM { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fs-icon-base--OMbBA, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-audio::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-archive::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-molecule::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dataset::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_canvas::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_failsafe::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-txt::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-kml::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-viewer3d::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-jupyter::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-graph::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-fits::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dicom::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord { display: block; position: relative; height: 80px; width: 100%; font-size: 11px; border-bottom: 1px solid rgba(70, 70, 70, 0.5); text-align: left; transition: background-color 0.2s ease-in; transform: translate3d(0, 0, 0); } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord:hover, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord:focus { background-color: #333; } .frontend-filesViewer-enlargedMode-fileListing-file-module__selectedFile--FCcDh { background-color: #333; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDetails--nntC8 { position: absolute; top: 0; left: 83px; right: 6px; bottom: 0; display: flex; flex-flow: column; justify-content: center; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileTitle--3nxoY { color: #fff; font-weight: bold; display: block; margin-bottom: 7px; text-overflow: ellipsis; white-space: nowrap; overflow: hidden; width: 100%; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileSize--fQuKP { font-weight: bold; color: #999; white-space: nowrap; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord:hover .frontend-filesViewer-enlargedMode-fileListing-file-module__fileSize--fQuKP, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileDisplay--Fsord:focus .frontend-filesViewer-enlargedMode-fileListing-file-module__fileSize--fQuKP, .frontend-filesViewer-enlargedMode-fileListing-file-module__selectedFile--FCcDh .frontend-filesViewer-enlargedMode-fileListing-file-module__fileSize--fQuKP { color: #bbb; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza { display: block; position: absolute; left: 6px; top: 6px; width: 68px; height: 68px; overflow: hidden; background: #fff; background-image: url('data:image/jpg;base64,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'); } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza div, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza img { width: 100%; height: 100%; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-generic-preview { background-image: url('data:image/jpg;base64,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'); } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-audio { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-audio::before { content: "\F186"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-archive { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-archive::before { content: "\F194"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-molecule { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-molecule::before { content: "\F193"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dataset { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dataset::before { content: "\F189"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_canvas, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_failsafe { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_canvas::before, .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-document_failsafe::before { content: "\F15D"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-txt { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-txt::before { content: "\F188"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-kml { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-kml::before { content: "\F191"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-viewer3d { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-viewer3d::before { content: "\F183"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-jupyter { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-jupyter::before { content: "\F18F"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-graph { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-graph::before { content: "\F18D"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-fits { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-fits::before { content: "\F139"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dicom { text-align: center; content: ""; background: #464646; color: #f0f0f0; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza .fs-dicom::before { content: "\F139"; text-align: center; font-size: 35px; vertical-align: middle; display: inline-block; line-height: 68px; } .frontend-filesViewer-enlargedMode-fileListing-file-module__fileThumbnail--BxEza img { max-width: 100%; max-height: 100%; -o-object-fit: scale-down; object-fit: scale-down; background: #fff; } .frontend-filesViewer-enlargedMode-fileListing-index-module__fileListing--VmRxq { color: #fff; position: absolute; top: 0; bottom: 0; left: 0; right: 0; overflow-y: auto; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fs-icon-base--OMw4b, .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-::after, .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI::after, .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page::before, .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fsButton--hjYQQ, .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-, .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fsButton--hjYQQ:focus, .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-:focus, .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fsButton--hjYQQ::before, .frontend-filesViewer-enlargedMode-enlargedMode-module__fsButton--hjYQQ::after, .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-::before, .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-::after, .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI::before, .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI::after { display: inline; font-size: 10px; } .frontend-filesViewer-enlargedMode-enlargedMode-module__enlargedMode---3Pwt { width: 100%; height: 100%; } .frontend-filesViewer-enlargedMode-enlargedMode-module__mainHeading--fPzt2 { position: absolute; left: -10000px; top: auto; width: 1px; height: 1px; overflow: hidden; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fileViewerContainer--aHeOh { margin-left: 300px; margin-right: 50px; height: 100%; display: flex; align-items: center; flex-direction: column; justify-content: center; } .frontend-filesViewer-enlargedMode-enlargedMode-module__fileViewerContainer--aHeOh .fs-figshare-viewer { margin: 0 auto; background: #fff; overflow: hidden; } .frontend-filesViewer-enlargedMode-enlargedMode-module__mobileView--pslHV .frontend-filesViewer-enlargedMode-enlargedMode-module__fileViewerContainer--aHeOh { position: fixed; top: 44px; margin: 0; height: auto; } .frontend-filesViewer-enlargedMode-enlargedMode-module__viewerWrapper--M5uiU { position: relative; overflow-y: hidden; } .frontend-filesViewer-enlargedMode-enlargedMode-module__viewerWrapper--M5uiU .frontend-filesViewer-enlargedMode-enlargedMode-module__skipViewerButton--yWEWB:focus { position: absolute; top: 7px; left: 7px; right: auto; z-index: 1; } .frontend-filesViewer-enlargedMode-enlargedMode-module__viewerContainer--1e-RP { border: 0 none transparent; } .frontend-filesViewer-enlargedMode-enlargedMode-module__viewerContainer--1e-RP .click-outside-wrapper { display: inline; } .frontend-filesViewer-enlargedMode-enlargedMode-module__filesList--shBsr { position: fixed; left: 0; top: 0; bottom: 0; width: 250px; overflow-y: auto; background: #000; } .frontend-filesViewer-enlargedMode-enlargedMode-module__mobileView--pslHV .frontend-filesViewer-enlargedMode-enlargedMode-module__filesList--shBsr { z-index: 3; width: 66%; top: 44px; } .frontend-filesViewer-enlargedMode-enlargedMode-module__mobileView--pslHV .frontend-filesViewer-enlargedMode-enlargedMode-module__filesList--shBsr::after { position: fixed; top: 0; left: 66%; right: 0; bottom: 0; content: " "; background: rgba(0, 0, 0, 0.85); } .frontend-filesViewer-enlargedMode-enlargedMode-module__menuBar--U40XK { z-index: 2; position: fixed; right: 0; top: 0; left: 0; padding: 0 20px; display: flex; flex-direction: row; justify-content: space-between; align-items: center; transform: translate3d(0, 0, 0); } .frontend-filesViewer-enlargedMode-enlargedMode-module__mobileView--pslHV .frontend-filesViewer-enlargedMode-enlargedMode-module__menuBar--U40XK { height: 44px; background: #000; } .frontend-filesViewer-enlargedMode-enlargedMode-module__desktopView--mv7vB .frontend-filesViewer-enlargedMode-enlargedMode-module__menuBar--U40XK { left: 250px; height: 75px; justify-content: flex-end; } .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb- { order: 2; padding: 0; } .frontend-filesViewer-enlargedMode-enlargedMode-module__closeOverlayButton--W9hb-::after { content: "\F124"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI { order: 1; padding: 0; } .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI::after { content: "\F152"; display: inline-block; width: 24px; line-height: 24px; text-align: center; } .frontend-filesViewer-enlargedMode-enlargedMode-module__desktopView--mv7vB .frontend-filesViewer-enlargedMode-enlargedMode-module__toggleListButton--sYzoI { display: none; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua { position: absolute; top: 50%; left: 250px; right: 0; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-pagination-info { display: none; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page, .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page { color: #999; position: absolute; height: 42px; font-size: 42px; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page:focus, .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page:hover, .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page:hover { color: #fff; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page[disabled], .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page[disabled] { color: #666; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page { left: 12px; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-prev-page::before { content: "\F108"; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page { right: 12px; } .frontend-filesViewer-enlargedMode-enlargedMode-module__navigationContainer--0U8Ua .fs-next-page::before { content: "\F109"; } .frontend-filesViewer-inlineMode-enlargeScreen-module__overlayContainer--nhxkt { width: 100%; height: 100%; } .frontend-filesViewer-inlineMode-enlargeScreen-module__overlayContainer--nhxkt .fs-overlay-content { position: absolute; left: 0; top: 0; bottom: 0; right: 0; } .frontend-filesViewer-inlineMode-enlargeScreen-module__overlayContainer--nhxkt .fs-overlay-content > div > div:first-child { width: 100%; height: 100%; } .frontend-filesViewer-inlineMode-enlargeScreen-module__overlayContainer--nhxkt .fs-overlay-content .fs-viewer-container { border: 0 none transparent; } .frontend-shared-components-searchInput-module__fs-icon-base--loB7A, .frontend-shared-components-searchInput-module__searchIcon--K3Tpq::after, .frontend-shared-components-searchInput-module__clearSearch--eZi0x::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-shared-components-searchInput-module__container--C2fcU { display: flex; align-items: center; position: relative; } .frontend-shared-components-searchInput-module__searchInput--8BR0k { width: inherit; height: inherit; box-sizing: border-box; font-size: 14px; line-height: 1; padding-left: 7px; padding-right: 31px; -webkit-appearance: textfield; } .frontend-shared-components-searchInput-module__searchInput--8BR0k:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-shared-components-searchInput-module__searchInput--8BR0k::-webkit-search-decoration, .frontend-shared-components-searchInput-module__searchInput--8BR0k::-webkit-search-cancel-button, .frontend-shared-components-searchInput-module__searchInput--8BR0k::-webkit-search-results-button, .frontend-shared-components-searchInput-module__searchInput--8BR0k::-webkit-search-results-decoration { display: none; } .frontend-shared-components-searchInput-module__searchInput--8BR0k::-ms-clear { display: none; } .frontend-shared-components-searchInput-module__icon--T-Pas { position: absolute; right: 7px; top: 0; bottom: 0; margin: auto; width: 24px; height: 24px; display: flex; align-items: center; justify-content: center; } .frontend-shared-components-searchInput-module__icon--T-Pas::after { font-size: 14px; color: #bbb; } .frontend-shared-components-searchInput-module__searchIcon--K3Tpq { } .frontend-shared-components-searchInput-module__searchIcon--K3Tpq::after { content: "\F171"; } .frontend-shared-components-searchInput-module__clearSearch--eZi0x { } .frontend-shared-components-searchInput-module__clearSearch--eZi0x:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-shared-components-searchInput-module__clearSearch--eZi0x::after { content: "\F112"; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw { box-sizing: border-box; font-size: 14px; line-height: 21px; padding: 7px 14px 7px 0; margin-left: 14px; border-top: 1px solid #ddd; width: calc(100% - 14px); text-align: left; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:first-of-type { border-top: 0 none transparent; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:last-of-type { border-bottom: 1px solid #ddd; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:focus, .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:hover { margin-left: 0; padding-left: 14px; width: 100%; background: #f8f8f8; } .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:focus + .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw, .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw:hover + .frontend-filesViewer-components-citationSelector-citationItem-module__option--BgpRw { margin-left: 0; padding-left: 14px; width: 100%; } .frontend-filesViewer-components-citationSelector-citationItem-module__selected--iaXKp { font-weight: bold; } .frontend-filesViewer-components-citationSelector-citationItem-module__defaultFlag--lb\+LQ { font-weight: normal; color: #999; } .frontend-filesViewer-components-citationSelector-loadingInfo-module__container--fOlcy { font-size: 12px; display: flex; align-items: center; justify-content: center; color: #464646; height: 84px; } .frontend-filesViewer-components-citationSelector-loadingInfo-module__container--fOlcy.frontend-filesViewer-components-citationSelector-loadingInfo-module__resultsInfo--SMtp8 { height: 63px; } .frontend-filesViewer-components-citationSelector-loadingInfo-module__loading--ACli7::before { width: 16px; height: 16px; background-image: url('data:image/gif;charset=utf-8;base64,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'); margin-right: 7px; content: ""; } .frontend-filesViewer-components-citationSelector-searchScreen-module__fs-icon-base--JAe4X, .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN::before, .frontend-filesViewer-components-citationSelector-searchScreen-module__searchButton--q59n3::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-components-citationSelector-searchScreen-module__fsButton--sdPlu, .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-components-citationSelector-searchScreen-module__fsButton--sdPlu:focus, .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-citationSelector-searchScreen-module__fsButton--sdPlu::before, .frontend-filesViewer-components-citationSelector-searchScreen-module__fsButton--sdPlu::after, .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN::before, .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN::after { display: inline; font-size: 10px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__container--NrclU { position: absolute; top: 0; left: 0; bottom: 0; width: 100%; background: #fff; border: 1px solid #ddd; border-bottom: 0px none transparent; box-sizing: border-box; display: flex; flex-direction: column; } .frontend-filesViewer-components-citationSelector-searchScreen-module__footer--rfpdK { width: 100%; height: 35px; box-sizing: border-box; padding: 0 14px; display: flex; justify-content: space-between; align-items: center; border-bottom: 1px solid #ddd; } .frontend-filesViewer-components-citationSelector-searchScreen-module__outsideArea--emujb { height: calc(100% - 35px); } .frontend-filesViewer-components-citationSelector-searchScreen-module__innerArea--DTEtB { max-height: calc(100% - 49px); } .frontend-filesViewer-components-citationSelector-searchScreen-module__activeScroll--fLN1x { overflow: auto; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchInput--IJPfp { height: 49px; width: 100%; box-sizing: border-box; border-bottom: 1px solid #ddd; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchInput--IJPfp input { padding-left: 14px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchInput--IJPfp input:focus { box-shadow: inset 0 0 1px 1px #ffa500; } .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN { border: none; margin-left: -6px; font-size: 14px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__backButton--MfrYN::before { content: "\F108"; margin-right: 6px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchButton--q59n3 { width: 24px; height: 24px; display: flex; align-items: center; justify-content: center; border-radius: 3px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchButton--q59n3::before { content: "\F171"; font-size: 14px; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchButton--q59n3:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-citationSelector-searchScreen-module__searchButton--q59n3[disabled] { display: none; } .frontend-filesViewer-components-citationSelector-app-module__fs-icon-base--InM2o, .frontend-filesViewer-components-citationSelector-app-module__trigger--0c1XS::after { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-components-citationSelector-app-module__container---8bf5 { width: 70%; height: 24px; display: flex; justify-content: flex-end; position: inherit; z-index: 1; } .frontend-filesViewer-components-citationSelector-app-module__trigger--0c1XS { border: none; max-width: 100%; display: flex; align-items: center; height: 24px; } .frontend-filesViewer-components-citationSelector-app-module__trigger--0c1XS:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-citationSelector-app-module__trigger--0c1XS::after { content: "\F109"; font-size: 11px; width: 12px; text-align: right; } .frontend-filesViewer-components-citationSelector-app-module__buttonText--5Tykm { font-size: 14px; margin-right: 0; color: #464646; flex: 0 1 auto; white-space: nowrap; overflow: hidden; text-overflow: ellipsis; } .frontend-filesViewer-components-citationContent-module__container--OF5zz { display: flex; flex-direction: column; background: #fff; font-size: 14px; } .frontend-filesViewer-components-citationContent-module__citationText---V0yO { line-height: 21px; word-wrap: break-word; } .frontend-filesViewer-components-citationContent-module__citationDoi--hmjWt { text-decoration: none; line-height: 21px; font-weight: bold; max-width: 100%; overflow: hidden; white-space: nowrap; text-overflow: ellipsis; box-sizing: border-box; border-top: 1px solid #ddd; border-bottom: 1px solid #ddd; margin-top: 14px; margin-right: auto; padding: 7px 0; } .frontend-filesViewer-components-citationScreen-app-module__fs-icon-base--TcxWI, .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-components-citationScreen-app-module__fsButton--5GZo9, .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-components-citationScreen-app-module__fsButton--5GZo9:focus, .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-components-citationScreen-app-module__fsButton--5GZo9::before, .frontend-filesViewer-components-citationScreen-app-module__fsButton--5GZo9::after, .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X::before, .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X::after { display: inline; font-size: 10px; } .frontend-filesViewer-components-citationScreen-app-module__container--nlNwG { font-size: 14px; padding: 28px 21px 21px 21px; background: #fff; width: 100%; max-height: 100%; border: 1px solid #ddd; box-sizing: border-box; box-shadow: 0 -4px 8px rgba(221, 221, 221, 0.88); overflow: hidden; } .frontend-filesViewer-components-citationScreen-app-module__container--nlNwG.frontend-filesViewer-components-citationScreen-app-module__selectorOpen--ep9QL { position: absolute; left: 0; bottom: 0; height: 90%; } .frontend-filesViewer-components-citationScreen-app-module__title--Rh\+AT { font-size: 18px; font-weight: normal; line-height: 28px; margin-bottom: 14px; } .frontend-filesViewer-components-citationScreen-app-module__citationContent--jIRLp { padding: 0; } .frontend-filesViewer-components-citationScreen-app-module__footer--fn4lP { display: flex; align-items: center; justify-content: space-between; margin-top: 42px; } .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X { border: none; margin-left: -6px; font-size: 14px; } .frontend-filesViewer-components-citationScreen-app-module__backButton--fEU0X::before { content: "\F108"; margin-right: 6px; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fs-icon-base--l8C-P, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fsButton--9BL1y, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fsButton--9BL1y:focus, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc:focus, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN:focus, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fsButton--9BL1y::before, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fsButton--9BL1y::after, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc::before, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc::after, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN::before, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN::after, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv::before, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__container--OGrNS { display: flex; flex-direction: column; width: 300px; margin: 0 auto; width: 100%; margin: 0; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__item--DTqSw { display: flex; align-items: center; height: 63px; box-sizing: border-box; border-bottom: 1px solid #ddd; justify-content: center; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__item--DTqSw:last-child { border-bottom: 0px none transparent; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc { display: flex; flex-direction: row; align-items: center; color: #fff; background: #3b5998; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc:focus, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__facebookButton--HyhQc:hover { background: #2d4373; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN { display: flex; flex-direction: row; align-items: center; color: #fff; background: #00aced; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN:focus, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twitterButton--1oQZN:hover { background: #0087ba; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__fbIcon--O-n8e, .frontend-filesViewer-inlineMode-shareScreen-shareList-module__twIcon--qqL87 { margin-right: 3px; fill: #fff; height: 16px; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv::before { content: "\F116"; margin-right: 6px; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__citeButton--PhdXv::before { font-size: 20px; line-height: 13px; position: relative; top: 1px; } .frontend-filesViewer-inlineMode-shareScreen-shareList-module__buttonText--7kz\+e { font-size: 13px; } .frontend-filesViewer-inlineMode-shareScreen-index-module__fs-icon-base--OakMD { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-shareScreen-index-module__fsButton--BU4T- { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-shareScreen-index-module__fsButton--BU4T-:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-shareScreen-index-module__fsButton--BU4T-::before, .frontend-filesViewer-inlineMode-shareScreen-index-module__fsButton--BU4T-::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-shareScreen-index-module__shareTitle--GVgtZ { width: 100%; padding: 28px 14px 14px 14px; font-weight: normal; font-size: 18px; line-height: 28px; text-align: center; box-sizing: border-box; } .frontend-filesViewer-inlineMode-shareScreen-index-module__arrowContainer--Hz6Pd { position: absolute; top: 0; left: 0; width: 100%; height: 100%; max-height: 100%; background: rgba(255, 255, 255, 0.75); display: flex; flex-direction: column; justify-content: flex-end; } .frontend-filesViewer-inlineMode-shareScreen-index-module__arrowShareContainer--MLaZ6 { width: 100%; max-height: 80%; background: #fff; border: 1px solid #ddd; box-sizing: border-box; box-shadow: 0 -4px 8px rgba(221, 221, 221, 0.88); z-index: 0; overflow: auto; } .frontend-filesViewer-inlineMode-mainSection-module__fs-icon-base--IY4SY, .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3::before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } .frontend-filesViewer-inlineMode-mainSection-module__fsButton--Fomyr, .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3 { display: inline-block; height: 24px; line-height: 24px; border: 1px solid #ddd; border-radius: 3px; padding: 0 6px; text-decoration: none; } .frontend-filesViewer-inlineMode-mainSection-module__fsButton--Fomyr:focus, .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-mainSection-module__fsButton--Fomyr::before, .frontend-filesViewer-inlineMode-mainSection-module__fsButton--Fomyr::after, .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3::before, .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3::after { display: inline; font-size: 10px; } .frontend-filesViewer-inlineMode-mainSection-module__mainSection--xdIHv { position: relative; } .frontend-filesViewer-inlineMode-mainSection-module__viewerContainer--HLGmA { position: relative; overflow: hidden; border-left: 1px solid #ddd; border-right: 1px solid #ddd; box-sizing: border-box; background: #fff; } .frontend-filesViewer-inlineMode-mainSection-module__viewerContainer--HLGmA .click-outside-wrapper { display: inline; } .frontend-filesViewer-inlineMode-mainSection-module__expandButton--E5RNi { position: absolute; top: 0; left: 0; width: 100%; height: 100%; padding: 0; margin: 0; border: 0 none transparent; display: flex; flex-flow: column; justify-content: flex-end; align-items: center; } .frontend-filesViewer-inlineMode-mainSection-module__expandButton--E5RNi:focus .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3 { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3 { width: 70px; height: 26px; border-radius: 14px; border: 1px solid #ddd; margin: 0 auto 21px auto; background: #f5f5f5; font-size: 11px; display: flex; align-items: center; justify-content: center; } .frontend-filesViewer-inlineMode-mainSection-module__expandButtonContent--w7pZ3::before { content: "\F173"; margin-right: 6px; } .frontend-filesViewer-inlineMode-index-module__container--LzxR7 { position: relative; } .frontend-filesViewer-inlineMode-index-module__fileDescription--55hIr { background: #f5f5f5; padding: 14px 14px 7px 14px; border-top: 1px solid #ddd; font-size: 13px; } .frontend-filesViewer-inlineMode-index-module__skipContainer--VO89A { position: relative; width: 100%; height: 0; z-index: 1; } .frontend-filesViewer-inlineMode-index-module__skipContainer--VO89A .frontend-filesViewer-inlineMode-index-module__skipButton--JbCgu:focus { position: absolute; top: 7px; left: 7px; right: auto; } /** * Load figshare icon font **/ .fs-icon-base, figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir::before, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir::before, figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir.fs-expanded::before, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir.fs-expanded::before, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play.fs-is-playing::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play.fs-is-playing::before, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button.fs-is-mute::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button.fs-is-mute::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play.fs-is-playing::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play.fs-is-playing::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button.fs-is-mute::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button.fs-is-mute::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen.fs-exit-fullscreen::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen.fs-exit-fullscreen::before, figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play::before, figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before, figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before, figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause::before, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control::before, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control::before, figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.next::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.next::after, figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.prev::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.prev::after, figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta::after, figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close::after, figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close::before, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-in:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-in:before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-out:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-out:before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-fit:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-fit:before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name::after, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-prev-layer-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-prev-layer-button:before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-next-layer-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-next-layer-button:before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button:before { font-family: "figIcon"; speak: none; font-style: normal; font-weight: normal; font-variant: normal; text-transform: none; line-height: 1; letter-spacing: 0; -ms-font-feature-settings: "liga" 1; -o-font-feature-settings: "liga"; font-feature-settings: "liga", normal; -webkit-font-smoothing: antialiased; -moz-osx-font-smoothing: grayscale; } @font-face { font-family: "figIcon"; src: url(data:application/vnd.ms-fontobject;base64,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 src: url(data:application/vnd.ms-fontobject;base64,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 format("embedded-opentype"), url(data:font/woff;base64,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 format('woff'), url(data:font/ttf;base64,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) format('truetype'), url(data:image/svg+xml;base64,<?xml version="1.0" standalone="no"?>
<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd" >
<svg xmlns="http://www.w3.org/2000/svg">
<defs>
  <font id="figIcon" horiz-adv-x="4333.333333333333">
    <font-face font-family="figIcon"
      units-per-em="1000" ascent="1000"
      descent="0" />
    <missing-glyph horiz-adv-x="0" />
    <glyph glyph-name="3D"
      unicode="&#xF101;"
      horiz-adv-x="999.9441964285716" d="M999.9441964285716 347.65625L869.3080357142859 282.4776785714286V326.0602678571428H369.4754464285715H346.372767857143L97.4888392857144 59.7656249999999L130.1339285714287 28.4598214285714L0 0L23.7165178571429 130.5245535714286L67.4665178571429 88.5602678571429L326.0044642857144 365.1785714285715V369.5870535714287V891.1830357142857H282.3660714285715L347.5446428571429 1000L412.7232142857142 891.1830357142857H369.4196428571428V369.5870535714286H869.2522321428571V412.8906249999999L999.9441964285716 347.65625z" />
    <glyph glyph-name="accessible"
      unicode="&#xF102;"
      horiz-adv-x="1500" d="M1209.7098214285716 622.8236607142858C1167.1875 837.8348214285714 977.5111607142856 1000 750 1000C569.3638392857142 1000 412.8348214285714 897.4888392857143 334.3750000000001 747.8236607142858C146.5401785714286 727.5111607142858 0 568.4151785714286 0 375C0 167.8013392857143 167.8013392857143 0 375 0H1187.5C1359.9888392857142 0 1500 140.0111607142858 1500 312.5C1500 477.5111607142858 1371.5401785714287 611.2723214285713 1209.7098214285716 622.8236607142858M625 187.5L406.25 406.25L494.6986607142857 494.6986607142857L625 364.3973214285715L948.4375 687.8348214285714L1036.8861607142858 599.3861607142858L625 187.5z" />
    <glyph glyph-name="achievements"
      unicode="&#xF103;"
      horiz-adv-x="730.9709821428572" d="M647.1540178571429 50.4464285714286L536.1049107142858 75.2790178571429L457.3660714285714 0.78125L390.4575892857144 256.6964285714286C390.4575892857144 256.6964285714286 454.1852678571429 259.9888392857144 488.8392857142857 268.9174107142858C528.4040178571429 279.1294642857144 580.3013392857143 306.3616071428571 580.3013392857143 306.3616071428571L647.1540178571429 50.4464285714286zM349.6651785714286 255.0781249999999C349.6651785714286 255.0781249999999 291.1272321428572 257.4776785714285 251.7299107142857 268.4151785714286C217.2991071428572 278.0133928571429 160.4352678571429 307.03125 160.4352678571429 307.03125L90.4575892857143 51.8973214285714L195.9263392857143 77.0089285714286L279.6875 0L349.6651785714286 255.0781249999999zM713.9508928571429 698.3816964285714L696.1495535714286 723.1584821428571C684.9888392857143 738.671875 679.0736607142858 757.3660714285714 679.1294642857143 776.4508928571429L679.296875 806.9754464285714C679.5200892857143 846.2053571428571 654.4642857142857 881.0825892857142 617.2433035714286 893.4151785714286L588.2812499999999 903.0133928571428C570.1450892857142 909.0401785714286 554.3526785714286 920.6473214285714 543.1919642857142 936.1607142857142L525.4464285714286 960.9375C502.5669642857143 992.8013392857144 461.7745535714286 1006.3058035714286 424.3861607142857 994.3638392857144L395.3125 985.1004464285714C377.1205357142857 979.296875 357.5334821428571 979.4084821428572 339.3415178571429 985.4352678571428L310.4352678571429 994.9776785714286C273.2142857142857 1007.3102678571428 232.2544642857143 994.2522321428572 209.0401785714286 962.6674107142856L190.9598214285714 938.1138392857142C179.6316964285714 922.7120535714286 163.7276785714286 911.2723214285714 145.5357142857143 905.46875L116.4620535714286 896.2053571428571C79.0736607142857 884.3191964285714 53.6272321428571 849.6651785714286 53.4040178571429 810.4352678571429L53.2366071428571 779.9107142857143C53.125 760.7700892857142 46.9866071428571 742.1875 35.6584821428571 726.7857142857142L17.578125 702.2321428571429C-5.6361607142857 670.6473214285714 -5.9151785714286 627.6785714285714 16.9642857142857 595.8147321428571L34.765625 571.0379464285714C45.9263392857143 555.5245535714287 51.8415178571429 536.8303571428571 51.7857142857143 517.7455357142858L51.6183035714286 487.2209821428571C51.3950892857143 447.9910714285715 76.4508928571429 413.1138392857142 113.671875 400.78125L142.6339285714286 391.1830357142857C160.7700892857143 385.15625 176.5625 373.5491071428572 187.7232142857143 358.0357142857142L205.5245535714286 333.2589285714285C228.4040178571429 301.3950892857142 269.1964285714286 287.8906249999999 306.5848214285715 299.8325892857141L335.6584821428572 309.0959821428569C353.8504464285715 314.8995535714285 373.4375000000001 314.7879464285712 391.6294642857143 308.7611607142856L420.5915178571429 299.1629464285712C457.8125 286.830357142857 498.7723214285714 299.8883928571427 521.9866071428571 331.4732142857141L540.0669642857142 356.0267857142856C551.3950892857142 371.4285714285712 567.299107142857 382.8683035714284 585.4910714285713 388.6718749999999L614.564732142857 397.9352678571427C651.9531249999999 409.8214285714284 677.3995535714284 444.4754464285712 677.6227678571428 483.705357142857L677.7901785714284 514.2299107142856C677.9017857142856 533.3705357142856 684.0401785714284 551.9531249999999 695.3683035714284 567.3549107142856L713.4486607142856 591.9084821428571C736.6071428571429 623.5491071428571 736.8303571428572 666.5178571428571 713.9508928571429 698.3816964285714zM632.1428571428571 639.453125C634.3191964285714 787.109375 516.40625 908.59375 368.75 910.7700892857142C221.09375 912.9464285714286 99.609375 795.0334821428571 97.4330357142857 647.3772321428571S213.1696428571429 378.2366071428571 360.8258928571429 376.0602678571428C508.4821428571429 373.828125 629.9665178571429 491.7968749999999 632.1428571428571 639.453125zM368.0245535714286 862.6116071428571C246.9866071428572 864.3973214285714 147.3772321428572 767.6897321428571 145.5915178571429 646.6517857142857S240.5133928571429 426.0044642857144 361.5513392857143 424.2187499999999C482.5892857142857 422.4330357142856 582.1986607142857 519.1406249999999 583.984375 640.1785714285713S489.1183035714286 860.8258928571429 368.0245535714286 862.6116071428571z" />
    <glyph glyph-name="activity"
      unicode="&#xF104;"
      horiz-adv-x="1333.2589285714284" d="M333.3147321428571 416.6852678571429H1333.3147321428573V583.3147321428571H333.3147321428573V416.6852678571429zM0 833.3147321428571H166.6852678571429V1000H0V833.3147321428571zM0 0H166.6852678571429V166.6852678571429H0V0zM0 416.6852678571429H166.6852678571429V583.3147321428571H0V416.6852678571429zM333.3147321428571 1000V833.3147321428571H1333.3147321428573V1000H333.3147321428571zM333.3147321428571 0H1333.3147321428573V166.6852678571429H333.3147321428573V0z" />
    <glyph glyph-name="api"
      unicode="&#xF105;"
      horiz-adv-x="1111.1049107142858" d="M1111.1049107142858 666.6852678571429V777.7901785714287H1000V888.8950892857143C1000 950 950.0000000000002 1000 888.8950892857143 1000H111.1049107142857C50 1000 0 950 0 888.8950892857142V111.1049107142857C0 49.9999999999999 50 0 111.1049107142857 0H888.8950892857142C950 0 999.9999999999998 49.9999999999999 999.9999999999998 111.1049107142857V222.2098214285715H1111.1049107142856V333.3147321428571H1000V444.4196428571428H1111.1049107142858V555.5803571428571H1000V666.6852678571429H1111.1049107142858zM888.8950892857143 111.1049107142857H111.1049107142857V888.8950892857142H888.8950892857143V111.1049107142857zM222.2098214285715 444.4196428571428H500V222.2098214285715H222.2098214285714V444.4196428571428zM555.5803571428571 777.7901785714286H777.7901785714286V611.1049107142858H555.5803571428571V777.7901785714286zM222.2098214285715 777.7901785714286H500V500H222.2098214285714V777.7901785714286zM555.5803571428571 555.5803571428571H777.7901785714286V222.2656250000001H555.5803571428571V555.5803571428571z" />
    <glyph glyph-name="arrow_back-new"
      unicode="&#xF106;"
      horiz-adv-x="1100" d="M13.7 465.9L465.8 13.7A47.8 47.8 0 0 1 532.8 13.7L643.8 124.7000000000001A47.8 47.8 0 0 1 643.8 191.6999999999999L464.4999999999999 373.3H1052.3999999999999C1079.2 373.3 1099.8999999999999 395.3 1099.8999999999999 420.8V579.2A47.6 47.6 0 0 1 1052.3999999999999 626.8H464.3999999999999L646 808.3A47.8 47.8 0 0 1 646 875.3L535.2 986.4A47.8 47.8 0 0 1 468.2 986.4L13.7 534.2A49.3 49.3 0 0 1 13.7 465.9z" />
    <glyph glyph-name="arrow_down"
      unicode="&#xF107;"
      horiz-adv-x="1792.5781250000002" d="M0 889.5089285714286C0 861.8861607142858 10.6026785714286 834.2633928571429 31.8080357142857 813.28125L750.7254464285714 94.1406249999999C872.0424107142858 -27.1763392857144 914.3415178571428 -33.3147321428573 1035.6026785714287 87.9464285714284L1760.2678571428573 812.6674107142858C1803.1250000000005 855.3013392857143 1803.4040178571431 924.7209821428572 1760.825892857143 967.6339285714286C1718.247767857143 1010.6026785714286 1648.9955357142858 1010.8258928571428 1606.0825892857142 968.1919642857144C1606.0825892857142 968.1919642857144 1133.091517857143 494.7544642857143 950.7254464285714 312.3883928571428C895.703125 257.3660714285715 890.6808035714286 259.375 842.96875 307.0870535714286C666.1830357142857 483.8727678571429 183.59375 965.7366071428572 183.59375 965.7366071428572C141.40625 1007.5334821428572 73.2142857142857 1007.3102678571428 31.3058035714286 965.234375C10.4352678571429 944.2522321428572 0 916.8526785714286 0 889.5089285714286z" />
    <glyph glyph-name="arrow_left"
      unicode="&#xF108;"
      horiz-adv-x="557.8683035714287" d="M538.4486607142858 17.4665178571428C561.9419642857143 40.8482142857143 562.0535714285714 78.9062499999999 538.7276785714287 102.4553571428571C538.7276785714287 102.4553571428571 269.9218750000001 371.6517857142857 171.3169642857144 470.3125C144.6986607142858 496.9308035714286 143.5825892857144 499.7209821428571 174.2745535714287 530.4129464285713C276.0044642857144 632.1428571428571 540.1227678571429 895.9821428571429 540.1227678571429 895.9821428571429C563.8950892857143 919.921875 563.7834821428572 958.5379464285714 539.8437500000001 982.3102678571428C515.9040178571429 1006.0825892857144 477.1763392857144 1005.9151785714286 453.4040178571429 982.03125L49.0513392857143 577.734375C-18.5825892857143 510.1004464285714 -15.1785714285714 486.4955357142857 52.5111607142857 418.8058035714286L453.6830357142857 17.7455357142857C465.4017857142857 5.9151785714284 480.8035714285714 0 496.2053571428572 0C511.4955357142857 0 526.7857142857143 5.8035714285714 538.4486607142858 17.4665178571428z" />
    <glyph glyph-name="arrow_right"
      unicode="&#xF109;"
      horiz-adv-x="557.8683035714287" d="M61.6629464285714 0C77.0647321428571 0 92.4665178571429 5.9151785714284 104.1852678571429 17.7455357142857L505.3571428571428 418.8058035714286C573.0468749999999 486.4955357142857 576.4508928571429 510.1004464285714 508.8169642857143 577.734375L104.5200892857143 981.9754464285714C80.7477678571429 1005.9151785714286 42.0200892857143 1006.0267857142856 18.0803571428571 982.2544642857142C-5.859375 958.4821428571428 -6.0267857142857 919.8660714285714 17.8013392857143 895.9263392857142C17.8013392857143 895.9263392857142 281.9196428571429 632.03125 383.6495535714286 530.3571428571429C414.3415178571429 499.6651785714286 413.2254464285715 496.875 386.6071428571429 470.2566964285713C287.9464285714286 371.6517857142857 19.140625 102.3995535714286 19.140625 102.3995535714286C-4.1852678571428 78.8504464285713 -4.0736607142857 40.7924107142857 19.4196428571429 17.4107142857143C31.1383928571429 5.8035714285714 46.3727678571429 0 61.6629464285714 0z" />
    <glyph glyph-name="arrow_up"
      unicode="&#xF10A;"
      horiz-adv-x="1792.5781250000002" d="M31.3058035714286 34.8214285714286C73.2142857142857 -7.2544642857143 141.40625 -7.5334821428572 183.59375 34.3191964285714C183.59375 34.3191964285714 666.1830357142857 516.1830357142858 842.96875 692.96875C890.6808035714286 740.6808035714286 895.703125 742.6897321428571 950.7254464285714 687.6674107142857C1133.091517857143 505.3013392857143 1606.0825892857142 31.8080357142857 1606.0825892857142 31.8080357142857C1648.9397321428573 -10.8258928571429 1718.247767857143 -10.6026785714286 1760.825892857143 32.3660714285714C1803.4040178571431 75.3348214285715 1803.180803571429 144.6986607142857 1760.2678571428573 187.3325892857143L1035.6026785714287 912.109375C914.3973214285716 1033.314732142857 872.0424107142859 1027.232142857143 750.7254464285716 905.9151785714286L31.8080357142857 186.7187499999999C10.6026785714286 165.7366071428571 0 138.1138392857142 0 110.4910714285715C0 83.1473214285715 10.4352678571429 55.7477678571429 31.3058035714286 34.8214285714286z" />
    <glyph glyph-name="arrow_up_down"
      unicode="&#xF10B;"
      horiz-adv-x="714.2857142857143" d="M357.1428571428572 1000L714.2857142857143 736.9128571428571L714.2857142857143 615.1357142857144L357.1428571428572 878.2228571428572L0 615.1357142857144L0 736.9128571428571L357.1428571428572 1000zM714.2857142857143 384.8642857142857L714.2857142857143 263.0857142857142L357.1428571428572 0L0 263.0857142857142L0 384.8642857142857L357.1428571428572 121.7785714285715L714.2857142857143 384.8642857142857z" />
    <glyph glyph-name="audio"
      unicode="&#xF10C;"
      horiz-adv-x="1120.5357142857142" d="M0 611.4955357142857C0 662.9464285714286 39.453125 702.9575892857142 91.4620535714286 702.9575892857142C140.1785714285714 702.9575892857142 257.2544642857143 702.9575892857142 257.2544642857143 702.9575892857142L617.1875 1000V0L251.5625 297.0424107142857C251.5625 297.0424107142857 122.7678571428571 297.3214285714286 85.7700892857143 297.0424107142857C40.0111607142857 296.7633928571428 0 340.8482142857142 0 382.8125S0 588.6160714285713 0 611.4955357142857zM754.4084821428572 651.5066964285714C754.4084821428572 651.5066964285714 784.654017857143 682.2544642857142 817.2991071428572 714.3973214285713C941.4620535714286 626.4508928571429 962.6674107142858 399.9441964285715 817.2991071428572 291.5736607142857C776.4508928571429 328.0691964285715 747.3214285714287 346.5959821428571 754.4084821428572 354.4642857142857C829.8549107142857 435.3236607142857 852.734375 553.2366071428571 754.4084821428572 651.5066964285714zM885.6584821428571 794.1964285714286C885.6584821428571 794.1964285714286 914.7879464285714 827.6785714285714 965.6808035714286 874.21875C1176.953125 672.4888392857142 1169.029017857143 331.3616071428571 959.9888392857144 136.9419642857142C932.1986607142858 164.174107142857 896.8191964285714 200.9486607142857 885.6584821428571 211.2723214285713C1048.2142857142858 353.6830357142857 1071.0379464285716 638.4486607142858 885.6584821428571 794.1964285714286z" />
    <glyph glyph-name="audio_mute"
      unicode="&#xF10D;"
      horiz-adv-x="1120.5357142857142" d="M1061.5513392857142 303.8504464285715L924.3303571428572 441.0714285714286L787.1093749999999 303.8504464285715L728.1808035714286 362.7790178571428L865.4017857142857 500L728.1808035714286 637.2209821428571L787.1093749999999 696.1495535714286L924.3303571428572 558.9285714285714L1061.5513392857142 696.1495535714286L1120.4799107142858 637.2209821428571L983.2589285714286 500L1120.4799107142858 362.7790178571428L1061.5513392857142 303.8504464285715zM0 611.4955357142857C0 662.9464285714286 39.453125 702.9575892857142 91.4620535714286 702.9575892857142C140.1785714285714 702.9575892857142 257.2544642857143 702.9575892857142 257.2544642857143 702.9575892857142L617.1875 1000V0L251.5625 297.0424107142857C251.5625 297.0424107142857 122.7678571428571 297.3214285714286 85.7700892857143 297.0424107142857C40.0111607142857 296.7633928571428 0 340.8482142857142 0 382.8125S0 588.6160714285713 0 611.4955357142857z" />
    <glyph glyph-name="back_arrow"
      unicode="&#xF10E;"
      horiz-adv-x="1499.9441964285716" d="M-194.7544642857143 611.6071428571429M1499.9441964285716 391.1272321428572H471.9866071428571V0L0 500L471.9866071428571 1000.0558035714286V608.984375H1499.9441964285713V391.1272321428572z" />
    <glyph glyph-name="book"
      unicode="&#xF10F;"
      horiz-adv-x="1556.0267857142858" d="M1462.6674107142858 1000H93.359375C41.5178571428571 1000 0 957.1428571428572 0 907.1428571428572V92.8571428571428C0 42.8571428571428 41.5178571428571 -1e-13 93.359375 -1e-13H1462.667410714286C1514.508928571429 -1e-13 1556.026785714286 42.8571428571427 1556.026785714286 89.2857142857141V907.1428571428572C1556.0267857142858 957.1428571428572 1514.5089285714287 1000 1462.6674107142858 1000zM1286.216517857143 165.9598214285715H847.0982142857143C833.2589285714287 141.7410714285715 805.5803571428571 127.9017857142857 777.9575892857142 127.9017857142857S722.65625 141.7410714285715 708.8169642857142 165.9598214285715H269.6986607142857C245.4799107142858 165.9598214285715 224.7209821428572 183.2589285714286 224.7209821428572 207.4776785714287V784.9330357142858C224.7209821428572 812.6116071428571 248.9397321428572 826.4508928571429 269.6986607142858 826.4508928571429H311.2165178571429V833.3705357142857C311.2165178571429 840.2901785714286 314.6763392857144 843.75 321.5959821428572 843.75H722.7120535714286C740.0111607142857 843.75 759.6540178571429 838.2254464285714 764.2299107142858 822.9910714285714V280.1897321428571C750.390625 269.8102678571429 726.171875 259.375 722.7120535714286 259.375H321.5959821428571H318.1361607142857H314.6763392857142C311.2165178571428 259.375 311.2165178571428 262.8348214285715 311.2165178571428 266.2946428571429V805.6919642857142H269.6986607142856C255.8593749999999 805.6919642857142 245.4799107142857 795.3125 245.4799107142857 784.9330357142858V207.4776785714287C245.4799107142857 193.638392857143 255.859375 186.71875 269.6986607142856 186.71875H719.2522321428571C726.1160714285713 165.9598214285715 750.3348214285714 148.6607142857143 774.5535714285713 148.6607142857143C798.7165178571428 148.6607142857143 822.9352678571428 165.9598214285715 829.8549107142856 186.71875H1282.8124999999998C1296.6517857142858 186.71875 1307.03125 197.0982142857143 1307.03125 207.4776785714287V784.9330357142858C1307.03125 798.7723214285714 1296.6517857142858 805.6919642857142 1282.8124999999998 805.6919642857142H1241.2946428571427V266.2946428571429C1241.2946428571427 259.375 1237.8348214285713 255.9151785714287 1230.9151785714284 255.9151785714287H829.7991071428571C826.3392857142857 255.9151785714287 802.1205357142858 269.7544642857144 788.28125 280.1339285714287V822.9910714285714C793.8616071428571 837.109375 812.5 843.75 829.7991071428571 843.75H1230.9151785714287C1237.8348214285716 843.75 1241.294642857143 840.2901785714286 1241.294642857143 833.3705357142857V826.4508928571429H1282.8125000000002C1307.0312500000002 826.4508928571429 1327.7901785714287 809.1517857142858 1327.7901785714287 784.9330357142858L1331.1941964285716 207.4776785714287C1331.1941964285716 179.7991071428571 1306.9754464285713 165.9598214285715 1286.216517857143 165.9598214285715zM1181.919642857143 504.9107142857142V470.3125C1181.919642857143 459.9330357142857 1175 453.0133928571428 1164.6205357142858 453.0133928571428H863.7834821428572C853.404017857143 453.0133928571428 846.4843750000001 459.9330357142857 846.4843750000001 470.3125V504.9107142857142C846.4843750000001 514.6205357142857 853.0691964285714 522.2098214285713 863.7834821428572 522.2098214285713H1164.6205357142858C1175 522.2098214285713 1181.919642857143 515.2901785714286 1181.919642857143 504.9107142857142zM1181.919642857143 771.1495535714286V736.5513392857142C1181.919642857143 726.171875 1175 719.2522321428571 1164.6205357142858 719.2522321428571H863.7834821428572C853.404017857143 719.2522321428571 846.4843750000001 726.171875 846.4843750000001 736.5513392857142V771.1495535714286C846.4843750000001 780.859375 853.0691964285714 788.4486607142857 863.7834821428572 788.4486607142857H1164.6205357142858C1175 788.4486607142857 1181.919642857143 781.5290178571429 1181.919642857143 771.1495535714286zM1181.919642857143 370.0334821428571V335.4352678571428C1181.919642857143 325.0558035714286 1175 318.1361607142857 1164.6205357142858 318.1361607142857H863.7834821428572C853.404017857143 318.1361607142857 846.4843750000001 325.0558035714286 846.4843750000001 335.4352678571428V370.0334821428571C846.4843750000001 379.7433035714286 853.0691964285714 387.3325892857142 863.7834821428572 387.3325892857142H1164.6205357142858C1175 387.3325892857142 1181.919642857143 380.4129464285715 1181.919642857143 370.0334821428571zM1181.919642857143 636.2165178571429V601.6183035714286C1181.919642857143 591.2388392857142 1175 584.3191964285714 1164.6205357142858 584.3191964285714H863.7834821428572C853.404017857143 584.3191964285714 846.4843750000001 591.2388392857142 846.4843750000001 601.6183035714286V636.2165178571429C846.4843750000001 645.9263392857142 853.0691964285714 653.515625 863.7834821428572 653.515625H1164.6205357142858C1175 653.515625 1181.919642857143 646.5959821428571 1181.919642857143 636.2165178571429zM373.4375000000001 736.5513392857142V771.1495535714286C373.4375000000001 780.859375 380.0223214285715 788.4486607142857 390.7366071428572 788.4486607142857H691.5736607142858C701.9531250000001 788.4486607142857 708.872767857143 781.5290178571429 708.872767857143 771.1495535714286V736.5513392857142C708.872767857143 726.171875 701.9531250000001 719.2522321428571 691.5736607142858 719.2522321428571H390.7366071428572C380.3571428571429 719.2522321428571 373.4375000000001 726.171875 373.4375000000001 736.5513392857142zM373.4375000000001 335.4352678571428V370.0334821428571C373.4375000000001 379.7433035714286 380.0223214285715 387.3325892857142 390.7366071428572 387.3325892857142H691.5736607142858C701.9531250000001 387.3325892857142 708.872767857143 380.4129464285714 708.872767857143 370.0334821428571V335.4352678571428C708.872767857143 325.0558035714286 701.9531250000001 318.1361607142857 691.5736607142858 318.1361607142857H390.7366071428572C380.3571428571429 318.1361607142857 373.4375000000001 325.0558035714286 373.4375000000001 335.4352678571428zM373.4375000000001 470.3125V504.9107142857142C373.4375000000001 514.6205357142857 380.0223214285715 522.2098214285713 390.7366071428572 522.2098214285713H691.5736607142858C701.9531250000001 522.2098214285713 708.872767857143 515.2901785714286 708.872767857143 504.9107142857142V470.3125C708.872767857143 459.9330357142857 701.9531250000001 453.0133928571428 691.5736607142858 453.0133928571428H390.7366071428572C380.3571428571429 453.0133928571428 373.4375000000001 459.9330357142858 373.4375000000001 470.3125zM373.4375000000001 601.6183035714286V636.2165178571429C373.4375000000001 645.9263392857142 380.0223214285715 653.515625 390.7366071428572 653.515625H691.5736607142858C701.9531250000001 653.515625 708.872767857143 646.5959821428571 708.872767857143 636.2165178571429V601.6183035714286C708.872767857143 591.2388392857142 701.9531250000001 584.3191964285714 691.5736607142858 584.3191964285714H390.7366071428572C380.3571428571429 584.3191964285714 373.4375000000001 591.2388392857142 373.4375000000001 601.6183035714286z" />
    <glyph glyph-name="burger_menu_icon"
      unicode="&#xF110;"
      horiz-adv-x="1000" d="M0 1000H1000V916.6666666666666H0V1000zM0 541.6666666666667H1000V458.3333333333334H0V541.6666666666667zM1000 83.3333333333334H0V0H1000V83.3333333333334z" />
    <glyph glyph-name="calendar"
      unicode="&#xF111;"
      horiz-adv-x="965.5133928571429" d="M241.4062500000001 413.7834821428571H301.7857142857143C335.15625 413.7834821428571 362.109375 444.6428571428571 362.109375 482.7566964285714S335.1004464285715 551.7299107142858 301.7857142857143 551.7299107142858H241.4062500000001C208.0357142857143 551.7299107142858 181.0825892857143 520.8705357142858 181.0825892857143 482.7566964285714C181.0267857142857 444.6428571428571 208.0357142857143 413.7834821428571 241.4062500000001 413.7834821428571zM663.7834821428571 413.7834821428571H724.1629464285714C757.5334821428571 413.7834821428571 784.4866071428571 444.6428571428571 784.4866071428571 482.7566964285714S757.4776785714286 551.7299107142858 724.1629464285714 551.7299107142858H663.7834821428571C630.4129464285714 551.7299107142858 603.4598214285714 520.8705357142858 603.4598214285714 482.7566964285714S630.4129464285714 413.7834821428571 663.7834821428571 413.7834821428571zM663.7834821428571 206.9196428571428H724.1629464285714C757.5334821428571 206.9196428571428 784.4866071428571 237.7790178571428 784.4866071428571 275.892857142857C784.4866071428571 314.0066964285714 757.4776785714286 344.8660714285714 724.1629464285714 344.8660714285714H663.7834821428571C630.4129464285714 344.8660714285714 603.4598214285714 314.0066964285714 603.4598214285714 275.892857142857C603.4598214285714 237.7232142857142 630.4129464285714 206.9196428571428 663.7834821428571 206.9196428571428zM241.4062500000001 206.9196428571428H301.7857142857143C335.15625 206.9196428571428 362.109375 237.7790178571428 362.109375 275.892857142857C362.109375 314.0066964285714 335.1004464285715 344.8660714285714 301.7857142857143 344.8660714285714H241.4062500000001C208.0357142857143 344.8660714285714 181.0825892857143 314.0066964285714 181.0825892857143 275.892857142857C181.0267857142857 237.7232142857142 208.0357142857143 206.9196428571428 241.4062500000001 206.9196428571428zM854.0178571428572 896.5401785714286H784.4866071428571V862.0535714285714C784.4866071428571 804.9665178571429 743.9174107142857 758.59375 693.9732142857142 758.59375C643.9732142857143 758.59375 603.4598214285713 804.9665178571429 603.4598214285713 862.0535714285714V896.5401785714286H362.0535714285714V862.0535714285713C362.0535714285714 804.9665178571429 321.484375 758.59375 271.5401785714286 758.59375C221.5401785714286 758.59375 181.0267857142858 804.9665178571429 181.0267857142858 862.0535714285714V896.5401785714286H111.4955357142857C46.3727678571429 896.5401785714286 0 839.84375 0 758.6495535714286V69.0290178571429C0 30.9151785714286 27.0089285714286 0.0558035714287 60.3236607142857 0.0558035714287H905.1339285714286C938.5044642857144 0.0558035714287 965.4575892857142 30.9151785714286 965.4575892857142 69.0290178571429V758.6495535714286C965.5133928571428 835.546875 911.6071428571428 896.5401785714286 854.0178571428572 896.5401785714286zM844.8102678571429 137.9464285714286H120.703125V620.703125H844.8660714285714L844.8102678571429 137.9464285714286L844.8102678571429 137.9464285714286zM446.9308035714286 413.7834821428571H507.2544642857143C540.625 413.7834821428571 567.578125 444.6428571428571 567.578125 482.7566964285714S540.625 551.7299107142858 507.2544642857143 551.7299107142858H446.9308035714286C413.5602678571429 551.7299107142858 386.5513392857143 520.8705357142858 386.5513392857143 482.7566964285714C386.6071428571429 444.6428571428571 413.5602678571429 413.7834821428571 446.9308035714286 413.7834821428571zM446.9308035714286 206.9196428571428H507.2544642857143C540.625 206.9196428571428 567.578125 237.7790178571428 567.578125 275.892857142857C567.578125 314.0066964285714 540.625 344.8660714285714 507.2544642857143 344.8660714285714H446.9308035714286C413.5602678571429 344.8660714285714 386.5513392857143 314.0066964285714 386.5513392857143 275.892857142857C386.6071428571429 237.7232142857142 413.5602678571429 206.9196428571428 446.9308035714286 206.9196428571428zM271.5401785714286 793.0803571428571C304.9107142857143 793.0803571428571 331.8638392857144 823.9397321428571 331.8638392857144 862.0535714285714V931.0267857142856C331.9196428571429 969.140625 304.9107142857143 1000 271.5401785714286 1000S211.2165178571429 969.140625 211.2165178571429 931.0267857142856V862.0535714285714C211.2165178571429 823.9397321428571 238.2254464285714 793.0803571428571 271.5401785714286 793.0803571428571zM693.9732142857142 793.0803571428571C727.34375 793.0803571428571 754.2968749999999 823.9397321428571 754.2968749999999 862.0535714285714V931.0267857142856C754.2968749999999 969.140625 727.2879464285713 1000 693.9732142857142 1000C660.6026785714286 1000 633.6495535714286 969.140625 633.6495535714286 931.0267857142856V862.0535714285714C633.6495535714286 823.9397321428571 660.6026785714286 793.0803571428571 693.9732142857142 793.0803571428571z" />
    <glyph glyph-name="cancel"
      unicode="&#xF112;"
      horiz-adv-x="1000" d="M1000 952.1763392857142L952.1763392857144 1000L500 547.8794642857142L47.8794642857143 1000L0 952.1763392857142L452.1205357142858 500L0 47.8794642857143L47.8794642857143 0L500 452.1763392857142L952.1763392857144 0L1000 47.8794642857143L547.8794642857144 500z" />
    <glyph glyph-name="checkbox"
      unicode="&#xF113;"
      horiz-adv-x="1001.5066964285714" d="M801.171875 861.9977678571429C823.1026785714286 861.9977678571429 837.5 850.8370535714286 844.140625 844.1964285714286C850.7812500000001 837.5558035714286 861.9419642857143 823.1026785714286 861.9419642857143 801.2276785714286V200.3348214285715C861.9419642857143 178.4040178571429 850.7812500000001 164.0066964285715 844.140625 157.3660714285715C837.5 150.7254464285713 823.0468750000001 139.5647321428571 801.171875 139.5647321428571H200.2790178571429C178.3482142857143 139.5647321428571 163.9508928571428 150.7254464285713 157.3102678571429 157.3660714285715C150.6696428571429 164.0066964285715 139.5089285714286 178.4598214285713 139.5089285714286 200.3348214285715V801.2276785714286C139.5089285714286 823.1026785714286 150.6696428571429 837.5558035714286 157.3102678571429 844.1964285714286C163.9508928571428 850.8370535714286 178.4040178571429 861.9977678571429 200.2790178571429 861.9977678571429H801.171875M801.171875 1001.5066964285714H200.2790178571429C89.6763392857143 1001.5066964285714 0 911.8303571428572 0 801.2276785714286V200.3348214285715C0 89.7321428571428 89.6763392857143 0.0558035714284 200.2790178571429 0.0558035714284H801.1718749999999C911.7745535714286 0.0558035714284 1001.4508928571428 89.7321428571428 1001.4508928571428 200.3348214285715V801.2276785714286C1001.5066964285714 911.8303571428572 911.8303571428572 1001.5066964285714 801.171875 1001.5066964285714L801.171875 1001.5066964285714z" />
    <glyph glyph-name="checkbox_marked"
      unicode="&#xF114;"
      horiz-adv-x="1001.5066964285714" d="M801.171875 861.9977678571429C823.1026785714286 861.9977678571429 837.5 850.8370535714286 844.140625 844.1964285714286S861.9419642857143 823.1026785714286 861.9419642857143 801.2276785714286V200.3348214285715C861.9419642857143 178.4040178571429 850.7812500000001 164.0066964285715 844.140625 157.3660714285715S823.0468750000001 139.5647321428571 801.171875 139.5647321428571H200.2790178571429C178.3482142857143 139.5647321428571 163.9508928571428 150.7254464285713 157.3102678571429 157.3660714285715S139.5089285714286 178.4598214285713 139.5089285714286 200.3348214285715V801.2276785714286C139.5089285714286 823.1026785714286 150.6696428571429 837.5558035714286 157.3102678571429 844.1964285714286S178.4040178571429 861.9977678571429 200.2790178571429 861.9977678571429H801.171875M801.171875 1001.5066964285714H200.2790178571429C89.6763392857143 1001.5066964285714 0 911.8303571428572 0 801.2276785714286V200.3348214285715C0 89.7321428571428 89.6763392857143 0.0558035714284 200.2790178571429 0.0558035714284H801.1718749999999C911.7745535714286 0.0558035714284 1001.4508928571428 89.7321428571428 1001.4508928571428 200.3348214285715V801.2276785714286C1001.5066964285714 911.8303571428572 911.8303571428572 1001.5066964285714 801.171875 1001.5066964285714L801.171875 1001.5066964285714zM845.4241071428571 865.5133928571429C849.9441964285714 870.7589285714286 858.9285714285714 875.6696428571429 867.0200892857143 867.4665178571429C876.3392857142858 858.0915178571429 990.9040178571428 746.1495535714286 996.7075892857144 738.7834821428571C1003.2366071428572 730.46875 1002.9575892857144 725.9486607142858 996.9866071428572 719.4196428571429C990.513392857143 712.3883928571429 540.513392857143 269.921875 528.0691964285716 256.5290178571429C517.0200892857144 244.6428571428572 512.0535714285716 243.9732142857144 500.5580357142858 254.5758928571429C487.2209821428573 266.8526785714286 181.138392857143 561.9419642857142 174.4419642857144 568.4151785714286C167.7455357142859 574.8883928571429 165.513392857143 580.9709821428571 172.4888392857144 587.890625C179.4642857142859 594.8660714285714 295.591517857143 712.0535714285714 300.1674107142859 716.5736607142858C305.6919642857145 722.0424107142858 314.0066964285716 726.0602678571429 321.7633928571431 718.359375C336.2723214285716 704.0178571428571 514.2857142857144 537.2209821428571 514.2857142857144 537.2209821428571C514.2857142857142 537.2767857142858 842.5223214285714 862.1651785714286 845.4241071428571 865.5133928571429z" />
    <glyph glyph-name="checkmark"
      unicode="&#xF115;"
      horiz-adv-x="1333.8169642857142" d="M1083.9285714285713 990.0111607142858C1091.127232142857 998.3816964285714 1105.5245535714284 1006.25 1118.5267857142858 993.1361607142858C1133.4263392857142 978.0691964285714 1316.796875 798.9955357142858 1326.0602678571427 787.2209821428571C1336.4955357142856 773.8839285714287 1336.049107142857 766.6294642857143 1326.5066964285713 756.25C1316.1830357142858 744.9776785714286 595.8705357142857 36.7745535714286 575.9486607142857 15.4017857142858C558.3147321428571 -3.5714285714286 550.3348214285714 -4.6875 531.9196428571429 12.2767857142858C510.546875 31.9754464285716 20.703125 504.2410714285716 9.9330357142857 514.6205357142858C-0.78125 525.0000000000002 -4.3526785714286 534.654017857143 6.8080357142857 545.8147321428573C17.9129464285714 556.9754464285716 203.8504464285714 744.4754464285716 211.2165178571428 751.7299107142858C220.0892857142857 760.435267857143 233.3147321428572 766.9642857142859 245.8147321428571 754.575892857143C269.0290178571429 731.5848214285716 553.9620535714287 464.6763392857144 553.9620535714287 464.6763392857144C553.90625 464.6205357142857 1079.296875 984.5982142857144 1083.9285714285713 990.0111607142858z" />
    <glyph glyph-name="cite"
      unicode="&#xF116;"
      horiz-adv-x="467.0758928571429" d="M467.0758928571429 889.5089285714286L400.1116071428572 889.5089285714286V246.09375L200.0558035714286 0L0 246.09375V1000H400.6696428571429L467.0758928571429 889.5089285714286z" />
    <glyph glyph-name="code"
      unicode="&#xF117;"
      horiz-adv-x="1571.4285714285716" d="M1476.6183035714287 1000.0558035714286H94.8660714285714C42.6897321428571 1000.0558035714286 0 957.3660714285714 0 905.1897321428572V94.8660714285715C0 42.6897321428571 42.6897321428571 0 94.8660714285714 0H1476.6183035714287C1528.794642857143 0 1571.484375 42.6897321428571 1571.484375 92.0758928571428V902.3995535714286C1571.484375 957.3660714285714 1528.794642857143 1000.0558035714286 1476.6183035714287 1000.0558035714286zM639.2857142857142 350.0558035714286C674.8325892857143 314.5089285714286 674.8325892857143 256.9196428571428 639.2857142857142 221.4285714285713C603.7388392857142 185.8816964285713 546.09375 185.8816964285713 510.5468749999999 221.4285714285713L296.7075892857144 435.1004464285715C296.4843750000001 435.3236607142857 296.1495535714286 435.3794642857144 295.9821428571429 435.546875C278.1250000000001 453.4040178571429 269.3080357142858 476.6741071428571 269.3080357142858 500C269.3080357142858 523.3258928571429 278.1250000000001 546.5959821428571 295.9821428571429 564.453125C296.1495535714286 564.6205357142857 296.4285714285715 564.6763392857142 296.7075892857144 564.9553571428571L510.546875 778.6272321428571C546.09375 814.1741071428571 603.7388392857143 814.1741071428571 639.2857142857142 778.6272321428571C674.8325892857143 743.0803571428571 674.8325892857143 685.546875 639.2857142857142 650L489.2299107142858 500L639.2857142857142 350.0558035714286zM1275.390625 435.546875C1275.2232142857142 435.3794642857144 1274.9441964285713 435.3236607142857 1274.6651785714284 435.1004464285715L1060.8258928571427 221.4285714285716C1025.2790178571427 185.8816964285715 967.6339285714284 185.8816964285715 932.0870535714284 221.4285714285716C896.5401785714284 256.9754464285716 896.5401785714284 314.5089285714287 932.0870535714284 350.0558035714287L1082.1986607142858 500L932.1428571428572 649.9441964285714C896.5959821428572 685.4910714285714 896.5959821428572 743.0245535714287 932.1428571428572 778.5714285714286C967.6897321428572 814.1183035714286 1025.3348214285716 814.1183035714286 1060.8816964285716 778.5714285714286L1274.7209821428573 564.8995535714287C1274.9441964285716 564.6763392857143 1275.2232142857144 564.6205357142858 1275.446428571429 564.3973214285714C1293.3035714285716 546.5959821428571 1302.120535714286 523.2700892857142 1302.120535714286 499.9441964285714C1302.0647321428573 476.6741071428571 1293.247767857143 453.4040178571429 1275.390625 435.546875z" />
    <glyph glyph-name="collaborative-spaces"
      unicode="&#xF118;"
      horiz-adv-x="1428.5714285714287" d="M1035.7142857142858 500C1134.263392857143 500 1213.560267857143 580.0223214285713 1213.560267857143 678.5714285714286S1134.263392857143 857.1428571428571 1035.7142857142858 857.1428571428571S857.1428571428571 777.1205357142858 857.1428571428571 678.5714285714286S937.1651785714286 500 1035.7142857142858 500M500 571.4285714285714C618.5825892857143 571.4285714285714 713.5602678571429 667.1316964285714 713.5602678571429 785.7142857142858S618.5825892857143 1000 500 1000S285.7142857142857 904.296875 285.7142857142857 785.7142857142858S381.4174107142857 571.4285714285714 500 571.4285714285714M1035.7142857142858 357.1428571428571C905.0223214285714 357.1428571428571 642.8571428571429 291.40625 642.8571428571429 160.7142857142857V0H1428.5714285714287V160.7142857142857C1428.5714285714287 291.40625 1166.40625 357.1428571428571 1035.7142857142858 357.1428571428571M500 428.5714285714286C333.59375 428.5714285714286 0 344.9776785714286 0 178.5714285714286V0H500V160.7142857142857C500 221.4285714285713 523.5491071428572 327.8459821428571 669.3080357142858 408.59375C607.1428571428571 421.4285714285715 547.1540178571429 428.5714285714286 500 428.5714285714286" />
    <glyph glyph-name="collections-folder"
      unicode="&#xF119;"
      horiz-adv-x="1200" d="M100 800H0V550H0.5022321428571L0 100C0 44.9776785714287 44.9776785714286 0 100 0H1000V100H100V800zM1100 900H700L600 1000H300C244.9776785714286 1000 200.5022321428572 955.0223214285714 200.5022321428572 900L200 299.9999999999999C200 244.9776785714286 244.9776785714286 199.9999999999999 300 199.9999999999999H1100C1155.0223214285713 199.9999999999999 1199.9999999999998 244.9776785714285 1199.9999999999998 299.9999999999999V800C1200 855.0223214285714 1155.0223214285716 900 1100 900M350.0000000000001 350L575.0000000000001 650L750 424.4977678571429L875 575L1050 350H350.0000000000001z" />
    <glyph glyph-name="collections"
      unicode="&#xF11A;"
      horiz-adv-x="1520.0334821428573" d="M0 1000H1209.9888392857144V308.59375H0V1000zM1371.4285714285713 148.046875H118.1919642857143V234.4866071428571H1284.9888392857144V882.7008928571429H1371.4285714285716V148.046875zM1520.0334821428573 0H266.8526785714286V86.4397321428572H1433.59375V734.6540178571429H1520.0334821428573V0z" />
    <glyph glyph-name="comment"
      unicode="&#xF11B;"
      horiz-adv-x="1145.1450892857142" d="M1073.1026785714287 1000.1116071428572H72.0424107142857C32.2544642857143 1000.1116071428572 0 967.8571428571428 0 928.0691964285714V216.0714285714286C0 176.2834821428572 32.2544642857143 144.0290178571429 72.0424107142857 144.0290178571429H300.3348214285715L270.5357142857143 0L497.1540178571429 144.0848214285715H1073.1026785714287C1112.890625 144.0848214285715 1145.1450892857142 176.3392857142857 1145.1450892857142 216.1272321428571V928.0691964285714C1145.1450892857142 967.8571428571428 1112.890625 1000.1116071428572 1073.1026785714287 1000.1116071428572z" />
    <glyph glyph-name="compact_controls"
      unicode="&#xF11C;"
      horiz-adv-x="1285.9933035714287" d="M419.3080357142858 1000H1285.9933035714287V0H419.3080357142858V1000zM1219.3080357142858 933.3147321428572H485.9933035714286V66.6852678571428H1219.3080357142858V933.3147321428572zM0 567.6897321428571L184.1517857142857 566.6294642857142V735.6584821428571L419.8660714285715 500L184.1517857142857 264.2857142857142V433.3705357142857L0 434.3749999999999V567.6897321428571z" />
    <glyph glyph-name="compress"
      unicode="&#xF11D;"
      horiz-adv-x="1549.3861607142858" d="M11.4955357142857 129.0736607142857C53.7388392857143 205.3013392857143 646.9866071428572 911.6629464285714 700.0558035714286 964.3973214285714C747.5446428571428 1011.6071428571428 801.5066964285714 1012.0535714285714 849.21875 964.3973214285714C922.4330357142858 891.1830357142857 1506.1941964285713 184.2075892857142 1536.216517857143 131.7522321428571C1573.6607142857142 66.2946428571428 1526.0602678571431 0 1462.0535714285716 0C1412.7232142857142 0 156.8080357142857 0 85.2678571428573 0C14.7321428571429 0 -19.53125 73.2142857142857 11.4955357142857 129.0736607142857z" />
    <glyph glyph-name="confidential"
      unicode="&#xF11E;"
      horiz-adv-x="1138.5044642857144" d="M553.90625 707.5334821428571C423.7165178571429 698.7723214285714 325.1674107142857 586.4955357142858 333.984375 455.9709821428572C334.2633928571428 451.6741071428572 335.6026785714285 447.7120535714287 336.1049107142856 443.4151785714287L598.7165178571428 706.0267857142858C584.0959821428572 707.8125 569.1406250000001 708.5379464285714 553.90625 707.5334821428571zM540.7924107142857 237.5558035714285C555.4129464285714 235.7700892857141 570.3683035714286 235.1004464285712 585.6026785714286 236.049107142857C715.7366071428571 244.7544642857141 814.1741071428571 357.4776785714284 805.46875 487.6674107142856C805.1897321428571 491.908482142857 803.90625 495.8705357142856 803.3482142857142 500.0558035714284L540.7924107142857 237.5558035714285zM1081.0825892857144 978.7388392857144C1052.7901785714287 1007.0870535714286 1006.8080357142858 1007.0870535714286 978.4598214285716 978.7388392857144L123.6607142857143 123.9397321428571C95.3125 95.5915178571428 95.3125 49.6093749999999 123.6607142857143 21.2611607142857C152.0089285714286 -7.0312500000001 197.9352678571429 -7.0312500000001 226.2834821428572 21.2611607142857L1081.0825892857144 876.1160714285714C1109.4308035714284 904.4642857142858 1109.4308035714284 950.390625 1081.0825892857144 978.7388392857144zM161.0491071428572 266.40625L293.2477678571429 398.6049107142858C286.9419642857144 422.3772321428572 282.5334821428572 446.7075892857145 282.5892857142857 472.4888392857143C282.9241071428572 631.1941964285716 411.6629464285715 759.3191964285716 570.3683035714286 758.984375C595.5915178571429 758.8727678571429 619.6428571428572 754.5758928571429 643.0245535714286 748.4375000000001L710.546875 815.9598214285716C452.9017857142857 876.7299107142858 167.1316964285715 770.3125000000001 9.0959821428571 495.9263392857143C9.0959821428571 495.9263392857143 0 489.9553571428572 0 476.3392857142858C0 462.7232142857143 9.0959821428571 450.8370535714286 9.0959821428571 450.8370535714286C51.2276785714286 376.8973214285715 103.3482142857143 316.2388392857144 161.0491071428572 266.40625zM1141.685267857143 471.8191964285714C1141.685267857143 483.203125 1120.5357142857142 509.7098214285714 1114.732142857143 519.0290178571429C1076.2276785714287 580.6361607142858 1030.5245535714287 631.9754464285714 980.6361607142858 675.4464285714287L847.154017857143 541.9642857142858C852.9017857142858 519.140625 856.8638392857144 495.7589285714286 856.8080357142858 471.1495535714286C856.4732142857144 312.6674107142858 727.5669642857144 184.3191964285716 569.0290178571429 184.6540178571429C544.8660714285714 184.7098214285715 521.8191964285714 188.6160714285715 499.4977678571429 194.3080357142857L432.8683035714286 127.6785714285713C683.9285714285714 70.2566964285713 960.7700892857144 170.1450892857142 1119.8660714285713 430.1897321428571C1118.5825892857142 428.0691964285714 1141.685267857143 465.1227678571429 1141.685267857143 471.8191964285714z" />
    <glyph glyph-name="curation"
      unicode="&#xF11F;"
      horiz-adv-x="900" d="M800 900H591.015625C569.9776785714287 957.9799107142858 515.0111607142858 1000 450.0000000000001 1000C384.9888392857144 1000 330.0223214285715 957.9799107142856 308.9843750000001 900H100C44.9776785714286 900 0 855.0223214285714 0 800V100C0 44.9776785714287 44.9776785714286 0 100 0H800.0000000000001C855.0223214285714 0 900.0000000000001 44.9776785714284 900.0000000000001 100V800C900.0000000000001 855.0223214285714 855.0223214285714 900 800 900M450.0000000000001 900C477.5111607142858 900 500.0000000000001 877.5111607142857 500.0000000000001 850C500.0000000000001 822.4888392857143 477.5111607142858 800 450.0000000000001 800S400.0000000000001 822.4888392857143 400.0000000000001 850C400 877.5111607142857 422.4888392857142 900 450.0000000000001 900M450.0000000000001 700C532.9799107142858 700 600 632.9799107142857 600 550S532.9799107142858 400 450.0000000000001 400S300.0000000000001 467.0200892857142 300.0000000000001 550S367.0200892857144 700 450.0000000000001 700M750 100H150V169.9776785714285C150 269.9776785714286 350 324.9999999999999 450 324.9999999999999S750 269.9776785714286 750 169.9776785714285V100z" />
    <glyph glyph-name="data-management"
      unicode="&#xF120;"
      horiz-adv-x="1000" d="M100 100H800.0000000000001V0H100C44.9776785714286 0 0 44.9776785714284 0 100V800H100V100zM900.0000000000001 1000C955.0223214285716 1000 1000.0000000000002 955.0223214285714 1000.0000000000002 900V299.9999999999999C1000.0000000000002 244.9776785714286 955.0223214285716 199.9999999999999 900.0000000000001 199.9999999999999H300C244.9776785714286 199.9999999999999 200 244.9776785714285 200 299.9999999999999V900C200 955.0223214285714 244.9776785714286 1000 300 1000H900.0000000000001M900.0000000000001 900H650.0000000000001V500L775.0000000000001 575L900.0000000000001 500V900z" />
    <glyph glyph-name="dataset"
      unicode="&#xF121;"
      horiz-adv-x="1571.4285714285716" d="M1465.1785714285713 1000.0558035714286H106.1383928571429C47.4888392857143 1000.0558035714286 -0.0558035714286 954.3526785714286 -0.0558035714286 897.9910714285714V102.0647321428572C-0.0558035714286 45.703125 47.4888392857143 0 106.1383928571429 0H1465.234375C1523.8839285714284 0 1571.4285714285716 45.703125 1571.4285714285716 102.0647321428572V897.9910714285714C1571.372767857143 954.3526785714286 1523.8839285714287 1000.0558035714286 1465.1785714285713 1000.0558035714286zM513.2254464285714 69.7544642857142C513.2254464285714 69.7544642857142 126.8415178571429 69.7544642857142 97.9910714285715 69.7544642857142C77.1763392857143 69.7544642857142 70.0892857142857 83.2031249999999 70.0892857142857 97.65625C70.0892857142857 100 70.0892857142857 141.1830357142857 70.0892857142857 141.1830357142857H513.2254464285714V69.7544642857142zM513.2254464285714 192.96875H70.0892857142857V264.3973214285714H513.2254464285714V192.96875zM513.2254464285714 316.1272321428571H70.0892857142857V387.5558035714286H513.2254464285714V316.1272321428571zM513.2254464285714 439.2857142857144H70.0892857142857V510.7142857142857H513.2254464285714V439.2857142857144zM513.2254464285714 562.5H70.0892857142857V633.9285714285714H513.2254464285714V562.5zM1008.7611607142858 69.7544642857142H567.2991071428572V141.1830357142857H1008.7611607142858V69.7544642857142zM1008.7611607142858 192.96875H567.2991071428572V264.3973214285714H1008.7611607142858V192.96875zM1008.7611607142858 316.1272321428571H567.2991071428572V387.5558035714286H1008.7611607142858V316.1272321428571zM1008.7611607142858 439.2857142857144H567.2991071428572V510.7142857142857H1008.7611607142858V439.2857142857144zM1008.7611607142858 562.5H567.2991071428572V633.9285714285714H1008.7611607142858V562.5zM1501.450892857143 97.65625C1501.450892857143 75.8928571428571 1494.6986607142858 69.7544642857142 1473.549107142857 69.7544642857142C1452.845982142857 69.7544642857142 1061.1049107142858 69.7544642857142 1061.1049107142858 69.7544642857142V141.1830357142857H1501.450892857143C1501.450892857143 141.1830357142857 1501.450892857143 107.3102678571428 1501.450892857143 97.65625zM1501.450892857143 192.96875H1061.1049107142858V264.3973214285714H1501.450892857143V192.96875zM1501.450892857143 316.1272321428571H1061.1049107142858V387.5558035714286H1501.450892857143V316.1272321428571zM1501.450892857143 439.2857142857144H1061.1049107142858V510.7142857142857H1501.450892857143V439.2857142857144zM1501.450892857143 562.5H1061.1049107142858V633.9285714285714H1501.450892857143V562.5z" />
    <glyph glyph-name="dataset_alternative"
      unicode="&#xF122;"
      horiz-adv-x="998.8839285714286" d="M0.0558035714286 1000.0558035714286L0 0L1000.0558035714286 0V1000.0558035714286H0.0558035714286zM320.3683035714286 86.2723214285713H40.7366071428571V157.7008928571428H320.3683035714286V86.2723214285713zM320.3683035714286 272.9910714285715H40.7366071428571V344.4196428571429H320.3683035714286V272.9910714285715zM320.3683035714286 459.6540178571429H40.7366071428571V531.0825892857142H320.3683035714286V459.6540178571429zM320.3683035714286 646.3727678571429H40.7366071428571V717.8013392857142H320.3683035714286V646.3727678571429zM639.84375 86.2723214285713H360.2120535714286V157.7008928571428H639.84375V86.2723214285713zM639.84375 272.9910714285715H360.2120535714286V344.4196428571429H639.84375V272.9910714285715zM639.84375 459.6540178571429H360.2120535714286V531.0825892857142H639.84375V459.6540178571429zM639.84375 646.3727678571429H360.2120535714286V717.8013392857142H639.84375V646.3727678571429zM959.3191964285714 86.2723214285713H679.6875V157.7008928571428H959.3191964285714V86.2723214285713zM959.3191964285714 272.9910714285715H679.6875V344.4196428571429H959.3191964285714V272.9910714285715zM959.3191964285714 459.6540178571429H679.6875V531.0825892857142H959.3191964285714V459.6540178571429zM959.3191964285714 646.3727678571429H679.6875V717.8013392857142H959.3191964285714V646.3727678571429z" />
    <glyph glyph-name="datastore"
      unicode="&#xF123;"
      horiz-adv-x="1000" d="M944.4196428571428 444.4196428571428H55.5803571428572C25 444.4196428571428 0 419.4196428571428 0 388.8392857142857V55.5245535714286C0 24.9441964285714 25 -0.0558035714286 55.5803571428572 -0.0558035714286H944.4754464285714C975.0558035714286 -0.0558035714286 1000.0558035714286 24.9441964285714 1000.0558035714286 55.5245535714286V388.8950892857144C1000 419.4196428571428 974.9999999999998 444.4196428571428 944.4196428571428 444.4196428571428M222.2098214285715 111.1049107142857C161.1049107142858 111.1049107142857 111.1049107142858 161.1049107142856 111.1049107142858 222.2098214285713S161.1049107142858 333.314732142857 222.2098214285715 333.314732142857S333.3147321428572 283.314732142857 333.3147321428572 222.2098214285713S283.3147321428572 111.1049107142857 222.2098214285715 111.1049107142857M944.4196428571428 1000H55.5803571428572C25 1000 0 975 0 944.4196428571428V611.1049107142858C0 580.5245535714287 25 555.5245535714286 55.5803571428572 555.5245535714286H944.4754464285714C975.0558035714286 555.5245535714286 1000.0558035714286 580.5245535714286 1000.0558035714286 611.1049107142858V944.4196428571428C1000 975 974.9999999999998 1000 944.4196428571428 1000M222.2098214285715 666.6852678571429C161.1049107142858 666.6852678571429 111.1049107142858 716.6852678571429 111.1049107142858 777.7901785714287S161.1049107142858 888.8950892857143 222.2098214285715 888.8950892857143S333.3147321428572 838.8950892857143 333.3147321428572 777.7901785714287S283.3147321428572 666.6852678571429 222.2098214285715 666.6852678571429" />
    <glyph glyph-name="delete"
      unicode="&#xF124;"
      horiz-adv-x="999.4419642857143" d="M726.171875 499.7209821428571L952.5669642857144 726.1160714285713C1015.0669642857144 788.6160714285714 1015.0669642857144 890.0111607142857 952.5669642857144 952.5111607142856C890.0669642857143 1015.0669642857142 788.671875 1015.0669642857142 726.171875 952.5111607142856L499.7209821428572 726.1160714285713L273.3258928571429 952.5669642857144C210.7700892857143 1015.0669642857144 109.4308035714286 1015.0669642857144 46.875 952.5669642857144C-15.625 890.0111607142857 -15.625 788.671875 46.875 726.1160714285713L273.2700892857143 499.7209821428571L46.875 273.3258928571428C-15.625 210.7700892857143 -15.625 109.4308035714284 46.875 46.875C109.375 -15.625 210.7700892857143 -15.625 273.2700892857143 46.875L499.6651785714286 273.2700892857144L726.0602678571428 46.875C788.5602678571428 -15.625 889.9553571428571 -15.625 952.4553571428572 46.875C1014.9553571428572 109.375 1014.9553571428572 210.7700892857143 952.4553571428572 273.2700892857144L726.171875 499.7209821428571z" />
    <glyph glyph-name="delete_sharp"
      unicode="&#xF125;"
      horiz-adv-x="1036.830357142857" d="M1036.830357142857 833.1473214285714L691.1272321428571 499.8325892857143L1036.3839285714287 166.9084821428571L863.7276785714287 0.3906249999999L518.4709821428572 333.314732142857L172.8236607142858 0L0 166.6294642857142L345.703125 500L0.4464285714286 832.8683035714286L173.1026785714286 999.3861607142856L518.3593750000001 666.4620535714286L864.0625000000001 999.8325892857142L1036.830357142857 833.1473214285714z" />
    <glyph glyph-name="desktop-uploader"
      unicode="&#xF126;"
      horiz-adv-x="1125" d="M1000 1000H125C55.9151785714286 1000 0 944.0848214285714 0 875V125C0 55.9151785714286 55.9151785714286 0 125 0H375V125H125V750H1000V125H750V0H1000C1069.0848214285713 0 1125 55.9151785714286 1125 125V875C1125 944.0848214285714 1069.0848214285716 1000 1000 1000M562.5 625L312.5 375H500V0H625V375H812.5L562.5 625z" />
    <glyph glyph-name="dissemination"
      unicode="&#xF127;"
      horiz-adv-x="1250" d="M1125 1000H125C56.25 1000 0.6138392857143 943.75 0.6138392857143 875L0 125C0 56.25 56.25 0 125 0H1125C1193.75 0 1250 56.25 1250 125V875C1250 943.75 1193.75 1000 1125 1000M812.5 125H125V375H812.5V125zM812.5 437.5H125V687.5H812.5V437.5zM1125 125H875V687.5H1125V125z" />
    <glyph glyph-name="divider_view"
      unicode="&#xF128;"
      horiz-adv-x="1363.8950892857142" d="M0 454.4642857142857H1363.8950892857142V545.2566964285714H0V454.4642857142857zM727.5111607142858 0H1363.8950892857144V363.6160714285714H727.5111607142858V0zM0 0H636.3839285714287V363.6160714285714H0V0zM727.5111607142858 1000V636.3839285714286H1363.8950892857144V1000H727.5111607142858zM636.3839285714287 1000H0V636.3839285714286H636.3839285714287V1000z" />
    <glyph glyph-name="doi"
      unicode="&#xF129;"
      horiz-adv-x="800" d="M500 1000H100C44.9776785714286 1000 0.5022321428572 955.0223214285714 0.5022321428572 900L0 100C0 44.9776785714287 44.4754464285714 0 99.4977678571429 0H700C755.0223214285713 0 800 44.9776785714284 800 100V700L500 1000zM600 299.9999999999999H450.0000000000001V150H350.0000000000001V299.9999999999999H200V400H350.0000000000001V550H450.0000000000001V400H600V299.9999999999999zM450.0000000000001 650V925L725 650H450.0000000000001z" />
    <glyph glyph-name="double_sharp_arrow_down"
      unicode="&#xF12A;"
      horiz-adv-x="750.5022321428572" d="M750.5022321428572 368.0245535714286L375.279017857143 0L0 368.0245535714286V583.0915178571429L375.2790178571429 215.0669642857143L750.5022321428572 583.0915178571429V368.0245535714286zM750.5022321428572 784.6540178571429L375.279017857143 416.6294642857142L0 784.6540178571429V999.7209821428572L375.2790178571429 631.6964285714286L750.5022321428572 999.7209821428572V784.6540178571429z" />
    <glyph glyph-name="double_sharp_arrow_up"
      unicode="&#xF12B;"
      horiz-adv-x="750.5022321428572" d="M750.5022321428572 416.6294642857142L375.279017857143 784.6540178571428L0 416.6294642857142V631.640625L375.2790178571429 999.6651785714286L750.5022321428572 631.640625V416.6294642857142zM750.5022321428572 0L375.279017857143 368.0245535714286L0 0V215.0669642857143L375.2790178571429 583.0915178571429L750.5022321428572 215.0669642857143V0z" />
    <glyph glyph-name="download"
      unicode="&#xF12C;"
      horiz-adv-x="799.7767857142858" d="M199.9441964285714 499.8883928571429V999.7209821428572H599.8325892857143V499.8883928571429H799.7767857142858L399.8883928571429 0.0558035714284L0 499.8883928571429H199.9441964285714z" />
    <glyph glyph-name="download_bold"
      unicode="&#xF12D;"
      horiz-adv-x="1083.314732142857" d="M1083.314732142857 416.6294642857142L541.2946428571429 0L0 416.6852678571429H222.2098214285714V1000H861.1049107142858V416.6294642857142H1083.314732142857z" />
    <glyph glyph-name="drag_handle"
      unicode="&#xF12E;"
      horiz-adv-x="2400" d="M0 200H2400V0H0V200zM0 1000H2400V800H0V1000z" />
    <glyph glyph-name="edit"
      unicode="&#xF12F;"
      horiz-adv-x="480.30133928571433" d="M111.0491071428572 255.5803571428571C60.4910714285714 277.0089285714286 3.90625 270.0892857142857 3.90625 270.0892857142857L0 -0.0558035714286L194.6986607142857 196.4285714285715C194.6986607142857 196.4285714285715 159.5982142857143 235.0446428571429 111.0491071428572 255.5803571428571zM235.3794642857143 829.5758928571429C168.4709821428572 670.4799107142858 45.2008928571429 377.34375 30.9151785714286 343.4709821428571C31.9754464285714 344.1964285714286 99.21875 338.28125 138.7834821428572 321.5401785714286C176.8973214285715 305.4129464285714 223.9955357142857 261.71875 223.7165178571429 261.8303571428571C237.2767857142857 294.140625 360.8816964285715 587.9464285714286 428.0133928571429 747.6004464285714C403.1808035714286 770.3683035714286 370.9821428571429 795.2566964285714 335.0446428571429 810.4910714285714C301.7857142857143 824.5535714285714 265.625 828.8504464285714 235.3794642857143 829.5758928571429zM417.0200892857143 991.5736607142856C361.6071428571429 1015.0669642857142 301.7299107142857 987.2767857142856 278.2924107142857 931.640625C278.2924107142857 931.640625 273.2142857142857 919.4754464285714 265.0669642857143 900.1674107142857C296.0379464285715 899.1071428571429 332.9241071428571 894.4196428571429 366.6294642857143 880.1339285714286C402.0089285714286 865.1785714285713 433.4263392857144 840.8482142857142 457.7566964285714 818.4151785714286C465.9040178571428 837.7790178571429 471.0379464285714 850 471.0379464285714 850C494.4196428571429 905.6919642857142 472.4888392857143 968.0803571428572 417.0200892857143 991.5736607142856z" />
    <glyph glyph-name="ellipsis-h"
      unicode="&#xF130;"
      horiz-adv-x="4333.333333333333" d="M0 500A500 500 0 0 1 1000 500A500 500 0 0 1 0 500M1666.6666666666665 500A500 500 0 0 1 2666.6666666666665 500A500 500 0 0 1 1666.6666666666665 500M3333.333333333333 500A500 500 0 0 1 4333.333333333333 500A500 500 0 0 1 3333.333333333333 500" />
    <glyph glyph-name="email"
      unicode="&#xF131;"
      horiz-adv-x="1363.950892857143" d="M0 761.8861607142858L675.1674107142858 289.2299107142857C707.8125000000001 266.3504464285714 751.6183035714287 267.4107142857144 783.1473214285714 291.9084821428571L1363.671875 743.4151785714284V90.9040178571429C1363.671875 40.6808035714286 1322.9910714285713 0 1272.767857142857 0H90.9040178571429C40.6808035714286 0 0 40.6808035714286 0 90.9040178571429V761.8861607142858zM21.9308035714286 968.1919642857144C38.6160714285714 987.5558035714286 63.28125 1000 90.9040178571429 1000H1272.7120535714287C1303.6272321428573 1000 1330.859375 984.7098214285714 1347.2656250000002 961.1049107142856L724.4419642857143 476.5625L21.9308035714286 968.1919642857144z" />
    <glyph glyph-name="exclamation_triangle"
      unicode="&#xF132;"
      horiz-adv-x="1144.7544642857144" d="M1133.1473214285713 107.421875L633.1473214285714 964.5647321428572C620.3125 986.5513392857144 596.8191964285714 1000 571.4285714285714 1000C546.0379464285714 1000 522.5446428571429 986.4955357142856 509.7098214285714 964.5647321428572L9.7098214285714 107.421875C-3.1808035714286 85.3236607142858 -3.2924107142857 58.0357142857142 9.4866071428571 35.8816964285713S45.8705357142857 0 71.4285714285714 0H1071.4285714285716C1096.986607142857 0 1120.6473214285716 13.671875 1133.3705357142858 35.8816964285713C1146.1495535714287 58.0357142857142 1146.0379464285713 85.3236607142858 1133.1473214285713 107.421875zM571.4285714285714 107.1428571428571C531.9754464285714 107.1428571428571 500 139.1183035714286 500 178.5714285714286C500 218.0245535714286 531.9754464285714 250 571.4285714285714 250C610.8816964285714 250 642.8571428571429 218.0245535714286 642.8571428571429 178.5714285714286C642.8571428571429 139.1183035714286 610.8816964285714 107.1428571428571 571.4285714285714 107.1428571428571zM642.8571428571429 392.8571428571429C642.8571428571429 353.4040178571428 610.8816964285714 321.4285714285715 571.4285714285714 321.4285714285715C531.9754464285714 321.4285714285715 500 353.4040178571428 500 392.8571428571429V642.8571428571429C500 682.3102678571429 531.9754464285714 714.2857142857142 571.4285714285714 714.2857142857142C610.8816964285714 714.2857142857142 642.8571428571429 682.3102678571429 642.8571428571429 642.8571428571429V392.8571428571429z" />
    <glyph glyph-name="expand"
      unicode="&#xF133;"
      horiz-adv-x="1549.3861607142858" d="M85.2120535714286 999.9441964285714C156.7522321428572 999.9441964285714 1412.6116071428569 999.9441964285714 1461.9977678571427 999.9441964285714C1526.004464285714 999.9441964285714 1573.660714285714 933.7053571428572 1536.1607142857142 868.1919642857142C1506.1941964285713 815.6808035714286 922.4330357142856 108.7611607142857 849.1629464285713 35.546875C801.4508928571428 -12.109375 747.4888392857142 -11.6629464285713 700 35.546875C646.9866071428572 88.28125 53.7388392857143 794.6428571428571 11.4955357142857 870.9263392857142C-19.53125 926.7299107142856 14.7321428571429 999.9441964285714 85.2120535714286 999.9441964285714z" />
    <glyph glyph-name="expand_controls"
      unicode="&#xF134;"
      horiz-adv-x="1285.9933035714287" d="M419.3080357142858 1000H1285.9933035714287V0H419.3080357142858V1000zM1219.3080357142858 933.3147321428572H485.9933035714286V66.6852678571428H1219.3080357142858V933.3147321428572zM419.8660714285715 432.2544642857144L235.7142857142858 433.3147321428571L235.7142857142858 264.2857142857142L0 499.9441964285714L235.7142857142857 735.6584821428571V566.5736607142857L419.8660714285715 565.5691964285713V432.2544642857144z" />
    <glyph glyph-name="fast_forward_backward"
      unicode="&#xF135;"
      horiz-adv-x="1141.685267857143" d="M1141.685267857143 499.7767857142857L523.2142857142858 999.609375L523.2142857142858 592.4107142857142L0 999.609375L0 0L523.2142857142858 407.1428571428571L523.2700892857143 0.0558035714284z" />
    <glyph glyph-name="fileset"
      unicode="&#xF136;"
      horiz-adv-x="1571.4285714285716" d="M1182.3660714285716 300.9486607142857L1294.8660714285716 224.4977678571428L1182.3660714285716 148.046875L1182.3660714285716 300.9486607142857L1182.3660714285716 300.9486607142857zM1571.4285714285716 404.0736607142857V45.9263392857142C1571.4285714285716 20.5357142857142 1550.0558035714284 0 1523.6607142857144 0H912.0535714285716C885.6584821428572 0 864.2857142857144 20.5915178571429 864.2857142857144 45.9263392857142V404.0736607142857C864.2857142857144 429.4642857142857 885.6584821428572 449.9999999999999 912.0535714285716 449.9999999999999H1523.6049107142858C1550 450 1571.4285714285716 429.4642857142857 1571.4285714285716 404.0736607142857zM1349.1629464285713 225C1349.1629464285713 297.5446428571429 1290.345982142857 356.3616071428572 1217.8013392857142 356.3616071428572S1086.4955357142858 297.5446428571429 1086.4955357142858 225S1145.3125 93.6383928571428 1217.857142857143 93.6383928571428S1349.1629464285713 152.4553571428571 1349.1629464285713 225zM659.375 1000H47.7678571428571C21.3727678571429 1000 0 979.4084821428572 0 954.0736607142856V595.9263392857143C0 570.5357142857143 21.3727678571429 550 47.7678571428571 550H659.375C685.7700892857142 550 707.1428571428571 570.5915178571429 707.1428571428571 595.9263392857143V954.0736607142858C707.1428571428572 979.4084821428572 685.7700892857143 1000 659.375 1000zM675.5580357142857 620.2008928571429C675.5580357142857 592.3549107142858 665.4575892857143 582.3660714285714 637.7232142857143 582.3660714285714C599.8325892857142 582.421875 197.2098214285714 582.5892857142858 69.4196428571429 582.5892857142858C42.4665178571429 582.5892857142858 31.5848214285714 593.6941964285713 31.5848214285714 620.4241071428571C31.5290178571429 629.0736607142857 31.5290178571429 629.0736607142857 31.5290178571429 642.3549107142857C112.7232142857143 706.7522321428571 230.4129464285714 800.1116071428571 230.4129464285714 800.1116071428571L391.1830357142857 705.5245535714286L514.1183035714286 894.6428571428571C514.1183035714286 894.6428571428571 593.6383928571429 839.453125 675.78125 782.421875C675.7254464285714 697.1540178571429 675.5580357142857 630.1339285714287 675.5580357142857 620.2008928571429zM659.375 450H47.7678571428571C21.3727678571429 450 0 429.4084821428571 0 404.0736607142857V45.9263392857142C0 20.5357142857142 21.3727678571429 0 47.7678571428571 0H659.375C685.7700892857142 0 707.1428571428571 20.5915178571429 707.1428571428571 45.9263392857142V404.0736607142857C707.1428571428572 429.4642857142857 685.7700892857143 450 659.375 450zM230.9709821428572 31.3616071428571C230.9709821428572 31.3616071428571 57.0870535714286 31.3616071428571 44.140625 31.3616071428571C34.765625 31.3616071428571 31.5848214285714 37.3883928571428 31.5848214285714 43.9174107142857C31.5848214285714 44.9776785714287 31.5848214285714 63.5044642857142 31.5848214285714 63.5044642857142H230.9709821428572V31.3616071428571zM230.9709821428572 86.8303571428571H31.5848214285714V118.9732142857141H230.9709821428572L230.9709821428572 86.8303571428571L230.9709821428572 86.8303571428571zM230.9709821428572 142.2433035714286H31.5848214285714V174.3861607142857H230.9709821428572L230.9709821428572 142.2433035714286L230.9709821428572 142.2433035714286zM230.9709821428572 197.65625H31.5848214285714V229.7991071428571H230.9709821428572L230.9709821428572 197.65625L230.9709821428572 197.65625zM230.9709821428572 253.1249999999999H31.5848214285714V285.267857142857H230.9709821428572L230.9709821428572 253.1249999999999L230.9709821428572 253.1249999999999zM453.9620535714286 31.3616071428571H255.3013392857143V63.5044642857142H453.9620535714286V31.3616071428571zM453.9620535714286 86.8303571428571H255.3013392857143V118.9732142857141H453.9620535714286V86.8303571428571zM453.9620535714286 142.2433035714286H255.3013392857143V174.3861607142857H453.9620535714286V142.2433035714286zM453.9620535714286 197.65625H255.3013392857143V229.7991071428571H453.9620535714286V197.65625zM453.9620535714286 253.1249999999999H255.3013392857143V285.267857142857H453.9620535714286V253.1249999999999zM675.6696428571429 43.9732142857142C675.6696428571429 34.2075892857142 672.6562499999999 31.4174107142857 663.1138392857143 31.4174107142857C653.7946428571428 31.4174107142857 477.5111607142857 31.4174107142857 477.5111607142857 31.4174107142857V63.5602678571428H675.6696428571429C675.6696428571429 63.5602678571429 675.6696428571429 48.2700892857142 675.6696428571429 43.9732142857142zM675.6696428571429 86.8303571428571H477.5111607142858V118.9732142857141H675.669642857143V86.8303571428571zM675.6696428571429 142.2433035714286H477.5111607142858V174.3861607142857H675.669642857143V142.2433035714286zM675.6696428571429 197.65625H477.5111607142858V229.7991071428571H675.669642857143V197.65625zM675.6696428571429 253.1249999999999H477.5111607142858V285.267857142857H675.669642857143V253.1249999999999zM1523.6607142857144 1000H912.0535714285716C885.7142857142858 1000 864.2857142857144 979.4084821428572 864.2857142857144 954.0736607142856V595.9263392857143C864.2857142857144 571.875 885.6584821428572 550 912.0535714285716 550H1523.6607142857144C1550.0558035714287 550 1571.4285714285716 570.5915178571429 1571.4285714285716 595.9263392857143V954.0736607142858C1571.4285714285716 979.4084821428572 1550.0558035714284 1000 1523.6607142857144 1000zM1192.075892857143 615.234375C1192.075892857143 613.7276785714286 1190.8482142857142 612.5 1189.3415178571427 612.5H1055.0223214285713C1053.5156249999998 612.5 1052.2879464285713 613.7276785714287 1052.2879464285713 615.234375V682.8683035714287C1052.2879464285713 685.4910714285714 1054.2968749999998 688.3370535714286 1054.520089285714 688.6160714285714C1055.9709821428569 690.2901785714286 1079.241071428571 710.8816964285713 1089.676339285714 721.3727678571429C1110.8258928571427 707.5334821428571 1134.9330357142856 708.3705357142857 1155.4687499999998 723.1584821428571C1166.0156249999998 712.6116071428571 1187.332589285714 691.5178571428571 1189.3973214285713 688.3370535714286C1189.8995535714284 687.5558035714286 1191.9084821428569 685.3236607142858 1192.020089285714 682.8683035714287V615.234375L1192.075892857143 615.234375zM1073.1026785714287 773.1026785714286C1073.1026785714287 800.1674107142857 1095.033482142857 822.0982142857142 1122.0982142857142 822.0982142857142C1149.1629464285716 822.0982142857142 1171.09375 800.1674107142857 1171.09375 773.1026785714286S1149.1629464285713 724.1071428571429 1122.0982142857142 724.1071428571429C1095.033482142857 724.1629464285713 1073.1026785714287 746.09375 1073.1026785714287 773.1026785714286zM1470.424107142857 739.3973214285714C1470.424107142857 712.0535714285714 1448.2700892857144 689.8995535714286 1420.9263392857144 689.8995535714286C1420.9263392857144 689.8995535714286 1211.7745535714287 689.7879464285713 1204.6316964285716 689.7879464285713C1195.089285714286 699.3303571428571 1173.0468750000002 722.4888392857142 1165.5133928571431 730.1339285714284C1180.9709821428573 745.6473214285713 1186.886160714286 768.8616071428571 1180.8593750000002 789.9553571428571C1175.837053571429 807.421875 1162.7790178571431 822.0424107142857 1146.0937500000002 829.1294642857142C1128.125 836.71875 1107.1428571428573 835.1004464285713 1090.5133928571431 824.8883928571428C1074.7209821428573 815.1785714285713 1063.9508928571431 798.2700892857142 1061.886160714286 779.7991071428571C1059.8214285714287 760.9374999999999 1066.7410714285716 741.9084821428571 1080.4687500000002 728.8504464285713C1064.0066964285716 712.6674107142857 1048.8281250000002 699.1071428571428 1039.0625000000002 689.8437499999999L1014.8995535714288 689.8437499999999C987.5558035714288 689.8437499999999 965.4017857142858 711.9977678571428 965.4017857142858 739.3415178571427V887.8906249999999C965.4017857142858 915.2343749999998 987.5558035714288 937.3883928571428 1014.8995535714288 937.3883928571428H1420.870535714286C1448.214285714286 937.3883928571428 1470.3683035714287 915.2343749999998 1470.4241071428573 887.8906249999999L1470.424107142857 739.3973214285714L1470.424107142857 739.3973214285714z" />
    <glyph glyph-name="fineart"
      unicode="&#xF137;"
      horiz-adv-x="1607.142857142857" d="M785.7142857142858 357.1428571428571L785.7142857142858 250L910.7142857142858 250L857.1428571428571 357.1428571428571zM625 250L750 250L750 357.1428571428571L678.5714285714286 357.1428571428571zM1500 1000H107.1428571428571C46.4285714285714 1000 0 953.5714285714286 0 892.8571428571429V107.1428571428571C0 46.4285714285714 46.4285714285714 0 107.1428571428571 0H1500C1560.7142857142858 0 1607.142857142857 46.4285714285714 1607.142857142857 107.1428571428571V892.8571428571429C1607.142857142857 953.5714285714286 1560.7142857142858 1000 1500 1000zM1107.142857142857 392.8571428571429C1107.142857142857 371.4285714285714 1092.857142857143 357.1428571428571 1071.4285714285716 357.1428571428571H892.8571428571429L946.4285714285714 250H1017.8571428571428C1028.5714285714287 250 1035.7142857142858 242.8571428571429 1035.7142857142858 232.1428571428571S1028.5714285714287 214.2857142857142 1017.8571428571428 214.2857142857142H964.2857142857144L1000 142.8571428571429H964.2857142857144L928.5714285714286 214.2857142857142H607.1428571428571L571.4285714285714 142.8571428571429H535.7142857142858L571.4285714285714 214.2857142857142H517.8571428571429C507.1428571428571 214.2857142857142 500 221.4285714285713 500 232.1428571428571S507.1428571428571 250 517.8571428571429 250H589.2857142857143L642.8571428571429 357.1428571428571H464.2857142857143C442.8571428571429 357.1428571428571 428.5714285714286 371.4285714285714 428.5714285714286 392.8571428571429V750C428.5714285714286 771.4285714285714 442.8571428571429 785.7142857142858 464.2857142857143 785.7142857142858H678.5714285714286C678.5714285714286 807.1428571428571 692.8571428571428 821.4285714285714 714.2857142857143 821.4285714285714H750V839.2857142857142C750 850 757.1428571428571 857.1428571428571 767.8571428571429 857.1428571428571C778.5714285714287 857.1428571428571 785.7142857142858 850 785.7142857142858 839.2857142857142V821.4285714285714H821.4285714285714C842.8571428571429 821.4285714285714 857.1428571428571 807.1428571428571 857.1428571428571 785.7142857142858H1071.4285714285716C1092.857142857143 785.7142857142858 1107.142857142857 771.4285714285714 1107.142857142857 750V392.8571428571429zM1053.5714285714287 750H482.1428571428572C471.4285714285714 750 464.2857142857143 742.8571428571429 464.2857142857143 732.1428571428571V410.7142857142857C464.2857142857143 410.7142857142857 464.2857142857143 410.7142857142857 464.2857142857143 410.7142857142857C464.2857142857143 407.1428571428571 464.2857142857143 403.5714285714286 467.8571428571428 400C467.8571428571428 400 467.8571428571428 400 471.4285714285714 396.4285714285714C475 392.8571428571427 478.5714285714286 392.8571428571427 482.1428571428572 392.8571428571427C482.1428571428572 392.8571428571427 482.1428571428572 392.8571428571427 485.7142857142857 392.8571428571427H1057.142857142857C1060.7142857142858 392.8571428571427 1064.2857142857144 392.8571428571427 1067.857142857143 396.4285714285714C1067.857142857143 396.4285714285714 1071.4285714285716 396.4285714285714 1071.4285714285716 400C1075.0000000000002 403.5714285714286 1075.0000000000002 407.1428571428571 1075.0000000000002 410.7142857142857C1075.0000000000002 410.7142857142857 1075.0000000000002 410.7142857142857 1075.0000000000002 410.7142857142857C1075.0000000000002 410.7142857142857 1075.0000000000002 410.7142857142857 1075.0000000000002 410.7142857142857L1075.0000000000002 410.7142857142857V732.1428571428571C1071.4285714285716 742.8571428571429 1064.2857142857144 750 1053.5714285714287 750zM521.4285714285714 428.5714285714286L678.5714285714286 600L807.1428571428572 471.4285714285714C810.7142857142857 464.2857142857142 817.8571428571428 464.2857142857142 821.4285714285714 464.2857142857142C825.0000000000001 464.2857142857142 832.1428571428572 467.8571428571428 835.7142857142857 471.4285714285714L910.7142857142858 560.7142857142857L1017.8571428571428 428.5714285714286H521.4285714285714zM1035.7142857142858 460.7142857142857L925 600C921.4285714285714 603.5714285714286 917.8571428571428 607.1428571428571 910.7142857142858 607.1428571428571C903.5714285714286 607.1428571428571 900 603.5714285714286 896.4285714285716 600L821.4285714285714 507.1428571428571L692.8571428571428 635.7142857142857C689.2857142857143 642.8571428571429 682.1428571428572 642.8571428571429 678.5714285714286 642.8571428571429C675 642.8571428571429 667.8571428571429 639.2857142857142 664.2857142857143 635.7142857142858L500 457.1428571428571V714.2857142857142H1035.7142857142858V460.7142857142857z" />
    <glyph glyph-name="fit_to_view"
      unicode="&#xF138;"
      horiz-adv-x="1200.0558035714287" d="M975.0558035714286 0H225.0558035714287C101.0044642857144 0 0.0558035714287 100.9486607142858 0.0558035714287 225V775C0.0558035714287 899.0513392857143 101.0044642857144 1000 225.0558035714287 1000H975.0558035714286C1099.107142857143 1000 1200.0558035714287 899.0513392857143 1200.0558035714287 775V225C1200.0558035714287 100.9486607142858 1099.107142857143 0 975.0558035714286 0zM225 950C128.515625 950 50 871.484375 50 775V224.9999999999999C50 128.5156249999998 128.515625 49.9999999999999 225 49.9999999999999H975C1071.484375 49.9999999999999 1150.0000000000002 128.515625 1150.0000000000002 224.9999999999999V774.9999999999999C1150.0000000000002 871.4843749999999 1071.484375 949.9999999999998 975.0000000000002 949.9999999999998H225zM651.4508928571429 445.3683035714286V275.78125H782.0312500000001L600 91.3504464285715L417.96875 275.78125L548.4933035714286 275.78125V445.3683035714286L651.4508928571429 445.3683035714286zM548.5491071428571 554.6875V724.2745535714286L418.0245535714286 724.2745535714286L600.0558035714286 908.7053571428572L782.0870535714286 724.2745535714286H651.5066964285714V554.6875L548.5491071428571 554.6875z" />
    <glyph glyph-name="fits"
      unicode="&#xF139;"
      horiz-adv-x="1571.4285714285716" d="M456.0825892857143 538.5044642857142C459.4308035714286 530.1897321428571 460.7142857142857 521.4285714285713 460.4352678571428 512.6116071428571C459.9888392857143 497.65625 455.0223214285714 483.7611607142857 445.9821428571429 472.1540178571429C441.9642857142857 467.0200892857142 438.2254464285714 463.28125 434.4866071428571 460.3794642857142C405.1339285714285 437.5558035714286 363.1696428571429 442.5223214285715 339.7879464285714 471.1495535714286L338.1138392857142 473.2700892857142C329.4642857142857 485.4910714285713 324.8325892857142 499.9441964285713 324.8883928571428 515.3459821428571C325.0558035714285 524.0513392857142 326.5624999999999 532.421875 329.9665178571428 540.234375C334.6540178571428 551.953125 341.6852678571428 561.3839285714284 350.7812499999999 568.4709821428571C354.5200892857142 571.4285714285713 358.3705357142856 573.828125 362.4999999999999 575.8928571428571C381.0825892857142 585.15625 402.7343749999999 585.0446428571428 422.0424107142857 575.6696428571428C432.0870535714285 570.8705357142857 440.2901785714285 564.3415178571429 446.5401785714285 556.3058035714284C450.1674107142858 551.6741071428571 453.125 545.9821428571429 456.0825892857143 538.5044642857142zM1571.4285714285716 897.9910714285714V102.0647321428572C1571.4285714285716 45.703125 1523.8839285714287 0 1465.234375 0H106.1383928571429C47.4888392857143 0 -0.0558035714286 45.703125 -0.0558035714286 102.0647321428572V897.9910714285714C-0.0558035714286 954.3526785714286 47.4888392857143 1000.0558035714286 106.1383928571429 1000.0558035714286H1465.1785714285713C1523.8839285714287 1000.0558035714286 1571.372767857143 954.3526785714286 1571.4285714285716 897.9910714285714zM509.0959821428571 121.2611607142857H65.9598214285714C65.9598214285714 121.2611607142857 65.9598214285714 105.7477678571428 65.9598214285714 103.4040178571428C65.9598214285714 88.9508928571428 73.046875 75.5022321428571 93.8616071428571 75.5022321428571C122.7120535714286 75.5022321428571 509.0959821428571 75.5022321428571 509.0959821428571 75.5022321428571V121.2611607142857zM509.0959821428571 214.2857142857142H65.9598214285714V169.0848214285715H509.0959821428572V214.2857142857142zM1004.6316964285714 121.2611607142857H563.1696428571429V75.5022321428571H1004.6316964285714V121.2611607142857zM1004.6316964285714 214.2857142857142H563.1696428571429V169.0848214285715H1004.6316964285714V214.2857142857142zM1497.3214285714284 121.2611607142857H1056.9754464285713V75.5022321428571C1056.9754464285713 75.5022321428571 1448.716517857143 75.5022321428571 1469.419642857143 75.5022321428571C1490.5691964285713 75.5022321428571 1497.3214285714284 81.640625 1497.3214285714284 103.4040178571428C1497.3214285714284 113.0580357142857 1497.3214285714284 121.2611607142857 1497.3214285714284 121.2611607142857zM1497.3214285714284 214.2857142857142H1056.9754464285713V169.0848214285715H1497.3214285714284V214.2857142857142zM1497.3214285714284 262.2767857142857V565.2901785714286C1474.8325892857142 509.5424107142857 1439.84375 459.4866071428571 1381.25 413.9508928571428C1158.4263392857142 240.7924107142857 841.6294642857142 271.986607142857 656.1383928571428 478.90625C799.7767857142857 503.3482142857142 965.7924107142856 543.6941964285713 1128.90625 594.7544642857142C1278.1249999999998 641.4620535714286 1383.2589285714284 678.2924107142858 1497.3214285714284 729.5200892857142V790.1227678571429C1382.8125 739.84375 1251.2276785714284 690.7366071428571 1112.3325892857142 647.3772321428571C944.1406249999998 594.6986607142857 768.1919642857141 552.0647321428571 616.5736607142856 528.2924107142857C581.361607142857 522.7120535714284 547.5446428571427 518.2477678571429 515.345982142857 515.0111607142857C515.4575892857141 527.4553571428571 513.4486607142856 539.6205357142858 509.7656249999998 551.3950892857142C536.3839285714284 554.0178571428571 564.6205357142856 557.5892857142858 594.6986607142856 562.0535714285714C584.5424107142854 578.8504464285714 575.5580357142856 595.9263392857143 567.5781249999998 613.5602678571429C548.4374999999998 655.1897321428571 535.1562499999998 698.4933035714287 527.2879464285712 742.578125C523.8839285714283 761.71875 521.4843749999998 781.0825892857142 520.2008928571425 800.4464285714286C435.0446428571426 758.3147321428571 364.1183035714284 716.6852678571429 314.3973214285711 678.0691964285713C263.6160714285711 638.6160714285713 237.0535714285711 603.0133928571428 243.4151785714283 583.0357142857142C246.5959821428569 572.6004464285713 258.3147321428569 564.2299107142857 277.6785714285711 557.7566964285713C273.5491071428569 546.484375 271.0937499999997 534.8214285714286 270.2566964285711 522.9910714285714C237.6116071428568 532.8683035714287 215.7366071428568 548.8839285714286 208.5937499999997 571.7075892857142C196.7075892857139 609.8214285714286 223.8839285714282 653.8504464285714 292.020089285714 706.8080357142858C346.2053571428569 748.8839285714286 424.6093749999997 794.5870535714286 519.2522321428568 840.3459821428571C519.3080357142853 861.328125 520.8147321428568 882.3102678571429 523.3816964285711 902.9575892857144C524.3303571428568 910.3236607142858 524.8883928571425 916.3504464285714 525.2232142857139 921.3169642857144H150.5580357142857C103.8504464285715 921.3169642857144 65.9598214285715 883.4263392857143 65.9598214285715 836.71875V262.2767857142857H1497.3214285714284z" />
    <glyph glyph-name="folder_fill"
      unicode="&#xF13A;"
      horiz-adv-x="1333.5937500000002" d="M1250.2232142857144 666.796875H412.8348214285714C322.8794642857142 666.796875 250.0558035714286 593.9732142857142 250.0558035714286 504.0178571428572V166.7410714285716H208.3705357142857C185.4352678571428 166.7410714285716 166.6852678571429 185.3794642857145 166.6852678571429 208.4263392857143V551.8415178571429C166.6852678571429 661.3839285714287 255.46875 750.1116071428571 364.9553571428572 750.1116071428571H1166.908482142857V833.4821428571429C1166.908482142857 879.5200892857143 1129.6316964285713 916.8526785714286 1083.5379464285713 916.8526785714286H500.1116071428572C500.1116071428572 962.9464285714286 462.8348214285715 1000.2232142857144 416.7410714285715 1000.2232142857144H166.6852678571429C120.6473214285714 1000.1674107142856 83.3705357142857 962.890625 83.3705357142857 916.8526785714286V917.1875C33.7053571428571 888.28125 0 835.15625 0 773.6049107142857V208.3705357142858C0 93.3035714285715 93.3035714285714 0 208.3705357142857 0H1166.908482142857C1258.984375 0 1333.5937499999998 74.6651785714286 1333.5937499999998 166.6852678571429V583.4263392857142C1333.5937500000002 629.5200892857142 1296.3169642857142 666.796875 1250.2232142857144 666.796875z" />
    <glyph glyph-name="folder_open"
      unicode="&#xF13B;"
      horiz-adv-x="1378.5714285714287" d="M1283.6495535714287 395.9263392857142H455.6361607142857C401.7299107142857 395.9263392857142 378.4040178571429 355.0223214285715 355.1339285714286 316.6294642857144C329.0178571428572 273.4933035714287 262.2209821428571 166.8526785714286 253.3482142857143 153.6830357142858C221.484375 106.3616071428572 180.3013392857143 112.6116071428572 155.0223214285715 126.8415178571429C127.734375 142.1875000000001 103.0133928571429 171.0379464285716 132.5892857142857 219.8660714285716C155.5803571428572 257.8125000000001 214.1183035714286 364.5647321428572 239.4531250000001 406.4174107142858C288.1138392857144 486.7745535714287 356.0825892857144 481.9754464285716 457.7566964285714 481.9754464285716H1202.9575892857142L1205.3013392857142 737.2767857142858C1205.3013392857142 784.8214285714287 1162.8348214285713 823.3816964285716 1110.4352678571427 823.3816964285716H497.2098214285714L443.9732142857144 959.9888392857144C426.953125 984.8772321428572 398.2700892857144 999.8325892857144 367.5223214285715 999.8325892857144H91.8526785714286C41.1272321428572 999.8325892857144 -0.0558035714286 959.7656250000002 -0.0558035714286 910.3794642857144V88.5044642857143C-0.0558035714286 43.4151785714287 33.0915178571429 0.279017857143 80.2455357142857 0.3348214285714C277.1763392857144 -0.1674107142857 1103.7388392857142 0.2232142857143 1120.200892857143 0.279017857143C1151.6183035714284 0.279017857143 1186.8861607142856 40.4017857142859 1208.1473214285716 65.3459821428572C1208.1473214285716 65.3459821428572 1365.5691964285713 252.622767857143 1374.5535714285713 279.6316964285715C1383.3147321428569 305.9151785714286 1377.0089285714287 336.8303571428572 1359.095982142857 359.7098214285715C1341.294642857143 382.4776785714286 1313.28125 395.9263392857142 1283.6495535714287 395.9263392857142z" />
    <glyph glyph-name="folder_outline"
      unicode="&#xF13C;"
      horiz-adv-x="1333.5937500000002" d="M416.7410714285714 944.3080357142856C426.6741071428571 944.3080357142856 433.203125 939.2857142857142 436.2165178571429 936.2723214285714C439.2299107142857 933.2589285714286 444.2522321428571 926.7299107142856 444.2522321428571 916.796875C444.2522321428571 902.0089285714286 450.1116071428571 887.7790178571429 460.6026785714285 877.34375C471.0937499999999 866.8526785714286 485.2678571428571 860.9933035714286 500.0558035714286 860.9933035714286H1083.482142857143C1093.4151785714284 860.9933035714286 1099.9441964285713 855.9151785714286 1102.9575892857142 852.9575892857142C1105.9709821428573 849.9441964285714 1110.9933035714287 843.4151785714286 1110.9933035714287 833.4821428571429V721.5959821428571H1166.796875V833.4263392857142C1166.796875 879.4642857142857 1129.5200892857144 916.796875 1083.4263392857142 916.796875H500.1116071428572C500.1116071428572 962.890625 462.8348214285715 1000.1674107142856 416.7410714285715 1000.1674107142856L416.7410714285715 1000.1674107142856H166.6852678571429C120.5915178571429 1000.1674107142856 83.3147321428571 962.890625 83.3147321428571 916.796875V917.1316964285714C33.6495535714286 888.2254464285714 -0.0558035714286 835.1004464285714 -0.0558035714286 773.5491071428571V208.2589285714286C-0.0558035714286 93.1919642857142 93.2477678571429 -0.1116071428571 208.3147321428571 -0.1116071428571H1166.8526785714287C1258.9285714285716 -0.1116071428571 1333.5379464285713 74.5535714285715 1333.5379464285713 166.5736607142858V638.5602678571429C1333.5379464285713 684.6540178571429 1296.2611607142856 721.9308035714284 1250.2232142857142 721.9308035714284H412.8348214285714C322.8794642857142 721.9308035714284 250.0558035714286 649.1071428571429 250.0558035714286 559.1517857142858C250.0558035714286 559.1517857142858 250.0558035714286 208.0357142857142 250.0558035714286 194.5312499999999C250.0558035714286 157.7566964285713 305.859375 157.7566964285713 305.859375 194.5312499999999C305.859375 207.3660714285713 305.859375 559.1517857142857 305.859375 559.1517857142857C305.859375 587.7790178571428 316.9642857142857 614.6205357142856 337.1651785714286 634.8214285714284C357.3660714285715 655.0223214285713 384.2075892857142 666.1272321428571 412.8348214285714 666.1272321428571H1250.2232142857142C1260.15625 666.1272321428571 1266.6852678571427 661.1049107142857 1269.6986607142856 658.0915178571429C1272.7120535714284 655.078125 1277.734375 648.5491071428571 1277.734375 638.5602678571429V166.5736607142857C1277.734375 136.9419642857143 1266.1830357142858 109.0959821428571 1245.2566964285716 88.1696428571428C1224.330357142857 67.2433035714286 1196.484375 55.6919642857142 1166.8526785714287 55.6919642857142H208.3705357142857C187.7790178571429 55.6919642857142 167.8013392857143 59.7098214285715 148.9955357142857 67.6339285714286C130.859375 75.3348214285715 114.5089285714286 86.328125 100.5022321428571 100.3348214285713C86.4955357142857 114.3415178571428 75.4464285714286 130.6919642857143 67.8013392857143 148.828125C59.8772321428572 167.578125 55.859375 187.5558035714287 55.859375 208.203125V773.4933035714286C55.859375 793.0803571428571 61.1049107142857 812.2767857142858 71.09375 829.0178571428571C80.0223214285714 844.0290178571429 92.578125 856.9196428571429 107.4776785714286 866.40625C113.1138392857143 869.0848214285714 118.359375 872.7678571428571 122.8794642857143 877.2879464285714C133.3705357142857 887.7790178571429 139.2299107142857 901.953125 139.2299107142857 916.7410714285714C139.2299107142857 926.6741071428572 144.2522321428572 933.203125 147.265625 936.2165178571428C150.2790178571428 939.2299107142858 156.8080357142857 944.2522321428572 166.7410714285714 944.2522321428572H416.7410714285714M-194.7544642857143 611.6071428571429" />
    <glyph glyph-name="forward_arrow"
      unicode="&#xF13D;"
      horiz-adv-x="1499.9441964285716" d="M749.9441964285714 500M0 608.9285714285714H1027.9575892857142V1000L1499.9441964285716 500L1027.9575892857144 0V391.1272321428572H0L0 608.9285714285714z" />
    <glyph glyph-name="ftp"
      unicode="&#xF13E;"
      horiz-adv-x="1601.7857142857144" d="M1338.950892857143 528.4040178571429H1310.1562500000002C1306.417410714286 696.09375 1168.75 831.25 1001.0602678571428 831.25C962.2767857142858 831.25 924.7209821428572 823.7165178571429 888.4486607142858 809.9888392857142C787.109375 988.9508928571428 560.6026785714286 1052.7901785714287 381.640625 952.6785714285714C230.1897321428572 867.578125 157.6450892857143 688.6160714285714 207.7008928571429 522.1540178571429C65.0669642857143 490.8482142857143 -25.0558035714286 349.4419642857144 6.25 208.0357142857143C32.5334821428571 86.6629464285716 138.8950892857143 0.279017857143 262.7790178571429 0.279017857143H951.060267857143H1339.0066964285716C1484.1517857142858 0.279017857143 1601.7857142857144 119.1406250000001 1601.7857142857144 265.5691964285715C1601.7857142857144 409.5424107142858 1484.1517857142858 527.1763392857142 1338.950892857143 528.4040178571429zM765.9598214285714 479.8549107142857L730.46875 445.8147321428571C724.609375 440.1785714285714 714.8995535714286 440.1785714285714 709.0401785714286 445.8147321428571L651.0044642857142 501.5066964285714V321.1495535714286C651.0044642857142 313.28125 644.3638392857142 306.5848214285715 635.8258928571428 306.5848214285715H585.15625C577.0089285714286 306.5848214285715 569.9776785714286 312.9464285714287 569.9776785714286 321.1495535714286V501.5624999999999L511.8861607142857 446.5959821428571C506.0267857142857 441.015625 496.3169642857142 441.015625 490.4575892857142 446.5959821428571L454.9665178571428 480.6361607142857C449.1071428571428 486.2723214285714 449.1071428571428 495.5915178571428 454.9665178571428 501.1718749999999L599.5535714285713 639.8995535714284C605.8035714285714 645.4799107142857 615.1227678571429 645.4799107142857 621.3727678571428 639.8995535714284L765.9040178571428 500.4464285714286C771.7633928571429 494.8102678571429 771.7633928571429 485.4910714285714 765.9598214285714 479.8549107142857zM1121.9308035714287 341.2388392857142L977.3437500000002 202.5669642857142C971.09375 196.9866071428571 961.7745535714286 196.9866071428571 955.5245535714286 202.5669642857142L810.9933035714287 342.0200892857144C805.1339285714287 347.6004464285715 805.1339285714287 356.9754464285715 810.9933035714287 362.5558035714286L846.4843750000001 396.5959821428571C852.3437500000001 402.2321428571428 862.0535714285716 402.2321428571428 867.9129464285716 396.5959821428571L925.9486607142858 340.9040178571429V521.2611607142858C925.9486607142858 529.1294642857142 932.5892857142858 535.8258928571429 941.1272321428575 535.8258928571429H991.7968750000002C999.9441964285716 535.8258928571429 1006.9754464285716 529.4642857142858 1006.9754464285716 521.2611607142858V340.9040178571429L1065.0669642857144 395.8705357142858C1070.9263392857144 401.4508928571429 1080.6361607142858 401.4508928571429 1086.4955357142858 395.8705357142858L1121.9866071428573 361.8303571428572C1127.7901785714287 356.1941964285714 1127.7901785714287 346.8749999999999 1121.9308035714287 341.2388392857142z" />
    <glyph glyph-name="fullscreen"
      unicode="&#xF13F;"
      horiz-adv-x="1000.6696428571429" d="M571.4285714285714 857.1428571428571H756.1383928571429L520.9263392857143 621.9308035714286L621.9308035714286 520.9263392857142L857.1428571428571 756.1383928571429V571.4285714285714H1000V857.1428571428571V1000L857.1428571428571 1000L571.4285714285714 1000V857.1428571428571zM0 142.8571428571429L0 428.5714285714286H142.8571428571429V243.8616071428571L378.0691964285715 479.0736607142857L479.0736607142857 378.0691964285715L243.8616071428572 142.8571428571429H428.5714285714286V0H142.8571428571429H0L0 142.8571428571429z" />
    <glyph glyph-name="fullscreen_exit"
      unicode="&#xF140;"
      horiz-adv-x="1000.6696428571429" d="M695.0837053571428 954.0776283482141H565.2287946428571V694.4236104910715V564.5686997767857H695.0837053571428H954.7377232142856V694.4236104910715H786.8805803571429L1000.6640625000002 908.2070926339286L908.8671875 1000.0039676339286L695.083705357143 786.2204854910715V954.0776283482141zM435.4296875 434.7695926339285V304.9146819196428V45.2606640624999H305.5747767857142V213.1178069196428L91.7912946428571 -0.6656752232144L-0.0055803571429 91.1311997767857L213.7779017857143 304.914681919643H45.9207589285714V434.7695926339286H305.5747767857142H435.4296875L435.4296875 434.7695926339285z" />
    <glyph glyph-name="generic_file"
      unicode="&#xF141;"
      horiz-adv-x="713.3928571428572" d="M428.0133928571429 951.8415178571428H47.4888392857143V48.1584821428571H665.7924107142857V714.0066964285714H428.0133928571429V951.8415178571428zM713.4486607142858 0.6138392857142H-0.0558035714286V999.3861607142856H451.7857142857143L713.3370535714286 734.9888392857142V0.6138392857142H713.4486607142858z" />
    <glyph glyph-name="github_logo"
      unicode="&#xF142;"
      horiz-adv-x="1025.279017857143" d="M512.6116071428572 1000C229.5758928571429 1000 0 770.4799107142857 0 487.2767857142858C0 260.8258928571429 146.875 68.6383928571428 350.6138392857143 0.8370535714286C376.2276785714286 -3.8504464285716 385.6026785714286 11.9419642857143 385.6026785714286 25.5580357142857C385.6026785714286 37.7232142857142 385.1562500000001 69.9776785714284 384.9330357142857 112.7232142857142C242.3549107142857 81.7522321428571 212.2209821428571 181.4174107142857 212.2209821428571 181.4174107142857C188.8950892857143 240.6249999999999 155.3013392857143 256.4174107142857 155.3013392857143 256.4174107142857C108.7611607142857 288.2254464285715 158.8169642857143 287.5558035714286 158.8169642857143 287.5558035714286C210.2678571428571 283.9285714285714 237.3325892857143 234.7098214285714 237.3325892857143 234.7098214285714C283.0357142857143 156.361607142857 357.3102678571429 179.0178571428571 386.5513392857143 192.1316964285713C391.1830357142857 225.2790178571429 404.4642857142857 247.8794642857144 419.0848214285714 260.6584821428571C305.2455357142857 273.6049107142857 185.546875 317.5781249999999 185.546875 514.0066964285713C185.546875 569.9776785714284 205.5245535714285 615.7366071428571 238.3370535714286 651.5625C233.0915178571429 664.5089285714286 215.4575892857143 716.6294642857142 243.359375 787.2209821428571C243.359375 787.2209821428571 286.3839285714286 801.0044642857142 384.3191964285714 734.6540178571429C425.2232142857142 746.0379464285713 469.0848214285714 751.7299107142857 512.6674107142857 751.8973214285713C556.1941964285714 751.6741071428571 600.0558035714286 746.0379464285713 641.0156249999999 734.6540178571429C738.8950892857142 801.0044642857142 781.8638392857142 787.2209821428571 781.8638392857142 787.2209821428571C809.8214285714284 716.6294642857142 792.2433035714284 664.5089285714286 786.9419642857142 651.5625C819.8102678571428 615.7366071428571 839.6205357142857 569.9776785714284 839.6205357142857 514.0066964285713C839.6205357142857 317.0758928571428 719.7544642857143 273.7165178571429 605.5803571428571 261.0491071428571C623.9955357142857 245.2008928571428 640.4017857142857 213.9508928571428 640.4017857142857 166.1272321428571C640.4017857142857 97.6004464285715 639.7879464285714 42.2991071428571 639.7879464285714 25.5022321428571C639.7879464285714 11.7745535714287 648.9955357142858 -4.1294642857142 675.0558035714287 0.8370535714286C878.5714285714287 68.75 1025.3348214285716 260.7700892857142 1025.3348214285716 487.2209821428571C1025.279017857143 770.4799107142857 795.7589285714286 1000 512.6116071428572 1000z" />
    <glyph glyph-name="go-to-link-alternative"
      unicode="&#xF143;"
      horiz-adv-x="1124.9441964285716" d="M937.4441964285716 0H187.5C84.0959821428571 0 0 84.0959821428571 0 187.5V812.4441964285714C0 915.8482142857142 84.0959821428571 999.9441964285714 187.5 999.9441964285714H562.5V874.9441964285714H187.5C153.0691964285715 874.9441964285714 125 846.875 125 812.4441964285714V187.5C125 153.0691964285713 153.0691964285715 125 187.5 125H937.4441964285716C971.9308035714286 125 999.9441964285716 153.0691964285713 999.9441964285716 187.5V437.5H1124.9441964285716V187.5C1124.9441964285716 84.0959821428571 1040.8482142857142 0 937.4441964285716 0zM749.609375 874.9441964285714H908.1473214285716L701.3950892857143 639.2857142857142L802.5111607142858 553.6830357142858L999.6093750000002 791.9642857142858V625H1124.6093750000002V1000H749.6093750000001V874.9441964285714z" />
    <glyph glyph-name="go_to_link"
      unicode="&#xF144;"
      horiz-adv-x="1000.3348214285714" d="M694.6986607142858 111.1607142857143H111.1607142857143V694.6986607142857H333.4821428571429V805.859375C172.8236607142858 805.859375 0.0558035714286 805.859375 0.0558035714286 805.859375V0H805.9151785714284V333.4263392857142H694.7544642857142V111.1607142857143zM611.328125 1000.3348214285714L472.3772321428572 889.1741071428571H790.9598214285716L358.3705357142857 456.5848214285714L456.6406250000001 358.3147321428572L889.2299107142858 790.9040178571429V500.1116071428572L1000.390625 611.2723214285714V1000.2790178571428H611.328125z" />
    <glyph glyph-name="grid_view"
      unicode="&#xF145;"
      horiz-adv-x="1363.8950892857142" d="M727.5111607142858 0H1363.8950892857144V454.4642857142857H727.5111607142858V0zM0 545.5357142857142H636.3839285714287V1000H0V545.5357142857142zM0 0H636.3839285714287V454.4642857142857H0V0zM727.5111607142858 1000V545.2566964285714H1363.8950892857144V1000H727.5111607142858z" />
    <glyph glyph-name="hide_details"
      unicode="&#xF146;"
      horiz-adv-x="961.9419642857143" d="M17.0758928571429 480.5803571428571C39.5647321428572 457.9799107142858 76.2276785714286 457.8125 98.8839285714286 480.3013392857143C98.8839285714286 480.3013392857143 358.1473214285715 739.1741071428571 453.125 834.1517857142858C478.7388392857143 859.765625 481.4732142857143 860.8816964285714 510.9933035714286 831.3058035714286C608.984375 733.3147321428571 863.1138392857143 478.9620535714286 863.1138392857143 478.9620535714286C886.1607142857143 456.0267857142857 923.3816964285714 456.1941964285714 946.2611607142858 479.2410714285714C969.140625 502.2879464285714 969.0290178571428 539.6205357142858 945.9821428571428 562.5L556.640625 951.8973214285714C491.5178571428571 1017.0200892857144 468.8058035714286 1013.7276785714286 403.5714285714286 948.5491071428572L17.2991071428571 562.1651785714287C5.9151785714286 550.8928571428571 0.2232142857143 536.0491071428572 0.2232142857143 521.2053571428571C0.2232142857143 506.5290178571429 5.859375 491.8526785714286 17.0758928571429 480.5803571428571zM15.6808035714286 18.6941964285714C38.1696428571429 -3.90625 74.8325892857143 -4.0736607142857 97.4888392857143 18.4151785714286C97.4888392857143 18.4151785714286 356.7522321428572 277.2879464285715 451.7299107142857 372.2656249999999C477.34375 397.8794642857142 480.078125 398.9955357142857 509.5982142857143 369.4196428571429C607.5892857142857 271.4285714285715 861.6629464285714 17.0758928571428 861.6629464285714 17.0758928571428C884.7098214285713 -5.859375 921.9308035714286 -5.6919642857143 944.8102678571428 17.3549107142857C967.6897321428572 40.4017857142857 967.578125 77.734375 944.53125 100.6138392857142L555.2455357142858 490.0111607142857C490.1227678571428 555.1339285714287 467.4107142857143 551.8415178571429 402.1763392857144 486.6629464285714L15.9040178571429 100.2790178571429C4.5200892857143 89.0066964285715 -1.171875 74.1629464285715 -1.171875 59.3191964285713C-1.1160714285714 44.6428571428571 4.4642857142857 29.9665178571429 15.6808035714286 18.6941964285714z" />
    <glyph glyph-name="histogram"
      unicode="&#xF147;"
      horiz-adv-x="1500" d="M1500 0V714.2857142857142L1285.7142857142858 500L1071.4285714285716 785.7142857142858L857.1428571428571 714.2857142857142L642.8571428571429 500L428.5714285714286 857.1428571428571L214.2857142857143 714.2857142857142L0 1000V0H1500z" />
    <glyph glyph-name="home"
      unicode="&#xF148;"
      horiz-adv-x="1199.609375" d="M0 466.6852678571428L599.7767857142857 1000L1199.5535714285713 466.6852678571428L999.7209821428572 466.6294642857142V0H733.0915178571429V333.3147321428571H466.4620535714286V0H199.8325892857143V466.6294642857143L0 466.6852678571428z" />
    <glyph glyph-name="home_manage"
      unicode="&#xF149;"
      horiz-adv-x="818.8616071428572" d="M725.6138392857143 0H165.6808035714286C114.1741071428571 0 72.8794642857143 41.8526785714286 72.8794642857143 93.5825892857143L73.2700892857143 726.1160714285714C72.65625 729.0736607142858 72.65625 732.03125 73.2700892857143 735.1004464285714L73.2700892857143 769.7544642857143C65.625 776.3392857142858 58.59375 783.5379464285714 53.0691964285714 791.7410714285714L31.1383928571428 782.6450892857142C19.7544642857143 777.9575892857143 6.6964285714285 783.3705357142858 1.953125 794.6986607142858C-2.734375 806.0825892857142 2.6785714285714 819.140625 14.0625 823.7723214285714L36.1049107142857 832.8683035714287C33.3705357142857 846.7075892857143 33.203125 861.1607142857143 35.9933035714286 875.390625L13.8392857142857 884.5424107142858C2.4553571428571 889.2857142857143 -2.9575892857143 902.2879464285714 1.7299107142857 913.6160714285714C6.4732142857143 925 19.53125 930.4129464285714 30.9151785714285 925.6696428571428L53.0133928571428 916.5736607142858C61.2723214285714 928.90625 71.7633928571428 939.1741071428572 83.6495535714285 947.1540178571428L74.4419642857143 969.3080357142856C69.6986607142857 980.6919642857144 75.1116071428571 993.6941964285714 86.4955357142857 998.3816964285714C97.8794642857143 1003.0691964285714 110.9375 997.65625 115.6808035714286 986.328125L124.8325892857143 964.3415178571428C138.7834821428572 967.0758928571428 153.2924107142857 967.2433035714286 167.6897321428571 964.453125L176.6741071428572 986.1049107142856C181.4174107142857 997.4888392857144 194.4754464285714 1002.9017857142856 205.859375 998.1584821428572C217.2433035714286 993.4151785714286 222.65625 980.4129464285714 217.96875 969.0848214285714L208.984375 947.4330357142858C214.1183035714285 943.9732142857144 218.6941964285714 939.9553571428572 223.1026785714286 935.8258928571428H539.1183035714286L818.8616071428571 655.078125V93.5825892857141C818.8616071428572 41.8526785714286 777.1205357142857 0 725.6138392857143 0zM263.1696428571429 190.0669642857142H628.7388392857142V277.7901785714286H263.1696428571428V190.0669642857142zM263.1696428571429 380.1339285714286H628.7388392857142V467.8571428571428H263.1696428571428V380.1339285714286zM106.6964285714286 870.703125C97.65625 848.828125 108.0915178571429 823.7723214285714 129.9665178571429 814.7321428571429C151.8973214285714 805.6919642857142 177.0089285714286 816.0714285714286 186.0491071428572 837.9464285714286C195.0892857142857 859.8214285714286 184.7098214285715 884.8772321428571 162.7790178571429 893.9174107142858C140.9040178571429 902.9017857142856 115.7924107142857 892.5223214285714 106.6964285714286 870.703125zM497.0982142857143 862.6674107142858V614.0625H745.703125L497.0982142857143 862.6674107142858z" />
    <glyph glyph-name="home_publish"
      unicode="&#xF14A;"
      horiz-adv-x="1408.0357142857142" d="M907.8125000000002 110.9375H111.1607142857143V777.9017857142857H907.8125V666.7410714285713H1018.9732142857144V889.0625C1018.9732142857144 958.1473214285714 963.1696428571428 1000 894.2522321428572 1000H124.7209821428572C55.8035714285714 1000 0 944.0290178571428 0 875V125C0 55.9151785714286 55.8035714285714 0 124.7209821428572 0H894.1964285714287C963.1138392857144 0 1018.9174107142858 41.8526785714286 1018.9174107142858 110.9375V222.0982142857142H907.7566964285714V110.9375zM1408.0357142857142 462.6674107142857L1148.6607142857142 703.7946428571429V500H648.4375V388.8392857142857H1148.6607142857144V203.5714285714286L1408.0357142857142 462.6674107142857z" />
    <glyph glyph-name="home_share"
      unicode="&#xF14B;"
      horiz-adv-x="935.9933035714286" d="M755.9709821428572 357.1428571428571C701.4508928571429 357.1428571428571 652.5111607142857 333.0357142857142 619.4754464285716 294.9776785714286L351.0044642857143 408.59375C356.8080357142857 426.1160714285715 359.9330357142857 444.8660714285715 359.9330357142857 464.3415178571429C359.9330357142857 491.9642857142857 353.6272321428572 518.1361607142858 342.3549107142857 541.5178571428571L623.2700892857143 700.8928571428572C656.1941964285714 665.2901785714287 703.4598214285716 642.96875 755.9709821428572 642.96875C855.4129464285716 642.96875 935.9933035714286 722.9352678571429 935.9933035714286 821.5401785714287C935.9933035714286 920.1450892857144 855.4129464285716 1000.1116071428572 755.9709821428573 1000.1116071428572C656.529017857143 1000.1116071428572 575.9486607142859 920.1450892857144 575.9486607142859 821.5401785714287C575.9486607142859 795.0334821428572 581.8080357142859 769.8660714285716 592.2433035714288 747.2098214285716L310.4910714285717 587.3883928571429C277.6785714285717 621.6517857142858 231.3616071428574 643.0245535714287 179.9665178571431 643.0245535714287C80.5803571428571 642.8013392857142 0 562.890625 0 464.2857142857142C0 365.6808035714286 80.5803571428572 285.7142857142857 180.0223214285715 285.7142857142857C240.2901785714286 285.7142857142857 293.6383928571429 315.1227678571429 326.2834821428572 360.2120535714286L590.2901785714287 248.5491071428572C581.0825892857144 227.0647321428572 575.9486607142858 203.4040178571429 575.9486607142858 178.5714285714286C575.9486607142858 79.9665178571428 656.529017857143 0 755.9709821428573 0S935.9933035714286 79.9665178571428 935.9933035714286 178.5714285714286C935.9933035714286 277.1763392857144 855.4129464285716 357.1428571428571 755.9709821428572 357.1428571428571z" />
    <glyph glyph-name="home_upload"
      unicode="&#xF14C;"
      horiz-adv-x="1470.2566964285713" d="M1185.7142857142858 624.8325892857142C1144.029017857143 838.671875 958.1473214285714 1000 735.1562500000001 1000C558.091517857143 1000 404.6316964285715 898.046875 327.7901785714286 749.1629464285714C143.6383928571429 728.9620535714286 0 570.7589285714286 0 378.4040178571428C0 172.3214285714286 153.6830357142857 0 356.7522321428571 0H681.0825892857143V306.3058035714286H482.8683035714286L742.3549107142857 567.578125L987.3883928571428 306.3058035714286H789.1741071428572V0H1167.5781250000002C1336.6629464285716 0 1470.2566964285716 144.6428571428571 1470.2566964285716 316.2388392857144C1470.2566964285713 480.3013392857143 1344.3638392857142 613.3370535714286 1185.7142857142858 624.8325892857142z" />
    <glyph glyph-name="info"
      unicode="&#xF14D;"
      horiz-adv-x="999.8883928571429" d="M499.9441964285715 999.8883928571428C223.828125 999.8883928571428 0 776.0602678571429 0 499.9441964285714S223.828125 0 499.9441964285714 0C776.0602678571429 0 999.8883928571428 223.8281249999999 999.8883928571428 499.9441964285714S776.0602678571429 999.8883928571428 499.9441964285715 999.8883928571428zM562.4441964285714 250C562.4441964285714 215.4575892857142 534.4866071428572 187.5 499.9441964285715 187.5C465.4017857142858 187.5 437.4441964285715 215.4575892857142 437.4441964285715 250V562.4441964285713C437.4441964285715 596.9866071428571 465.4017857142858 624.9441964285713 499.9441964285715 624.9441964285713C534.4308035714286 624.9441964285713 562.4441964285714 596.9866071428571 562.4441964285714 562.4441964285713V250zM501.0602678571429 686.328125C465.9040178571429 686.328125 437.4441964285715 714.7879464285713 437.4441964285715 749.9441964285714C437.4441964285715 785.1004464285714 465.9040178571429 813.5602678571429 501.0602678571429 813.5602678571429C536.1607142857143 813.5602678571429 564.6763392857143 785.1004464285714 564.6763392857143 749.9441964285714C564.6763392857143 714.7879464285713 536.2165178571429 686.328125 501.0602678571429 686.328125z" />
    <glyph glyph-name="institutional_account"
      unicode="&#xF14E;"
      horiz-adv-x="904.7433035714286" d="M904.7433035714286 333.3147321428571V583.3147321428571C904.7433035714286 629.4084821428571 867.9687500000001 666.6294642857142 822.4888392857143 666.6294642857142H740.234375V750C740.234375 842.0758928571429 666.5736607142858 916.6852678571428 575.7254464285714 916.6852678571428C575.7254464285714 962.7232142857142 538.9508928571429 1000 493.4709821428572 1000H411.2723214285715C365.7924107142857 1000 329.0178571428572 962.7232142857144 329.0178571428572 916.6852678571428C238.1696428571429 916.6852678571428 164.5089285714286 842.0758928571429 164.5089285714286 750V666.6852678571429H82.2544642857143C36.7745535714286 666.6852678571429 0 629.4084821428571 0 583.3705357142858V333.3705357142858V0H82.2544642857143H329.0178571428571H575.78125H822.5446428571429H904.7991071428572M287.8906250000001 833.3147321428571C310.5468750000001 833.3147321428571 329.0178571428572 814.6205357142858 329.0178571428572 791.6294642857143S310.546875 750 287.8906250000001 750C265.2343750000001 750 246.7633928571429 768.6941964285714 246.7633928571429 791.6852678571429S265.234375 833.3147321428571 287.8906250000001 833.3147321428571zM452.3995535714286 833.3147321428571C475.1116071428572 833.3147321428571 493.5267857142858 814.6205357142858 493.5267857142858 791.6294642857143S475.1116071428572 749.9441964285714 452.3995535714286 749.9441964285714C429.7433035714286 749.9441964285714 411.2723214285715 768.6383928571429 411.2723214285715 791.6294642857143S429.7433035714286 833.3147321428571 452.3995535714286 833.3147321428571zM616.8526785714287 833.3147321428571C639.6205357142858 833.3147321428571 657.9799107142858 814.6205357142858 657.9799107142858 791.6294642857143S639.5647321428572 749.9441964285714 616.8526785714287 749.9441964285714C594.140625 749.9441964285714 575.7254464285714 768.6383928571429 575.7254464285714 791.6294642857143S594.140625 833.3147321428571 616.8526785714287 833.3147321428571zM616.8526785714287 666.6852678571429C639.6205357142858 666.6852678571429 657.9799107142858 647.9910714285714 657.9799107142858 625C657.9799107142858 601.953125 639.5647321428572 583.3147321428571 616.8526785714287 583.3147321428571C594.140625 583.3147321428571 575.7254464285714 601.953125 575.7254464285714 625C575.7254464285714 647.9352678571429 594.140625 666.6852678571429 616.8526785714287 666.6852678571429zM452.3995535714286 666.6852678571429C475.1116071428572 666.6852678571429 493.5267857142858 647.9910714285714 493.5267857142858 625C493.5267857142858 601.953125 475.1116071428572 583.3147321428571 452.3995535714286 583.3147321428571C429.7433035714286 583.3147321428571 411.2723214285715 601.953125 411.2723214285715 625C411.2723214285715 647.9352678571429 429.7433035714286 666.6852678571429 452.3995535714286 666.6852678571429zM287.8906250000001 666.6852678571429C310.5468750000001 666.6852678571429 329.0178571428572 647.9910714285714 329.0178571428572 625C329.0178571428572 601.953125 310.5468750000001 583.3147321428571 287.8906250000001 583.3147321428571C265.2343750000001 583.3147321428571 246.7633928571429 601.953125 246.7633928571429 625C246.7633928571429 647.9352678571429 265.234375 666.6852678571429 287.8906250000001 666.6852678571429zM287.8906250000001 500C310.5468750000001 500 329.0178571428572 481.3616071428571 329.0178571428572 458.3147321428571C329.0178571428572 435.2678571428571 310.5468750000001 416.6852678571429 287.8906250000001 416.6852678571429C265.2343750000001 416.6852678571429 246.7633928571429 435.3236607142857 246.7633928571429 458.3147321428571C246.7633928571429 481.3616071428571 265.234375 500 287.8906250000001 500zM452.3995535714286 500C475.1116071428572 500 493.5267857142858 481.3616071428571 493.5267857142858 458.3147321428571C493.5267857142858 435.2678571428571 475.1116071428572 416.6852678571429 452.3995535714286 416.6852678571429C429.7433035714286 416.6852678571429 411.2723214285715 435.3236607142857 411.2723214285715 458.3147321428571C411.2723214285715 481.3616071428571 429.7433035714286 500 452.3995535714286 500zM616.8526785714287 500C639.6205357142858 500 657.9799107142858 481.3616071428571 657.9799107142858 458.3147321428571C657.9799107142858 435.2678571428571 639.5647321428572 416.6852678571429 616.8526785714287 416.6852678571429C594.140625 416.6852678571429 575.7254464285714 435.3236607142857 575.7254464285714 458.3147321428571C575.7254464285714 481.3616071428571 594.140625 500 616.8526785714287 500zM123.3816964285715 583.3147321428571C146.0379464285715 583.3147321428571 164.5089285714286 564.6763392857142 164.5089285714286 541.6294642857142C164.5089285714286 518.5825892857142 146.0379464285714 499.9441964285713 123.3816964285715 499.9441964285713C100.7254464285715 499.9441964285713 82.2544642857143 518.5825892857142 82.2544642857143 541.6294642857142C82.2544642857143 564.6763392857142 100.7254464285714 583.3147321428571 123.3816964285715 583.3147321428571zM123.3816964285715 416.6852678571429C146.0379464285715 416.6852678571429 164.5089285714286 398.046875 164.5089285714286 375C164.5089285714286 351.953125 146.0379464285714 333.3147321428571 123.3816964285715 333.3147321428571C100.7254464285715 333.3147321428571 82.2544642857143 351.953125 82.2544642857143 375C82.2544642857143 398.046875 100.7254464285714 416.6852678571429 123.3816964285715 416.6852678571429zM781.3616071428572 583.3147321428571C804.0736607142858 583.3147321428571 822.4888392857143 564.6763392857142 822.4888392857143 541.6294642857142C822.4888392857143 518.5825892857142 804.0736607142858 499.9441964285713 781.3616071428572 499.9441964285713C758.6495535714286 499.9441964285713 740.234375 518.5825892857142 740.234375 541.6294642857142C740.234375 564.6763392857142 758.6495535714286 583.3147321428571 781.3616071428572 583.3147321428571zM781.3616071428572 416.6852678571429C804.0736607142858 416.6852678571429 822.4888392857143 398.046875 822.4888392857143 375C822.4888392857143 351.953125 804.0736607142858 333.3147321428571 781.3616071428572 333.3147321428571C758.6495535714286 333.3147321428571 740.234375 351.953125 740.234375 375C740.234375 398.046875 758.6495535714286 416.6852678571429 781.3616071428572 416.6852678571429zM616.8526785714287 333.3147321428571C639.6205357142858 333.3147321428571 657.9799107142858 314.6763392857142 657.9799107142858 291.6294642857142C657.9799107142858 268.5825892857142 639.5647321428572 249.9441964285714 616.8526785714287 249.9441964285714C594.140625 249.9441964285714 575.7254464285714 268.5825892857142 575.7254464285714 291.6294642857142C575.7254464285714 314.6763392857144 594.140625 333.3147321428571 616.8526785714287 333.3147321428571zM452.3995535714286 333.3147321428571C475.1116071428572 333.3147321428571 493.5267857142858 314.6763392857142 493.5267857142858 291.6294642857142C493.5267857142858 268.5825892857142 475.1116071428572 249.9441964285714 452.3995535714286 249.9441964285714C429.7433035714286 249.9441964285714 411.2723214285715 268.5825892857142 411.2723214285715 291.6294642857142C411.2723214285715 314.6763392857144 429.7433035714286 333.3147321428571 452.3995535714286 333.3147321428571zM287.8906250000001 333.3147321428571C310.5468750000001 333.3147321428571 329.0178571428572 314.6763392857142 329.0178571428572 291.6294642857142C329.0178571428572 268.5825892857142 310.5468750000001 249.9441964285714 287.8906250000001 249.9441964285714C265.2343750000001 249.9441964285714 246.7633928571429 268.5825892857142 246.7633928571429 291.6294642857142C246.7633928571429 314.6763392857144 265.234375 333.3147321428571 287.8906250000001 333.3147321428571zM287.8906250000001 166.6294642857142H616.9084821428571C639.6763392857143 166.6294642857142 658.0357142857143 147.9910714285713 658.0357142857143 125C658.0357142857143 101.953125 639.6205357142858 83.3147321428571 616.9084821428571 83.3147321428571H287.890625C265.234375 83.3147321428571 246.7633928571429 101.953125 246.7633928571429 125C246.7633928571429 148.046875 265.234375 166.6294642857142 287.8906250000001 166.6294642857142zM123.3816964285715 250C146.0379464285715 250 164.5089285714286 231.3616071428571 164.5089285714286 208.3147321428571C164.5089285714286 185.2678571428571 146.0379464285714 166.6294642857142 123.3816964285715 166.6294642857142C100.7254464285715 166.6294642857142 82.2544642857143 185.2678571428571 82.2544642857143 208.3147321428571C82.2544642857143 231.3616071428571 100.7254464285714 250 123.3816964285715 250zM781.3616071428572 250C804.0736607142858 250 822.4888392857143 231.3616071428571 822.4888392857143 208.3147321428571C822.4888392857143 185.2678571428571 804.0736607142858 166.6294642857142 781.3616071428572 166.6294642857142C758.6495535714286 166.6294642857142 740.234375 185.2678571428571 740.234375 208.3147321428571C740.234375 231.3616071428571 758.6495535714286 250 781.3616071428572 250z" />
    <glyph glyph-name="linked"
      unicode="&#xF14F;"
      horiz-adv-x="1278.404017857143" d="M1197.3214285714284 918.0245535714286C1089.2299107142858 1027.2879464285716 913.9508928571428 1027.232142857143 805.8035714285714 918.0803571428572L582.9241071428571 692.6897321428571C512.1651785714286 621.2611607142857 488.671875 520.8705357142857 510.4910714285714 429.2410714285714L903.6830357142858 819.140625C957.7566964285716 873.7165178571429 1045.3683035714287 873.7723214285714 1099.497767857143 819.0848214285713C1153.5156250000002 764.453125 1153.5156250000002 675.8928571428571 1099.497767857143 621.2611607142857L702.1763392857144 227.34375C795.8147321428572 200.1116071428572 900.6138392857143 222.4888392857143 974.3861607142858 297.0982142857144L1197.3772321428573 522.4330357142858C1305.4129464285716 631.5848214285713 1305.46875 808.7611607142858 1197.3214285714284 918.0245535714286zM767.9129464285714 570.6473214285714L374.7209821428572 180.8035714285715C320.7031250000001 126.2276785714286 233.0357142857143 126.2276785714286 178.9620535714286 180.8593749999999C124.9441964285714 235.4910714285715 124.9441964285714 324.0513392857142 178.9620535714286 378.7388392857142L576.2834821428572 772.6004464285713C482.6450892857143 799.7767857142857 377.8459821428572 777.4553571428571 304.0736607142857 702.8459821428571L81.0825892857143 477.5111607142858C-27.0089285714286 368.3035714285715 -27.0647321428572 191.1830357142857 81.0825892857143 81.9196428571429C189.1741071428571 -27.34375 364.5089285714286 -27.2879464285713 472.6004464285714 81.8638392857145L695.5357142857143 307.1986607142858C766.2388392857142 378.6830357142857 789.7879464285714 479.0736607142857 767.9129464285714 570.6473214285714z" />
    <glyph glyph-name="list"
      unicode="&#xF150;"
      horiz-adv-x="1333.2589285714284" d="M83.3147321428572 1000C37.2767857142858 1000 0 962.6674107142856 0 916.6852678571428C0 870.6473214285714 37.3325892857143 833.3705357142857 83.3147321428572 833.3705357142857C129.3526785714286 833.3705357142857 166.6294642857143 870.703125 166.6294642857143 916.6852678571428C166.6852678571429 962.6674107142858 129.3526785714286 1000 83.3147321428572 1000zM83.3147321428572 583.3705357142857C37.2767857142858 583.3705357142857 0 546.09375 0 500.0558035714285S37.3325892857143 416.7410714285714 83.3147321428572 416.7410714285714C129.3526785714286 416.7410714285714 166.6294642857143 454.0736607142857 166.6294642857143 500.0558035714285S129.3526785714286 583.3705357142857 83.3147321428572 583.3705357142857zM83.3147321428572 166.6852678571429C37.2767857142858 166.6852678571429 0 129.3526785714286 0 83.3705357142858C0 37.3883928571428 37.3325892857143 0.0558035714287 83.3147321428572 0.0558035714287C129.3526785714286 0.0558035714287 166.6294642857143 37.388392857143 166.6294642857143 83.3705357142858C166.6852678571429 129.4084821428572 129.3526785714286 166.6852678571429 83.3147321428572 166.6852678571429zM416.6294642857144 833.3147321428571H1249.9441964285716C1295.982142857143 833.3147321428571 1333.2589285714287 870.6473214285714 1333.2589285714287 916.6294642857144C1333.2589285714287 962.6674107142858 1295.9263392857144 999.9441964285714 1249.9441964285716 999.9441964285714H416.6294642857144C370.5915178571429 999.9441964285714 333.3147321428571 962.6116071428572 333.3147321428571 916.6294642857144C333.3147321428571 870.6473214285714 370.6473214285715 833.3147321428571 416.6294642857144 833.3147321428571zM1249.9441964285716 583.3705357142857H416.6294642857144C370.5915178571429 583.3705357142857 333.3147321428571 546.09375 333.3147321428571 500.0558035714285S370.6473214285714 416.7410714285714 416.6294642857142 416.7410714285714H1249.9441964285713C1295.9821428571427 416.7410714285714 1333.2589285714284 454.0736607142857 1333.2589285714284 500.0558035714285S1295.982142857143 583.3705357142857 1249.9441964285716 583.3705357142857zM1249.9441964285716 166.6852678571429H416.6294642857144C370.5915178571429 166.6852678571429 333.3147321428571 129.3526785714286 333.3147321428571 83.3705357142858C333.3147321428571 37.3883928571428 370.6473214285714 0.0558035714287 416.6294642857142 0.0558035714287H1249.9441964285713C1295.9821428571427 0.0558035714287 1333.2589285714284 37.388392857143 1333.2589285714284 83.3705357142858C1333.2589285714284 129.4084821428572 1295.982142857143 166.6852678571429 1249.9441964285716 166.6852678571429z" />
    <glyph glyph-name="list_view-alternative"
      unicode="&#xF151;"
      horiz-adv-x="999.8325892857143" d="M932.2544642857142 1000H67.578125C30.2455357142857 1000 0 970.1450892857144 0 933.3147321428572V866.6294642857142C0 829.7991071428571 30.2455357142857 799.9441964285714 67.578125 799.9441964285714H932.2544642857142C969.5870535714286 799.9441964285714 999.8325892857142 829.7991071428571 999.8325892857142 866.6294642857142V933.3147321428572C999.8325892857142 970.1450892857144 969.5870535714286 1000 932.2544642857142 1000zM932.2544642857142 200H67.578125C30.2455357142857 200 0 170.1450892857143 0 133.3147321428572V66.6294642857143C0 29.7991071428571 30.2455357142857 -0.0558035714286 67.578125 -0.0558035714286H932.2544642857142C969.5870535714286 -0.0558035714286 999.8325892857142 29.7991071428571 999.8325892857142 66.6294642857143V133.3147321428572C999.8325892857142 170.1450892857143 969.5870535714286 200 932.2544642857142 200zM932.2544642857142 600H67.578125C30.2455357142857 600 0 570.1450892857142 0 533.3147321428571V466.6294642857143C0 429.7991071428571 30.2455357142857 399.9441964285715 67.578125 399.9441964285715H932.2544642857142C969.5870535714286 399.9441964285715 999.8325892857142 429.7991071428571 999.8325892857142 466.6294642857143V533.3147321428571C999.8325892857142 570.1450892857142 969.5870535714286 600 932.2544642857142 600z" />
    <glyph glyph-name="list_view"
      unicode="&#xF152;"
      horiz-adv-x="1285.7142857142858" d="M0 742.8571428571429H1285.7142857142858V1000H0V742.8571428571429zM1285.7142857142858 0V257.1428571428571H0V0H1285.7142857142858zM0 371.4285714285714H1285.7142857142858V628.5714285714286H0V371.4285714285714z" />
    <glyph glyph-name="load_more"
      unicode="&#xF153;"
      horiz-adv-x="961.9419642857143" d="M0.2232142857143 477.9017857142857C0.2232142857143 463.0580357142857 5.9151785714286 448.2142857142857 17.2991071428571 436.9419642857142L403.5714285714286 50.5580357142857C468.75 -14.6205357142858 491.4620535714286 -17.9129464285716 556.640625 47.2098214285713L945.9263392857142 436.6071428571428C968.9732142857142 459.5424107142857 969.0848214285714 496.8191964285713 946.2053571428572 519.8660714285713C923.3258928571428 542.9129464285713 886.1049107142857 543.0803571428571 863.0580357142857 520.1450892857142C863.0580357142857 520.1450892857142 608.9285714285713 265.7924107142857 510.9933035714286 167.8013392857141C481.4174107142857 138.2254464285713 478.7388392857143 139.2857142857142 453.1249999999999 164.9553571428571C358.1473214285714 259.9330357142857 98.8839285714286 518.8058035714286 98.8839285714286 518.8058035714286C76.2276785714286 541.2946428571429 39.5647321428571 541.1272321428571 17.0758928571429 518.5267857142858C5.859375 507.3102678571428 0.2232142857143 492.578125 0.2232142857143 477.9017857142857zM-1.1160714285714 939.7321428571428C-1.1160714285714 924.8883928571428 4.5758928571429 910.0446428571428 15.9598214285714 898.7723214285714L402.2321428571429 512.4441964285713C467.4107142857142 447.265625 490.1227678571428 443.9732142857142 555.3013392857142 509.0959821428571L944.5870535714284 898.4933035714286C967.6339285714284 921.4285714285714 967.7455357142856 958.7053571428572 944.8660714285714 981.7522321428572C921.9866071428572 1004.7991071428572 884.7656249999999 1004.9665178571428 861.7187499999999 982.03125C861.7187499999999 982.03125 607.5892857142856 727.6785714285713 509.6540178571428 629.6874999999999C480.0781249999999 600.1116071428571 477.3995535714285 601.1718749999999 451.7857142857142 626.8415178571428C356.8080357142857 721.8191964285714 97.4888392857143 980.6919642857142 97.4888392857143 980.6919642857142C74.8325892857143 1003.1808035714286 38.1696428571429 1003.0133928571428 15.6808035714286 980.4129464285714C4.4642857142857 969.1964285714286 -1.1160714285714 954.4642857142856 -1.1160714285714 939.7321428571428z" />
    <glyph glyph-name="location_pin"
      unicode="&#xF154;"
      horiz-adv-x="562.1651785714286" d="M282.1428571428571 999.609375C78.3482142857143 999.609375 -39.6205357142858 800.7254464285714 12.1651785714285 634.7098214285713C73.4375 438.3928571428571 282.1428571428571 0 282.1428571428571 0S484.7098214285714 439.3415178571428 548.9955357142857 634.7098214285714C604.4084821428571 803.0691964285714 478.9062500000001 999.609375 282.1428571428571 999.609375zM282.1428571428571 522.4330357142858C172.9910714285714 522.4330357142858 84.375 610.9933035714287 84.375 720.2008928571429S172.9352678571428 917.96875 282.1428571428571 917.96875C391.2946428571429 917.96875 479.9107142857142 829.4084821428571 479.9107142857142 720.2008928571429S391.2946428571429 522.4330357142858 282.1428571428571 522.4330357142858z" />
    <glyph glyph-name="locked"
      unicode="&#xF155;"
      horiz-adv-x="778.4598214285714" d="M681.1383928571428 566.7410714285714H648.7165178571429V766.7410714285714C648.7165178571429 895.6473214285714 532.5334821428571 1000.1116071428572 389.2299107142857 1000.1116071428572C245.9263392857143 1000.1116071428572 129.7433035714286 895.6473214285714 129.7433035714286 766.7410714285714V566.7410714285714H97.3214285714286C43.5825892857143 566.7410714285714 0 521.9866071428571 0 466.7410714285714V100C0 44.7544642857143 43.5825892857143 0 97.3214285714286 0H681.138392857143C734.8772321428572 0 778.4598214285716 44.7544642857143 778.4598214285716 100V466.7410714285714C778.4598214285714 521.9866071428571 734.8772321428572 566.7410714285714 681.1383928571428 566.7410714285714zM387.3883928571429 201.0044642857143C315.7366071428572 201.0602678571428 257.7008928571429 260.7700892857142 257.7008928571429 334.4308035714286C257.7566964285715 408.0915178571428 315.8482142857144 467.7455357142857 387.5000000000001 467.6897321428571C459.1517857142858 467.6339285714284 517.1875 407.9241071428571 517.1875 334.2633928571429C517.1316964285714 260.6584821428571 459.0401785714286 201.0044642857143 387.3883928571429 201.0044642857143zM518.9732142857143 600.0558035714286C518.9732142857143 588.3928571428571 516.3504464285714 577.1763392857142 511.6071428571428 566.7410714285713H266.8526785714286C262.0535714285715 577.1763392857142 259.4866071428572 588.3928571428571 259.4866071428572 600.0558035714286V766.7410714285714C259.4866071428572 821.9866071428571 317.578125 866.7410714285714 389.2299107142857 866.7410714285714C460.8816964285715 866.796875 518.9732142857143 821.9866071428571 518.9732142857143 766.7410714285714V600.0558035714286z" />
    <glyph glyph-name="metadata"
      unicode="&#xF156;"
      horiz-adv-x="874.8883928571429" d="M824.8883928571429 0H50C22.3772321428572 0 0 22.3772321428571 0 49.9999999999999V824.8883928571429C0 852.5111607142857 22.3772321428572 874.8883928571429 50 874.8883928571429H304.3526785714286C310.7700892857144 944.8660714285714 367.5223214285715 999.8883928571428 437.4441964285715 999.8883928571428S564.1183035714287 944.921875 570.5357142857143 874.8883928571429H824.8883928571429C852.5111607142858 874.8883928571429 874.8883928571429 852.5111607142857 874.8883928571429 824.8883928571429V49.9999999999999C874.888392857143 22.3772321428571 852.5111607142857 0 824.8883928571429 0zM199.9441964285714 199.9441964285713H524.8883928571428V299.9441964285715H199.9441964285714V199.9441964285713zM699.8883928571429 649.8883928571429H199.9441964285714V549.8883928571429H699.8883928571428V649.8883928571429zM699.8883928571429 474.9441964285714H199.9441964285714V374.9441964285715H699.8883928571428V474.9441964285714zM437.4441964285715 931.0825892857142C474.5535714285715 931.0825892857142 504.6316964285715 900.1116071428571 504.6316964285715 861.9419642857142S474.5535714285715 792.8013392857142 437.4441964285715 792.8013392857142C400.3348214285715 792.8013392857142 370.2566964285715 823.7723214285714 370.2566964285715 861.9419642857142S400.3348214285715 931.0825892857142 437.4441964285715 931.0825892857142z" />
    <glyph glyph-name="minus_icon"
      unicode="&#xF157;"
      horiz-adv-x="999.8883928571429" d="M999.8883928571428 400.4464285714286C666.5922433035714 400.4464285714286 333.2961328125 400.4464285714286 0 400.4464285714286L0 600.2232142857142C333.3147321428571 600.2232142857142 666.6294642857142 600.2232142857142 999.9441964285716 600.2232142857142L999.9441964285716 400.4464285714286zM1.0080636160714 598.7903404017857H1002.0160993303572V401.2096986607143H1.0080636160715V598.7903404017857z" />
    <glyph glyph-name="molecule"
      unicode="&#xF158;"
      horiz-adv-x="881.1383928571429" d="M789.8995535714286 548.3258928571429C748.1026785714286 549.8883928571429 711.6071428571429 524.1629464285714 697.65625 487.0535714285714C695.7589285714286 487.6116071428571 693.8058035714286 488.1138392857143 691.796875 488.6160714285714L544.5870535714287 514.7321428571429L544.5870535714287 511.1607142857143C544.5870535714287 511.4397321428571 544.6428571428572 511.71875 544.6428571428572 511.9977678571428C544.6428571428572 561.2723214285713 511.6629464285715 600.78125 464.7321428571429 609.5424107142857L468.3593750000001 608.8169642857142L489.1741071428572 785.6026785714286C547.8236607142858 786.2165178571429 595.200892857143 833.8727678571429 595.200892857143 892.6339285714286C595.200892857143 951.8415178571428 547.2098214285716 999.7767857142858 488.0580357142858 999.7767857142858C428.8504464285715 999.7767857142858 380.9151785714287 951.7857142857144 380.9151785714287 892.6339285714286C380.9151785714287 843.2477678571429 414.5089285714287 802.1205357142858 459.9330357142858 789.6763392857143L442.1316964285716 610.9933035714287L445.1450892857144 611.1049107142858C391.4620535714287 609.9330357142858 349.7209821428572 566.0714285714287 349.7209821428572 511.9419642857143C349.7209821428572 508.9843750000001 350.3348214285716 506.1941964285715 350.5580357142859 503.3482142857143L350.0000000000001 508.0357142857143L207.2544642857144 477.2879464285714C192.075892857143 518.0803571428571 153.1808035714287 547.265625 107.0870535714287 547.265625C47.8794642857144 547.265625 -0.0558035714285 499.2745535714286 -0.0558035714285 440.1227678571429S47.9352678571429 332.9799107142857 107.0870535714287 332.9799107142857C166.2946428571429 332.9799107142857 214.2299107142858 380.9709821428571 214.2299107142858 440.1227678571429C214.2299107142858 443.6941964285715 213.5044642857144 446.9866071428571 213.169642857143 450.4464285714286L354.5200892857144 482.5334821428571L353.7388392857144 485.2120535714286C365.2343750000001 443.4151785714286 402.5111607142859 412.5558035714286 447.2098214285716 412.5558035714286C449.888392857143 412.5558035714286 452.4553571428573 413.0580357142858 455.0781250000001 413.2812499999999L450.279017857143 412.7232142857142L482.3660714285716 257.8683035714286C424.3303571428573 244.4196428571429 380.9151785714287 192.6897321428571 380.9151785714287 130.5245535714285C380.9151785714287 58.203125 439.5647321428572 -0.4464285714287 511.8861607142858 -0.4464285714287S642.8571428571429 58.3147321428571 642.8571428571429 130.6361607142857S584.2075892857143 261.6071428571428 511.8861607142857 261.6071428571428C510.7142857142857 261.6071428571428 509.6540178571428 261.2723214285715 508.4821428571428 261.2723214285715L475.3348214285714 417.3549107142858L472.3214285714286 416.4620535714287C508.59375 426.5625 536.9977678571429 459.3750000000001 542.96875 495.3125000000001L542.3549107142857 492.0758928571429L691.6294642857143 464.0625000000001C691.796875 464.0066964285714 691.9084821428572 463.9508928571429 692.0758928571429 463.9508928571429C691.8526785714286 461.7745535714287 691.6852678571429 459.5424107142858 691.5736607142857 457.2544642857143C689.6205357142857 404.9665178571429 730.3571428571429 360.9933035714286 782.6450892857143 359.0401785714286S878.90625 397.8236607142858 880.859375 450.1116071428572S842.1316964285714 546.3169642857142 789.8995535714286 548.3258928571429z" />
    <glyph glyph-name="next"
      unicode="&#xF159;"
      horiz-adv-x="645.3936469062883" d="M83.3033206079378 992.5808542822156C132.5001800763524 965.317294532882 588.3814737448679 582.4389541165455 622.4159043434416 548.1884318951236C652.884823165022 517.5394367211697 653.1729453288195 482.712670172153 622.4159043434416 451.9196139163005C575.1638694806599 404.6675790535187 118.8864078369228 27.911834617878 85.0320535907224 8.5356191024996C42.7861413239213 -15.6306273860115 0 15.0903983288915 0 56.399913563351C0 88.2374126629693 0 898.7970899661458 0 944.9686667146872C0 990.4919685946842 47.2520348627818 1012.6053446661384 83.3033206079378 992.5808542822156z" />
    <glyph glyph-name="note"
      unicode="&#xF15A;"
      horiz-adv-x="1030.9151785714287" d="M368.5825892857143 341.8526785714286L585.3236607142858 403.3482142857142L430.46875 557.1986607142858L368.5825892857143 341.8526785714286zM947.4888392857144 763.2254464285714L792.6339285714287 917.0200892857144L452.0089285714286 578.6272321428571L606.8638392857143 424.8325892857142L947.4888392857144 763.2254464285714zM902.34375 193.6383928571429C902.34375 140.1785714285715 858.984375 96.8191964285715 805.4687500000001 96.8191964285715H193.6941964285715C140.234375 96.8191964285715 96.875 140.1785714285715 96.875 193.6383928571429V774.4419642857142C96.875 827.9017857142858 140.234375 871.2611607142858 193.6941964285715 871.2611607142858H611.8303571428572L708.6495535714287 968.0803571428572H161.3839285714286C72.265625 968.0803571428572 0 895.8147321428571 0 806.7522321428571V161.3839285714286C0 72.265625 72.265625 0.0558035714284 161.3839285714286 0.0558035714284H837.7790178571429C926.8973214285714 0.0558035714284 999.1629464285714 72.265625 999.1629464285714 161.3839285714285V677.6227678571429L902.34375 580.859375V193.6383928571429zM1019.4196428571428 892.4107142857142L922.65625 988.5602678571428C906.6406249999998 1004.4642857142856 879.7433035714284 1003.6272321428572 862.6674107142857 986.6629464285714L816.2388392857142 940.5133928571428L971.0937499999998 786.71875L1017.5223214285714 832.8683035714286C1034.5982142857142 849.7767857142858 1035.4910714285713 876.4508928571429 1019.4196428571428 892.4107142857142z" />
    <glyph glyph-name="notification_bell"
      unicode="&#xF15B;"
      horiz-adv-x="857.1428571428571" d="M756.1941964285714 571.4285714285714V294.2522321428571L857.1428571428571 193.359375V142.8571428571429H0V193.359375L100.9486607142857 294.2522321428571V571.4285714285714C100.9486607142857 726.171875 208.0915178571429 855.7477678571429 352.8459821428571 889.9553571428571V924.2745535714286C352.8459821428571 966.1272321428572 386.6629464285715 1000 428.5714285714286 1000S504.2968750000001 966.1272321428572 504.2968750000001 924.2745535714286V889.9553571428571C649.0513392857142 855.7477678571429 756.1941964285714 726.171875 756.1941964285714 571.4285714285714zM428.5714285714286 0C376.171875 0 333.8169642857144 42.4107142857142 333.8169642857144 94.7544642857142H523.3816964285714C523.3258928571429 42.4107142857142 480.9151785714286 0 428.5714285714286 0z" />
    <glyph glyph-name="onlineresource"
      unicode="&#xF15C;"
      horiz-adv-x="1607.142857142857" d="M839.2857142857143 464.2857142857142H1125C1135.7142857142858 464.2857142857142 1142.857142857143 471.4285714285714 1142.857142857143 482.1428571428571S1135.7142857142858 500 1125 500H839.2857142857143C828.5714285714286 500 821.4285714285714 492.8571428571429 821.4285714285714 482.1428571428571S828.5714285714286 464.2857142857142 839.2857142857143 464.2857142857142zM839.2857142857143 571.4285714285714H1017.8571428571428C1028.5714285714287 571.4285714285714 1035.7142857142858 578.5714285714286 1035.7142857142858 589.2857142857142S1028.5714285714287 607.1428571428571 1017.8571428571428 607.1428571428571H839.2857142857143C828.5714285714286 607.1428571428571 821.4285714285714 600 821.4285714285714 589.2857142857142S828.5714285714286 571.4285714285714 839.2857142857143 571.4285714285714zM482.1428571428572 464.2857142857142H696.4285714285714C707.1428571428572 464.2857142857142 714.2857142857143 471.4285714285714 714.2857142857143 482.1428571428571V696.4285714285714C714.2857142857143 707.1428571428571 707.1428571428572 714.2857142857142 696.4285714285714 714.2857142857142H482.1428571428572C471.4285714285714 714.2857142857142 464.2857142857143 707.1428571428571 464.2857142857143 696.4285714285714V482.1428571428571C464.2857142857143 471.4285714285714 471.4285714285714 464.2857142857142 482.1428571428572 464.2857142857142zM500 678.5714285714286H678.5714285714286V500H500V678.5714285714286zM839.2857142857143 678.5714285714286H1125C1135.7142857142858 678.5714285714286 1142.857142857143 685.7142857142857 1142.857142857143 696.4285714285714S1135.7142857142858 714.2857142857142 1125 714.2857142857142H839.2857142857143C828.5714285714286 714.2857142857142 821.4285714285714 707.1428571428571 821.4285714285714 696.4285714285714S828.5714285714286 678.5714285714286 839.2857142857143 678.5714285714286zM1500 1000H107.1428571428571C46.4285714285714 1000 0 953.5714285714286 0 892.8571428571429V107.1428571428571C0 46.4285714285714 46.4285714285714 0 107.1428571428571 0H1500C1560.7142857142858 0 1607.142857142857 46.4285714285714 1607.142857142857 107.1428571428571V892.8571428571429C1607.142857142857 953.5714285714286 1560.7142857142858 1000 1500 1000zM285.7142857142857 839.2857142857142C285.7142857142857 839.2857142857142 285.7142857142857 839.2857142857142 285.7142857142857 839.2857142857142C285.7142857142857 842.8571428571429 285.7142857142857 846.4285714285714 289.2857142857143 846.4285714285714C289.2857142857143 846.4285714285714 289.2857142857143 850 292.8571428571429 850C292.8571428571429 853.5714285714286 300 857.1428571428571 303.5714285714286 857.1428571428571H1303.5714285714287C1314.2857142857142 857.1428571428571 1321.4285714285716 850 1321.4285714285716 839.2857142857142V321.4285714285715H1285.7142857142858H1250H1214.2857142857142H321.4285714285715H285.7142857142857V839.2857142857142zM1307.1428571428573 142.8571428571429H300C253.5714285714286 142.8571428571429 214.2857142857143 182.1428571428572 214.2857142857143 228.5714285714286C214.2857142857143 239.2857142857142 225 250 235.7142857142857 250H678.5714285714286V214.2857142857142C678.5714285714286 192.8571428571428 692.8571428571428 178.5714285714286 714.2857142857143 178.5714285714286H892.8571428571429C914.2857142857144 178.5714285714286 928.5714285714286 192.8571428571428 928.5714285714286 214.2857142857142V250H1371.4285714285713C1382.142857142857 250 1392.857142857143 239.2857142857142 1392.857142857143 228.5714285714286C1392.857142857143 182.1428571428572 1353.5714285714287 142.8571428571429 1307.1428571428573 142.8571428571429z" />
    <glyph glyph-name="paper"
      unicode="&#xF15D;"
      horiz-adv-x="1571.4285714285716" d="M1476.6183035714287 1000H94.8102678571429C42.6339285714286 1000 -0.0558035714286 957.3102678571428 -0.0558035714286 905.1339285714286V94.8102678571429C-0.0558035714286 42.6339285714286 42.6339285714286 -0.0558035714286 94.8102678571429 -0.0558035714286H1476.5625C1528.7388392857142 -0.0558035714286 1571.4285714285716 42.6339285714286 1571.4285714285716 92.0200892857143V902.34375C1571.484375 957.3102678571428 1528.794642857143 1000 1476.6183035714287 1000zM830.8593750000001 928.5714285714286C830.8593750000001 928.5714285714286 1444.419642857143 928.5714285714286 1473.2700892857144 928.5714285714286C1494.0848214285718 928.5714285714286 1501.1718750000002 915.1227678571428 1501.1718750000002 900.6696428571429C1501.1718750000002 898.3258928571429 1501.1718750000002 857.1428571428571 1501.1718750000002 857.1428571428571H830.8593750000001V928.5714285714286zM830.9709821428571 385.0446428571429V456.4732142857142H1501.1160714285716V385.0446428571429H830.9709821428571zM1501.060267857143 297.3214285714286V225.8928571428571H830.9709821428571V297.3214285714286H1501.060267857143zM830.9709821428571 537.9464285714286V609.375H1501.1160714285716V537.9464285714286H830.9709821428571zM744.3638392857143 71.3727678571429C744.3638392857143 71.3727678571429 126.8973214285715 71.3727678571429 98.046875 71.3727678571429C77.2321428571429 71.3727678571429 70.1450892857143 84.8214285714286 70.1450892857143 99.2745535714286C70.1450892857143 101.6183035714287 70.1450892857143 142.8013392857143 70.1450892857143 142.8013392857143H744.3638392857143V71.3727678571429zM744.3638392857143 296.875V225.4464285714286H70.2566964285714V296.875H744.3638392857143zM70.2566964285714 385.0446428571429V456.4732142857142H744.3080357142859V385.0446428571429H70.2566964285714zM744.3638392857143 539.0625H70.1450892857143C70.1450892857143 539.0625 70.1450892857143 899.4419642857142 70.1450892857143 901.7857142857142C70.1450892857143 916.2388392857144 77.2321428571429 929.6875 98.046875 929.6875C126.8973214285714 929.6875 744.3638392857143 929.6875 744.3638392857143 929.6875L744.3638392857143 539.0625L744.3638392857143 539.0625zM1501.171875 99.2745535714286C1501.171875 84.8214285714286 1494.0848214285716 71.3727678571429 1473.2700892857142 71.3727678571429C1444.419642857143 71.3727678571429 830.859375 71.3727678571429 830.859375 71.3727678571429V142.8013392857143H1501.171875C1501.171875 142.8013392857143 1501.171875 101.6183035714284 1501.171875 99.2745535714286zM1501.6183035714287 696.9308035714286H831.5290178571428V768.359375H1501.674107142857V696.9308035714286z" />
    <glyph glyph-name="pause"
      unicode="&#xF15E;"
      horiz-adv-x="636.1607142857143" d="M0 1000H244.8102678571429V0H0V1000zM391.7410714285715 1000H636.1607142857143V0H391.7410714285715V1000z" />
    <glyph glyph-name="pause_video"
      unicode="&#xF15F;"
      horiz-adv-x="683.8727678571429" d="M0 1000H263.1696428571429L263.1696428571429 0L0 0V1000zM421.09375 1000H683.8727678571429V0L421.09375 0L421.09375 1000z" />
    <glyph glyph-name="pdf-file"
      unicode="&#xF160;"
      horiz-adv-x="1000" d="M900.0000000000001 1000H300C244.9776785714286 1000 200 955.0223214285714 200 900V299.9999999999999C200 244.9776785714286 244.9776785714286 199.9999999999999 300 199.9999999999999H900C955.0223214285714 199.9999999999999 1000 244.9776785714285 1000 299.9999999999999V900C1000 955.0223214285714 955.0223214285716 1000 900.0000000000001 1000M475.0000000000001 625C475.0000000000001 583.4821428571429 441.5178571428572 550 400.0000000000001 550H350.0000000000001V450H275.0000000000001V750H400.0000000000001C441.5178571428572 750 475.0000000000001 716.5178571428571 475.0000000000001 675V625zM725 525C725 483.4821428571428 691.5178571428572 450 650 450H525V750H650C691.5178571428572 750 725 716.5178571428571 725 675V525zM925 675H849.9999999999999V625H925V550H849.9999999999999V450H774.9999999999999V750H924.9999999999998V675zM350.0000000000001 625H400.0000000000001V675H350.0000000000001V625zM100 800H0V100C0 44.9776785714287 44.9776785714286 0 100 0H800.0000000000001V100H100V800zM600 525H650V675H600V525z" />
    <glyph glyph-name="pictures"
      unicode="&#xF161;"
      horiz-adv-x="1571.4285714285716" d="M1465.1785714285713 1000.0558035714286H106.1383928571429C47.4888392857143 1000.0558035714286 -0.0558035714286 954.3526785714286 -0.0558035714286 897.9910714285714V102.0647321428572C-0.0558035714286 45.703125 47.4888392857143 0 106.1383928571429 0H1465.234375C1523.8839285714284 0 1571.4285714285716 45.703125 1571.4285714285716 102.0647321428572V897.9910714285714C1571.372767857143 954.3526785714286 1523.8839285714287 1000.0558035714286 1465.1785714285713 1000.0558035714286zM1501.171875 156.0267857142857C1501.171875 94.1406249999999 1478.7388392857144 71.9866071428571 1417.1316964285716 71.9866071428571C1332.9799107142856 72.1540178571428 438.2254464285714 72.5446428571428 154.2410714285715 72.5446428571428C94.3080357142858 72.5446428571428 70.2008928571429 97.265625 70.2008928571429 156.5848214285713C70.0334821428572 175.78125 70.0334821428572 175.78125 70.0334821428572 205.3013392857142C250.4464285714286 348.3816964285714 511.9977678571429 555.8035714285713 511.9977678571429 555.8035714285713L869.1964285714286 345.703125L1142.3549107142858 765.9040178571429C1142.3549107142858 765.9040178571429 1319.029017857143 643.2477678571429 1501.5625000000002 516.4620535714286C1501.6183035714287 327.0089285714286 1501.171875 178.125 1501.171875 156.0267857142857z" />
    <glyph glyph-name="play"
      unicode="&#xF162;"
      horiz-adv-x="636.1607142857143" d="M0 0L636.1607142857143 499.9441964285714L0 1000V0z" />
    <glyph glyph-name="plus_icon"
      unicode="&#xF163;"
      horiz-adv-x="999.8883928571429" d="M999.8883928571428 400.4464285714286H599.4419642857142V0H399.6651785714286V400.4464285714286H0V600.2232142857142H399.6651785714286V999.8883928571428H599.497767857143V600.2232142857142H999.9441964285716V400.4464285714286z" />
    <glyph glyph-name="portal"
      unicode="&#xF164;"
      horiz-adv-x="1222.2098214285713" d="M1111.1049107142858 1000H111.1049107142857C50 1000 0 950 0 888.8950892857142V222.2098214285715C0 161.1049107142857 50 111.1049107142858 111.1049107142857 111.1049107142858H388.8950892857143V0H833.3147321428571V111.1049107142857H1111.1049107142858C1172.2098214285713 111.1049107142857 1221.6517857142856 161.1049107142856 1221.6517857142856 222.2098214285713L1222.2098214285713 888.8950892857142C1222.2098214285713 950 1172.2098214285713 1000 1111.1049107142858 1000M1111.1049107142858 222.2098214285715H111.1049107142857V888.8950892857142H1111.1049107142858V222.2098214285715zM1000 722.2098214285713H388.8950892857144V611.1049107142858H1000V722.2098214285713zM1000 500H388.8950892857144V388.8950892857144H1000V500zM333.3147321428571 722.2098214285713H222.2098214285715V611.1049107142858H333.3147321428572V722.2098214285713zM333.3147321428571 500H222.2098214285715V388.8950892857144H333.3147321428572V500z" />
    <glyph glyph-name="poster"
      unicode="&#xF165;"
      horiz-adv-x="1571.4285714285716" d="M1465.234375 1000.0558035714286H106.0825892857143C47.4888392857143 1000.0558035714286 -0.0558035714286 954.3526785714286 -0.0558035714286 897.9910714285714V102.0647321428572C-0.0558035714286 48.6607142857143 47.4888392857143 0 106.1383928571429 0H1465.234375C1523.8839285714284 0 1571.4285714285716 45.703125 1571.4285714285716 102.0647321428572V897.9910714285714C1571.372767857143 954.3526785714286 1523.9397321428573 1000.0558035714286 1465.234375 1000.0558035714286zM728.4598214285714 144.9776785714286C728.4598214285714 141.6294642857142 725.7254464285714 138.8950892857142 722.3772321428571 138.8950892857142H423.8839285714286C420.5357142857144 138.8950892857142 417.8013392857144 141.6294642857142 417.8013392857144 144.9776785714286V295.3683035714286C417.8013392857144 301.2276785714286 422.2656250000001 307.5334821428571 422.7678571428572 308.0915178571428C426.0044642857143 311.8303571428571 477.734375 357.533482142857 500.9486607142858 380.9151785714286C547.9352678571429 350.1674107142857 601.5625 352.0089285714286 647.1540178571429 384.9330357142857C670.5357142857143 361.4397321428571 718.0245535714287 314.5647321428571 722.6004464285714 307.5334821428571C723.7723214285714 305.7477678571428 728.1250000000001 300.8370535714286 728.4598214285714 295.3683035714286V144.9776785714286zM464.0625 495.8705357142857C464.0625 555.9709821428571 512.7790178571429 604.6875 572.8794642857142 604.6875C632.9799107142857 604.6875 681.6964285714286 555.9709821428571 681.6964285714286 495.8705357142857S632.9799107142857 387.0535714285714 572.8794642857142 387.0535714285714S464.0625 435.7700892857142 464.0625 495.8705357142857zM1346.9308035714284 420.9263392857142C1346.9308035714284 360.1562499999999 1297.6562499999998 310.8816964285714 1236.8861607142858 310.8816964285714C1236.8861607142858 310.8816964285714 772.1540178571429 310.658482142857 756.25 310.658482142857C734.9888392857143 331.8080357142857 686.1049107142858 383.314732142857 669.3080357142858 400.3348214285714C703.6272321428572 434.8214285714285 716.8526785714287 486.439732142857 703.4040178571429 533.2589285714284C692.1875000000001 572.0982142857142 663.28125 604.5758928571428 626.1160714285714 620.3125C586.1607142857143 637.2209821428571 539.5089285714286 633.5937499999999 502.5669642857143 610.9375C467.4107142857143 589.3415178571428 443.5267857142858 551.7857142857142 439.0066964285715 510.7142857142857C434.3750000000001 468.8616071428571 449.8325892857144 426.5066964285714 480.3571428571429 397.4888392857142C443.7500000000001 361.4955357142857 410.1004464285715 331.3616071428571 388.3928571428572 310.7700892857144L334.6540178571429 310.8258928571429C273.8839285714286 310.8258928571429 224.609375 360.1004464285714 224.609375 420.8705357142858V751.0044642857142C224.609375 811.7745535714286 273.8839285714286 861.0491071428571 334.6540178571429 861.0491071428571H1236.7745535714287C1297.5446428571431 861.0491071428571 1346.8191964285716 811.7745535714286 1346.8750000000002 751.0044642857142L1346.9308035714284 420.9263392857142L1346.9308035714284 420.9263392857142z" />
    <glyph glyph-name="preprint"
      unicode="&#xF166;"
      horiz-adv-x="1571.4285714285716" d="M1476.674107142857 1000H94.8102678571429C42.6339285714286 1000 -0.0558035714286 957.3102678571428 -0.0558035714286 905.1339285714286V94.8102678571429C-0.0558035714286 42.6339285714286 42.6339285714286 -0.0558035714286 94.8102678571429 -0.0558035714286H1476.5625C1528.7388392857142 -0.0558035714286 1571.4285714285716 42.6339285714286 1571.4285714285716 92.0200892857143V902.34375C1571.5401785714284 957.3102678571428 1528.8504464285713 1000 1476.674107142857 1000zM168.0803571428572 693.8616071428571L342.2433035714286 735.0446428571429C346.7633928571429 736.1049107142857 351.5066964285715 736.0491071428571 355.9709821428572 734.8214285714286L500.6696428571429 695.4799107142857L668.5267857142858 734.9888392857142C672.9910714285714 736.0491071428571 677.6785714285714 735.9933035714286 682.1428571428572 734.765625L835.6584821428571 693.6383928571429C850.5580357142858 689.6763392857142 859.375 674.3303571428571 855.4129464285716 659.4866071428571C852.0647321428572 647.0424107142858 840.7924107142858 638.7834821428571 828.4598214285716 638.7834821428571C826.060267857143 638.7834821428571 823.6607142857144 639.1183035714286 821.2053571428572 639.7321428571429L674.497767857143 679.0178571428571L506.5848214285716 639.5089285714286C502.0647321428573 638.4486607142858 497.3772321428573 638.5044642857142 492.857142857143 639.7321428571429L348.2142857142857 679.1294642857142L180.9709821428572 639.5647321428571C165.9598214285715 635.9933035714287 150.9486607142857 645.3125 147.3772321428572 660.3236607142858S153.125 690.3459821428571 168.0803571428572 693.8616071428571zM168.0803571428572 492.4665178571428L342.2433035714286 533.6495535714286C346.7633928571429 534.7098214285713 351.5066964285715 534.6540178571428 355.9709821428572 533.4263392857142L507.4776785714286 492.2433035714286C522.3214285714286 488.2254464285714 531.1383928571429 472.8794642857142 527.0647321428571 457.9799107142858C523.7165178571429 445.5357142857144 512.4441964285714 437.3883928571429 500.1674107142857 437.3883928571429C497.7678571428572 437.3883928571429 495.2566964285714 437.7232142857144 492.8571428571428 438.3928571428571L348.2142857142857 477.734375L180.9709821428572 438.1696428571429C165.9598214285715 434.5982142857144 150.9486607142857 443.9174107142857 147.3772321428572 458.9285714285714S153.125 488.9508928571429 168.0803571428572 492.4665178571428zM828.4040178571429 235.9933035714286C826.0044642857143 235.9933035714286 823.6049107142858 236.328125 821.1495535714286 236.9419642857144L674.4419642857142 276.2276785714287L506.5290178571428 236.7187500000001C502.0089285714286 235.6584821428572 497.3214285714286 235.7142857142859 492.8013392857142 236.9419642857145L348.2142857142857 276.3392857142857L180.9709821428572 236.7745535714286C165.9598214285715 233.2589285714286 150.9486607142857 242.5223214285714 147.3772321428572 257.5334821428571C143.8058035714286 272.5446428571429 153.125 287.5558035714286 168.1361607142857 291.1272321428571L342.2991071428572 332.3102678571429C346.8191964285715 333.3705357142858 351.5625000000001 333.3147321428571 356.0267857142858 332.0870535714286L500.7254464285716 292.7455357142857L668.5825892857143 332.2544642857142C673.0468750000001 333.3147321428571 677.7343750000001 333.2589285714285 682.1986607142859 332.0312499999999L835.7142857142859 290.9040178571427C845.0334821428575 288.392857142857 852.0089285714288 281.4732142857141 854.9107142857146 272.9352678571427L902.2321428571432 437.611607142857L1035.1562500000002 305.5245535714285L854.6875000000002 254.2968749999999C850.5580357142858 243.1919642857142 839.9553571428572 236.0491071428571 828.4040178571429 235.9933035714286zM1053.5714285714287 323.9397321428571L920.6473214285714 455.9709821428571L1213.0580357142858 746.4285714285714L1345.982142857143 614.3973214285714L1053.5714285714287 323.9397321428571zM1406.0825892857142 674.21875L1366.2388392857142 634.5982142857142L1233.314732142857 766.6294642857143L1273.158482142857 806.25C1287.8348214285716 820.8147321428572 1310.8816964285713 821.5401785714286 1324.6651785714287 807.8683035714286L1407.7008928571431 725.3348214285714C1421.4843750000002 711.6071428571429 1420.703125 688.7276785714287 1406.0825892857142 674.21875z" />
    <glyph glyph-name="preprint_i"
      unicode="&#xF167;"
      horiz-adv-x="2185.212053571429" d="M36.9419642857143 787.2209821428571L336.3281250000001 858.0357142857142C344.0848214285715 859.8772321428571 352.2321428571429 859.765625 359.9330357142857 857.6450892857142L608.7053571428571 790.0111607142857L897.3214285714286 857.9241071428571C905.0223214285714 859.765625 913.0580357142858 859.5982142857142 920.7031250000002 857.5892857142858L1184.5982142857144 786.8861607142858C1210.2120535714287 780.0223214285714 1225.390625 753.7388392857143 1218.526785714286 728.125C1212.779017857143 706.6964285714286 1193.415178571429 692.578125 1172.2098214285716 692.578125C1168.0803571428573 692.578125 1163.950892857143 693.1361607142858 1159.7656250000002 694.2522321428571L907.5334821428572 761.8303571428571L618.8616071428571 693.8616071428571C611.1049107142857 692.0200892857142 603.0133928571428 692.1875 595.3125 694.2522321428571L346.5401785714286 761.8303571428571L59.0401785714286 693.8616071428571C33.2589285714286 687.7790178571429 7.421875 703.7388392857142 1.3392857142857 729.5200892857142S11.1607142857143 781.1383928571429 36.9419642857143 787.2209821428571zM36.9419642857143 440.9598214285715L336.3281250000001 511.71875C344.0848214285715 513.5602678571429 352.2321428571429 513.4486607142858 359.9330357142857 511.3281249999999L620.4241071428571 440.5133928571428C645.9821428571428 433.5379464285715 661.1049107142858 407.1986607142858 654.1294642857143 381.640625C648.3258928571429 360.2678571428572 628.9620535714287 346.2611607142857 607.8683035714286 346.2611607142857C603.6830357142857 346.2611607142857 599.4977678571429 346.8191964285715 595.2566964285714 347.9352678571429L346.5401785714286 415.5691964285715L58.984375 347.6004464285715C33.203125 341.5178571428571 7.3660714285715 357.4776785714287 1.2834821428572 383.2589285714287C-4.7991071428571 408.984375 11.1607142857143 434.8214285714286 36.9419642857143 440.9598214285715zM1172.265625 0C1168.1361607142856 0 1164.0066964285713 0.5580357142857 1159.8214285714284 1.6183035714286L907.5892857142856 69.1964285714286L618.8616071428571 1.2834821428571C611.1049107142857 -0.5580357142858 603.0133928571428 -0.390625 595.3125 1.6741071428571L346.5401785714286 69.3080357142857L58.984375 1.3392857142858C33.203125 -4.7433035714286 7.3660714285715 11.2165178571429 1.2834821428572 36.9977678571429C-4.7991071428571 62.779017857143 11.1607142857143 88.6160714285716 36.9419642857143 94.6986607142859L336.3281250000001 165.513392857143C344.0848214285715 167.3549107142859 352.2321428571429 167.2433035714287 359.9330357142857 165.122767857143L608.7053571428571 97.4888392857145L897.2656250000001 165.4017857142859C904.9665178571428 167.2433035714287 913.0022321428572 167.075892857143 920.6473214285716 165.0669642857145L1184.542410714286 94.3638392857145C1200.6138392857147 90.0669642857145 1212.5558035714287 78.1250000000002 1217.5223214285716 63.5044642857145L1298.883928571429 346.5959821428574L1527.399553571429 119.5312500000001L1217.1316964285716 31.4732142857145C1210.2678571428573 12.3325892857142 1192.0200892857142 0 1172.265625 0zM1559.3191964285716 151.171875L1330.7477678571431 378.1250000000001L1833.4821428571431 877.5669642857143L2062.0535714285716 650.6138392857143L1559.3191964285716 151.171875zM2165.401785714286 753.4040178571429L2096.875 685.2678571428571L1868.359375 912.2209821428572L1936.8861607142856 980.3571428571428C1962.109375 1005.4129464285714 2001.7857142857144 1006.640625 2025.446428571429 983.1473214285714L2168.247767857143 841.2388392857142C2191.9084821428573 817.7455357142857 2190.625 778.3482142857142 2165.401785714286 753.4040178571429z" />
    <glyph glyph-name="presentation"
      unicode="&#xF168;"
      horiz-adv-x="1571.4285714285716" d="M1465.1785714285713 1000.0558035714286H106.0825892857143C47.4888392857143 1000.0558035714286 -0.0558035714286 954.3526785714286 -0.0558035714286 897.9910714285714V102.0647321428572C-0.0558035714286 45.703125 47.4888392857143 0 106.1383928571429 0H1465.234375C1523.8839285714284 0 1571.4285714285716 45.703125 1571.4285714285716 102.0647321428572V897.9352678571429C1571.3169642857144 954.296875 1523.8839285714287 1000.0558035714286 1465.1785714285713 1000.0558035714286zM1357.8125 134.1517857142858C1328.9062499999998 126.1160714285715 1303.627232142857 107.1986607142858 1287.8906249999998 81.5290178571429C1271.9866071428569 107.3102678571429 1246.484375 126.5066964285715 1217.0758928571427 134.4308035714286C1231.2499999999998 158.5379464285715 1257.4218749999998 174.7209821428572 1287.3883928571427 174.7209821428572C1317.4107142857144 174.6651785714286 1343.6941964285716 158.3705357142857 1357.8125 134.1517857142858zM1508.2589285714287 104.5758928571429C1504.4642857142858 97.7678571428571 1494.9776785714287 89.9553571428571 1489.7321428571431 81.5290178571429C1473.8281250000002 107.3102678571429 1448.3258928571431 126.5066964285715 1418.917410714286 134.4308035714286C1433.0915178571431 158.5379464285715 1459.2633928571431 174.7209821428572 1489.229910714286 174.7209821428572C1499.7209821428573 174.7209821428572 1509.821428571429 172.7120535714287 1519.0290178571431 169.0848214285716C1519.1964285714287 161.0491071428571 1522.8236607142858 131.3058035714286 1508.2589285714287 104.5758928571429zM52.1763392857143 169.0290178571429C61.3839285714286 172.65625 71.4285714285714 174.6651785714286 81.9754464285714 174.6651785714286C111.9419642857143 174.6651785714286 138.1138392857143 158.4821428571429 152.2879464285714 134.375C122.8794642857143 126.4508928571428 97.3772321428571 107.2544642857143 81.4732142857143 81.4732142857142C76.2834821428571 89.9553571428571 66.796875 97.6004464285715 62.9464285714286 104.5200892857142C48.3258928571429 131.3058035714286 51.953125 161.0491071428571 52.1763392857143 169.0290178571429zM224.2745535714286 734.2633928571429C224.2745535714286 795.0334821428571 273.4933035714286 844.3080357142858 334.3191964285714 844.3080357142858H1236.439732142857C1297.2098214285713 844.3080357142858 1346.4843749999998 795.0334821428571 1346.5401785714287 734.2633928571429V482.3102678571428C1346.5401785714287 421.5401785714285 1297.265625 372.2656249999999 1236.4955357142858 372.2656249999999H334.3191964285715C273.5491071428572 372.2656249999999 224.2745535714286 421.5401785714285 224.2745535714286 482.3102678571428V734.2633928571429zM1155.8035714285713 134.1517857142858C1126.8973214285713 126.1160714285715 1101.6183035714284 107.1986607142858 1085.8258928571427 81.5290178571429C1069.9218749999998 107.3102678571429 1044.4196428571427 126.5066964285715 1015.0111607142856 134.4308035714286C1029.1852678571427 158.5379464285715 1055.3571428571427 174.7209821428572 1085.3236607142856 174.7209821428572C1115.4017857142858 174.6651785714286 1141.685267857143 158.3705357142857 1155.8035714285713 134.1517857142858zM953.7946428571428 134.1517857142858C924.8883928571428 126.1160714285715 899.6093750000001 107.1986607142858 883.8169642857143 81.5290178571429C867.9129464285714 107.3102678571429 842.4107142857142 126.5066964285715 813.0022321428571 134.4308035714286C827.1763392857143 158.5379464285715 853.3482142857142 174.7209821428572 883.3147321428571 174.7209821428572C913.3928571428572 174.6651785714286 939.6763392857144 158.3705357142857 953.7946428571428 134.1517857142858zM751.7857142857143 134.1517857142858C722.8794642857143 126.1160714285715 697.6004464285716 107.1986607142858 681.8080357142857 81.5290178571429C665.9040178571429 107.3102678571429 640.4017857142857 126.5066964285715 610.9933035714286 134.4308035714286C625.1674107142857 158.5379464285715 651.3392857142857 174.7209821428572 681.3058035714286 174.7209821428572C711.3839285714286 174.6651785714286 737.6674107142858 158.3705357142857 751.7857142857143 134.1517857142858zM549.7209821428571 134.1517857142858C520.8147321428571 126.1160714285715 495.5357142857142 107.1986607142858 479.7991071428571 81.5290178571429C463.8950892857143 107.3102678571429 438.3928571428571 126.5066964285715 408.984375 134.4308035714286C423.1584821428571 158.5379464285715 449.3303571428571 174.7209821428572 479.296875 174.7209821428572C509.375 174.6651785714286 535.6026785714286 158.3705357142857 549.7209821428571 134.1517857142858zM277.2879464285715 174.6651785714286C307.3660714285714 174.6651785714286 333.59375 158.3705357142857 347.7120535714286 134.1517857142858C318.8058035714286 126.1160714285715 293.5267857142857 107.1986607142858 277.7901785714286 81.5290178571429C261.8861607142857 107.3102678571429 236.3839285714286 126.5066964285715 206.9754464285714 134.4308035714286C221.1495535714286 158.4821428571429 247.3214285714286 174.6651785714286 277.2879464285715 174.6651785714286zM299.5535714285714 52.1763392857142C311.9977678571429 81.1941964285715 340.8482142857143 101.5625 374.4419642857143 101.5625C409.9330357142857 101.5625 440.0669642857144 83.9285714285715 451.2834821428572 52.2879464285714M502.7901785714286 52.1763392857142C515.234375 81.1941964285715 544.0848214285714 101.5625 577.6785714285714 101.5625C613.1696428571429 101.5625 643.3035714285714 83.9285714285715 654.5200892857143 52.2879464285714M706.0267857142858 52.1763392857142C718.4709821428571 81.1941964285715 747.3214285714287 101.5625 780.9151785714287 101.5625C816.40625 101.5625 846.5401785714286 83.9285714285715 857.7566964285716 52.2879464285714M905.1339285714286 52.1763392857142C917.578125 81.1941964285715 946.4285714285714 101.5625 980.0223214285714 101.5625C1015.5133928571428 101.5625 1045.6473214285713 83.9285714285715 1056.8638392857144 52.2879464285714M1105.9151785714287 52.1763392857142C1118.359375 81.1941964285715 1147.2098214285716 101.5625 1180.8035714285716 101.5625C1216.294642857143 101.5625 1246.4285714285713 83.9285714285715 1257.6450892857142 52.2879464285714M1311.607142857143 52.1763392857142C1324.0513392857144 81.1941964285715 1352.9017857142858 101.5625 1386.4955357142858 101.5625C1418.861607142857 101.5625 1446.763392857143 86.8861607142857 1459.9330357142856 60.3236607142857C1458.0357142857142 59.3749999999999 1440.5133928571427 53.1808035714286 1429.017857142857 52.2879464285713C1422.767857142857 51.8973214285714 1311.607142857143 52.1763392857142 1311.607142857143 52.1763392857142zM135.7700892857143 52.2879464285714C123.4375 53.4040178571429 106.7522321428572 59.375 104.8549107142857 60.3236607142858C118.0245535714286 86.8861607142857 145.9263392857143 101.5625000000001 178.2924107142857 101.5625000000001C211.8861607142857 101.5625000000001 240.7366071428572 81.25 253.1808035714286 52.1763392857143C253.1808035714286 52.1763392857142 142.0758928571429 51.8973214285714 135.7700892857143 52.2879464285714z" />
    <glyph glyph-name="previous"
      unicode="&#xF169;"
      horiz-adv-x="645.3936469062883" d="M645.3936469062883 944.968666714687C645.3936469062883 898.7970899661457 645.3936469062883 88.237412662969 645.3936469062883 56.3999135633509C645.3936469062883 15.0903983288915 602.6075055823669 -15.6306273860117 560.3615933155659 8.5356191024994C526.5072390693655 27.911834617878 70.2297774256285 404.6675790535187 22.9777425628467 451.9196139163005C-7.7792984225311 482.7126701721529 -7.4911762587336 517.5394367211697 22.9777425628467 548.1884318951234C57.0121731614205 582.4389541165453 512.893466829936 965.3172945328818 562.0903262983506 992.5808542822156C598.1416120435065 1012.6053446661384 645.3936469062883 990.4919685946842 645.3936469062883 944.968666714687z" />
    <glyph glyph-name="private-space"
      unicode="&#xF16A;"
      horiz-adv-x="1250" d="M125 187.5H250.0558035714286V62.4441964285715H125V187.5zM0 0H1250V250H0V0zM125 562.5H250.0558035714286V437.5H125V562.5zM0 375H1250V625H0V375zM125 937.5H250.0558035714286V812.5H125V937.5zM0 750H1250V1000H0V750z" />
    <glyph glyph-name="public-space"
      unicode="&#xF16B;"
      horiz-adv-x="1000" d="M100 800H0V100C0 44.7544642857143 44.7544642857143 0 100 0H800.0000000000001V100H100V800zM900.0000000000001 1000H300C244.7544642857143 1000 200 955.2455357142856 200 900V299.9999999999999C200 244.7544642857142 244.7544642857143 199.9999999999999 300 199.9999999999999H900C955.2455357142856 199.9999999999999 1000 244.7544642857142 1000 299.9999999999999V900C1000 955.2455357142856 955.2455357142858 1000 900.0000000000001 1000M849.9999999999999 550H649.9999999999999V350H549.9999999999999V550H349.9999999999999V650H549.9999999999999V850H649.9999999999999V650H849.9999999999999V550z" />
    <glyph glyph-name="published-paper"
      unicode="&#xF16C;"
      horiz-adv-x="814.9553571428572" d="M181.0825892857143 954.5200892857144C181.0825892857143 979.6316964285714 201.4508928571429 1000 226.5625 1000H769.53125C794.6428571428571 1000 815.0111607142858 979.6316964285714 815.0111607142858 954.5200892857144V227.2321428571429C815.0111607142858 202.1205357142857 794.6428571428571 181.7522321428571 769.53125 181.7522321428571H226.5625C201.4508928571429 181.7522321428571 181.0825892857143 202.1205357142857 181.0825892857143 227.2321428571429V954.5200892857144zM271.6517857142857 795.4799107142858C271.6517857142857 808.0357142857142 281.8080357142857 818.1919642857142 294.3638392857143 818.1919642857142H701.6741071428572C714.2299107142858 818.1919642857142 724.3861607142858 808.0357142857142 724.3861607142858 795.4799107142858S714.2299107142858 772.7678571428571 701.6741071428572 772.7678571428571H294.3638392857143C281.8638392857144 772.7120535714286 271.6517857142857 782.9241071428571 271.6517857142857 795.4799107142858zM294.3638392857143 681.8080357142857C281.8080357142857 681.8080357142857 271.6517857142857 671.6517857142857 271.6517857142857 659.0959821428571S281.8080357142857 636.3839285714284 294.3638392857143 636.3839285714284H701.6741071428572C714.2299107142858 636.3839285714284 724.3861607142858 646.5401785714284 724.3861607142858 659.0959821428571S714.2299107142858 681.8080357142857 701.6741071428572 681.8080357142857H294.3638392857143zM271.6517857142857 522.7120535714286C271.6517857142857 535.2678571428571 281.8080357142857 545.4241071428571 294.3638392857143 545.4241071428571H701.6741071428572C714.2299107142858 545.4241071428571 724.3861607142858 535.2678571428571 724.3861607142858 522.7120535714286S714.2299107142858 500 701.6741071428572 500H294.3638392857143C281.8638392857144 500 271.6517857142857 510.15625 271.6517857142857 522.7120535714286zM294.3638392857143 409.0959821428571C281.8080357142857 409.0959821428571 271.6517857142857 398.939732142857 271.6517857142857 386.3839285714285S281.8080357142857 363.6718749999999 294.3638392857143 363.6718749999999H701.6741071428572C714.2299107142858 363.6718749999999 724.3861607142858 373.8281249999999 724.3861607142858 386.3839285714285S714.2299107142858 409.0959821428571 701.6741071428572 409.0959821428571H294.3638392857143zM90.5691964285714 909.0959821428572C90.5691964285714 934.1517857142856 110.8258928571429 954.4642857142856 135.8258928571429 954.5758928571428V181.8638392857142C135.8258928571429 156.752232142857 156.1941964285714 136.3839285714285 181.3058035714286 136.3839285714285H769.6986607142859C769.6986607142859 111.2723214285713 749.3303571428572 90.9040178571428 724.2187500000001 90.9040178571428H135.9933035714286C110.8816964285714 90.9040178571428 90.5133928571429 111.2723214285713 90.5133928571429 136.3839285714285V909.0959821428572H90.5691964285714zM45.2566964285714 863.6160714285714C20.2566964285714 863.5602678571429 0 843.2477678571429 0 818.1919642857142V45.4799107142857C0 20.3683035714286 20.3683035714286 0 45.4799107142857 0H633.7053571428571C658.8169642857142 0 679.1852678571428 20.3683035714286 679.1852678571428 45.4799107142857H90.7366071428571C65.625 45.4799107142857 45.2566964285714 65.8482142857142 45.2566964285714 90.9598214285715C45.2566964285714 90.9598214285715 45.2566964285714 863.6160714285714 45.2566964285714 863.6160714285714z" />
    <glyph glyph-name="quote"
      unicode="&#xF16D;"
      horiz-adv-x="1305.2455357142858" d="M563.9508928571429 882.7566964285714C517.4107142857143 852.1763392857143 474.2745535714286 822.3214285714286 434.4308035714286 793.0803571428571C394.53125 763.8392857142858 359.9888392857144 733.3147321428571 330.8035714285714 701.4508928571429C301.5625 669.5870535714286 278.6272321428571 634.9888392857142 262.0535714285715 597.8236607142857C245.4241071428571 560.6026785714284 237.1651785714286 518.1361607142858 237.1651785714286 470.3125C237.1651785714286 455.6919642857142 238.4486607142857 441.40625 241.1272321428572 427.4553571428571C243.75 413.5044642857142 247.0982142857143 398.5491071428571 251.1160714285714 382.6450892857144C259.0959821428571 389.2857142857144 274.3303571428571 396.5959821428571 296.9308035714286 404.5758928571429C319.53125 412.5558035714286 344.0848214285714 416.5178571428572 370.6473214285714 416.5178571428572C427.734375 416.5178571428572 471.2611607142857 397.265625 501.1718749999999 358.7053571428572C531.0825892857142 320.1450892857145 545.9821428571428 273.6607142857144 545.9821428571428 219.1964285714287C545.9821428571428 188.6160714285716 540.0111607142857 160.0446428571429 528.0691964285713 133.482142857143C516.1272321428571 106.8638392857143 499.4977678571428 83.6495535714287 478.2366071428571 63.7276785714287C456.9754464285714 43.8058035714287 432.7008928571428 28.1808035714287 405.5245535714285 16.9084821428572C378.2924107142857 5.5803571428572 348.7165178571428 1e-13 316.8526785714285 1e-13C262.3883928571428 1e-13 215.2343749999999 10.6026785714287 175.3906249999999 31.8638392857143C135.5468749999999 53.0691964285714 102.6785714285713 80.0223214285715 76.7299107142856 112.5558035714286C50.8370535714285 145.0892857142857 31.5290178571428 181.3058035714286 18.9174107142856 221.1495535714286C6.3058035714286 261.0491071428571 0 299.5535714285714 0 336.7745535714286C0 428.4598214285714 12.6116071428571 508.1473214285714 37.8348214285714 575.8928571428571C63.0580357142857 643.6383928571429 97.65625 703.4040178571429 141.4620535714286 755.2455357142857C185.3236607142857 807.03125 235.7700892857143 852.2321428571429 292.9129464285715 890.7366071428571C350 929.2410714285714 410.4910714285714 965.7924107142856 474.2745535714286 1000.3348214285714L563.9508928571429 882.7566964285714zM1305.2455357142858 882.7566964285714C1258.705357142857 852.1763392857143 1215.5691964285716 822.265625 1175.7254464285716 793.0803571428571C1135.8816964285716 763.8392857142858 1101.3392857142858 733.3147321428571 1072.0982142857144 701.3950892857143C1042.857142857143 669.53125 1019.921875 634.9330357142858 1003.3482142857144 597.7678571428571C986.71875 560.546875 978.4598214285716 518.0803571428571 978.4598214285716 470.2566964285714C978.4598214285716 455.6361607142857 979.7433035714286 441.3504464285715 982.421875 427.3995535714286C985.0446428571428 413.4486607142858 988.3928571428572 398.4933035714287 992.4107142857144 382.5892857142858C1000.390625 389.2299107142858 1015.625 396.5401785714287 1038.2254464285716 404.5200892857144C1060.825892857143 412.5 1085.3794642857142 416.4620535714287 1111.9419642857142 416.4620535714287C1169.029017857143 416.4620535714287 1212.5558035714287 397.2098214285716 1242.466517857143 358.6495535714287C1272.377232142857 320.0892857142859 1287.2767857142858 273.6049107142858 1287.2767857142858 219.1406250000001C1287.2767857142858 188.560267857143 1281.305803571429 159.9888392857145 1269.3638392857144 133.4263392857145C1257.4218750000002 106.8080357142859 1240.792410714286 83.5937500000002 1219.53125 63.6718750000002C1198.2700892857144 43.7500000000001 1173.9955357142858 28.1250000000001 1146.8191964285713 16.8526785714287C1119.5870535714287 5.5245535714287 1090.0111607142858 -0.0558035714284 1058.1473214285713 -0.0558035714284C1003.6830357142856 -0.0558035714284 956.5290178571428 10.5468750000001 916.6852678571428 31.8080357142858C876.8415178571428 53.013392857143 843.9732142857142 79.9665178571429 818.0245535714286 112.5C792.1316964285713 145.0334821428572 772.8236607142857 181.2500000000001 760.2120535714286 221.09375C747.5446428571428 260.9375 741.2946428571428 299.4419642857144 741.2946428571428 336.6629464285714C741.2946428571428 428.3482142857142 753.9062499999999 508.0357142857143 779.1294642857142 575.78125C804.3526785714286 643.5267857142857 838.8950892857141 703.2924107142857 882.7566964285713 755.1339285714286C926.6183035714284 806.9196428571429 977.0647321428572 852.1205357142857 1034.2075892857142 890.625C1091.2946428571427 929.1294642857142 1151.785714285714 965.6808035714286 1215.5691964285713 1000.2232142857142L1305.2455357142858 882.7566964285714z" />
    <glyph glyph-name="reserve-doi"
      unicode="&#xF16E;"
      horiz-adv-x="800" d="M500 1000H100C44.7544642857143 1000 0.5022321428572 955.2455357142856 0.5022321428572 900L0 100C0 44.7544642857143 44.2522321428571 0 99.4977678571429 0H700C755.2455357142857 0 800 44.7544642857143 800 100V700L500 1000zM600 200H200V299.9999999999999H599.9999999999999V200zM600 400H200V500H599.9999999999999V400zM450.0000000000001 650V925L725 650H450.0000000000001z" />
    <glyph glyph-name="retry"
      unicode="&#xF16F;"
      horiz-adv-x="1000.8370535714286" d="M148.9955357142857 851.171875C239.2857142857143 940.234375 363.1696428571429 995.5915178571428 500.3906250000001 995.5915178571428C776.7299107142858 995.5915178571428 1000.8370535714286 772.7120535714286 1000.8370535714286 497.7678571428571C1000.8370535714286 222.8794642857143 776.7857142857143 0 500.3906250000001 0C347.0424107142858 0 209.9330357142858 68.8058035714286 118.1361607142857 176.8973214285715L206.7522321428572 265.6808035714286C275.5580357142858 179.7433035714285 381.4174107142857 124.4419642857142 500.3906250000001 124.4419642857142C707.6450892857142 124.4419642857142 875.7254464285716 291.6294642857142 875.7254464285716 497.7678571428571C875.7254464285716 703.9620535714286 707.700892857143 871.09375 500.3906250000001 871.09375C397.7120535714287 871.09375 305.6919642857144 829.0736607142857 238.0022321428573 762.5L441.5736607142859 559.9330357142857H130.7477678571428H125.1116071428571V560.2678571428571H62.5558035714286C28.0133928571429 560.2678571428571 0 588.1138392857142 0 622.4888392857142V682.4776785714286V684.8214285714287V999.3861607142856L148.9955357142857 851.171875z" />
    <glyph glyph-name="rss_feed"
      unicode="&#xF170;"
      horiz-adv-x="1000" d="M911.4955357142858 1000H88.4486607142857C40.4017857142857 1000 0 959.375 0 911.0491071428572V89.0066964285715C0 38.7276785714286 40.4017857142857 0.0558035714284 88.4486607142857 0.0558035714284H911.4955357142858C961.4955357142858 0.0558035714284 999.9441964285716 40.6808035714284 999.9441964285716 89.0066964285715V911.0491071428572C1001.8415178571428 959.375 961.4955357142858 1000 911.4955357142858 1000zM263.4486607142857 148.9397321428572C213.4486607142857 148.9397321428572 171.1495535714286 189.5647321428571 171.1495535714286 241.7968750000001C171.1495535714286 292.0758928571429 211.5513392857143 334.654017857143 263.4486607142857 334.654017857143C313.4486607142857 334.654017857143 355.7477678571429 294.029017857143 355.7477678571429 241.7968750000001C353.8504464285715 189.5647321428571 313.4486607142857 148.9397321428572 263.4486607142857 148.9397321428572zM613.4486607142857 148.9397321428572H519.1964285714286C519.1964285714286 156.6964285714287 521.09375 164.3973214285715 521.09375 174.1071428571429C521.09375 367.5223214285714 363.3928571428571 526.1160714285713 171.09375 526.1160714285713V620.8705357142858C415.2901785714286 620.8705357142858 615.2901785714286 419.6986607142857 615.2901785714286 174.0513392857143C613.4486607142857 166.3504464285713 613.4486607142857 156.6964285714286 613.4486607142857 148.9397321428572zM823.0468750000001 148.9397321428572H734.5982142857143C740.3459821428572 179.9107142857143 744.1964285714287 208.9285714285715 744.1964285714287 241.7968750000001C744.1964285714287 528.0691964285716 513.4486607142858 760.1562500000001 228.8504464285715 760.1562500000001C209.5982142857144 760.1562500000001 190.4017857142858 758.2031250000001 171.1495535714287 756.3058035714287V847.2098214285714C190.4017857142858 849.1629464285714 209.5982142857144 851.0602678571429 228.8504464285715 851.0602678571429C561.4955357142858 851.0602678571429 832.6450892857143 578.3482142857142 832.6450892857143 243.6941964285714C830.6919642857142 210.8258928571428 828.7946428571429 179.9107142857143 823.0468750000001 148.9397321428572z" />
    <glyph glyph-name="search"
      unicode="&#xF171;"
      horiz-adv-x="987.4441964285714" d="M607.5334821428571 240.5133928571429C449.6651785714285 240.5133928571429 304.7433035714285 341.7410714285715 250.5580357142856 490.0111607142857C195.5357142857142 640.5133928571429 243.3593749999999 813.3928571428571 367.1316964285713 914.84375C489.9553571428571 1015.4575892857144 668.3035714285714 1027.9575892857142 804.1294642857142 945.7589285714286C939.1183035714286 864.0625 1010.7700892857142 702.5111607142857 980.5245535714286 547.6004464285713C946.0937500000002 371.2611607142857 787.1651785714286 240.5133928571429 607.5334821428571 240.5133928571429zM607.5334821428571 933.3147321428572C477.9017857142857 933.3147321428572 358.8169642857142 850.4464285714286 314.0066964285714 728.7946428571429C268.6383928571428 605.6919642857142 306.8080357142856 464.1741071428571 407.2544642857142 380.1897321428571C508.4821428571428 295.5357142857142 657.0312499999999 284.4308035714286 769.8660714285714 352.9575892857142C880.8035714285713 420.3683035714286 939.3973214285714 553.125 914.6205357142858 680.5245535714286C886.3839285714287 825.6696428571429 755.3013392857143 933.3147321428572 607.5334821428571 933.3147321428572zM359.9330357142858 360.3236607142857C386.216517857143 334.0401785714287 386.216517857143 291.3504464285715 359.9330357142858 265.0669642857142L115.0111607142858 20.1450892857142C88.7276785714287 -6.1383928571429 46.0379464285715 -6.1383928571429 19.7544642857144 20.1450892857142C-6.5290178571428 46.4285714285713 -6.5290178571428 89.1183035714284 19.7544642857144 115.4017857142857L264.6763392857144 360.3236607142856C290.9598214285715 386.6071428571428 333.5937500000001 386.6071428571428 359.9330357142858 360.3236607142857z" />
    <glyph glyph-name="search_bar"
      unicode="&#xF172;"
      horiz-adv-x="1000" d="M990.8214285714286 141.9285714285715L812.5357142857142 320.1428571428571C856.25 388.3571428571428 874.1071428571428 474.9642857142857 874.1071428571428 562C874.1071428571429 803.8571428571429 680.0357142857143 1000 438.0357142857144 1000C196.1785714285715 1000.0714285714286 0 800 0 558.0714285714286C0 316.1071428571428 196.1428571428572 125.9285714285715 438.0714285714286 125.9285714285715C525.0714285714286 125.9285714285715 611.7142857142858 143.7857142857142 679.8571428571429 187.4285714285715L858.1428571428572 9.0714285714286A31.4285714285714 31.4285714285714 0 0 1 902.357142857143 9.0714285714286L990.8928571428572 97.6428571428572A31.4285714285714 31.4285714285714 0 0 1 990.8214285714286 141.9285714285715zM438.0714285714285 249.0714285714286C265.5714285714285 249.0714285714286 125.2142857142857 391.4642857142857 125.2142857142857 563.9285714285714C125.2142857142857 736.3571428571429 265.6428571428571 876.7857142857143 438.0714285714285 876.7857142857143C610.6428571428571 876.7857142857143 750.9642857142857 736.3571428571429 750.9642857142857 563.9285714285714C750.9642857142857 391.4285714285715 610.6071428571428 249.0714285714286 438.0714285714285 249.0714285714286z" />
    <glyph glyph-name="search_left"
      unicode="&#xF173;"
      horiz-adv-x="987.4441964285714" d="M6.9196428571428 547.4888392857142C-23.3258928571429 702.3995535714287 48.3258928571429 863.9508928571429 183.3147321428572 945.6473214285714C319.140625 1027.845982142857 497.4888392857143 1015.3459821428572 620.3125 914.7321428571428C744.0848214285713 813.28125 791.9084821428571 640.4017857142858 736.8861607142858 489.8995535714286C682.7008928571429 341.6294642857144 537.7790178571429 240.4017857142858 379.9107142857143 240.4017857142858C200.2790178571429 240.4017857142857 41.3504464285714 371.1495535714286 6.9196428571428 547.4888392857142zM72.8236607142857 680.4129464285713C48.046875 553.0133928571429 106.640625 420.2566964285714 217.578125 352.8459821428571C330.4129464285714 284.3191964285715 478.9620535714286 295.424107142857 580.1897321428571 380.0781249999999C680.6361607142857 464.0624999999999 718.8058035714286 605.5803571428571 673.4374999999999 728.6830357142857C628.627232142857 850.3348214285713 509.5424107142856 933.203125 379.9107142857142 933.203125C232.1428571428572 933.203125 101.0602678571429 825.5580357142857 72.8236607142857 680.4129464285713zM722.7678571428572 360.2120535714286L967.6897321428572 115.2901785714285C993.9732142857142 89.0066964285715 993.9732142857142 46.3169642857142 967.6897321428572 20.033482142857S898.7165178571429 -6.25 872.4330357142857 20.033482142857L627.5111607142858 264.9553571428571C601.2276785714287 291.2388392857142 601.2276785714287 333.9285714285714 627.5111607142858 360.2120535714286C653.8504464285714 386.4955357142857 696.484375 386.4955357142857 722.7678571428572 360.2120535714286z" />
    <glyph glyph-name="settings"
      unicode="&#xF174;"
      horiz-adv-x="988.3928571428572" d="M941.796875 394.921875L867.8571428571429 425.8928571428571C877.6227678571429 475.6696428571429 877.2321428571429 525.78125 868.0803571428572 573.9397321428571L942.8013392857144 605.3013392857142C981.3616071428572 621.484375 999.5535714285716 666.0714285714286 983.7611607142858 705.0223214285714C967.7455357142858 744.0290178571429 923.6049107142856 762.4441964285714 885.0446428571429 746.4285714285714L810.2678571428571 715.2901785714286C783.3147321428572 755.6361607142858 748.7165178571429 791.40625 707.1428571428572 819.5870535714286L737.5000000000001 893.75C753.4598214285714 932.7008928571428 735.1004464285716 977.2879464285714 696.5401785714287 993.4709821428572S613.7834821428572 991.0714285714286 597.8236607142857 952.1205357142858L567.4665178571429 877.9575892857143C518.6941964285714 887.5558035714286 469.53125 886.9419642857143 422.3772321428572 877.5669642857143L391.40625 952.9017857142858C375.4464285714286 991.8526785714286 331.25 1010.2678571428572 292.6897321428572 994.2522321428572C254.1294642857143 978.0691964285714 235.9375 933.4821428571428 251.8973214285714 894.53125L282.9241071428572 818.9732142857143C242.578125 791.5736607142858 207.1986607142857 756.25 179.1852678571429 714.0625L104.4642857142857 745.2566964285714C65.9040178571429 761.4397321428571 21.7075892857143 742.8571428571429 5.7477678571428 703.90625S8.1473214285714 620.3683035714287 46.7075892857143 604.1852678571429L121.6517857142857 572.7678571428571C112.2767857142857 523.828125 112.890625 474.2745535714286 122.0982142857143 426.8973214285715L47.5446428571429 395.7031249999999C8.984375 379.7433035714286 -9.2075892857143 334.9330357142856 6.5848214285714 295.9821428571428C22.5446428571429 257.0312499999999 66.7410714285714 238.6160714285714 105.3013392857143 254.6316964285714L179.8549107142857 285.8258928571429C206.640625 245.6473214285715 241.015625 210.1004464285715 282.1428571428572 181.9196428571429L250.7812500000001 105.5803571428571C234.8214285714286 66.6294642857143 253.1808035714286 22.0424107142857 291.7410714285715 5.859375S374.497767857143 8.2589285714286 390.4575892857144 47.2098214285713L421.6517857142857 123.3258928571428C470.8147321428572 113.5602678571428 520.1450892857143 114.1183035714286 567.5223214285714 123.3258928571428L598.4933035714286 47.7678571428571C614.453125 8.8169642857142 658.6495535714286 -9.5982142857143 697.2098214285714 6.5848214285713C735.7700892857143 22.5446428571428 754.1852678571429 67.3549107142857 738.169642857143 106.3058035714286L707.1986607142858 181.8638392857143C746.9866071428572 208.8169642857143 781.919642857143 243.1919642857144 809.4866071428572 284.5982142857142L883.8169642857144 253.4040178571428C922.2098214285716 237.4441964285715 966.5736607142858 255.8035714285714 982.5334821428575 294.7544642857142C998.5491071428572 334.2075892857144 980.3571428571428 378.9620535714286 941.796875 394.921875zM629.6316964285714 444.0848214285714C598.8839285714287 369.140625 513.9508928571429 333.59375 439.7879464285714 364.5647321428571S330.4687500000001 481.4732142857142 361.0491071428572 556.4174107142858C391.796875 631.3616071428571 476.7299107142857 666.9084821428571 550.8928571428572 635.9375S660.2120535714286 519.0290178571429 629.6316964285714 444.0848214285714z" />
    <glyph glyph-name="share"
      unicode="&#xF175;"
      horiz-adv-x="928.7946428571429" d="M750.1674107142857 357.1986607142857C696.0379464285714 357.1986607142857 647.4888392857143 333.0915178571428 614.7321428571428 295.033482142857L348.3258928571429 408.6495535714285C354.0736607142857 426.1718749999999 357.1986607142857 444.9218749999999 357.1986607142857 464.3973214285712C357.1986607142857 492.0200892857141 350.9486607142857 518.1919642857141 339.7321428571429 541.5736607142856L618.4709821428571 701.0044642857142C651.1160714285714 665.4017857142856 698.046875 643.0803571428571 750.1674107142857 643.0803571428571C848.828125 643.0803571428571 928.7946428571428 723.046875 928.7946428571428 821.7075892857142C928.7946428571428 920.3683035714286 848.8281250000001 1000.3348214285714 750.1674107142859 1000.3348214285714C651.5066964285716 1000.3348214285714 571.5401785714288 920.3683035714286 571.5401785714288 821.7075892857142C571.5401785714288 795.2008928571428 577.3437500000002 770.0334821428571 587.7232142857144 747.3772321428571L308.1473214285716 587.5C275.6138392857145 621.7633928571429 229.6316964285716 643.1361607142858 178.6272321428573 643.1361607142858C79.9665178571428 642.96875 0 563.0022321428571 0 464.3973214285714C0 365.7366071428572 79.9665178571429 285.7700892857142 178.6272321428572 285.7700892857142C238.4486607142857 285.7700892857142 291.3504464285715 315.1785714285714 323.7723214285715 360.2678571428571L585.7700892857143 248.549107142857C576.6183035714286 227.064732142857 571.5401785714287 203.4040178571427 571.5401785714287 178.5714285714285C571.5401785714287 79.9107142857141 651.5066964285714 -0.0558035714288 750.1674107142859 -0.0558035714288C848.8281250000001 -0.0558035714288 928.7946428571428 79.910714285714 928.7946428571428 178.5714285714283C928.7946428571428 277.2321428571428 848.828125 357.1986607142857 750.1674107142857 357.1986607142857z" />
    <glyph glyph-name="sharp_arrow_down"
      unicode="&#xF176;"
      horiz-adv-x="1295.0997398091936" d="M1295.0997398091936 631.1795316565481L647.5860075166233 0L0 631.1795316565481L0 999.9999999999998L647.5860075166233 368.8927435675049L1295.0997398091936 999.9999999999998V631.1795316565481z" />
    <glyph glyph-name="sharp_arrow_down2"
      unicode="&#xF177;"
      horiz-adv-x="1500.7254464285716" d="M750.1674107142857 480.2455357142857L1264.5089285714287 1000L1500.7254464285716 760.8816964285714L749.21875 0L0 758.4821428571429L237.6674107142857 999.1071428571428L750.1674107142857 480.2455357142857z" />
    <glyph glyph-name="sharp_arrow_left"
      unicode="&#xF178;"
      horiz-adv-x="772.0982142857142" d="M487.3325892857143 0L0 499.9441964285714L487.3325892857143 999.9441964285714H772.0982142857142L284.8214285714285 499.9441964285714L772.0982142857142 0H487.3325892857143z" />
    <glyph glyph-name="sharp_arrow_left2"
      unicode="&#xF179;"
      horiz-adv-x="666.5736607142858" d="M665.9598214285714 158.4263392857142L505.5803571428572 0L0 499.3861607142857L507.1428571428571 1000.3348214285714L666.5736607142858 842.9129464285714L320.1450892857144 500.0558035714285L665.9598214285714 158.4263392857142z" />
    <glyph glyph-name="sharp_arrow_right"
      unicode="&#xF17A;"
      horiz-adv-x="772.0982142857142" d="M0 0L487.2767857142858 499.9441964285714L0 1000H284.765625L772.0982142857142 500L284.765625 0H0z" />
    <glyph glyph-name="sharp_arrow_right2"
      unicode="&#xF17B;"
      horiz-adv-x="666.5736607142858" d="M346.4285714285714 500.0558035714285L0 842.9129464285714L159.4308035714286 1000.390625L666.5736607142858 499.4419642857143L160.9933035714286 0.0558035714284L0.6138392857143 158.4821428571429L346.4285714285714 500.0558035714285z" />
    <glyph glyph-name="sharp_arrow_up"
      unicode="&#xF17C;"
      horiz-adv-x="1295.0997398091936" d="M1295.0997398091936 0L647.5860075166233 631.1072564324948L0 0L0 368.8204683434518L647.5860075166233 999.9999999999998L1295.0997398091936 368.8204683434518V0z" />
    <glyph glyph-name="sharp_arrow_up2"
      unicode="&#xF17D;"
      horiz-adv-x="1500.7254464285716" d="M237.6674107142857 0.8928571428571L0 241.5178571428571L749.21875 1000L1500.7812499999998 239.1183035714286L1264.5089285714287 0L750.1674107142857 519.7544642857143L237.6674107142857 0.8928571428571z" />
    <glyph glyph-name="stats"
      unicode="&#xF17E;"
      horiz-adv-x="1000" d="M888.8950892857143 1000H111.1049107142857C50 1000 0 950 0 888.8950892857142V111.1049107142857C0 49.9999999999999 50 0 111.1049107142857 0H888.8950892857142C950 0 999.9999999999998 49.9999999999999 999.9999999999998 111.1049107142857V888.8950892857142C1000 950 950.0000000000002 1000 888.8950892857143 1000M333.3147321428571 222.2098214285715H222.2098214285715V611.1049107142858H333.3147321428572V222.2098214285715zM555.5803571428571 222.2098214285715H444.4754464285714V777.7901785714287H555.5803571428571V222.2098214285715zM777.7901785714286 222.2098214285715H666.6852678571429V444.4196428571428H777.7901785714286V222.2098214285715z" />
    <glyph glyph-name="thesis"
      unicode="&#xF17F;"
      horiz-adv-x="1571.4285714285716" d="M1476.674107142857 1000H94.8102678571429C42.6339285714286 1000 -0.0558035714286 957.3102678571428 -0.0558035714286 905.1339285714286V94.8102678571429C-0.0558035714286 42.6339285714286 42.6339285714286 -0.0558035714286 94.8102678571429 -0.0558035714286H1476.5625C1528.7388392857142 -0.0558035714286 1571.4285714285716 42.6339285714286 1571.4285714285716 92.0200892857143V902.34375C1571.5401785714284 957.3102678571428 1528.8504464285713 1000 1476.674107142857 1000zM1116.5736607142858 170.8147321428571C880.9151785714287 134.5982142857141 730.6361607142859 120.1450892857142 424.3861607142857 170.8147321428571C406.6406250000001 295.8147321428571 424.3861607142857 368.1919642857142 424.3861607142857 368.1919642857142L784.2633928571429 198.9955357142857L1116.5178571428573 368.1361607142856C1116.5736607142858 368.0803571428571 1125.6138392857142 300.4464285714286 1116.5736607142858 170.8147321428571zM784.375 255.3571428571428L96.0379464285714 573.8839285714286L784.3191964285714 860.6026785714286L1472.65625 573.8839285714286L784.375 255.3571428571428z" />
    <glyph glyph-name="thin_arrow_down"
      unicode="&#xF180;"
      horiz-adv-x="624.7209821428571" d="M274.8883928571429 1000.1116071428572V132.9799107142857L0 271.9308035714287V173.7165178571429L312.3325892857144 0.0558035714287L624.7209821428571 173.7165178571429V271.9308035714287L349.8325892857143 133.0357142857143V1000.1116071428572H274.8883928571429z" />
    <glyph glyph-name="thin_arrow_up"
      unicode="&#xF181;"
      horiz-adv-x="624.7209821428571" d="M349.8325892857144 0.111607142857V867.1875L624.7209821428571 728.2924107142858V826.5066964285714L312.3325892857144 1000.1674107142856L0 826.5066964285714V728.2924107142858L274.8883928571429 867.1875V0.0558035714284H349.8325892857144z" />
    <glyph glyph-name="thumb_3dviewer"
      unicode="&#xF182;"
      horiz-adv-x="1556.0267857142858" d="M1450.7254464285713 1000.0558035714286H105.1339285714286C47.0424107142857 1000.0558035714286 0 954.3526785714286 0 897.9910714285714V102.0647321428572C0 45.703125 47.0982142857143 0 105.1339285714286 0H1450.8370535714287C1508.9285714285716 0 1555.9709821428573 45.703125 1555.9709821428573 102.0647321428572V897.9910714285714C1555.9151785714284 954.3526785714286 1508.872767857143 1000.0558035714286 1450.7254464285713 1000.0558035714286zM1057.9799107142858 318.8058035714286V355.1339285714285H669.1964285714287H649.9441964285714L470.3125000000001 160.9374999999999L497.5446428571429 134.8214285714285L389.0625000000001 111.1607142857141L408.8169642857144 219.9218749999999L445.2566964285715 184.9330357142857L632.9241071428572 387.6674107142856V391.3504464285712V798.3258928571427H596.5959821428572L650.9486607142859 889.0624999999998L705.3013392857144 798.3258928571427H669.1964285714288V391.3504464285712H1057.979910714286V427.4553571428571L1166.852678571429 373.1026785714286L1057.9799107142858 318.8058035714286zM741.2388392857143 518.4151785714286C746.484375 515.0111607142857 758.7053571428571 509.765625 771.5959821428572 509.765625C795.3683035714286 509.765625 802.7901785714287 524.9441964285714 802.5669642857143 536.328125C802.3995535714287 555.46875 785.1004464285714 563.7276785714287 767.1875 563.7276785714287H756.8638392857143V577.6227678571429H767.1875C780.6919642857143 577.6227678571429 797.7120535714286 584.5424107142858 797.7120535714286 600.78125C797.7120535714286 611.71875 790.7366071428571 621.4285714285714 773.6607142857143 621.4285714285714C762.7232142857144 621.4285714285714 752.1763392857143 616.5736607142857 746.2611607142859 612.3883928571429L741.4062500000001 625.8928571428571C748.5491071428572 631.1383928571429 762.5000000000001 636.4397321428571 777.232142857143 636.4397321428571C804.185267857143 636.4397321428571 816.4620535714287 620.4241071428571 816.4620535714287 603.7946428571429C816.4620535714287 589.6763392857142 808.0357142857144 577.6785714285714 791.1830357142859 571.5401785714287V571.09375C808.0357142857144 567.7455357142858 821.7075892857144 555.078125 821.7075892857144 535.9375C821.7075892857144 514.0066964285716 804.6316964285717 494.8660714285714 771.763392857143 494.8660714285714C756.3616071428573 494.8660714285714 742.9129464285716 499.7209821428572 736.1607142857144 504.1294642857143L741.2388392857143 518.4151785714286zM896.0379464285713 637.2209821428571C907.1986607142856 638.8950892857142 920.4799107142856 640.1785714285713 935.0446428571428 640.1785714285713C961.3839285714286 640.1785714285713 980.1339285714286 634.0401785714284 992.5781249999998 622.4888392857142C1005.2455357142856 610.8816964285713 1012.6116071428572 594.4754464285713 1012.6116071428572 571.484375C1012.6116071428572 548.3258928571428 1005.46875 529.3526785714284 992.1316964285714 516.2946428571429C978.8504464285714 503.0133928571428 956.9196428571428 495.8705357142857 929.3526785714286 495.8705357142857C916.2946428571428 495.8705357142857 905.3571428571428 496.4843749999999 896.0379464285713 497.5446428571428V637.2209821428571zM914.3973214285714 511.6629464285714C919.0290178571428 510.8258928571428 925.78125 510.6026785714286 932.9241071428572 510.6026785714286C972.1540178571428 510.6026785714286 993.4151785714286 532.5334821428571 993.4151785714286 570.8705357142858C993.638392857143 604.3526785714286 974.6651785714286 625.6696428571429 935.8816964285716 625.6696428571429C926.3950892857144 625.6696428571429 919.1964285714286 624.8325892857142 914.3973214285716 623.7723214285714V511.6629464285714z" />
    <glyph glyph-name="thumb_3dviewer_i"
      unicode="&#xF183;"
      horiz-adv-x="1556.0267857142858" d="M1057.9799107142858 318.75V355.078125H669.1964285714287H649.9441964285714L470.3125000000001 160.8816964285713L497.5446428571429 134.765625L389.0625000000001 111.1049107142857L408.8169642857144 219.8660714285715L445.2566964285715 184.8772321428571L632.9241071428572 387.6116071428571V391.2946428571428V798.2700892857142H596.5959821428572L650.9486607142859 889.0066964285714L705.3571428571429 798.2700892857142H669.2522321428571V391.2946428571429H1058.0357142857142V427.4553571428571L1166.908482142857 373.1026785714286L1057.9799107142858 318.75zM741.2946428571429 518.4151785714286C746.5401785714287 515.0111607142857 758.7611607142858 509.765625 771.6517857142859 509.765625C795.4241071428572 509.765625 802.8459821428573 524.9441964285714 802.622767857143 536.328125C802.4553571428573 555.46875 785.1562500000001 563.7276785714287 767.2433035714287 563.7276785714287H756.9196428571429V577.5669642857142H767.2433035714287C780.747767857143 577.5669642857142 797.7678571428572 584.4866071428571 797.7678571428572 600.7254464285714C797.7678571428572 611.6629464285714 790.7924107142858 621.3727678571429 773.716517857143 621.3727678571429C762.779017857143 621.3727678571429 752.232142857143 616.5178571428571 746.3169642857144 612.3325892857142L741.4620535714288 625.8370535714286C748.6049107142859 631.0825892857142 762.5558035714287 636.3839285714286 777.2879464285717 636.3839285714286C804.2410714285716 636.3839285714286 816.5178571428573 620.3683035714284 816.5178571428573 603.7388392857142C816.5178571428573 589.6205357142858 808.0915178571431 577.6227678571429 791.2388392857144 571.484375V571.0379464285714C808.0915178571431 567.6897321428571 821.7633928571431 555.0223214285714 821.7633928571431 535.8816964285714C821.7633928571431 513.9508928571429 804.6875000000002 494.8102678571429 771.8191964285717 494.8102678571429C756.417410714286 494.8102678571429 742.9687500000002 499.6651785714286 736.2165178571431 504.0736607142857L741.2946428571429 518.4151785714286zM896.09375 637.2209821428571C907.2544642857144 638.8950892857142 920.5357142857142 640.1785714285713 935.1004464285714 640.1785714285713C961.4397321428572 640.1785714285713 980.1897321428572 634.0401785714284 992.6339285714286 622.4888392857142C1005.3013392857144 610.8816964285713 1012.6674107142858 594.4754464285713 1012.6674107142858 571.484375C1012.6674107142858 548.3258928571428 1005.5245535714286 529.3526785714284 992.1875 516.2946428571429C978.90625 503.0133928571428 956.9754464285716 495.8705357142857 929.4084821428572 495.8705357142857C916.3504464285714 495.8705357142857 905.4129464285714 496.4843749999999 896.09375 497.5446428571428V637.2209821428571zM914.4531249999998 511.6629464285714C919.0848214285714 510.8258928571428 925.8370535714286 510.6026785714286 932.9799107142856 510.6026785714286C972.2098214285714 510.6026785714286 993.4709821428572 532.5334821428571 993.4709821428572 570.8705357142858C993.6941964285714 604.3526785714286 974.7209821428572 625.6696428571429 935.9375 625.6696428571429C926.4508928571428 625.6696428571429 919.2522321428572 624.8325892857142 914.453125 623.7723214285714L914.4531249999998 511.6629464285714L914.4531249999998 511.6629464285714z" />
    <glyph glyph-name="thumb_archive"
      unicode="&#xF184;"
      horiz-adv-x="1555.859375" d="M1450.6138392857142 1000H105.1339285714286C47.0982142857143 1000 0 954.296875 0 897.9352678571429V102.0647321428572C0 45.703125 47.0424107142858 0 105.1339285714286 0H1450.7812500000002C1508.8169642857144 0 1555.9151785714287 45.703125 1555.9151785714287 102.0647321428572V897.8794642857142C1555.8035714285716 954.296875 1508.7611607142856 1000 1450.6138392857142 1000zM694.53125 822.65625V848.9955357142858V855.5803571428571H791.4062499999999V848.9955357142858H823.7165178571428V822.65625H791.4062499999999V816.0714285714286H694.53125V822.65625zM694.53125 700.390625V726.7299107142858V733.2589285714286C694.53125 733.2589285714286 791.4062499999999 733.2589285714286 791.4062499999999 733.203125V726.6183035714286H823.7165178571428V700.2790178571429H791.4062499999999V693.75H694.53125V700.390625zM694.53125 578.3482142857142V604.6875V611.2723214285713H791.4062499999999V604.6875H823.7165178571428V578.3482142857142H791.4062499999999V571.7633928571429H694.53125V578.3482142857142zM694.53125 456.3616071428571V482.7008928571429V489.2299107142857H791.4062499999999V482.6450892857143H823.7165178571428V456.3058035714286H791.4062499999999V449.7209821428572H694.53125V456.3616071428571zM694.53125 334.2075892857144V360.4910714285715V367.0758928571428H791.4062499999999V360.4910714285715H823.7165178571428V334.1517857142857H791.4062499999999V327.5669642857142H694.53125V334.2075892857144zM694.53125 212.1651785714286V238.5044642857144V245.0892857142857H791.4062499999999V238.5044642857144H823.7165178571428V212.1651785714286H791.4062499999999V205.5803571428572H694.53125V212.1651785714286zM823.6049107142858 90.1227678571428H791.2946428571429V83.5379464285715H694.4196428571429V90.1227678571428V116.4620535714286V123.046875H791.2946428571429V116.4620535714286H823.6049107142858V90.1227678571428zM861.216517857143 177.4553571428571V151.1160714285713V144.53125H764.341517857143V151.1160714285713H732.0312500000001V177.4553571428571H764.341517857143V184.0401785714286H861.216517857143V177.4553571428571zM861.216517857143 299.4419642857142V273.1026785714285V266.5178571428571H764.341517857143V273.1026785714285H732.0312500000001V299.4419642857142H764.341517857143V306.0267857142857H861.216517857143V299.4419642857142zM861.216517857143 421.7075892857144V395.3683035714286V388.8392857142857H764.341517857143V395.3683035714286H732.0312500000001V421.7075892857144H764.341517857143V428.2924107142857H861.216517857143V421.7075892857144zM861.216517857143 543.6941964285713V517.3549107142857V510.7700892857142H764.341517857143V517.3549107142857H732.0312500000001V543.6941964285713H764.341517857143V550.2790178571428H861.216517857143V543.6941964285713zM861.216517857143 665.6808035714286V639.3415178571429V632.7566964285713H764.341517857143V639.3415178571429H732.0312500000001V665.6808035714286H764.341517857143V672.265625H861.216517857143V665.6808035714286zM861.216517857143 787.9464285714286V761.6071428571429V755.0223214285714H764.341517857143V761.6071428571429H732.0312500000001V787.9464285714286H764.341517857143V794.53125H861.216517857143V787.9464285714286zM861.216517857143 909.9330357142858V883.59375V877.0089285714286H764.341517857143V883.59375H732.0312500000001V909.9330357142858H764.341517857143V916.5178571428572H861.216517857143V909.9330357142858z" />
    <glyph glyph-name="thumb_audio"
      unicode="&#xF185;"
      horiz-adv-x="1555.859375" d="M1450.6138392857142 999.9441964285714H105.1339285714286C47.0424107142858 999.9441964285714 0 954.2410714285714 0 897.8794642857142V102.0647321428572C0 45.703125 47.0424107142858 0 105.1339285714286 0H1450.7254464285716C1508.8169642857144 0 1555.8593750000002 45.703125 1555.8593750000002 102.0647321428572V897.9352678571429C1555.7477678571427 954.2410714285714 1508.7611607142856 999.9441964285714 1450.6138392857142 999.9441964285714zM847.3214285714287 166.6294642857142L597.3214285714287 361.1049107142857C597.2656250000001 360.3236607142856 493.3593750000001 360.9933035714285 458.4263392857143 361.1049107142857S402.8459821428572 384.4308035714285 402.8459821428572 421.875C402.8459821428572 449.8325892857144 402.8459821428572 559.0959821428571 402.8459821428572 574.3303571428571C402.8459821428572 609.375 423.3816964285715 637.890625 458.4263392857143 638.8950892857143C497.6562500000001 640.0111607142858 597.3214285714287 638.8950892857143 597.3214285714287 638.8950892857143L847.3214285714287 833.3705357142858V166.6294642857142zM950.0558035714286 361.0491071428571C922.7120535714286 385.3794642857142 903.2366071428572 397.7120535714286 907.9241071428572 402.9575892857142C958.4263392857144 456.8638392857142 973.7165178571428 535.4910714285713 907.9241071428572 601.0044642857142C907.9241071428572 601.0044642857142 928.1808035714286 621.5401785714284 950.0558035714286 642.9129464285713C1033.1473214285716 584.3191964285714 1047.377232142857 433.3147321428571 950.0558035714286 361.0491071428571zM1045.5357142857142 257.9241071428571C1026.953125 276.0602678571429 1003.2366071428572 300.5580357142858 995.7589285714286 307.4776785714286C1104.575892857143 402.3995535714286 1119.8660714285713 592.2991071428571 995.7589285714286 696.09375C995.7589285714286 696.09375 1015.2901785714286 718.4151785714286 1049.330357142857 749.4419642857142C1190.8482142857142 615.0111607142858 1185.546875 387.5558035714286 1045.5357142857142 257.9241071428571z" />
    <glyph glyph-name="thumb_audio_i"
      unicode="&#xF186;"
      horiz-adv-x="1555.859375" d="M847.3214285714287 166.6294642857142L597.3214285714287 361.1049107142857C597.2656250000001 360.3236607142856 493.3593750000001 360.9933035714285 458.4263392857143 361.1049107142857S402.8459821428572 384.4308035714285 402.8459821428572 421.875C402.8459821428572 449.8325892857144 402.8459821428572 559.0959821428571 402.8459821428572 574.3303571428571C402.8459821428572 609.375 423.3816964285715 637.890625 458.4263392857143 638.8950892857143C497.6562500000001 640.0111607142858 597.3214285714287 638.8950892857143 597.3214285714287 638.8950892857143L847.3214285714287 833.3705357142858V166.6294642857142zM950.0558035714286 361.0491071428571C922.7120535714286 385.3794642857142 903.2366071428572 397.7120535714286 907.9241071428572 402.9575892857142C958.4263392857144 456.8638392857142 973.7165178571428 535.4910714285713 907.9241071428572 601.0044642857142C907.9241071428572 601.0044642857142 928.1808035714286 621.5401785714284 950.0558035714286 642.9129464285713C1033.1473214285716 584.3191964285714 1047.377232142857 433.3147321428571 950.0558035714286 361.0491071428571zM1045.5357142857142 257.9241071428571C1026.953125 276.0602678571429 1003.2366071428572 300.5580357142858 995.7589285714286 307.4776785714286C1104.575892857143 402.3995535714286 1119.8660714285713 592.2991071428571 995.7589285714286 696.09375C995.7589285714286 696.09375 1015.2901785714286 718.4151785714286 1049.330357142857 749.4419642857142C1190.8482142857142 615.0111607142858 1185.546875 387.5558035714286 1045.5357142857142 257.9241071428571z" />
    <glyph glyph-name="thumb_code_text"
      unicode="&#xF187;"
      horiz-adv-x="1555.9709821428573" d="M1450.7254464285713 1000H105.1339285714286C47.0424107142857 1000 0 954.296875 0 897.9352678571429V102.0089285714286C0 45.6473214285713 47.0982142857143 -0.0558035714287 105.1339285714286 -0.0558035714287H1450.8370535714287C1508.9285714285716 -0.0558035714287 1555.9709821428573 45.6473214285713 1555.9709821428573 102.0089285714286V897.9352678571428C1555.9151785714284 954.296875 1508.872767857143 1000 1450.7254464285713 1000zM498.4933035714286 360.3236607142857C511.71875 347.265625 511.6629464285714 325.8928571428571 498.4375 312.8348214285715L498.4375 312.8348214285715C485.3236607142857 299.8883928571429 464.1741071428571 300.0558035714286 451.2276785714286 313.1696428571429L283.8727678571429 482.9241071428572C271.0379464285714 495.9263392857143 271.0379464285714 516.796875 283.8727678571429 529.7991071428571L451.2276785714286 699.5535714285716C464.1741071428571 712.6674107142858 485.3236607142857 712.8348214285714 498.4375 699.8883928571429L498.4933035714286 699.8325892857143C511.71875 686.7745535714287 511.7745535714286 665.4575892857143 498.6049107142857 652.3995535714287L351.3950892857144 505.9709821428572L498.4933035714286 360.3236607142857zM866.9642857142857 482.9241071428571L699.6093749999999 313.1696428571429C686.6629464285713 300.0558035714286 665.5133928571428 299.8883928571429 652.3995535714286 312.8348214285715H652.3995535714286C639.1741071428571 325.8928571428572 639.1183035714286 347.2656250000001 652.34375 360.3236607142857L799.4419642857142 505.9709821428572L652.2321428571429 652.3995535714286C639.0625 665.5133928571429 639.1183035714286 686.8303571428571 652.34375 699.8325892857142L652.3995535714286 699.8883928571429C665.5133928571428 712.8348214285714 686.6629464285714 712.6674107142858 699.6093749999999 699.5535714285714L866.9642857142857 529.7991071428571C879.7991071428571 516.796875 879.7991071428571 495.9263392857143 866.9642857142857 482.9241071428571zM1257.1986607142858 337.2209821428571C1257.1986607142858 318.8058035714286 1242.2433035714287 303.8504464285715 1223.828125 303.8504464285715H982.7566964285714C964.3415178571428 303.8504464285715 949.3861607142856 318.8058035714286 949.3861607142856 337.2209821428571V338.392857142857C949.3861607142856 356.8080357142857 964.3415178571428 371.7633928571428 982.7566964285714 371.7633928571428H1223.828125C1242.2433035714287 371.7633928571428 1257.1986607142858 356.8080357142857 1257.1986607142858 338.392857142857V337.2209821428571zM1257.1986607142858 505.8035714285714C1257.1986607142858 487.3883928571428 1242.2433035714287 472.4330357142857 1223.828125 472.4330357142857H1043.638392857143C1025.2232142857142 472.4330357142857 1010.2678571428572 487.3883928571428 1010.2678571428572 505.8035714285714V506.9754464285714C1010.2678571428572 525.390625 1025.2232142857142 540.3459821428571 1043.638392857143 540.3459821428571H1223.7723214285716C1242.1875 540.3459821428571 1257.1428571428573 525.390625 1257.1428571428573 506.9754464285714V505.8035714285714zM1257.1986607142858 676.0044642857143C1257.1986607142858 657.5892857142858 1242.2433035714287 642.6339285714287 1223.828125 642.6339285714287H982.7566964285714C964.3415178571428 642.6339285714287 949.3861607142856 657.5892857142858 949.3861607142856 676.0044642857143V677.1763392857143C949.3861607142856 695.5915178571429 964.3415178571428 710.546875 982.7566964285714 710.546875H1223.828125C1242.2433035714287 710.546875 1257.1986607142858 695.5915178571429 1257.1986607142858 677.1763392857143V676.0044642857143z" />
    <glyph glyph-name="thumb_code_text_i"
      unicode="&#xF188;"
      horiz-adv-x="1555.9709821428573" d="M498.4933035714286 360.3794642857142C511.71875 347.3214285714285 511.6629464285714 325.9486607142857 498.4375 312.890625L498.4375 312.890625C485.3236607142857 299.9441964285715 464.1741071428571 300.1116071428571 451.2276785714286 313.2254464285714L283.8727678571429 482.9799107142857C271.0379464285715 495.9821428571428 271.0379464285715 516.8526785714286 283.8727678571429 529.8549107142858L451.2276785714286 699.609375C464.1741071428572 712.7232142857142 485.3236607142858 712.890625 498.4375000000001 699.9441964285714L498.4933035714286 699.8883928571429C511.7187500000001 686.8303571428571 511.7745535714286 665.5133928571429 498.6049107142858 652.4553571428571L351.3950892857144 506.0267857142857L498.4933035714286 360.3794642857142zM866.9642857142857 482.9799107142857L699.6093749999999 313.2254464285714C686.6629464285713 300.1116071428571 665.5133928571428 299.9441964285715 652.3995535714286 312.890625L652.3995535714286 312.890625C639.1741071428571 325.9486607142858 639.1183035714286 347.3214285714286 652.34375 360.3794642857142L799.4419642857142 506.0267857142857L652.2321428571429 652.4553571428571C639.0625 665.5691964285713 639.1183035714286 686.8861607142857 652.34375 699.8883928571429L652.3995535714286 699.9441964285713C665.5133928571428 712.890625 686.6629464285714 712.7232142857142 699.6093749999999 699.609375L866.9642857142857 529.8549107142858C879.7991071428571 516.796875 879.7991071428571 495.9821428571428 866.9642857142857 482.9799107142857zM1257.1986607142858 337.2209821428571C1257.1986607142858 318.8058035714286 1242.2433035714287 303.8504464285715 1223.828125 303.8504464285715H982.7566964285714C964.3415178571428 303.8504464285715 949.3861607142856 318.8058035714286 949.3861607142856 337.2209821428571V338.392857142857C949.3861607142856 356.8080357142857 964.3415178571428 371.7633928571428 982.7566964285714 371.7633928571428H1223.828125C1242.2433035714287 371.7633928571428 1257.1986607142858 356.8080357142857 1257.1986607142858 338.392857142857V337.2209821428571zM1257.1986607142858 505.859375C1257.1986607142858 487.4441964285714 1242.2433035714287 472.4888392857143 1223.828125 472.4888392857143H1043.638392857143C1025.2232142857142 472.4888392857143 1010.2678571428572 487.4441964285714 1010.2678571428572 505.859375V507.03125C1010.2678571428572 525.4464285714286 1025.2232142857142 540.4017857142857 1043.638392857143 540.4017857142857H1223.7723214285716C1242.1875 540.4017857142857 1257.1428571428573 525.4464285714286 1257.1428571428573 507.03125L1257.1986607142858 505.859375L1257.1986607142858 505.859375zM1257.1986607142858 676.0602678571429C1257.1986607142858 657.6450892857142 1242.2433035714287 642.6897321428571 1223.828125 642.6897321428571H982.7566964285714C964.3415178571428 642.6897321428571 949.3861607142856 657.6450892857142 949.3861607142856 676.0602678571429V677.2321428571429C949.3861607142856 695.6473214285714 964.3415178571428 710.6026785714286 982.7566964285714 710.6026785714286H1223.828125C1242.2433035714287 710.6026785714286 1257.1986607142858 695.6473214285714 1257.1986607142858 677.2321428571429V676.0602678571429z" />
    <glyph glyph-name="thumb_dataset_i"
      unicode="&#xF189;"
      horiz-adv-x="1571.4285714285716" d="M567.4107142857142 141.1830357142857H1008.8727678571428V69.7544642857142H567.4107142857143V141.1830357142857zM567.4107142857142 192.96875H1008.8727678571428V264.3973214285714H567.4107142857143V192.96875zM70.2008928571428 562.5H513.3370535714286V633.9285714285714H70.2008928571428V562.5zM70.2008928571428 97.65625C70.2008928571428 83.2031249999999 77.2879464285714 69.7544642857142 98.1026785714286 69.7544642857142C126.953125 69.7544642857142 513.3370535714286 69.7544642857142 513.3370535714286 69.7544642857142V141.1830357142857H70.2008928571428C70.2008928571428 141.1830357142857 70.2008928571428 100 70.2008928571428 97.65625zM70.2008928571428 192.96875H513.3370535714286V264.3973214285714H70.2008928571428V192.96875zM567.4107142857142 316.1272321428571H1008.8727678571428V387.5558035714286H567.4107142857143V316.1272321428571zM70.2008928571428 439.2857142857144H513.3370535714286V510.7142857142857H70.2008928571428V439.2857142857144zM70.2008928571428 316.1272321428571H513.3370535714286V387.5558035714286H70.2008928571428V316.1272321428571zM1061.216517857143 439.2857142857144H1501.5625H1501.5625V510.7142857142857H1061.216517857143V439.2857142857144zM1061.216517857143 562.5H1501.5625H1501.5625V633.9285714285714H1061.216517857143V562.5zM1061.216517857143 316.1272321428571H1501.5625H1501.5625V387.5558035714286H1061.216517857143V316.1272321428571zM567.4107142857142 439.2857142857144H1008.8727678571428V510.7142857142857H567.4107142857143V439.2857142857144zM567.4107142857142 562.5H1008.8727678571428V633.9285714285714H567.4107142857143V562.5zM1061.216517857143 192.96875H1501.5625H1501.5625V264.3973214285714H1061.216517857143V192.96875zM1061.216517857143 69.7544642857142C1061.216517857143 69.7544642857142 1452.9575892857142 69.7544642857142 1473.6607142857142 69.7544642857142C1494.810267857143 69.7544642857142 1501.5625 75.8928571428571 1501.5625 97.65625C1501.5625 107.3102678571428 1501.5625 141.1830357142857 1501.5625 141.1830357142857H1061.216517857143V69.7544642857142zM1571.484375 897.9910714285714V102.0647321428572C1571.484375 45.703125 1523.9397321428573 0 1465.2901785714284 0H106.1383928571429C47.4888392857143 0 -0.0558035714286 45.703125 -0.0558035714286 102.0647321428572V897.9910714285714C-0.0558035714286 954.3526785714286 47.4888392857143 1000.0558035714286 106.1383928571429 1000.0558035714286H1465.1785714285718C1523.9397321428573 1000.0558035714286 1571.372767857143 954.3526785714286 1571.484375 897.9910714285714zM1543.5825892857142 897.9910714285714C1543.5825892857142 917.578125 1535.546875 936.1049107142856 1521.0379464285713 950.0558035714286C1506.1941964285713 964.2857142857142 1486.3839285714287 972.1540178571428 1465.234375 972.1540178571428H106.1383928571429C62.9464285714286 972.1540178571428 27.8459821428572 938.8950892857142 27.8459821428572 897.9910714285714V102.0647321428572C27.8459821428572 61.1607142857143 62.9464285714286 27.9017857142857 106.1383928571429 27.9017857142857H1465.234375C1508.4263392857142 27.9017857142857 1543.5267857142858 61.1607142857142 1543.5267857142858 102.0647321428572V897.9910714285714z" />
    <glyph glyph-name="thumb_generic"
      unicode="&#xF18A;"
      horiz-adv-x="1555.859375" d="M831.3616071428571 660.2678571428571H1009.4308035714286V161.6071428571428H546.372767857143V838.3928571428571H831.3616071428572V660.2678571428571zM1450.6138392857142 1000H105.1339285714286C47.0424107142857 1000 0 954.296875 0 897.9352678571429V102.064732142857C0 45.7031249999999 47.0424107142857 -1e-13 105.1339285714286 -1e-13H1450.7254464285716C1508.8169642857144 -1e-13 1555.8593750000002 45.7031249999999 1555.8593750000002 102.064732142857V897.9352678571429C1555.747767857143 954.296875 1508.7611607142856 1000 1450.6138392857142 1000zM510.7700892857143 126.0044642857142V873.9955357142857H849.1629464285713L1045.033482142857 676.0044642857142V126.0044642857142H510.7700892857143z" />
    <glyph glyph-name="thumb_generic_i"
      unicode="&#xF18B;"
      horiz-adv-x="1555.859375" d="M849.1629464285714 873.9955357142857H510.7700892857143V126.0044642857142H1045.033482142857V676.0044642857142L849.1629464285714 873.9955357142857zM1009.4308035714286 161.6071428571428H546.372767857143V838.3928571428571H831.3616071428572V660.2678571428571H1009.4308035714286V161.6071428571428z" />
    <glyph glyph-name="thumb_graph"
      unicode="&#xF18C;"
      horiz-adv-x="1555.859375" d="M988.1138392857144 354.5200892857142C955.9709821428572 373.7723214285714 921.3169642857144 384.4866071428571 889.6763392857143 390.2901785714285C889.2299107142858 386.6071428571428 888.5602678571429 382.8683035714286 887.7232142857143 379.1852678571427C887.6674107142858 379.017857142857 887.6674107142858 378.9062499999999 887.6116071428571 378.7388392857142C887.4441964285714 378.0691964285714 887.2767857142858 377.3995535714285 887.1093749999999 376.7299107142857C887.0535714285714 376.3950892857142 886.9419642857142 376.1160714285715 886.8861607142856 375.7812499999999C886.7745535714284 375.2790178571427 886.6071428571428 374.8325892857142 886.4955357142856 374.3303571428571C886.3839285714284 373.8281249999999 886.2165178571427 373.3816964285714 886.1049107142856 372.8794642857144C885.9933035714283 372.5446428571429 885.9374999999999 372.2656250000001 885.8258928571427 371.9308035714286C885.658482142857 371.2611607142857 885.4352678571427 370.6473214285715 885.2120535714284 369.9776785714286C885.1562499999999 369.8102678571429 885.1004464285713 369.6986607142857 885.0446428571428 369.53125C883.0357142857142 363.3928571428571 880.580357142857 357.5334821428571 877.6227678571428 351.8973214285715C877.5669642857142 351.7857142857144 877.5111607142857 351.6741071428571 877.4553571428571 351.5625000000001C877.1205357142857 350.9486607142859 876.8415178571429 350.3906250000001 876.5066964285713 349.7767857142858C876.3392857142857 349.4419642857145 876.1160714285713 349.1071428571429 875.9486607142857 348.7723214285716C875.78125 348.4375000000001 875.5580357142857 348.1026785714287 875.3348214285714 347.7678571428572C875 347.2098214285716 874.6651785714286 346.5959821428574 874.3303571428571 346.0379464285716C874.2745535714287 345.9263392857145 874.21875 345.8147321428572 874.1629464285714 345.7589285714288C870.3683035714287 339.5647321428574 866.0156250000001 333.6495535714288 860.9933035714287 328.1808035714288C860.9375000000001 328.1250000000003 860.8816964285714 328.0691964285716 860.8816964285714 328.0133928571431C860.3794642857143 327.4553571428574 859.8772321428572 326.8973214285717 859.3191964285716 326.3950892857145C859.0959821428572 326.1718750000001 858.9285714285714 325.9486607142859 858.7053571428572 325.7812500000003C858.3147321428572 325.3906250000003 857.9799107142859 325.0558035714288 857.5892857142859 324.6651785714288C857.1986607142858 324.2745535714288 856.8638392857144 323.9397321428574 856.4732142857144 323.5491071428574C856.2500000000001 323.325892857143 856.0825892857143 323.1584821428574 855.8593750000002 322.9352678571431C855.3013392857144 322.4330357142859 854.7433035714288 321.8750000000003 854.2410714285716 321.3727678571431C854.185267857143 321.3169642857147 854.1294642857143 321.261160714286 854.0736607142859 321.2053571428575C850.6138392857144 318.0245535714289 846.9308035714288 315.0669642857146 843.1361607142859 312.3325892857146C862.6116071428575 288.0022321428575 888.5602678571431 262.0535714285718 921.2053571428575 242.5223214285718C963.5044642857146 217.2433035714289 1009.9888392857144 206.6406250000004 1048.3816964285718 202.6227678571431C1049.7209821428576 213.9508928571431 1053.4598214285718 225.1674107142861 1059.7098214285718 235.6026785714288C1065.9598214285718 246.0379464285717 1074.1071428571431 254.6316964285717 1083.4821428571431 261.2165178571431C1063.1138392857142 291.2388392857144 1031.529017857143 328.5714285714286 988.1138392857144 354.5200892857142zM811.8861607142858 740.234375C801.4508928571429 735.6026785714286 789.8995535714286 732.9799107142858 777.7901785714287 732.9799107142858C765.6250000000001 732.9799107142858 754.0736607142859 735.6026785714286 743.638392857143 740.234375C728.4598214285714 707.4776785714286 712.6116071428573 661.2165178571429 712.6116071428573 610.6026785714287C712.6116071428573 573.1026785714286 721.1495535714287 537.890625 732.4218750000001 507.7566964285714C735.044642857143 508.8727678571429 737.6674107142859 509.9330357142857 740.3459821428573 510.8816964285715C740.3459821428573 510.8816964285715 740.3459821428573 510.8816964285715 740.3459821428573 510.8816964285715C741.1272321428573 511.1607142857143 741.9642857142859 511.4397321428572 742.7455357142859 511.7187500000001C742.9129464285716 511.7745535714286 743.0803571428572 511.8303571428572 743.247767857143 511.8861607142857C743.9174107142859 512.109375 744.5312500000002 512.2767857142858 745.200892857143 512.5C745.5357142857144 512.6116071428572 745.8705357142859 512.6674107142858 746.1495535714288 512.7790178571429C746.651785714286 512.9464285714286 747.0982142857146 513.0580357142858 747.6004464285716 513.1696428571429C748.1026785714288 513.2812500000001 748.6049107142859 513.4486607142858 749.0513392857144 513.5602678571429C749.3861607142858 513.6718750000001 749.7209821428573 513.7276785714287 750.0000000000001 513.7834821428572C750.669642857143 513.9508928571429 751.3392857142859 514.1183035714287 751.9531250000001 514.2857142857143C752.1205357142859 514.3415178571429 752.2879464285716 514.3415178571429 752.4553571428573 514.3973214285716C753.2924107142859 514.5647321428572 754.1294642857144 514.7879464285716 754.966517857143 514.9553571428572C754.966517857143 514.9553571428572 754.966517857143 514.9553571428572 754.966517857143 514.9553571428572C761.3281250000002 516.2388392857143 767.745535714286 517.0200892857144 774.1629464285717 517.1875000000001C774.2745535714289 517.1875000000001 774.386160714286 517.1875000000001 774.4419642857146 517.1875000000001C775.1116071428575 517.1875000000001 775.7812500000003 517.2433035714287 776.450892857143 517.2433035714287C776.8415178571431 517.2433035714287 777.2321428571431 517.2433035714287 777.622767857143 517.2433035714287C778.013392857143 517.2433035714287 778.404017857143 517.2433035714287 778.794642857143 517.2433035714287C779.4642857142859 517.2433035714287 780.1339285714287 517.2433035714287 780.8035714285714 517.1875000000001C780.9151785714287 517.1875000000001 781.0267857142858 517.1875000000001 781.0825892857143 517.1875000000001C788.3928571428571 516.9642857142858 795.7031250000001 516.0714285714287 802.9575892857143 514.3973214285716C803.125 514.3415178571429 803.2366071428572 514.3415178571429 803.4040178571429 514.2857142857143C804.0736607142858 514.1183035714287 804.7433035714287 513.9508928571429 805.4129464285714 513.7834821428572C805.7477678571428 513.7276785714287 806.0267857142857 513.6160714285716 806.3616071428571 513.5602678571429C806.8638392857143 513.4486607142858 807.3102678571429 513.2812500000001 807.8125 513.1696428571429C808.3147321428571 513.0580357142858 808.7611607142857 512.890625 809.2633928571428 512.7790178571429C809.5982142857141 512.6674107142858 809.877232142857 512.6116071428572 810.2120535714286 512.5C810.8816964285714 512.3325892857143 811.4955357142857 512.109375 812.1651785714286 511.8861607142857C812.3325892857142 511.8303571428572 812.4441964285713 511.7745535714286 812.6116071428571 511.7187500000001C816.1830357142857 510.5468750000001 819.6986607142857 509.2075892857143 823.1584821428571 507.7008928571429C834.0401785714286 536.9419642857142 842.96875 572.4888392857143 842.96875 610.546875C842.9129464285714 659.8772321428571 828.125 705.1897321428571 811.8861607142858 740.234375zM701.4508928571429 321.3727678571429C700.8928571428571 321.8750000000001 700.3348214285714 322.3772321428571 699.8325892857142 322.9352678571429C699.6093749999999 323.1584821428572 699.3861607142857 323.3258928571429 699.21875 323.5491071428571C698.828125 323.9397321428571 698.4933035714287 324.2745535714286 698.1026785714286 324.6651785714286C697.7120535714286 325.0558035714286 697.3772321428572 325.3906249999999 696.9866071428571 325.78125C696.7633928571428 326.0044642857144 696.5959821428571 326.171875 696.3727678571429 326.3950892857142C695.8705357142858 326.9531249999999 695.3125 327.5111607142857 694.810267857143 328.0133928571428C694.7544642857144 328.0691964285714 694.6986607142858 328.125 694.6428571428572 328.1808035714286C689.6205357142858 333.6495535714285 685.1562500000001 339.5647321428571 681.3616071428573 345.8705357142857C681.3058035714288 345.9821428571429 681.2500000000001 346.0379464285714 681.1941964285716 346.1495535714286C680.8593750000002 346.7075892857142 680.5245535714287 347.265625 680.1897321428573 347.8236607142857C680.0223214285717 348.158482142857 679.7991071428573 348.4933035714286 679.6316964285717 348.8281249999999C679.4642857142859 349.1629464285714 679.2410714285716 349.4977678571428 679.0736607142859 349.8325892857142C678.7388392857146 350.3906249999999 678.4598214285717 351.0044642857141 678.1250000000002 351.5624999999999C678.0691964285717 351.674107142857 678.013392857143 351.7857142857142 677.9575892857144 351.8415178571427C675.3906250000002 356.6964285714285 673.1584821428573 361.7745535714285 671.3169642857144 367.0200892857142C671.3169642857144 367.0200892857142 671.3169642857144 367.0200892857142 671.3169642857144 367.0200892857142C671.0379464285716 367.8013392857142 670.7589285714287 368.6383928571428 670.4799107142859 369.4196428571427C670.4241071428573 369.5870535714285 670.3683035714287 369.7544642857141 670.3125000000002 369.9218749999999C670.0892857142858 370.5915178571428 669.9218750000001 371.205357142857 669.6986607142859 371.8749999999999C669.5870535714288 372.2098214285714 669.5312500000002 372.5446428571428 669.4196428571431 372.8236607142857C669.2522321428573 373.3258928571429 669.1406250000002 373.7723214285714 669.029017857143 374.2745535714285C668.9174107142859 374.7767857142857 668.7500000000002 375.2790178571427 668.638392857143 375.7254464285712C668.5267857142859 376.0602678571427 668.4709821428573 376.3950892857141 668.4151785714287 376.674107142857C668.247767857143 377.3437499999999 668.0803571428573 378.0133928571428 667.9129464285714 378.627232142857C667.857142857143 378.7946428571427 667.857142857143 378.9620535714284 667.8013392857143 379.1294642857142C667.6339285714287 379.9665178571428 667.4107142857143 380.8035714285714 667.2433035714287 381.6406249999999C667.2433035714287 381.6406249999999 667.2433035714287 381.6406249999999 667.2433035714287 381.6406249999999C666.6852678571429 384.4308035714285 666.1830357142857 387.2767857142856 665.8482142857143 390.0669642857141C635.15625 384.3191964285714 600.1116071428572 373.7165178571428 567.5223214285714 354.1294642857141C525.2790178571429 328.7946428571427 494.0290178571428 292.7455357142856 472.3772321428572 260.7700892857141C481.7522321428571 254.2410714285712 489.8995535714286 245.6473214285714 496.1495535714286 235.267857142857C502.3995535714286 224.8325892857141 506.1383928571429 213.5602678571427 507.5334821428572 202.2321428571427C543.4709821428571 206.0825892857141 591.2388392857143 216.3504464285712 634.6540178571429 242.4107142857141C666.7968749999999 261.7187499999998 692.578125 287.1651785714284 712.6116071428571 312.332589285714C708.8169642857143 315.0111607142855 705.1339285714286 317.9687499999997 701.6741071428572 321.1495535714284C701.5066964285714 321.2611607142857 701.4508928571429 321.3169642857142 701.4508928571429 321.3727678571429zM1450.669642857143 1000H105.1339285714286C47.0424107142857 1000 0 954.296875 0 897.9352678571429V102.064732142857C0 45.7031249999999 47.0982142857143 -1e-13 105.1339285714286 -1e-13H1450.7812500000002C1508.872767857143 -1e-13 1555.9151785714287 45.7031249999999 1555.9151785714287 102.064732142857V897.9352678571429C1555.8035714285716 954.296875 1508.8169642857144 1000 1450.669642857143 1000zM1203.4598214285713 149.7767857142858C1179.6875 110.1004464285716 1128.236607142857 97.1540178571429 1088.560267857143 120.9263392857143C1065.625 134.6540178571428 1051.6183035714284 157.6450892857142 1048.4374999999998 182.3102678571429C1009.9888392857142 186.1607142857143 958.1473214285714 196.9308035714286 910.7142857142856 225.2790178571429C887.611607142857 239.1183035714286 865.7366071428571 256.9754464285715 845.7031249999999 278.3482142857142C838.671875 285.8816964285714 831.8638392857142 293.8058035714286 825.2790178571428 302.1763392857142C821.986607142857 300.6696428571428 818.6383928571428 299.2745535714286 815.1785714285714 298.0468749999999C815.1785714285714 298.0468749999999 815.1785714285714 298.0468749999999 815.1785714285714 298.0468749999999C814.3973214285713 297.7678571428571 813.5602678571428 297.4888392857142 812.7790178571429 297.2098214285714C812.6116071428572 297.1540178571429 812.4441964285716 297.0982142857142 812.2767857142857 297.0424107142857C811.6071428571428 296.8191964285714 810.9933035714286 296.6517857142857 810.3236607142857 296.4285714285715C809.9888392857143 296.3169642857142 809.6540178571428 296.2611607142858 809.375 296.1495535714286C808.8727678571428 295.9821428571429 808.4263392857142 295.8705357142857 807.9241071428571 295.7589285714286C807.421875 295.6473214285714 806.9196428571429 295.4799107142857 806.4732142857143 295.3683035714286C806.138392857143 295.2566964285714 805.8035714285714 295.2008928571429 805.5245535714286 295.1450892857142C804.8549107142857 294.9776785714286 804.1852678571428 294.8102678571428 803.5714285714286 294.642857142857C803.4040178571429 294.5870535714285 803.2366071428572 294.5870535714285 803.0691964285714 294.5312499999998C802.2321428571428 294.3638392857141 801.3950892857142 294.1406249999998 800.5580357142857 293.9732142857141C800.5580357142857 293.9732142857141 800.5580357142857 293.9732142857141 800.5580357142857 293.9732142857141C794.1964285714284 292.689732142857 787.7790178571428 291.908482142857 781.3616071428571 291.7410714285712C781.2499999999999 291.7410714285712 781.1383928571428 291.7410714285712 781.0825892857142 291.7410714285712C780.4129464285713 291.7410714285712 779.7433035714284 291.6852678571428 779.0736607142857 291.6852678571428C778.6830357142857 291.6852678571428 778.2924107142857 291.6852678571428 777.9017857142858 291.6852678571428C777.5111607142857 291.6852678571428 777.1205357142857 291.6852678571428 776.7299107142858 291.6852678571428C776.0602678571429 291.6852678571428 775.390625 291.6852678571428 774.7209821428572 291.7410714285712C774.6093750000001 291.7410714285712 774.4977678571429 291.7410714285712 774.4419642857144 291.7410714285712C767.1316964285716 291.9642857142856 759.8214285714287 292.8571428571427 752.5669642857143 294.5312499999998C752.3995535714287 294.5870535714284 752.2879464285716 294.5870535714284 752.1205357142859 294.642857142857C751.450892857143 294.8102678571427 750.7812500000001 294.9776785714284 750.1116071428573 295.1450892857142C749.7767857142859 295.2008928571427 749.4977678571431 295.3124999999999 749.1629464285716 295.3683035714286C748.6607142857143 295.4799107142857 748.2142857142859 295.6473214285714 747.7120535714288 295.7589285714286C747.2098214285716 295.8705357142857 746.763392857143 296.0379464285715 746.261160714286 296.1495535714286C745.9263392857146 296.2611607142858 745.6473214285717 296.3169642857142 745.3125000000002 296.4285714285715C744.6428571428573 296.5959821428571 744.0290178571431 296.8191964285715 743.3593750000002 297.0424107142857C743.1919642857146 297.0982142857142 743.0803571428573 297.1540178571429 742.9129464285717 297.2098214285714C738.6160714285717 298.6049107142857 734.4308035714288 300.2790178571428 730.3571428571431 302.1205357142857C709.4308035714288 275.4464285714286 681.1383928571431 246.7075892857142 645.089285714286 225.0558035714286C621.9866071428573 211.2165178571429 595.9263392857146 200.3348214285715 567.6339285714289 192.7455357142857C548.4933035714289 187.611607142857 528.292410714286 183.9843749999999 507.4776785714289 181.8638392857142C504.4084821428575 157.2544642857142 490.401785714286 134.2075892857141 467.4665178571432 120.4241071428571C427.7901785714289 96.5959821428571 376.339285714286 109.4308035714284 352.511160714286 149.1071428571429C328.683035714286 188.7834821428571 341.5178571428574 240.234375 381.1941964285718 264.0625C404.1294642857146 277.8459821428572 431.026785714286 279.3526785714287 454.2410714285717 270.5357142857142C475.7254464285717 302.6227678571428 509.7098214285717 343.3035714285715 557.0312500000003 371.7075892857142C580.1339285714289 385.5468749999999 606.1941964285717 396.4285714285714 634.4866071428575 404.0178571428571C644.4196428571431 406.6964285714286 654.6875000000003 408.9285714285714 665.1227678571432 410.8258928571429C665.4575892857146 417.1316964285715 666.3504464285718 423.4933035714286 667.8013392857146 429.7433035714286C667.8571428571431 429.9107142857142 667.8571428571431 430.0223214285715 667.9129464285718 430.1897321428571C668.0803571428575 430.859375 668.2477678571431 431.5290178571428 668.4151785714289 432.1986607142857C668.4709821428575 432.5334821428571 668.5825892857147 432.8125 668.6383928571432 433.1473214285715C668.7500000000005 433.6495535714286 668.9174107142861 434.0959821428572 669.0290178571433 434.5982142857144C669.1406250000005 435.1004464285715 669.3080357142861 435.546875 669.4196428571433 436.0491071428572C669.5312500000005 436.3839285714287 669.587053571429 436.6629464285715 669.6986607142862 436.9977678571429C669.8660714285719 437.6674107142858 670.0892857142862 438.28125 670.3125000000005 438.9508928571429C670.3683035714289 439.1183035714286 670.4241071428576 439.2299107142858 670.4799107142861 439.3973214285715C672.4888392857147 445.5357142857144 674.9441964285718 451.3950892857143 677.9017857142861 457.03125C677.9575892857146 457.1428571428572 678.0133928571432 457.2544642857143 678.0691964285718 457.3660714285714C678.4040178571431 457.9799107142858 678.683035714286 458.5379464285716 679.0178571428576 459.1517857142858C679.1852678571432 459.4866071428572 679.4084821428576 459.8214285714287 679.5758928571432 460.1562500000001C679.7433035714289 460.4910714285716 679.9665178571432 460.8258928571429 680.1897321428575 461.1607142857143C680.5245535714289 461.71875 680.8593750000003 462.3325892857143 681.1941964285717 462.890625C681.2500000000002 463.0022321428572 681.3058035714289 463.1138392857143 681.3616071428575 463.1696428571429C685.1562500000002 469.3638392857143 689.5089285714288 475.2790178571429 694.5312500000002 480.7477678571429C694.5870535714288 480.8035714285714 694.6428571428573 480.8593750000001 694.6428571428573 480.9151785714286C695.1450892857146 481.4732142857143 695.6473214285717 482.03125 696.2053571428573 482.5334821428571C696.4285714285717 482.7566964285714 696.5959821428573 482.9799107142857 696.8191964285716 483.1473214285714C697.2098214285717 483.5379464285714 697.544642857143 483.8727678571429 697.935267857143 484.2633928571429C698.325892857143 484.6540178571429 698.6607142857144 484.9888392857143 699.0513392857144 485.3794642857143C699.2745535714288 485.6026785714286 699.4419642857144 485.7700892857143 699.6651785714287 485.9933035714286C700.2232142857144 486.4955357142858 700.7812500000001 487.0535714285714 701.2834821428573 487.5558035714286C701.3392857142858 487.6116071428571 701.3950892857144 487.6674107142858 701.450892857143 487.7232142857142C705.5245535714287 491.4620535714286 709.8772321428572 494.8660714285714 714.3973214285716 497.9352678571428C702.3437500000002 529.6316964285713 692.2433035714287 568.6383928571429 692.2433035714287 610.7142857142857C692.2433035714287 637.6116071428571 696.3169642857144 665.5691964285713 704.4084821428572 693.75C709.8772321428572 712.7790178571429 717.1316964285716 731.9754464285714 726.060267857143 750.9486607142858C706.529017857143 766.2946428571429 693.9732142857143 790.1227678571429 693.9732142857143 816.8526785714287C693.9732142857143 863.1138392857143 731.4732142857144 900.6138392857143 777.734375 900.6138392857143C823.9955357142858 900.6138392857143 861.4955357142857 863.1138392857143 861.4955357142857 816.8526785714287C861.4955357142857 790.1227678571429 848.9397321428571 766.2946428571429 829.4084821428571 750.9486607142858C845.8705357142857 716.015625 863.2254464285713 665.9598214285714 863.2254464285713 610.7142857142858C863.2254464285713 583.8169642857142 859.1517857142857 555.859375 851.0602678571428 527.6785714285716C848.2142857142858 517.8013392857143 844.921875 507.8125000000001 841.0714285714284 497.9352678571429C845.5357142857142 494.9218750000001 849.7767857142856 491.5178571428572 853.8504464285714 487.8348214285716C853.9062499999999 487.7790178571429 853.9620535714286 487.7232142857143 854.0178571428571 487.7232142857143C854.5758928571428 487.2209821428572 855.1339285714286 486.7187500000001 855.6361607142857 486.1607142857143C855.859375 485.9375000000001 856.0825892857142 485.7700892857143 856.25 485.5468750000001C856.640625 485.15625 856.9754464285713 484.8214285714287 857.3660714285713 484.4308035714287C857.7566964285714 484.0401785714286 858.0915178571428 483.7053571428572 858.4821428571428 483.3147321428572C858.7053571428571 483.0915178571429 858.8727678571428 482.9241071428572 859.095982142857 482.7008928571429C859.5982142857142 482.1428571428572 860.15625 481.5848214285714 860.658482142857 481.0825892857143C860.7142857142856 481.0267857142858 860.7700892857142 480.9709821428572 860.8258928571427 480.9151785714287C865.8482142857141 475.4464285714287 870.3124999999999 469.5312500000001 874.1071428571427 463.2254464285714C874.1629464285712 463.1138392857143 874.2187499999999 463.0580357142858 874.2745535714283 462.9464285714287C874.6093749999998 462.3883928571429 874.9441964285712 461.8303571428572 875.2790178571427 461.2723214285714C875.4464285714283 460.9375 875.6696428571427 460.6026785714286 875.8370535714283 460.2678571428572C876.004464285714 459.9330357142858 876.2276785714283 459.5982142857143 876.3950892857141 459.263392857143C876.7299107142854 458.7053571428572 877.0089285714283 458.091517857143 877.3437499999998 457.5334821428572C877.3995535714283 457.4218750000001 877.455357142857 457.310267857143 877.5111607142854 457.2544642857143C880.0781249999998 452.3995535714287 882.3102678571425 447.3214285714287 884.1517857142856 442.0758928571429C884.1517857142856 442.0758928571429 884.1517857142856 442.0758928571429 884.1517857142856 442.0758928571429C884.4308035714284 441.2946428571429 884.7098214285712 440.4575892857144 884.9888392857141 439.6763392857144C885.0446428571427 439.5089285714287 885.1004464285713 439.3415178571429 885.1562499999998 439.1741071428572C885.3794642857141 438.5044642857144 885.5468749999998 437.8906250000001 885.770089285714 437.2209821428572C885.8816964285712 436.8861607142858 885.9374999999998 436.5513392857144 886.0491071428569 436.2723214285716C886.2165178571425 435.7700892857144 886.3281249999998 435.3236607142858 886.4397321428569 434.8214285714287C886.5513392857141 434.3191964285716 886.7187499999998 433.8169642857144 886.8303571428569 433.3705357142858C886.9419642857141 433.0357142857144 886.9977678571427 432.7008928571429 887.0535714285713 432.421875C887.220982142857 431.7522321428571 887.3883928571427 431.0825892857144 887.5558035714284 430.46875C887.611607142857 430.3013392857144 887.611607142857 430.1339285714286 887.6674107142857 429.9665178571429C887.8348214285713 429.1294642857144 888.0580357142857 428.2924107142857 888.2254464285713 427.4553571428571C888.2254464285713 427.4553571428571 888.2254464285713 427.4553571428571 888.2254464285713 427.4553571428571C889.3415178571428 422.0424107142857 890.0111607142857 416.6294642857142 890.345982142857 411.1607142857142C923.7165178571428 405.2455357142858 962.4441964285714 393.8616071428571 998.4933035714284 372.2656249999999C1021.5959821428572 358.4263392857142 1043.4709821428569 340.5691964285714 1063.5044642857142 319.1964285714286C1077.0647321428569 304.7433035714286 1089.7879464285713 288.671875 1101.4508928571427 271.2611607142857C1124.6651785714284 280.1339285714286 1151.506696428571 278.6830357142857 1174.497767857143 264.9553571428571C1214.2857142857142 240.9040178571429 1227.1763392857142 189.453125 1203.4598214285713 149.7767857142858z" />
    <glyph glyph-name="thumb_graph_i"
      unicode="&#xF18D;"
      horiz-adv-x="1555.859375" d="M1174.5535714285716 264.6763392857142C1151.6183035714287 278.4040178571428 1124.7209821428573 279.9107142857142 1101.5066964285716 270.9821428571428C1089.84375 288.3928571428571 1077.1205357142858 304.4642857142857 1063.560267857143 318.9174107142857C1043.5267857142858 340.3459821428572 1021.6517857142858 358.203125 998.5491071428572 371.9866071428571C962.4441964285714 393.5825892857142 923.7723214285716 404.9665178571428 890.4017857142858 410.8816964285714C890.122767857143 416.2946428571429 889.3973214285716 421.7633928571428 888.2812500000001 427.1763392857142C888.2812500000001 427.1763392857142 888.2812500000001 427.1763392857142 888.2812500000001 427.1763392857142C888.1138392857144 428.0133928571429 887.9464285714288 428.8504464285715 887.7232142857144 429.6875C887.6674107142859 429.8549107142857 887.6674107142859 430.0223214285715 887.6116071428572 430.1897321428571C887.4441964285716 430.859375 887.2767857142859 431.5290178571428 887.109375 432.1428571428571C887.0535714285716 432.4776785714286 886.9419642857143 432.8125 886.8861607142857 433.0915178571429C886.7745535714286 433.59375 886.6071428571429 434.0959821428571 886.4955357142857 434.5424107142858C886.3839285714286 435.0446428571429 886.2165178571428 435.4910714285715 886.1049107142857 435.9933035714286C885.9933035714284 436.328125 885.9375 436.6629464285715 885.8258928571428 436.9419642857144C885.6026785714284 437.6116071428572 885.4352678571428 438.2254464285715 885.2120535714286 438.8950892857144C885.15625 439.0625 885.1004464285714 439.2299107142858 885.0446428571429 439.3973214285715C884.765625 440.1785714285715 884.4866071428572 441.015625 884.2075892857143 441.796875C884.2075892857143 441.796875 884.2075892857143 441.796875 884.2075892857143 441.796875C882.3660714285714 447.0424107142857 880.1339285714287 452.1205357142858 877.5669642857142 456.9754464285714C877.5111607142858 457.0870535714287 877.4553571428571 457.1986607142858 877.3995535714286 457.2544642857143C877.0647321428572 457.8125 876.7857142857143 458.4263392857143 876.4508928571428 458.984375C876.2834821428571 459.3191964285714 876.0602678571428 459.6540178571429 875.8928571428571 459.9888392857143C875.7254464285714 460.3236607142858 875.5022321428571 460.6584821428572 875.3348214285714 460.9933035714286C875 461.5513392857143 874.6651785714286 462.109375 874.3303571428571 462.6674107142857C874.2745535714287 462.7790178571429 874.21875 462.8348214285714 874.1629464285714 462.9464285714286C870.3683035714287 469.1964285714286 865.9040178571429 475.1116071428571 860.8816964285714 480.6361607142858C860.825892857143 480.6919642857143 860.7700892857143 480.7477678571429 860.7142857142858 480.8035714285714C860.2120535714286 481.3616071428571 859.7098214285716 481.9196428571429 859.1517857142858 482.421875C858.9285714285714 482.6450892857142 858.7611607142858 482.8683035714286 858.5379464285716 483.0357142857143C858.1473214285716 483.4263392857143 857.8125000000002 483.7611607142857 857.4218750000001 484.1517857142857C857.0312500000001 484.5424107142858 856.6964285714288 484.8772321428571 856.3058035714287 485.2678571428571C856.0825892857143 485.4910714285714 855.8593750000002 485.6584821428571 855.6919642857144 485.8816964285714C855.1339285714288 486.3839285714286 854.575892857143 486.9419642857143 854.0736607142859 487.4441964285714C854.0178571428573 487.5 853.9620535714287 487.5558035714286 853.9062500000002 487.5558035714286C849.888392857143 491.2388392857143 845.591517857143 494.6428571428572 841.1272321428572 497.65625C844.9218750000001 507.5892857142857 848.2700892857143 517.5223214285714 851.1160714285716 527.3995535714286C859.2075892857144 555.5803571428571 863.2812500000001 583.5379464285714 863.2812500000001 610.4352678571429C863.2812500000001 665.6808035714287 845.8705357142859 715.6808035714287 829.4642857142859 750.6696428571429C848.9955357142859 766.015625 861.5513392857144 789.84375 861.5513392857144 816.5736607142858C861.5513392857144 862.8348214285714 824.0513392857144 900.3348214285714 777.7901785714288 900.3348214285714C731.529017857143 900.3348214285714 694.0290178571431 862.8348214285714 694.0290178571431 816.5736607142858C694.0290178571431 789.84375 706.5848214285717 766.015625 726.1160714285717 750.6696428571429C717.1875000000003 731.6964285714287 709.8772321428575 712.5 704.464285714286 693.4709821428571C696.3727678571432 665.2901785714287 692.2991071428575 637.3325892857142 692.2991071428575 610.4352678571429C692.2991071428575 568.359375 702.3995535714289 529.296875 714.4531250000003 497.65625C709.9330357142861 494.5870535714286 705.5803571428575 491.1830357142857 701.5066964285718 487.4441964285714C701.4508928571432 487.3883928571429 701.3950892857146 487.3325892857143 701.3392857142861 487.2767857142858C700.7812500000003 486.7745535714287 700.2232142857147 486.2723214285716 699.7209821428575 485.7142857142858C699.4977678571431 485.4910714285716 699.2745535714289 485.3236607142858 699.1071428571432 485.1004464285716C698.7165178571432 484.7098214285714 698.3816964285718 484.3750000000001 697.9910714285718 483.9843750000001C697.6004464285718 483.5937500000001 697.2656250000005 483.2589285714287 696.8750000000003 482.8683035714287C696.651785714286 482.6450892857143 696.4843750000003 482.4218750000001 696.2611607142861 482.2544642857143C695.7589285714289 481.6964285714287 695.2008928571432 481.1383928571429 694.6986607142861 480.6361607142858C694.6428571428577 480.5803571428572 694.587053571429 480.5245535714287 694.587053571429 480.4687500000001C689.5647321428576 475.0000000000001 685.212053571429 469.1406250000001 681.4174107142862 462.8906250000001C681.3616071428577 462.7790178571429 681.305803571429 462.6674107142859 681.2500000000005 462.6116071428572C680.9151785714291 462.0535714285716 680.5803571428576 461.4397321428572 680.2455357142862 460.8816964285716C680.0223214285719 460.5468750000001 679.8549107142862 460.2120535714287 679.631696428572 459.8772321428572C679.4642857142863 459.5424107142858 679.241071428572 459.2075892857143 679.0736607142863 458.872767857143C678.7388392857149 458.2589285714287 678.4040178571435 457.700892857143 678.1250000000006 457.0870535714287C678.069196428572 456.9754464285716 678.0133928571433 456.8638392857144 677.9575892857149 456.7522321428572C675.0000000000006 451.1160714285716 672.4888392857149 445.2566964285716 670.5357142857149 439.1183035714287C670.4799107142863 438.950892857143 670.4241071428577 438.8392857142858 670.3683035714292 438.6718750000001C670.1450892857149 438.0022321428572 669.9776785714291 437.388392857143 669.754464285715 436.7187500000001C669.6428571428578 436.3839285714287 669.5870535714292 436.1049107142858 669.4754464285721 435.7700892857144C669.3638392857149 435.2678571428572 669.1964285714292 434.8214285714287 669.0848214285721 434.3191964285716C668.9732142857149 433.8169642857144 668.8058035714292 433.3705357142858 668.694196428572 432.8683035714287C668.6383928571436 432.5334821428572 668.5267857142863 432.2544642857144 668.4709821428577 431.9196428571429C668.303571428572 431.25 668.1361607142863 430.5803571428572 667.9687500000006 429.9107142857144C667.912946428572 429.7433035714287 667.912946428572 429.6316964285715 667.8571428571433 429.4642857142858C666.4062500000006 423.2142857142858 665.5133928571433 416.8526785714287 665.1785714285719 410.546875C654.6875000000006 408.7053571428572 644.475446428572 406.4174107142857 634.5424107142862 403.7388392857142C606.1941964285719 396.1495535714286 580.133928571429 385.2678571428572 557.0870535714291 371.4285714285714C509.709821428572 342.9687499999999 475.7812500000005 302.34375 454.2968750000005 270.2566964285714C431.0825892857148 279.0736607142857 404.2410714285719 277.5669642857142 381.2500000000005 263.7834821428571C341.5736607142862 239.9553571428571 328.7388392857148 188.5044642857142 352.5669642857148 148.828125C376.3950892857148 109.1517857142858 427.8459821428576 96.3169642857143 467.5223214285719 120.1450892857142C490.4575892857148 133.9285714285715 504.4084821428577 156.9196428571429 507.5334821428577 181.5848214285713C528.4040178571433 183.705357142857 548.6049107142862 187.3325892857141 567.6897321428577 192.4665178571428C596.0379464285719 200.0558035714285 622.0982142857148 210.9374999999999 645.1450892857148 224.7767857142857C681.1941964285719 246.4285714285714 709.4866071428576 275.1674107142857 730.4129464285719 301.8415178571428C734.4866071428576 299.9441964285714 738.6718750000005 298.3258928571428 742.9687500000005 296.9308035714286C743.1361607142861 296.875 743.2477678571433 296.8191964285714 743.415178571429 296.7633928571428C744.0848214285719 296.5401785714285 744.6986607142861 296.3727678571428 745.368303571429 296.1495535714286C745.7031250000003 296.0379464285715 745.9821428571432 295.9821428571429 746.3169642857148 295.8705357142857C746.8191964285719 295.7589285714286 747.2656250000005 295.5915178571429 747.7678571428576 295.4799107142857C748.2700892857148 295.3683035714286 748.7165178571432 295.2008928571429 749.2187500000003 295.0892857142857C749.5535714285717 295.0334821428571 749.8325892857146 294.921875 750.1674107142861 294.8660714285714C750.837053571429 294.6986607142857 751.5066964285719 294.53125 752.1763392857146 294.3638392857141C752.3437500000003 294.3080357142857 752.4553571428575 294.3080357142857 752.6227678571431 294.252232142857C759.8772321428575 292.5781249999998 767.1875000000002 291.6852678571428 774.4977678571432 291.4620535714285C774.6093750000003 291.4620535714285 774.7209821428576 291.4620535714285 774.776785714286 291.4620535714285C775.4464285714289 291.4620535714285 776.1160714285718 291.4062499999999 776.7857142857146 291.4062499999999C777.1763392857146 291.4062499999999 777.5669642857146 291.4062499999999 777.9575892857146 291.4062499999999C778.3482142857146 291.4062499999999 778.7388392857146 291.4062499999999 779.1294642857144 291.4062499999999C779.7991071428573 291.4062499999999 780.4687500000002 291.4062499999999 781.138392857143 291.4620535714285C781.2500000000001 291.4620535714285 781.3616071428573 291.4620535714285 781.4174107142859 291.4620535714285C787.8348214285716 291.6294642857141 794.2522321428572 292.4107142857141 800.6138392857144 293.6941964285712C800.6138392857144 293.6941964285712 800.6138392857144 293.6941964285712 800.6138392857144 293.6941964285712C801.450892857143 293.861607142857 802.2879464285716 294.0290178571427 803.1250000000002 294.252232142857C803.2924107142859 294.3080357142855 803.4598214285716 294.3080357142855 803.6272321428573 294.3638392857141C804.2968750000002 294.5312499999998 804.9665178571431 294.6986607142856 805.5803571428573 294.8660714285714C805.9151785714288 294.9218749999999 806.2500000000002 295.033482142857 806.5290178571431 295.0892857142857C807.0312500000003 295.2008928571429 807.5334821428573 295.3683035714286 807.9799107142859 295.4799107142857C808.4821428571431 295.5915178571429 808.9285714285717 295.7589285714286 809.4308035714287 295.8705357142857C809.7656250000001 295.9821428571429 810.1004464285716 296.0379464285715 810.3794642857144 296.1495535714286C811.0491071428573 296.3727678571429 811.6629464285716 296.5401785714286 812.3325892857144 296.7633928571428C812.5000000000001 296.8191964285714 812.6674107142859 296.875 812.8348214285717 296.9308035714286C813.6160714285717 297.2098214285714 814.4531250000002 297.4888392857142 815.2343750000002 297.7678571428571C815.2343750000002 297.7678571428571 815.2343750000002 297.7678571428571 815.2343750000002 297.7678571428571C818.6941964285717 298.9955357142857 822.0424107142859 300.3348214285714 825.3348214285716 301.8973214285715C831.919642857143 293.5267857142857 838.7276785714287 285.6026785714286 845.7589285714287 278.0691964285714C865.7924107142859 256.6406249999999 887.6674107142858 238.783482142857 910.7700892857144 225C958.1473214285716 196.6517857142858 1010.044642857143 185.8816964285715 1048.4933035714287 182.03125C1051.6183035714287 157.421875 1065.625 134.4308035714286 1088.6160714285716 120.6473214285715C1128.2924107142858 96.875 1179.743303571429 109.8214285714286 1203.5156250000002 149.4977678571429C1227.1763392857142 189.453125 1214.2857142857142 240.9040178571429 1174.5535714285716 264.6763392857142zM743.5825892857143 740.234375C754.0178571428572 735.546875 765.5691964285714 732.9799107142858 777.734375 732.9799107142858C789.8995535714286 732.9799107142858 801.4508928571429 735.546875 811.8303571428571 740.234375C828.125 705.2455357142857 842.9129464285714 659.8772321428571 842.9129464285714 610.6026785714287C842.9129464285714 572.5446428571429 833.984375 536.9977678571429 823.1026785714286 507.7566964285714C819.6986607142858 509.2633928571429 816.1830357142857 510.6026785714286 812.5558035714286 511.7745535714286C812.3883928571429 511.8303571428572 812.2767857142857 511.8861607142858 812.109375 511.9419642857143C811.4397321428571 512.1651785714287 810.8258928571429 512.3325892857143 810.15625 512.5558035714287C809.8214285714287 512.6674107142858 809.5424107142858 512.7232142857143 809.2075892857142 512.8348214285716C808.7053571428571 512.9464285714287 808.2589285714286 513.1138392857143 807.7566964285714 513.2254464285716C807.2544642857142 513.3370535714287 806.8080357142857 513.5044642857143 806.3058035714287 513.6160714285716C805.9709821428572 513.6718750000001 805.6919642857144 513.7834821428572 805.3571428571429 513.8392857142858C804.6875 514.0066964285716 804.0178571428571 514.1741071428572 803.3482142857143 514.3415178571429C803.1808035714287 514.3973214285716 803.0691964285716 514.3973214285716 802.9017857142859 514.4531250000001C795.6473214285716 516.1272321428572 788.3370535714288 517.0200892857144 781.0267857142858 517.2433035714287C780.9151785714287 517.2433035714287 780.8035714285714 517.2433035714287 780.747767857143 517.2433035714287C780.0781250000001 517.2433035714287 779.4084821428572 517.2991071428573 778.7388392857144 517.2991071428573C778.3482142857144 517.2991071428573 777.9575892857143 517.2991071428573 777.5669642857144 517.2991071428573C777.1763392857144 517.2991071428573 776.7857142857144 517.2991071428573 776.3950892857146 517.2991071428573C775.7254464285717 517.2991071428573 775.0558035714288 517.2991071428573 774.386160714286 517.2433035714287C774.2745535714289 517.2433035714287 774.1629464285717 517.2433035714287 774.1071428571431 517.2433035714287C767.6897321428575 517.075892857143 761.2723214285718 516.2946428571429 754.9107142857146 515.0111607142858C754.9107142857146 515.0111607142858 754.9107142857146 515.0111607142858 754.9107142857146 515.0111607142858C754.073660714286 514.8437500000001 753.2366071428573 514.6763392857144 752.3995535714288 514.4531250000001C752.2321428571431 514.3973214285716 752.0647321428575 514.3973214285716 751.8973214285717 514.3415178571429C751.2276785714288 514.1741071428572 750.5580357142859 514.0066964285716 749.9441964285717 513.8392857142858C749.6093750000002 513.7834821428572 749.2745535714288 513.6718750000001 748.9955357142859 513.6160714285716C748.4933035714287 513.5044642857143 747.9910714285717 513.3370535714287 747.5446428571431 513.2254464285716C747.0424107142859 513.1138392857143 746.5959821428573 512.9464285714287 746.0937500000002 512.8348214285716C745.7589285714289 512.7232142857143 745.4241071428573 512.6674107142858 745.1450892857146 512.5558035714287C744.4754464285717 512.3325892857143 743.8616071428575 512.1651785714287 743.1919642857146 511.9419642857143C743.0245535714289 511.8861607142857 742.8571428571431 511.8303571428572 742.6897321428573 511.7745535714286C741.9084821428573 511.4955357142858 741.0714285714287 511.2165178571429 740.2901785714288 510.9375000000001C740.2901785714288 510.9375000000001 740.2901785714288 510.9375000000001 740.2901785714288 510.9375000000001C737.6116071428573 509.9888392857143 734.933035714286 508.9285714285715 732.3660714285717 507.8125000000001C721.0937500000002 537.9464285714287 712.5558035714288 573.1584821428572 712.5558035714288 610.6584821428572C712.5558035714287 661.2165178571429 728.4040178571429 707.4776785714286 743.5825892857143 740.234375zM634.5982142857143 242.4665178571429C591.1830357142858 216.40625 543.4151785714287 206.1383928571429 507.4776785714286 202.2879464285715C506.0825892857143 213.6160714285713 502.3995535714286 224.8883928571429 496.0937500000001 235.3236607142857C489.8437500000001 245.7589285714286 481.6964285714287 254.296875 472.3214285714286 260.8258928571429C493.9732142857143 292.8013392857144 525.2232142857143 328.7946428571429 567.4665178571429 354.1852678571429C600.0558035714287 373.7723214285714 635.15625 384.4308035714286 665.7924107142858 390.1227678571429C666.1272321428572 387.3325892857144 666.6294642857143 384.4866071428572 667.1875000000001 381.6964285714287C667.1875000000001 381.6964285714287 667.1875000000001 381.6964285714287 667.1875000000001 381.6964285714287C667.3549107142858 380.8593750000001 667.5223214285714 380.0223214285716 667.7455357142858 379.185267857143C667.8013392857143 379.0178571428572 667.8013392857143 378.8504464285716 667.857142857143 378.6830357142858C668.0245535714287 378.0133928571429 668.1919642857143 377.34375 668.3593750000002 376.7299107142858C668.4151785714287 376.3950892857145 668.5267857142859 376.0602678571429 668.5825892857146 375.78125C668.6941964285717 375.2790178571429 668.8616071428573 374.7767857142858 668.9732142857146 374.3303571428572C669.0848214285717 373.828125 669.2522321428573 373.3816964285715 669.3638392857146 372.8794642857145C669.4754464285717 372.544642857143 669.5312500000002 372.2098214285716 669.6428571428575 371.9308035714287C669.8660714285718 371.2611607142858 670.0334821428575 370.6473214285716 670.2566964285717 369.9776785714287C670.3125000000002 369.810267857143 670.3683035714288 369.6428571428574 670.4241071428573 369.4754464285715C670.7031250000002 368.6941964285715 670.982142857143 367.8571428571429 671.2611607142859 367.0758928571429C671.2611607142859 367.0758928571429 671.2611607142859 367.0758928571429 671.2611607142859 367.0758928571429C673.1026785714288 361.8303571428572 675.3348214285716 356.7522321428572 677.901785714286 351.8973214285715C677.9575892857144 351.7857142857144 678.0133928571431 351.6741071428571 678.0691964285717 351.6183035714287C678.404017857143 351.0602678571429 678.6830357142859 350.4464285714287 679.0178571428573 349.888392857143C679.1852678571431 349.5535714285716 679.4084821428575 349.2187500000001 679.5758928571431 348.8839285714287C679.7433035714288 348.5491071428574 679.9665178571431 348.2142857142858 680.1339285714288 347.8794642857145C680.4687500000002 347.3214285714288 680.8035714285717 346.763392857143 681.1383928571431 346.2053571428574C681.1941964285716 346.0937500000001 681.2500000000002 346.0379464285716 681.3058035714288 345.9263392857145C685.1004464285716 339.6763392857145 689.5647321428573 333.7611607142859 694.5870535714288 328.2366071428572C694.6428571428572 328.1808035714288 694.6986607142859 328.1250000000001 694.7544642857144 328.0691964285716C695.2566964285717 327.5111607142859 695.7589285714287 326.9531250000001 696.3169642857144 326.450892857143C696.5401785714288 326.2276785714287 696.7075892857144 326.0044642857145 696.9308035714287 325.8370535714288C697.3214285714287 325.4464285714287 697.65625 325.1116071428574 698.0468750000001 324.7209821428574C698.4375000000001 324.3303571428572 698.7723214285714 323.9955357142859 699.1629464285714 323.6049107142859C699.3861607142858 323.3816964285716 699.609375 323.2142857142859 699.7767857142858 322.9910714285717C700.3348214285714 322.4888392857145 700.8928571428571 321.9308035714288 701.3950892857143 321.4285714285717C701.4508928571429 321.3727678571431 701.5066964285716 321.3169642857145 701.5625 321.3169642857145C705.0223214285714 318.1361607142859 708.7053571428571 315.1785714285717 712.5 312.5000000000003C692.5223214285714 287.2209821428571 666.7410714285714 261.7745535714286 634.5982142857143 242.4665178571429zM843.1361607142858 312.5C846.9308035714287 315.1785714285714 850.6138392857144 318.1361607142857 854.0736607142858 321.3727678571429C854.1294642857142 321.4285714285715 854.1852678571429 321.4843750000001 854.2410714285714 321.5401785714286C854.7991071428571 322.0424107142858 855.3571428571429 322.5446428571429 855.859375 323.1026785714286C856.0825892857143 323.3258928571429 856.3058035714286 323.4933035714286 856.4732142857143 323.7165178571428C856.8638392857143 324.1071428571428 857.1986607142857 324.4419642857142 857.5892857142857 324.8325892857142C857.9799107142858 325.2232142857142 858.3147321428571 325.5580357142857 858.7053571428571 325.9486607142857C858.9285714285714 326.171875 859.0959821428572 326.3950892857142 859.3191964285713 326.5624999999999C859.8214285714286 327.1205357142856 860.3794642857143 327.6785714285714 860.8816964285713 328.1808035714286C860.9374999999999 328.236607142857 860.9933035714286 328.2924107142857 860.9933035714286 328.3482142857142C866.015625 333.8169642857142 870.3683035714284 339.6763392857142 874.1629464285713 345.9263392857142C874.2187499999999 346.0379464285714 874.2745535714284 346.1495535714286 874.330357142857 346.205357142857C874.6651785714284 346.7633928571428 874.9999999999999 347.377232142857 875.3348214285713 347.9352678571427C875.5580357142857 348.2700892857141 875.7254464285713 348.6049107142856 875.9486607142856 348.939732142857C876.1160714285712 349.2745535714284 876.3392857142856 349.6093749999999 876.5066964285712 349.9441964285712C876.8415178571425 350.5580357142855 877.1763392857141 351.1160714285712 877.455357142857 351.7299107142856C877.5111607142854 351.8415178571427 877.5669642857141 351.9531249999999 877.6227678571427 352.0647321428569C880.5803571428569 357.7008928571426 883.0915178571427 363.5602678571426 885.0446428571427 369.6986607142855C885.1004464285711 369.8660714285711 885.1562499999998 369.9776785714284 885.2120535714283 370.145089285714C885.4352678571427 370.8147321428569 885.6026785714283 371.4285714285711 885.8258928571425 372.098214285714C885.9374999999998 372.4330357142853 885.9933035714282 372.7120535714282 886.1049107142854 373.0468749999998C886.2165178571425 373.5491071428569 886.3839285714282 373.9955357142855 886.4955357142854 374.4977678571426C886.6071428571425 374.9999999999998 886.7745535714283 375.4464285714284 886.8861607142854 375.9486607142853C886.941964285714 376.2834821428568 887.0535714285711 376.5624999999996 887.1093749999998 376.8973214285711C887.2767857142854 377.566964285714 887.4441964285712 378.2366071428569 887.611607142857 378.9062499999996C887.6674107142856 379.0736607142853 887.6674107142856 379.1852678571425 887.7232142857141 379.3526785714281C888.5602678571428 383.0357142857138 889.2299107142857 386.7187499999996 889.6763392857141 390.4575892857139C921.3169642857142 384.6540178571425 955.9151785714284 373.9397321428568 988.1138392857142 354.6874999999997C1031.5290178571427 328.6830357142855 1063.1138392857142 291.4062499999997 1083.4263392857142 261.5513392857139C1074.0513392857142 254.9665178571426 1065.9040178571427 246.4285714285711 1059.654017857143 235.9374999999996C1053.404017857143 225.5022321428567 1049.720982142857 214.2857142857139 1048.325892857143 202.9575892857139C1009.9330357142856 206.9754464285711 963.4486607142858 217.5781249999997 921.1495535714284 242.8571428571425C888.5602678571429 262.2209821428572 862.6116071428571 288.1138392857142 843.1361607142858 312.5z" />
    <glyph glyph-name="thumb_ipynb"
      unicode="&#xF18E;"
      horiz-adv-x="1555.859375" d="M831.25 349.7209821428571V413.9508928571428C834.5982142857142 422.265625 839.6205357142857 429.0178571428571 846.3169642857142 434.2633928571428C853.0133928571428 439.5089285714286 861.8303571428571 442.1316964285714 872.7120535714286 442.1316964285714C886.5513392857142 442.1316964285714 896.9308035714286 436.5513392857142 904.017857142857 425.3348214285715C911.049107142857 414.1183035714287 914.5647321428572 399.5535714285715 914.5647321428572 381.640625V378.90625C914.5647321428572 361.9419642857142 911.1049107142856 348.4375 904.1294642857142 338.3928571428572C897.1540178571428 328.3482142857144 886.7745535714286 323.3258928571429 872.9910714285713 323.3258928571429C862.5 323.3258928571429 853.8504464285714 325.6696428571429 846.9308035714284 330.4129464285715S834.7656250000001 341.5736607142857 831.25 349.7209821428571zM1555.859375 897.3772321428571V101.5066964285713C1555.859375 45.1450892857142 1508.8169642857142 -0.5580357142859 1450.7254464285713 -0.5580357142859H105.1339285714286C47.0424107142857 -0.5580357142858 0 45.1450892857143 0 101.5066964285715V897.3772321428571C0 953.7388392857144 47.0424107142857 999.4419642857144 105.1339285714286 999.4419642857144H1450.6138392857142C1508.7611607142856 999.4419642857144 1555.747767857143 953.7388392857144 1555.859375 897.3772321428571zM1045.033482142857 675.4464285714286L849.1629464285714 873.4375H510.7700892857143V125.4464285714286H1045.033482142857V675.4464285714286zM831.3616071428571 659.7098214285714H1009.4308035714286V161.0491071428571H546.3727678571429V837.8348214285714H831.3616071428571V659.7098214285714zM769.9776785714286 501.6183035714286H754.1294642857142V340.4017857142857L753.3482142857142 340.2901785714286L644.1964285714286 501.6183035714286H628.3482142857143V312.6116071428571H644.1964285714287V474.21875L644.9776785714287 474.3303571428571L754.1294642857143 312.6116071428571H769.9776785714287V501.6183035714286zM815.4017857142858 515.0669642857142V312.5558035714286H827.7901785714287L830.4129464285716 333.59375C835.1562500000001 326.0044642857144 841.3504464285714 320.0892857142857 848.9397321428572 316.015625C856.5290178571429 311.8861607142857 865.5691964285716 309.8772321428571 875.9486607142859 309.8772321428571C892.9129464285717 309.8772321428571 906.2500000000002 316.1272321428571 915.9598214285716 328.627232142857C925.6138392857144 341.1272321428571 930.4687500000002 357.924107142857 930.4687500000002 378.9062499999999V381.6406249999999C930.4687500000002 404.0736607142857 925.6138392857144 421.986607142857 915.9598214285716 435.4352678571428C906.3058035714288 448.8839285714285 892.857142857143 455.6361607142857 875.7254464285717 455.6361607142857C865.6250000000003 455.6361607142857 856.8638392857146 453.515625 849.4419642857146 449.3303571428571C842.0758928571431 445.1450892857142 835.9933035714289 439.1741071428571 831.2500000000003 431.4732142857142V515.0669642857142H815.4017857142858z" />
    <glyph glyph-name="thumb_ipynb_i"
      unicode="&#xF18F;"
      horiz-adv-x="713.3928571428572" d="M428.0133928571429 951.8415178571428H47.4888392857143V48.1584821428571H665.7924107142857V714.0066964285714H428.0133928571429V951.8415178571428zM713.4486607142858 0.6138392857142H-0.0558035714286V999.3861607142856H451.7857142857143L713.3370535714286 734.9888392857142V0.6138392857142H713.4486607142858zM348.2700892857144 233.5937499999999H326.171875L174.1629464285715 462.4441964285714L173.046875 462.2209821428571V233.5379464285715H150.9486607142857V501.0044642857142H173.046875L325.1116071428572 272.6562499999999L326.2276785714286 272.8794642857142V501.0044642857142H348.3258928571429L348.2700892857144 233.5937499999999L348.2700892857144 233.5937499999999zM571.875 327.3995535714286C571.875 297.65625 565.1227678571429 273.9397321428572 551.6741071428571 256.25C538.2254464285713 238.5602678571429 519.5870535714286 229.7433035714286 495.9821428571428 229.7433035714286C481.5290178571428 229.7433035714286 468.9732142857143 232.6450892857144 458.3705357142857 238.4486607142857C447.7678571428572 244.2522321428571 439.1183035714286 252.5669642857142 432.5892857142857 263.3370535714285L428.90625 233.5937499999999H411.6071428571429V520.0892857142858H433.7053571428571V401.7857142857142C440.3459821428571 412.6674107142857 448.7723214285714 421.0937499999999 459.0959821428571 427.0089285714286C469.4196428571428 432.9799107142858 481.5848214285714 435.9375 495.703125 435.9375C519.5870535714286 435.9375 538.28125 426.3950892857144 551.7857142857142 407.3660714285715C565.234375 388.3370535714286 571.9866071428571 362.9464285714286 571.9866071428571 331.2499999999999V327.3995535714286zM549.609375 331.2499999999999C549.609375 356.5848214285715 544.6986607142858 377.2321428571429 534.8772321428571 393.080357142857S510.546875 416.8526785714285 491.2946428571428 416.8526785714285C476.1160714285714 416.8526785714285 463.8392857142857 413.1696428571428 454.5200892857143 405.7477678571427S438.1696428571429 388.7276785714284 433.59375 377.0089285714284V286.049107142857C438.5044642857143 274.5535714285714 445.7589285714286 265.4017857142856 455.3571428571429 258.8169642857141C465.0111607142857 252.1205357142856 477.0647321428572 248.8281249999998 491.6852678571429 248.8281249999998C510.9375 248.8281249999998 525.390625 255.9151785714284 535.0446428571429 270.1450892857141C544.7544642857143 284.3191964285714 549.5535714285714 303.4598214285712 549.5535714285714 327.455357142857L549.609375 331.2499999999999L549.609375 331.2499999999999z" />
    <glyph glyph-name="thumb_map"
      unicode="&#xF190;"
      horiz-adv-x="1555.859375" d="M1555.8593750000002 897.9352678571429V102.0647321428572C1555.8593750000002 45.703125 1508.8169642857144 0 1450.7254464285716 0H105.1339285714286C47.0424107142858 0 0 45.703125 0 102.0647321428572V897.9352678571429C0 954.296875 47.0424107142858 1000 105.1339285714286 1000H1450.6138392857142C1508.7611607142856 1000 1555.7477678571427 954.296875 1555.8593750000002 897.9352678571429zM764.6205357142857 861.3839285714286C613.6160714285714 861.3839285714286 526.2276785714284 712.109375 564.6205357142857 587.5C609.9888392857142 440.1785714285714 764.6205357142857 111.1049107142857 764.6205357142857 111.1049107142857S914.6763392857142 440.8482142857142 962.3325892857142 587.5C1003.4040178571428 713.8392857142858 910.3794642857144 861.3839285714286 764.6205357142857 861.3839285714286zM764.6205357142857 503.2366071428571C683.7611607142856 503.2366071428571 618.0803571428571 569.6986607142857 618.0803571428571 651.6741071428571S683.7053571428571 800.1116071428571 764.6205357142857 800.1116071428571C845.4799107142857 800.1116071428571 911.1607142857142 733.6495535714284 911.1607142857142 651.6741071428571S845.4799107142857 503.2366071428571 764.6205357142857 503.2366071428571z" />
    <glyph glyph-name="thumb_map_i"
      unicode="&#xF191;"
      horiz-adv-x="1555.859375" d="M764.6205357142858 861.3839285714286C613.6160714285714 861.3839285714286 526.2276785714286 712.109375 564.6205357142858 587.5C609.9888392857143 440.1785714285714 764.6205357142858 111.1049107142857 764.6205357142858 111.1049107142857S914.6763392857142 440.8482142857142 962.3325892857144 587.5C1003.4040178571428 713.8392857142858 910.3794642857144 861.3839285714286 764.6205357142858 861.3839285714286zM764.6205357142858 503.2366071428571C683.7611607142857 503.2366071428571 618.0803571428571 569.6986607142857 618.0803571428571 651.6741071428571S683.7053571428571 800.1116071428571 764.6205357142857 800.1116071428571C845.4799107142857 800.1116071428571 911.1607142857142 733.6495535714284 911.1607142857142 651.6741071428571S845.4799107142857 503.2366071428571 764.6205357142858 503.2366071428571z" />
    <glyph glyph-name="thumb_molecule"
      unicode="&#xF192;"
      horiz-adv-x="1555.859375" d="M1450.669642857143 1000H105.1339285714286C47.0424107142858 1000 0 954.296875 0 897.9352678571429V102.0647321428572C0 45.703125 47.0982142857143 0 105.1339285714286 0H1450.7812500000002C1508.872767857143 0 1555.9151785714287 45.703125 1555.9151785714287 102.0647321428572V897.9352678571429C1555.8035714285716 954.296875 1508.8169642857144 1000 1450.669642857143 1000zM1124.21875 460.3794642857142C1122.65625 418.6941964285714 1087.611607142857 386.1607142857142 1045.8705357142856 387.7232142857142C1004.1852678571428 389.2857142857142 971.6517857142856 424.330357142857 973.2142857142856 466.0714285714286C973.3258928571428 467.9129464285713 973.4375 469.6986607142857 973.6049107142856 471.4285714285714C973.4933035714284 471.4285714285714 973.3816964285714 471.484375 973.2700892857142 471.5401785714286L854.1852678571428 493.8616071428571L854.6875 496.4285714285714C849.9441964285714 467.7455357142858 827.2879464285714 441.5736607142858 798.3258928571428 433.5379464285715L800.7254464285713 434.2633928571428L827.1763392857143 309.7656249999999C828.125 309.7656249999999 828.9620535714286 310.0446428571428 829.9107142857143 310.0446428571428C887.6116071428572 310.0446428571428 934.3750000000002 263.28125 934.3750000000002 205.580357142857S887.6116071428572 101.1160714285713 829.9107142857143 101.1160714285713S725.4464285714286 147.8794642857141 725.4464285714286 205.580357142857C725.4464285714286 255.1897321428571 760.1004464285714 296.4285714285714 806.3616071428571 307.142857142857L780.7477678571428 430.6919642857142C779.9107142857142 430.580357142857 779.1294642857142 430.5245535714285 778.2924107142857 430.5245535714285C743.359375 430.5245535714285 714.1183035714284 454.1294642857142 704.296875 486.4397321428571L704.3526785714284 486.2723214285714L591.6294642857142 460.6584821428571C591.9084821428571 457.9241071428571 592.4665178571428 455.2455357142857 592.4665178571428 452.3995535714286C592.4665178571428 405.1897321428572 554.2410714285713 366.9084821428571 506.9754464285713 366.9084821428571C459.7656249999999 366.9084821428571 421.4843749999999 405.1897321428571 421.4843749999999 452.3995535714286S459.7656250000001 537.9464285714286 506.9754464285714 537.9464285714286C543.75 537.9464285714286 574.7767857142858 514.6763392857142 586.8861607142858 482.1428571428571L700.78125 506.6964285714286L700.78125 506.640625C700.6696428571428 507.7008928571428 700.5580357142857 508.7611607142857 700.5580357142857 509.8214285714286C700.5580357142857 552.2321428571429 732.7008928571428 586.71875 774.3861607142857 588.8392857142858L774.3303571428571 588.8392857142858L788.5044642857143 731.3616071428571C752.2879464285714 741.2946428571429 725.4464285714286 774.1071428571429 725.4464285714286 813.5044642857142C725.4464285714286 860.7142857142858 763.671875 898.9955357142857 810.9375 898.9955357142857C858.1473214285714 898.9955357142857 896.4285714285716 860.7700892857142 896.4285714285716 813.5044642857142C896.4285714285716 766.6294642857142 858.6495535714287 728.6272321428571 811.8303571428572 728.125L795.2008928571429 587.0535714285714L795.1450892857143 587.0535714285714C830.3571428571429 579.1852678571429 855.0781250000001 549.2745535714286 855.9709821428572 511.9419642857143L973.3258928571428 491.1272321428572C974.9441964285716 490.7366071428571 976.5066964285716 490.3459821428572 978.013392857143 489.8995535714286C989.1183035714288 519.4754464285713 1018.2477678571428 540.0111607142857 1051.6183035714287 538.7834821428571C1093.3035714285713 537.1651785714286 1125.78125 502.1205357142857 1124.21875 460.3794642857142z" />
    <glyph glyph-name="thumb_molecule_i"
      unicode="&#xF193;"
      horiz-adv-x="1555.859375" d="M1124.21875 460.3794642857142C1122.65625 418.6941964285714 1087.611607142857 386.1607142857142 1045.8705357142856 387.7232142857142C1004.1852678571428 389.2857142857142 971.6517857142856 424.330357142857 973.2142857142856 466.0714285714286C973.3258928571428 467.9129464285713 973.4375 469.6986607142857 973.6049107142856 471.4285714285714C973.4933035714284 471.4285714285714 973.3816964285714 471.484375 973.2700892857142 471.5401785714286L854.1852678571428 493.8616071428571L854.6875 496.4285714285714C849.9441964285714 467.7455357142858 827.2879464285714 441.5736607142858 798.3258928571428 433.5379464285715L800.7254464285713 434.2633928571428L827.1763392857143 309.7656249999999C828.125 309.7656249999999 828.9620535714286 310.0446428571428 829.9107142857143 310.0446428571428C887.6116071428572 310.0446428571428 934.3750000000002 263.28125 934.3750000000002 205.580357142857C934.3750000000002 147.8794642857141 887.6116071428572 101.1160714285713 829.9107142857143 101.1160714285713S725.4464285714286 147.8794642857142 725.4464285714286 205.5803571428572C725.4464285714286 255.1897321428572 760.1004464285714 296.4285714285715 806.3616071428571 307.1428571428572L780.7477678571428 430.6919642857144C779.9107142857142 430.5803571428571 779.1294642857142 430.5245535714287 778.2924107142857 430.5245535714287C743.359375 430.5245535714287 714.1183035714284 454.1294642857143 704.296875 486.4397321428572L704.3526785714284 486.2723214285716L591.6294642857142 460.6584821428572C591.9084821428571 457.9241071428572 592.4665178571428 455.2455357142858 592.4665178571428 452.3995535714287C592.4665178571428 405.1897321428574 554.2410714285713 366.9084821428572 506.9754464285713 366.9084821428572C459.7656249999999 366.9084821428572 421.4843749999999 405.1897321428572 421.4843749999999 452.3995535714287S459.7656250000001 537.9464285714286 506.9754464285714 537.9464285714286C543.75 537.9464285714286 574.7767857142858 514.6763392857142 586.8861607142858 482.1428571428571L700.78125 506.6964285714286V506.640625C700.6696428571428 507.7008928571428 700.5580357142857 508.7611607142857 700.5580357142857 509.8214285714286C700.5580357142857 552.2321428571429 732.7008928571428 586.71875 774.3861607142857 588.8392857142858H774.3303571428571L788.5044642857143 731.3616071428571C752.2879464285714 741.2946428571429 725.4464285714286 774.1071428571429 725.4464285714286 813.5044642857142C725.4464285714286 860.7142857142858 763.671875 898.9955357142857 810.9375 898.9955357142857C858.1473214285714 898.9955357142857 896.4285714285716 860.7700892857142 896.4285714285716 813.5044642857142C896.4285714285716 766.6294642857142 858.6495535714287 728.6272321428571 811.8303571428572 728.125L795.2008928571429 587.0535714285714H795.1450892857143C830.3571428571429 579.1852678571429 855.0781250000001 549.2745535714286 855.9709821428572 511.9419642857143L973.3258928571428 491.1272321428572C974.9441964285716 490.7366071428571 976.5066964285716 490.3459821428572 978.013392857143 489.8995535714286C989.1183035714288 519.4754464285713 1018.2477678571428 540.0111607142857 1051.6183035714287 538.7834821428571C1093.3035714285713 537.1651785714286 1125.78125 502.1205357142857 1124.21875 460.3794642857142z" />
    <glyph glyph-name="thumb_zip_i"
      unicode="&#xF194;"
      horiz-adv-x="1555.859375" d="M694.53125 822.65625V848.9955357142857V855.5803571428571H791.4062499999999V848.9955357142857H823.7165178571428V822.65625H791.4062499999999V816.0714285714286H694.53125V822.65625zM694.53125 700.390625V726.7299107142858V733.2589285714286C694.53125 733.2589285714286 791.4062499999999 733.2589285714286 791.4062499999999 733.203125V726.6183035714286H823.7165178571428V700.2790178571429H791.4062499999999V693.75H694.53125V700.390625zM694.53125 578.3482142857142V604.6875V611.2723214285713H791.4062499999999V604.6875H823.7165178571428V578.3482142857142H791.4062499999999V571.7633928571429H694.53125V578.3482142857142zM694.53125 456.3616071428571V482.7008928571429V489.2299107142857H791.4062499999999V482.6450892857143H823.7165178571428V456.3058035714286H791.4062499999999V449.7209821428572H694.53125V456.3616071428571zM694.53125 334.2075892857144V360.4910714285715V367.0758928571428H791.4062499999999V360.4910714285715H823.7165178571428V334.1517857142857H791.4062499999999V327.5669642857142H694.53125V334.2075892857144zM694.53125 212.1651785714286V238.5044642857144V245.0892857142857H791.4062499999999V238.5044642857144H823.7165178571428V212.1651785714286H791.4062499999999V205.5803571428572H694.53125V212.1651785714286zM823.6049107142858 90.1227678571428H791.2946428571429V83.5379464285715H694.4196428571429V90.1227678571428V116.4620535714286V123.046875H791.2946428571429V116.4620535714286H823.6049107142858V90.1227678571428zM861.2165178571429 177.4553571428571V151.1160714285713V144.53125H764.3415178571429V151.1160714285713H732.03125V177.4553571428571H764.3415178571429V184.0401785714286H861.2165178571429V177.4553571428571zM861.2165178571429 299.4419642857142V273.1026785714285V266.5178571428571H764.3415178571429V273.1026785714285H732.03125V299.4419642857142H764.3415178571429V306.0267857142857H861.2165178571429V299.4419642857142zM861.2165178571429 421.7075892857144V395.3683035714286V388.8392857142857H764.3415178571429V395.3683035714286H732.03125V421.7075892857144H764.3415178571429V428.2924107142857H861.2165178571429V421.7075892857144zM861.2165178571429 543.6941964285713V517.3549107142857V510.7700892857142H764.3415178571429V517.3549107142857H732.03125V543.6941964285713H764.3415178571429V550.2790178571428H861.2165178571429V543.6941964285713zM861.2165178571429 665.6808035714286V639.3415178571429V632.7566964285713H764.3415178571429V639.3415178571429H732.03125V665.6808035714286H764.3415178571429V672.265625H861.2165178571429V665.6808035714286zM861.2165178571429 787.9464285714286V761.6071428571429V755.0223214285714H764.3415178571429V761.6071428571429H732.03125V787.9464285714286H764.3415178571429V794.53125H861.2165178571429V787.9464285714286zM861.2165178571429 909.9330357142856V883.59375V877.0089285714286H764.3415178571429V883.59375H732.03125V909.9330357142856H764.3415178571429V916.5178571428572H861.2165178571429V909.9330357142856z" />
    <glyph glyph-name="timeline_view"
      unicode="&#xF195;"
      horiz-adv-x="1272.544642857143" d="M0 0.1116071428571H181.8080357142857V1000H0V0.1116071428571zM272.7120535714286 0.1116071428571H1272.544642857143V454.6316964285714H272.7120535714286V0.1116071428571zM272.7120535714286 1000V545.5357142857142H1272.544642857143V1000H272.7120535714286z" />
    <glyph glyph-name="timer"
      unicode="&#xF196;"
      horiz-adv-x="876.5066964285714" d="M481.1383928571429 393.8616071428571L619.1964285714287 622.9910714285713L390.0669642857143 484.9330357142857L481.1383928571429 393.8616071428571zM500.1116071428572 869.9776785714286V875H562.6116071428572C597.154017857143 875 625.1674107142857 902.9575892857142 625.1674107142857 937.5C625.1674107142857 971.9866071428572 597.2098214285714 1000 562.6116071428571 1000H312.5558035714286C278.0133928571429 1000 250.0558035714286 971.9866071428572 250.0558035714286 937.5C250.0558035714286 902.9575892857142 278.0133928571429 875 312.5558035714286 875H375.0558035714286V870.0334821428571C163.1138392857143 839.6205357142858 0 657.8125 0 437.4441964285715C0 195.8147321428572 195.8705357142857 -0.0558035714286 437.5558035714286 -0.0558035714286S875.1116071428572 195.8147321428572 875.1116071428572 437.4441964285715C875.1674107142857 657.8125 712.0535714285714 839.6205357142858 500.1116071428572 869.9776785714286zM437.5558035714286 124.9441964285715C264.8995535714286 124.9441964285715 125 264.84375 125 437.4441964285715S264.8995535714286 749.9441964285714 437.5558035714286 749.9441964285714S750.1116071428572 610.0446428571429 750.1116071428572 437.4441964285715S610.2120535714286 124.9441964285715 437.5558035714286 124.9441964285715z" />
    <glyph glyph-name="trash"
      unicode="&#xF197;"
      horiz-adv-x="937.5" d="M875 785.7142857142858H62.5C27.9575892857143 785.7142857142858 0 751.0602678571429 0 716.5178571428571C0 681.9754464285714 28.0133928571429 645.0892857142858 62.5 645.0892857142858H75.8928571428571V142.8571428571429C75.8928571428571 73.828125 131.8638392857143 0 200.8928571428572 0H732.1428571428571C801.171875 0 857.1428571428571 73.828125 857.1428571428571 142.8571428571429V645.0892857142858H875C909.5424107142858 645.0892857142858 937.5 681.9754464285714 937.5 716.5178571428571C937.5 751.0602678571429 909.5424107142858 785.7142857142858 875 785.7142857142858zM714.2857142857143 142.8571428571429H214.2857142857143V645.0892857142858H714.2857142857143V142.8571428571429zM375 930.8035714285714H562.5C597.0424107142858 930.8035714285714 625 893.9174107142858 625 859.375H687.5C687.5 928.4040178571428 631.5290178571429 1000 562.5 1000H375C305.9709821428571 1000 250 928.4040178571428 250 859.375H312.5C312.5 893.9174107142857 340.4575892857144 930.8035714285714 375 930.8035714285714zM321.4285714285715 211.7745535714286C341.0714285714286 211.7745535714286 357.1428571428572 223.2700892857142 357.1428571428572 237.4441964285715V545.7589285714286C357.1428571428572 559.9330357142857 341.0714285714286 571.4285714285714 321.4285714285715 571.4285714285714C301.7299107142857 571.4285714285714 285.7142857142857 559.8772321428571 285.7142857142857 545.7589285714286V237.4999999999999C285.7142857142857 223.2700892857142 301.7299107142857 211.7745535714286 321.4285714285715 211.7745535714286zM464.2857142857143 211.7745535714286C483.984375 211.7745535714286 500 223.2700892857142 500 237.4441964285715V545.7589285714286C500 559.9330357142857 483.9285714285714 571.4285714285714 464.2857142857143 571.4285714285714S428.5714285714286 559.8772321428571 428.5714285714286 545.7589285714286V237.4999999999999C428.5714285714286 223.2700892857142 444.5870535714286 211.7745535714286 464.2857142857143 211.7745535714286zM607.1428571428571 211.7745535714286C626.8973214285714 211.7745535714286 642.8571428571429 223.2700892857142 642.8571428571429 237.4441964285715V545.7589285714286C642.8571428571429 559.9330357142857 626.8973214285714 571.4285714285714 607.1428571428571 571.4285714285714C587.3883928571428 571.4285714285714 571.4285714285714 559.8772321428571 571.4285714285714 545.7589285714286V237.4999999999999C571.4285714285714 223.2700892857142 587.3883928571428 211.7745535714286 607.1428571428571 211.7745535714286z" />
    <glyph glyph-name="update"
      unicode="&#xF198;"
      horiz-adv-x="1000.8370535714286" d="M1000.8370535714286 999.3861607142856V684.8214285714284V682.4776785714284V622.4888392857142C1000.8370535714286 588.1138392857142 972.8236607142856 560.2678571428571 938.2812500000002 560.2678571428571H875.7254464285716V559.9330357142857H870.0892857142859H559.263392857143L762.8348214285716 762.5C695.1450892857143 829.0736607142857 603.125 871.09375 500.4464285714286 871.09375C293.1361607142857 871.09375 125.1116071428571 703.9620535714286 125.1116071428571 497.7678571428571C125.1116071428571 291.6294642857142 293.1919642857142 124.4419642857142 500.4464285714286 124.4419642857142C619.4196428571429 124.4419642857142 725.2790178571428 179.7433035714285 794.0848214285714 265.6808035714286L882.7008928571429 176.8973214285715C790.9040178571429 68.8058035714286 653.7946428571428 0 500.4464285714286 0C224.0513392857143 0 0 222.8794642857143 0 497.7678571428571C0 772.7120535714286 224.1071428571429 995.5915178571428 500.4464285714286 995.5915178571428C637.6674107142858 995.5915178571428 761.5513392857142 940.234375 851.8415178571429 851.171875L1000.8370535714286 999.3861607142856z" />
    <glyph glyph-name="upload"
      unicode="&#xF199;"
      horiz-adv-x="799.7767857142858" d="M0 499.8883928571429L399.8883928571429 999.7209821428572L799.7767857142858 499.8883928571429H599.8325892857143V0.0558035714284H199.9441964285714V499.8883928571429H0z" />
    <glyph glyph-name="user"
      unicode="&#xF19A;"
      horiz-adv-x="1142.9129464285713" d="M995.5357142857144 202.734375C995.5357142857144 202.734375 910.7142857142858 228.90625 773.1026785714287 277.34375C771.8191964285716 301.2834821428572 753.8504464285714 333.4821428571428 722.154017857143 333.4821428571428H718.75H677.5111607142857L684.0959821428572 429.1294642857142C752.5111607142858 461.2723214285714 783.0357142857143 511.6629464285714 796.0379464285714 578.7946428571429C805.8035714285714 629.1852678571429 816.8526785714286 685.9933035714287 816.8526785714286 755.5245535714286C816.8526785714286 824.1071428571429 798.4933035714286 999.9441964285714 569.6428571428571 999.9441964285714C568.9174107142858 999.9441964285714 567.96875 999.9441964285714 567.1316964285714 999.9441964285714C565.9040178571428 999.9441964285714 564.8995535714286 999.9441964285714 563.8950892857143 999.9441964285714C359.0401785714286 1000 327.1205357142857 828.2924107142858 327.1205357142857 755.5803571428571C327.1205357142857 711.8303571428571 338.9508928571429 627.2879464285714 347.8794642857144 578.8504464285713C358.872767857143 519.53125 386.8303571428572 454.5200892857142 459.7656250000001 429.1852678571428L466.2946428571429 333.5379464285714H425.0000000000001H421.6517857142858C389.8995535714287 333.5379464285714 372.4330357142858 299.1071428571427 370.6473214285715 275.1674107142857C237.5000000000001 224.4977678571428 132.6450892857143 196.09375 132.6450892857143 196.09375C33.0357142857143 157.421875 2.8459821428571 182.7566964285715 0 132.3660714285715V93.1919642857142V46.3169642857142L1.7299107142857 34.0401785714286C3.4598214285714 27.3995535714284 6.7522321428571 21.7633928571428 10.7142857142857 16.6294642857142C16.40625 9.1517857142857 23.9955357142857 4.408482142857 32.9799107142857 1.953125C36.328125 1.1160714285714 39.6763392857143 0.0558035714284 43.359375 0.0558035714284L43.359375 0.0558035714284H1098.2142857142858C1110.4910714285716 0.0558035714284 1121.0379464285716 5.4129464285713 1128.7388392857144 14.2299107142857C1140.345982142857 24.9441964285714 1142.9129464285713 46.3169642857142 1142.9129464285713 46.3169642857142V93.1919642857142C1142.9129464285713 93.1919642857142 1142.9129464285713 118.8616071428571 1142.9129464285713 128.6272321428571C1142.9129464285713 130.0223214285715 1142.2433035714284 134.765625 1141.3504464285713 137.7232142857142C1130.46875 172.8236607142857 1100.7254464285716 155.4687499999999 995.5357142857144 202.734375z" />
    <glyph glyph-name="video_player"
      unicode="&#xF19B;"
      horiz-adv-x="1571.4285714285716" d="M706.9196428571429 668.6941964285713L956.9754464285716 498.8839285714286L706.919642857143 329.0178571428571L706.9196428571429 668.6941964285713L706.9196428571429 668.6941964285713zM1571.484375 897.9910714285714V102.0647321428572C1571.484375 45.703125 1523.9397321428573 0 1465.2901785714284 0H106.1383928571429C47.4888392857143 0 -0.0558035714286 45.703125 -0.0558035714286 102.0647321428572V897.9910714285714C-0.0558035714286 954.3526785714286 47.4888392857143 1000.0558035714286 106.1383928571429 1000.0558035714286H1465.1785714285713C1523.8839285714287 1000.0558035714286 1571.372767857143 954.3526785714286 1571.484375 897.9910714285714zM1077.5669642857142 500C1077.5669642857142 661.1607142857142 946.875 791.8526785714286 785.6584821428572 791.8526785714286S493.8058035714286 661.1607142857142 493.8058035714286 500S624.497767857143 208.1473214285715 785.6584821428572 208.1473214285715S1077.5669642857142 338.7834821428571 1077.5669642857142 500z" />
    <glyph glyph-name="view_public"
      unicode="&#xF19C;"
      horiz-adv-x="1559.0401785714287" d="M1559.2075892857142 502.3995535714286C1559.2075892857142 425.5022321428571 1302.9575892857144 0 779.6316964285714 0C304.4642857142857 0 0.0558035714285 427.3995535714286 0.0558035714285 502.3995535714286C0.0558035714285 569.3080357142857 297.7678571428571 998.9955357142856 775.1674107142857 998.9955357142856C1286.216517857143 998.9955357142856 1559.2075892857142 569.3080357142858 1559.2075892857142 502.3995535714286zM779.296875 0M1178.2366071428573 498.2700892857143C1178.2366071428573 719.5870535714286 998.8281250000002 898.9955357142857 777.5111607142859 898.9955357142857S376.7857142857143 719.5870535714286 376.7857142857143 498.2700892857143S556.1941964285714 97.5446428571428 777.5111607142857 97.5446428571428S1178.2366071428573 276.953125 1178.2366071428573 498.2700892857143zM777.5111607142859 824.2745535714286C597.4330357142858 824.2745535714286 451.450892857143 678.2924107142858 451.450892857143 498.2142857142857S597.4330357142859 172.1540178571429 777.5111607142859 172.1540178571429S1103.5714285714287 318.1361607142857 1103.5714285714287 498.2142857142857S957.5334821428572 824.2745535714286 777.5111607142859 824.2745535714286z" />
    <glyph glyph-name="viewer"
      unicode="&#xF19D;"
      horiz-adv-x="1000" d="M100 900H450.0000000000001V1000H100C44.9776785714286 1000 0 955.0223214285714 0 900V550H100V900zM400 450L200 200H800L650 400.0000000000001L548.4933035714287 264.5089285714286L400 450zM750 675C750 716.5178571428571 716.5178571428571 750 675 750C633.4821428571428 750 599.9999999999999 716.5178571428571 599.9999999999999 675S633.4821428571428 600 675 600C716.5178571428571 600 750 633.4821428571429 750 675M900.0000000000001 1000H550.0000000000001V900H900.0000000000001V550H1000V900C1000 955.0223214285714 955.0223214285716 1000 900.0000000000001 1000M900.0000000000001 100H550.0000000000001V0H900.0000000000001C955.0223214285716 0 1000.0000000000002 44.9776785714284 1000.0000000000002 100V450H900.0000000000001V100zM100 450H0V99.9999999999999C0 44.9776785714284 44.9776785714286 -1e-13 100 -1e-13H450.0000000000001V100H100V450z" />
    <glyph glyph-name="viewer_folder"
      unicode="&#xF19E;"
      horiz-adv-x="823.7165178571428" d="M706.0267857142858 1000.2232142857142H117.6897321428572C52.6785714285714 1000.2232142857142 0 947.5446428571428 0 882.5334821428571V117.6897321428572C0 52.6785714285714 52.6785714285714 0 117.6897321428572 0H706.0267857142857C771.0379464285713 0 823.7165178571428 52.6785714285714 823.7165178571428 117.6897321428572V882.5334821428571C823.7165178571428 947.4888392857142 770.9821428571428 1000.2232142857142 706.0267857142858 1000.2232142857142zM647.2098214285714 294.1964285714286H176.5066964285714V353.0133928571428H647.2098214285714V294.1964285714286zM647.2098214285714 470.703125H176.5066964285714V529.5200892857142H647.2098214285714V470.703125zM647.2098214285714 647.2098214285713H176.5066964285714V706.0267857142857H647.2098214285714V647.2098214285713z" />
    <glyph glyph-name="warning"
      unicode="&#xF19F;"
      horiz-adv-x="999.8883928571429" d="M499.9441964285715 0C776.0602678571429 0 999.888392857143 223.8281249999999 999.888392857143 499.9441964285714C999.888392857143 776.0602678571429 776.060267857143 999.8883928571428 499.9441964285716 999.8883928571428C223.8281250000001 999.8883928571428 1e-13 776.0602678571428 1e-13 499.9441964285714C1e-13 223.8281249999999 223.8281250000002 0 499.9441964285716 0zM409.5424107142858 739.84375C405.1171316964286 774.1015290178572 465.4017857142858 812.3883928571429 499.9441964285715 812.3883928571429C534.4866071428572 812.3883928571429 594.7712611607144 774.1015290178572 590.3459821428572 739.84375L551.2834821428572 437.4441964285715C546.8582589285716 403.1864341517857 534.4866071428572 374.9441964285715 499.9441964285715 374.9441964285715C465.4575892857144 374.9441964285715 453.0301339285715 403.1864341517857 448.6049107142858 437.4441964285715zM498.8281250000001 313.5602678571429C533.9843750000001 313.5602678571429 562.4441964285714 285.1004464285715 562.4441964285714 249.9441964285715C562.4441964285714 214.7879464285715 533.9843750000001 186.328125 498.8281250000001 186.328125C463.7276785714287 186.328125 435.2120535714286 214.7879464285715 435.2120535714286 249.9441964285715C435.2120535714286 285.1004464285715 463.6718750000001 313.5602678571429 498.8281250000001 313.5602678571429z" />
    <glyph glyph-name="wesm"
      unicode="&#xF1A0;"
      horiz-adv-x="1607.142857142857" d="M714.2857142857143 357.1428571428571L714.2857142857143 178.5714285714286L892.8571428571429 178.5714285714286L892.8571428571429 357.1428571428571L803.5714285714286 357.1428571428571zM357.1428571428572 303.5714285714286C357.1428571428572 253.5714285714286 396.4285714285715 214.2857142857142 446.4285714285715 214.2857142857142C496.4285714285714 214.2857142857142 535.7142857142858 253.5714285714286 535.7142857142858 303.5714285714286S496.4285714285714 392.8571428571429 446.4285714285715 392.8571428571429C396.4285714285715 392.8571428571429 357.1428571428572 353.5714285714286 357.1428571428572 303.5714285714286zM500 303.5714285714286C500 275 475.0000000000001 250 446.4285714285715 250C417.8571428571429 250 392.8571428571429 275 392.8571428571429 303.5714285714286C392.8571428571429 332.1428571428571 417.8571428571429 357.1428571428571 446.4285714285715 357.1428571428571C475.0000000000001 357.1428571428571 500 332.1428571428571 500 303.5714285714286zM1500 1000H107.1428571428571C46.4285714285714 1000 0 953.5714285714286 0 892.8571428571429V107.1428571428571C0 46.4285714285714 46.4285714285714 0 107.1428571428571 0H1500C1560.7142857142858 0 1607.142857142857 46.4285714285714 1607.142857142857 107.1428571428571V892.8571428571429C1607.142857142857 953.5714285714286 1560.7142857142858 1000 1500 1000zM1160.7142857142858 178.5714285714286C1092.857142857143 178.5714285714286 1035.7142857142858 235.7142857142858 1035.7142857142858 303.5714285714286C1035.7142857142858 367.8571428571429 1082.142857142857 417.8571428571428 1142.857142857143 428.5714285714286V500H821.4285714285714V392.8571428571429H892.8571428571429C914.2857142857144 392.8571428571429 928.5714285714286 378.5714285714286 928.5714285714286 357.1428571428571V178.5714285714286C928.5714285714286 157.1428571428571 914.2857142857144 142.8571428571429 892.8571428571429 142.8571428571429H714.2857142857143C692.8571428571428 142.8571428571429 678.5714285714286 157.1428571428571 678.5714285714286 178.5714285714286V357.1428571428571C678.5714285714286 378.5714285714286 692.8571428571428 392.8571428571429 714.2857142857143 392.8571428571429H785.7142857142858V500H464.2857142857143V428.5714285714286C525 421.4285714285715 571.4285714285714 367.8571428571429 571.4285714285714 303.5714285714286C571.4285714285714 235.7142857142858 514.2857142857143 178.5714285714286 446.4285714285715 178.5714285714286C378.5714285714286 178.5714285714286 321.4285714285715 235.7142857142858 321.4285714285715 303.5714285714286C321.4285714285715 367.8571428571429 367.8571428571429 417.8571428571428 428.5714285714286 428.5714285714286V517.8571428571429C428.5714285714286 528.5714285714286 435.7142857142857 535.7142857142858 446.4285714285715 535.7142857142858H785.7142857142858V642.8571428571429H642.8571428571429C621.4285714285714 642.8571428571429 607.1428571428571 657.1428571428571 607.1428571428571 678.5714285714286V821.4285714285714C607.1428571428571 842.8571428571429 621.4285714285714 857.1428571428571 642.8571428571429 857.1428571428571H964.2857142857144C985.7142857142858 857.1428571428571 1000 842.8571428571429 1000 821.4285714285714V678.5714285714286C1000 657.1428571428571 985.7142857142858 642.8571428571429 964.2857142857144 642.8571428571429H821.4285714285714V535.7142857142858H1160.7142857142858C1171.4285714285713 535.7142857142858 1178.5714285714287 528.5714285714286 1178.5714285714287 517.8571428571429V428.5714285714286C1239.2857142857144 421.4285714285715 1285.7142857142858 367.8571428571429 1285.7142857142858 303.5714285714286C1285.7142857142858 235.7142857142858 1228.5714285714287 178.5714285714286 1160.7142857142858 178.5714285714286zM964.2857142857144 696.4285714285714V803.5714285714286C964.2857142857144 814.2857142857142 957.1428571428572 821.4285714285714 946.4285714285714 821.4285714285714H660.7142857142858C650 821.4285714285714 642.8571428571429 814.2857142857142 642.8571428571429 803.5714285714286V696.4285714285714C642.8571428571429 685.7142857142857 650 678.5714285714286 660.7142857142858 678.5714285714286H803.5714285714286H946.4285714285714C957.1428571428572 678.5714285714286 964.2857142857144 685.7142857142857 964.2857142857144 696.4285714285714zM928.5714285714286 714.2857142857142H678.5714285714286V785.7142857142858H928.5714285714286V714.2857142857142zM1160.7142857142858 392.8571428571429C1110.7142857142858 392.8571428571429 1071.4285714285716 353.5714285714286 1071.4285714285716 303.5714285714286S1110.7142857142858 214.2857142857142 1160.7142857142858 214.2857142857142C1210.7142857142858 214.2857142857142 1250 253.5714285714286 1250 303.5714285714286S1210.7142857142858 392.8571428571429 1160.7142857142858 392.8571428571429zM1160.7142857142858 250C1132.142857142857 250 1107.142857142857 275 1107.142857142857 303.5714285714286C1107.142857142857 332.1428571428571 1132.142857142857 357.1428571428571 1160.7142857142858 357.1428571428571C1189.2857142857142 357.1428571428571 1214.2857142857142 332.1428571428571 1214.2857142857142 303.5714285714286C1214.2857142857142 275 1189.2857142857142 250 1160.7142857142858 250z" />
    <glyph glyph-name="zoom_in"
      unicode="&#xF1A1;"
      horiz-adv-x="999.7209821428572" d="M990.5691964285714 141.8526785714286L812.3325892857142 320.0334821428571C856.0267857142858 388.1696428571428 873.8839285714286 474.7767857142857 873.8839285714286 561.7745535714287C873.8839285714286 803.5714285714286 679.8549107142857 999.6651785714286 437.9464285714286 999.6651785714286C196.09375 999.7209821428572 0 799.7209821428571 0 557.8683035714287C0 315.9598214285715 196.09375 125.8370535714286 437.9464285714286 125.8370535714286C524.9441964285714 125.8370535714286 611.5513392857143 143.6941964285715 679.6875 187.3325892857143L857.9241071428572 9.0401785714286C870.0334821428572 -3.0691964285714 890.0111607142858 -3.0691964285714 902.1205357142858 9.0401785714286L990.625 97.5446428571428C1002.7901785714286 109.765625 1002.7901785714286 129.6875 990.5691964285714 141.8526785714286zM437.9464285714286 248.9397321428571C265.5133928571429 248.9397321428571 125.1674107142857 391.2946428571428 125.1674107142857 563.7276785714284C125.1674107142857 736.1049107142857 265.5691964285715 876.5066964285713 437.9464285714286 876.5066964285713C610.4910714285714 876.5066964285713 750.7812500000001 736.1049107142857 750.7812500000001 563.7276785714284C750.7812500000001 391.2946428571429 610.4352678571429 248.9397321428571 437.9464285714286 248.9397321428571zM499.8883928571428 624.9441964285713H625V499.8325892857143H499.8883928571428V374.7209821428571L374.7767857142857 374.7209821428571L374.7767857142857 499.8325892857143H249.609375V624.9441964285713H374.7209821428572V750.0558035714286H499.8325892857143V624.9441964285713z" />
    <glyph glyph-name="zoom_out"
      unicode="&#xF1A2;"
      horiz-adv-x="999.7209821428572" d="M990.5691964285714 141.8526785714286L812.3325892857142 320.0334821428571C856.0267857142858 388.1696428571428 873.8839285714286 474.7767857142857 873.8839285714286 561.7745535714287C873.8839285714286 803.5714285714286 679.8549107142857 999.6651785714286 437.9464285714286 999.6651785714286C196.09375 999.7209821428572 0 799.7209821428571 0 557.8683035714287C0 315.9598214285715 196.09375 125.8370535714286 437.9464285714286 125.8370535714286C524.9441964285714 125.8370535714286 611.5513392857143 143.6941964285715 679.6875 187.3325892857143L857.9241071428572 9.0401785714286C870.0334821428572 -3.0691964285714 890.0111607142858 -3.0691964285714 902.1205357142858 9.0401785714286L990.625 97.5446428571428C1002.7901785714286 109.765625 1002.7901785714286 129.6875 990.5691964285714 141.8526785714286zM437.9464285714286 248.9397321428571C265.5133928571429 248.9397321428571 125.1674107142857 391.2946428571428 125.1674107142857 563.7276785714284C125.1674107142857 736.1049107142857 265.5691964285715 876.5066964285713 437.9464285714286 876.5066964285713C610.4910714285714 876.5066964285713 750.7812500000001 736.1049107142857 750.7812500000001 563.7276785714284C750.7812500000001 391.2946428571429 610.4352678571429 248.9397321428571 437.9464285714286 248.9397321428571zM249.609375 499.8325892857143H625V624.9441964285713H249.609375V499.8325892857143z" />
  </font>
</defs>
</svg>
#figIcon) format('svg'); } /** * Apply resets only where needed **/ body.fs-no-overflow { overflow: hidden; } figshare-widget, figshare-overlay { line-height: 1; box-sizing: content-box; color: #464646; /* HTML5 display-role reset for older browsers */ color: #444; transform: translate3d(0, 0, 0); } figshare-widget *, figshare-overlay * { box-sizing: content-box; } figshare-widget *:focus, figshare-overlay *:focus { outline: none; } figshare-widget *::-moz-focus-inner, figshare-overlay *::-moz-focus-inner { border: 0; } figshare-widget div, figshare-overlay div, figshare-widget span, figshare-overlay span, figshare-widget applet, figshare-overlay applet, figshare-widget object, figshare-overlay object, figshare-widget iframe, figshare-overlay iframe, figshare-widget h1, figshare-overlay h1, figshare-widget h2, figshare-overlay h2, figshare-widget h3, figshare-overlay h3, figshare-widget h4, figshare-overlay h4, figshare-widget h5, figshare-overlay h5, figshare-widget h6, figshare-overlay h6, figshare-widget p, figshare-overlay p, figshare-widget blockquote, figshare-overlay blockquote, figshare-widget pre, figshare-overlay pre, figshare-widget a, figshare-overlay a, figshare-widget abbr, figshare-overlay abbr, figshare-widget acronym, figshare-overlay acronym, figshare-widget address, figshare-overlay address, figshare-widget big, figshare-overlay big, figshare-widget cite, figshare-overlay cite, figshare-widget code, figshare-overlay code, figshare-widget del, figshare-overlay del, figshare-widget dfn, figshare-overlay dfn, figshare-widget em, figshare-overlay em, figshare-widget img, figshare-overlay img, figshare-widget ins, figshare-overlay ins, figshare-widget kbd, figshare-overlay kbd, figshare-widget q, figshare-overlay q, figshare-widget s, figshare-overlay s, figshare-widget samp, figshare-overlay samp, figshare-widget small, figshare-overlay small, figshare-widget strike, figshare-overlay strike, figshare-widget strong, figshare-overlay strong, figshare-widget tt, figshare-overlay tt, figshare-widget var, figshare-overlay var, figshare-widget b, figshare-overlay b, figshare-widget u, figshare-overlay u, figshare-widget i, figshare-overlay i, figshare-widget center, figshare-overlay center, figshare-widget dl, figshare-overlay dl, figshare-widget dt, figshare-overlay dt, figshare-widget dd, figshare-overlay dd, figshare-widget ol, figshare-overlay ol, figshare-widget ul, figshare-overlay ul, figshare-widget li, figshare-overlay li, figshare-widget fieldset, figshare-overlay fieldset, figshare-widget form, figshare-overlay form, figshare-widget label, figshare-overlay label, figshare-widget legend, figshare-overlay legend, figshare-widget table, figshare-overlay table, figshare-widget caption, figshare-overlay caption, figshare-widget tbody, figshare-overlay tbody, figshare-widget tfoot, figshare-overlay tfoot, figshare-widget thead, figshare-overlay thead, figshare-widget tr, figshare-overlay tr, figshare-widget th, figshare-overlay th, figshare-widget td, figshare-overlay td, figshare-widget article, figshare-overlay article, figshare-widget aside, figshare-overlay aside, figshare-widget canvas, figshare-overlay canvas, figshare-widget details, figshare-overlay details, figshare-widget embed, figshare-overlay embed, figshare-widget figure, figshare-overlay figure, figshare-widget figcaption, figshare-overlay figcaption, figshare-widget footer, figshare-overlay footer, figshare-widget header, figshare-overlay header, figshare-widget hgroup, figshare-overlay hgroup, figshare-widget menu, figshare-overlay menu, figshare-widget nav, figshare-overlay nav, figshare-widget output, figshare-overlay output, figshare-widget ruby, figshare-overlay ruby, figshare-widget section, figshare-overlay section, figshare-widget summary, figshare-overlay summary, figshare-widget time, figshare-overlay time, figshare-widget mark, figshare-overlay mark, figshare-widget audio, figshare-overlay audio, figshare-widget video, figshare-overlay video { margin: 0; padding: 0; border: 0; font-size: 100%; font-family: Arial, Helvetica, sans-serif; vertical-align: baseline; } figshare-widget sub, figshare-overlay sub, figshare-widget sup, figshare-overlay sup, figshare-widget button, figshare-overlay button { margin: 0; padding: 0; border: 0; font-family: Arial, Helvetica, sans-serif; } figshare-widget, figshare-overlay, figshare-widget article, figshare-overlay article, figshare-widget aside, figshare-overlay aside, figshare-widget details, figshare-overlay details, figshare-widget figcaption, figshare-overlay figcaption, figshare-widget figure, figshare-overlay figure, figshare-widget footer, figshare-overlay footer, figshare-widget header, figshare-overlay header, figshare-widget hgroup, figshare-overlay hgroup, figshare-widget menu, figshare-overlay menu, figshare-widget nav, figshare-overlay nav, figshare-widget section, figshare-overlay section { display: block; } figshare-widget ol, figshare-overlay ol, figshare-widget ul, figshare-overlay ul { list-style: none; } figshare-widget blockquote, figshare-overlay blockquote, figshare-widget q, figshare-overlay q { quotes: none; } figshare-widget blockquote::before, figshare-overlay blockquote::before, figshare-widget q::before, figshare-overlay q::before, figshare-widget blockquote::after, figshare-overlay blockquote::after, figshare-widget q::after, figshare-overlay q::after { content: ''; content: none; } figshare-widget table, figshare-overlay table { border-collapse: collapse; border-spacing: 0; } figshare-widget input, figshare-overlay input, figshare-widget select, figshare-overlay select, figshare-widget textarea, figshare-overlay textarea { font-family: Arial, Helvetica, sans-serif; outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget input, figshare-overlay input, figshare-widget textarea, figshare-overlay textarea { border: 0 none; padding: 0; font-size: 13px; background-color: transparent; } figshare-widget strong, figshare-overlay strong { font-weight: bold; } figshare-widget button, figshare-overlay button { cursor: default; background: transparent; } figshare-widget button:focus, figshare-overlay button:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .figshare-loader, figshare-overlay .figshare-loader { width: 100%; position: relative; } figshare-widget .figshare-loader .fs-figshare-loader-holder, figshare-overlay .figshare-loader .fs-figshare-loader-holder { display: inline-block; vertical-align: middle; text-align: center; position: absolute; width: 100%; height: 100%; top: 0; bottom: 0; left: 0; right: 0; } figshare-widget .figshare-loader .fs-figshare-loader-holder:before, figshare-overlay .figshare-loader .fs-figshare-loader-holder:before { content: ''; height: 100%; vertical-align: middle; display: inline-block; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message { padding: 14px 28px; display: inline-block; vertical-align: middle; position: relative; z-index: 200; border: 1px solid #ddd; background: #fff; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message:before, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message:before { content: ''; height: 100%; vertical-align: middle; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .figshare-logo, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .figshare-logo { width: 21px; height: 21px; } @-webkit-keyframes opacityPulse { 0% { opacity: 0; } 10% { opacity: 1; } 100% { opacity: 0; } } @keyframes opacityPulse { 0% { opacity: 0; } 10% { opacity: 1; } 100% { opacity: 0; } } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message g[class^='group'], figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message g[class^='group'] { opacity: 1; -webkit-animation: 0.8s opacityPulse infinite linear; animation: 0.8s opacityPulse infinite linear; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-20, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-20 { -webkit-animation-delay: 0.76s !important; animation-delay: 0.76s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-19, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-19 { -webkit-animation-delay: 0.72s !important; animation-delay: 0.72s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-18, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-18 { -webkit-animation-delay: 0.6799999999999999s !important; animation-delay: 0.6799999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-17, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-17 { -webkit-animation-delay: 0.6399999999999999s !important; animation-delay: 0.6399999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-16, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-16 { -webkit-animation-delay: 0.5999999999999999s !important; animation-delay: 0.5999999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-15, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-15 { -webkit-animation-delay: 0.5599999999999998s !important; animation-delay: 0.5599999999999998s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-14, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-14 { -webkit-animation-delay: 0.5199999999999998s !important; animation-delay: 0.5199999999999998s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-13, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-13 { -webkit-animation-delay: 0.4799999999999998s !important; animation-delay: 0.4799999999999998s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-12, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-12 { -webkit-animation-delay: 0.43999999999999984s !important; animation-delay: 0.43999999999999984s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-11, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-11 { -webkit-animation-delay: 0.39999999999999986s !important; animation-delay: 0.39999999999999986s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-10, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-10 { -webkit-animation-delay: 0.3599999999999999s !important; animation-delay: 0.3599999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-9, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-9 { -webkit-animation-delay: 0.3199999999999999s !important; animation-delay: 0.3199999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-8, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-8 { -webkit-animation-delay: 0.2799999999999999s !important; animation-delay: 0.2799999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-7, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-7 { -webkit-animation-delay: 0.2399999999999999s !important; animation-delay: 0.2399999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-6, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-6 { -webkit-animation-delay: 0.1999999999999999s !important; animation-delay: 0.1999999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-5, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-5 { -webkit-animation-delay: 0.1599999999999999s !important; animation-delay: 0.1599999999999999s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-4, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-4 { -webkit-animation-delay: 0.11999999999999988s !important; animation-delay: 0.11999999999999988s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-3, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-3 { -webkit-animation-delay: 0.07999999999999988s !important; animation-delay: 0.07999999999999988s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-2, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-2 { -webkit-animation-delay: 0.039999999999999876s !important; animation-delay: 0.039999999999999876s !important; } figshare-widget .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-1, figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message .group-1 { -webkit-animation-delay: -0.0000000000000001249s !important; animation-delay: -0.0000000000000001249s !important; } figshare-widget .figshare-loader.fs-loading, figshare-overlay .figshare-loader.fs-loading { position: relative; } figshare-widget .figshare-loader.fs-loading:before, figshare-overlay .figshare-loader.fs-loading:before { content: ''; opacity: 0.7; background: #fff; position: absolute; width: 100%; height: 100%; top: 0; bottom: 0; left: 0; right: 0; z-index: 100; } figshare-widget .figshare-loader .fs-logo, figshare-overlay .figshare-loader .fs-logo, figshare-widget .figshare-loader .fs-message-wrap, figshare-overlay .figshare-loader .fs-message-wrap, figshare-widget .figshare-loader .fs-retry-action, figshare-overlay .figshare-loader .fs-retry-action { display: inline-block; vertical-align: middle; } figshare-widget .figshare-loader .fs-message-wrap, figshare-overlay .figshare-loader .fs-message-wrap, figshare-widget .figshare-loader .fs-retry-action, figshare-overlay .figshare-loader .fs-retry-action { margin-left: 7px; } figshare-widget .figshare-loader .fs-retry-action, figshare-overlay .figshare-loader .fs-retry-action { color: #c74d5a; margin-left: 7px; } figshare-widget .figshare-loader .fs-retry-action:hover, figshare-overlay .figshare-loader .fs-retry-action:hover { text-decoration: underline; } figshare-widget .figshare-loader.fs-error .fs-figshare-loader-message, figshare-overlay .figshare-loader.fs-error .fs-figshare-loader-message { background-color: #ffdbdb; border: none; border-radius: 3px; } figshare-widget .fs-figshare-viewer, figshare-overlay .fs-figshare-viewer { text-align: left; } figshare-widget .fs-figshare-viewer .fs-not-previewable, figshare-overlay .fs-figshare-viewer .fs-not-previewable { height: 100%; position: relative; z-index: 1; } figshare-widget .fs-figshare-viewer .fs-not-previewable:before, figshare-overlay .fs-figshare-viewer .fs-not-previewable:before { content: ''; position: absolute; width: 106%; height: 1px; background-color: #eee; top: 50%; left: -3%; transform: rotate(23deg); } figshare-widget .fs-figshare-viewer .fs-not-previewable:after, figshare-overlay .fs-figshare-viewer .fs-not-previewable:after { content: ''; position: absolute; width: 106%; height: 1px; background-color: #eee; top: 50%; left: -3%; transform: rotate(-23deg); } figshare-widget .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-content, figshare-overlay .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-content { display: inline-block; height: 100%; width: 100%; vertical-align: middle; text-align: center; position: relative; z-index: 10; white-space: nowrap; } figshare-widget .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-content:before, figshare-overlay .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-content:before { content: ' '; display: inline-block; vertical-align: middle; height: 100%; } figshare-widget .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-message, figshare-overlay .fs-figshare-viewer .fs-not-previewable .fs-not-previewable-message { background: #fff; box-shadow: 0 0 50px 30px #fff; white-space: normal; } figshare-widget .fs-figshare-viewer .image-display, figshare-overlay .fs-figshare-viewer .image-display { overflow: hidden; position: absolute; width: 100%; height: 100%; } figshare-widget .fs-figshare-viewer .fs-image-display img, figshare-overlay .fs-figshare-viewer .fs-image-display img { box-shadow: 0px 0px 7px #ccc; position: absolute; top: 50%; left: 50%; transform: translateY(-50%) translateX(-50%); } figshare-widget .fs-figshare-viewer .fs-archive-display, figshare-overlay .fs-figshare-viewer .fs-archive-display { height: 100%; overflow: auto; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-table, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-table { display: table; width: 100%; height: 100%; text-align: left; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row { margin: 0 7px; height: 28px; line-height: 28px; white-space: nowrap; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row:first-child, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row:first-child { border-top: 7px solid #fff; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row:nth-child(even), figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row:nth-child(even) { background: #fff; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row:nth-child(odd), figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row:nth-child(odd) { background: #f5f5f5; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir::before, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir::before { content: "\F13A"; color: #bbb; font-size: 11px; margin-right: 7px; } figshare-widget .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir.fs-expanded::before, figshare-overlay .fs-figshare-viewer .fs-archive-display .fs-archive-row > .fs-archive-dir.fs-expanded::before { content: "\F13B"; color: #bbb; font-size: 11px; margin-right: 7px; } figshare-widget .fs-figshare-viewer .fv-slider-track, figshare-overlay .fs-figshare-viewer .fv-slider-track { position: relative; } figshare-widget .fs-figshare-viewer .fv-slider-track.horizontal .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fv-slider-track.horizontal .fv-slider-progress { width: 100%; } figshare-widget .fs-figshare-viewer .fv-slider-track.vertical .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fv-slider-track.vertical .fv-slider-progress { height: 100%; } figshare-widget .fs-figshare-viewer .fv-slider-track .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fv-slider-track .fv-slider-progress { position: absolute; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper { max-width: 100%; max-height: 100%; width: 100%; height: 100%; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper audio, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper audio { margin: auto; background: transparent; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-controls, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-controls { position: absolute; width: 100%; height: 100%; top: 0; left: 0; background: transparent; margin: 0; padding: 0; border: 0 none transparent; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-controls-bar, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-controls-bar { position: absolute; top: 50%; left: 20px; right: 20px; height: 42px; margin: 20px auto; background-color: #fff; box-sizing: border-box; border: 1px solid #ddd; display: flex; flex-direction: row; align-items: center; justify-content: stretch; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline-container, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline-container { position: relative; margin: 0 7px; height: 42px; width: auto; flex-grow: 1; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline-container::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline-container::before { position: absolute; left: 0; top: 0; width: 100%; height: 42px; background: #f8f8f8; content: ""; border: 1px solid #ddd; border-left: 0; border-right: 0; box-sizing: border-box; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-container, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-container { position: relative; width: 49px; height: 42px; margin-right: 21px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-container::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-container::before { position: absolute; left: 0px; top: 14px; width: 100%; height: 14px; content: ""; border: 1px solid #ddd; box-sizing: border-box; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume { position: absolute; left: 0px; top: 14px; width: 100%; height: 42px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume .fv-slider-progress { left: 0px; height: 14px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume .fv-slider-progress { background: #bbb; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-time, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-time { min-width: 30px; font-size: 12px; text-align: center; color: #464646; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline { position: absolute; left: 0px; top: 0px; width: 100%; height: 42px; display: flex; align-items: center; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline .fv-slider-progress, figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline .fv-slider-progress { top: 0; left: 0px; height: 42px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-timeline .fv-slider-progress { background: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-bufferline .fv-slider-progress { background: #ddd; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play { background: transparent; font-size: 21px; color: #bbb; text-align: center; height: 42px; width: 49px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play[disabled], figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play:focus, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play::before { content: "\F162"; line-height: 42px; color: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play.fs-is-playing::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-play.fs-is-playing::before { content: "\F15E"; line-height: 42px; color: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button { background: transparent; font-size: 21px; color: #bbb; text-align: center; height: 42px; width: 35px; margin-left: 7px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button[disabled], figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button:focus, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button::before { content: "\F10C"; line-height: 42px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button.fs-is-mute::before, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-volume-button.fs-is-mute::before { content: "\F10D"; line-height: 42px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper .fs-audio-glider, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper .fs-audio-glider { position: absolute; bottom: 49px; height: 14px; line-height: 14px; width: auto; color: #999; transform: translateX(-50%); } figshare-widget .fs-figshare-viewer .fs-audio-wrapper.native-controls .fs-audio-loader, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper.native-controls .fs-audio-loader { display: flex; align-items: center; } figshare-widget .fs-figshare-viewer .fs-audio-wrapper.native-controls audio, figshare-overlay .fs-figshare-viewer .fs-audio-wrapper.native-controls audio { width: calc(100% - 28px); padding: 0 14px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper, figshare-overlay .fs-figshare-viewer .fs-media-wrapper { max-width: 100%; max-height: 100%; width: 100%; height: 100%; } figshare-widget .fs-figshare-viewer .fs-media-wrapper video, figshare-overlay .fs-figshare-viewer .fs-media-wrapper video { margin: auto; background: #000; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-controls-shortcuts, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-controls-shortcuts { position: absolute; width: 100%; height: 100%; top: 0; left: 0; background: transparent; margin: 0; padding: 0; border: 0 none transparent; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay { background: transparent; font-size: 21px; color: #ddd; position: absolute; top: 50%; left: 50%; height: 49px; width: 140px; padding: 0 10px; background: rgba(70, 70, 70, 0.95); border-radius: 3px; box-sizing: border-box; font-size: 14px; font-weight: bold; line-height: 49px; margin-left: -70px; margin-top: -24.5px; transition: opacity 0.3s ease-out; opacity: 0.9; text-align: center; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay[disabled], figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:focus, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:hover { color: #fff; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay::before { content: "\F16F"; font-size: 21px; line-height: 49px; margin-right: 10px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-replay:hover { transition: opacity 0.3s ease-in; opacity: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container { position: absolute; bottom: 0; width: 100%; padding: 49px 20px 0 20px; transition: opacity 0.3s ease-out 1s; opacity: 0.9; box-sizing: border-box; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container:hover { transition: opacity 0.3s ease-in 0s; opacity: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-hide-controls, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-hide-controls { transition: opacity 0.3s ease-out 1s; opacity: 0; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-hide-controls:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-hide-controls:hover { transition: opacity 0.3s ease-in 0s; opacity: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-force-visible, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-force-visible { opacity: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-force-visible.fs-hide-controls, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-container.fs-force-visible.fs-hide-controls { opacity: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-bar, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-controls-bar { position: relative; max-width: 800px; height: 49px; margin: 20px auto; background-color: rgba(70, 70, 70, 0.95); color: #fff; border-radius: 3px; display: flex; flex-direction: row; align-items: center; justify-content: stretch; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline-container, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline-container { position: relative; margin: 0 7px; height: 49px; width: auto; flex-grow: 1; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline-container::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline-container::before { position: absolute; top: 21px; left: 0px; height: 7px; width: 100%; background: #fff; content: ""; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-container, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-container { position: relative; width: 70px; height: 49px; margin-right: 7px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-container::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-container::before { position: absolute; top: 21px; left: 0px; height: 7px; width: 100%; background: #fff; content: ""; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-time, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-time { min-width: 42px; font-size: 12px; font-weight: bold; text-align: center; color: #fff; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume { position: absolute; left: 0px; top: 0px; width: 100%; height: 49px; display: flex; align-items: center; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline .fv-slider-progress, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline .fv-slider-progress, figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume .fv-slider-progress { left: 0px; height: 7px; top: 21px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-timeline .fv-slider-progress { background: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-bufferline .fv-slider-progress { background: #ddd; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume .fv-slider-progress, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume .fv-slider-progress { background: #bbb; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play { background: transparent; font-size: 21px; color: #ddd; text-align: center; height: 49px; width: 49px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play[disabled], figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play:focus, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play:hover { color: #fff; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play::before { content: "\F162"; line-height: 49px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-play.fs-is-playing::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-play.fs-is-playing::before { content: "\F15E"; line-height: 49px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button { background: transparent; font-size: 21px; color: #ddd; text-align: center; height: 49px; width: 35px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button[disabled], figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button:focus, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button:hover { color: #fff; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button::before { content: "\F10C"; line-height: 49px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button.fs-is-mute::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-volume-button.fs-is-mute::before { content: "\F10D"; line-height: 49px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen { background: transparent; font-size: 21px; color: #ddd; text-align: center; height: 49px; width: 42px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen[disabled], figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen:focus, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen:hover, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen:hover { color: #fff; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen::before { content: "\F13F"; line-height: 49px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen.fs-exit-fullscreen::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-fullscreen.fs-exit-fullscreen::before { content: "\F140"; line-height: 49px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-glider, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-glider { position: absolute; bottom: 63px; min-height: 28px; width: auto; border-radius: 3px; transform: translateX(-50%); } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-glider.fs-has-preview, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-glider.fs-has-preview { background-position: center; background-repeat: no-repeat; background-color: rgba(70, 70, 70, 0.95); } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-glider.fs-has-preview .fs-video-glider-text, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-glider.fs-has-preview .fs-video-glider-text { border-radius: 0 0 3px 3px; transform: none; width: 100%; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-glider-text, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-glider-text { position: absolute; bottom: 0px; top: auto; padding: 0 10px; white-space: nowrap; height: 28px; width: auto; border-radius: 3px; box-sizing: border-box; text-align: center; transform: translateX(-50%); background: rgba(70, 70, 70, 0.95); color: #fff; font-size: 12px; font-weight: bold; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering { position: absolute; width: 65px; height: 65px; left: 50%; top: 50%; transform: translateX(-50%) translateY(-50%); } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering::before { background: transparent; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering .fs-figshare-loader-message, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fv-loader.fs-video-buffering .fs-figshare-loader-message { background: rgba(255, 255, 255, 0.95); } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-cue, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-cue { position: absolute; box-sizing: content-box; top: 0px; width: 15px; height: 49px; transform: translateX(-50%); } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-cue::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-cue::before { position: absolute; top: 21px; left: 5px; background: rgba(70, 70, 70, 0.5); content: ""; width: 5px; height: 7px; } figshare-widget .fs-figshare-viewer .fs-media-wrapper .fs-video-cue:hover::before, figshare-overlay .fs-figshare-viewer .fs-media-wrapper .fs-video-cue:hover::before { position: absolute; top: 0px; left: 5px; background-color: rgba(187, 187, 187, 0.5); content: ""; width: 5px; height: 28px; } figshare-widget .fs-figshare-viewer .fs-molecule-display, figshare-overlay .fs-figshare-viewer .fs-molecule-display { position: relative; } figshare-widget .fs-figshare-viewer .fs-molecule-display canvas, figshare-overlay .fs-figshare-viewer .fs-molecule-display canvas { background: linear-gradient(to top, #202020 0%, #000 29%, #131313 29%, #000 55%); } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-speed-controls, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-speed-controls { position: absolute; left: 0; bottom: 28px; width: 100%; height: 48px; display: flex; align-items: center; justify-content: center; margin: 0 auto; transition: opacity 0.3s ease-out; opacity: 0; } figshare-widget .fs-figshare-viewer .fs-molecule-display.fs-mode-mobile .fs-speed-controls, figshare-overlay .fs-figshare-viewer .fs-molecule-display.fs-mode-mobile .fs-speed-controls { opacity: 1; bottom: 21px; } figshare-widget .fs-figshare-viewer .fs-molecule-display:hover .fs-speed-controls, figshare-overlay .fs-figshare-viewer .fs-molecule-display:hover .fs-speed-controls { transition: opacity 0.3s ease-in; opacity: 1; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play { height: 48px; background-color: rgba(70, 70, 70, 0.55); font-size: 21px; color: #fff; text-align: center; width: 76px; border-radius: 3px; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play[disabled], figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play:focus, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-play::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-play::before { content: "\F159"; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed { height: 48px; background-color: rgba(70, 70, 70, 0.55); font-size: 21px; color: #fff; text-align: center; width: 56px; border-top-left-radius: 3px; border-bottom-left-radius: 3px; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed[disabled], figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed:focus, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-decrease-speed::before { line-height: 48px; content: "\F179"; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed { height: 48px; background-color: rgba(70, 70, 70, 0.55); font-size: 21px; color: #fff; text-align: center; width: 56px; border-top-right-radius: 3px; border-bottom-right-radius: 3px; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed[disabled], figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed:focus, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-increase-speed::before { line-height: 48px; content: "\F17B"; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause { height: 48px; background-color: rgba(70, 70, 70, 0.55); font-size: 21px; color: #fff; text-align: center; width: 76px; display: flex; flex-flow: column; align-items: center; justify-content: center; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause::before { display: inline-block; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause[disabled], figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause[disabled] { color: #464646; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause:focus, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause:focus { outline: none; box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause::before, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause::before { font-size: 10px; margin-bottom: 4px; content: "\F15F"; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause .fs-speed-indicator, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause .fs-speed-indicator { line-height: 14px; font-size: 13px; } figshare-widget .fs-figshare-viewer .fs-molecule-display .fs-pause .fs-speed-indicator.default-speed, figshare-overlay .fs-figshare-viewer .fs-molecule-display .fs-pause .fs-speed-indicator.default-speed { font-size: 11px; } figshare-widget .fs-figshare-viewer .fs-document-display, figshare-overlay .fs-figshare-viewer .fs-document-display { overflow: auto !important; position: relative; } figshare-widget .fs-figshare-viewer .fs-document-display .figshare-loader, figshare-overlay .fs-figshare-viewer .fs-document-display .figshare-loader { overflow: hidden !important; } figshare-widget .fs-figshare-viewer .fs-document-display .fs-image-display, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-image-display { position: relative; margin: 30px auto; border: 1px solid #ddd; box-shadow: 0px 0px 7px #ccc; opacity: 0.9; transition: opacity 1s, transform 1s, transform-origin 1s; } figshare-widget .fs-figshare-viewer .fs-document-display .fs-image-display.fs-prev-page, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-image-display.fs-prev-page { transform-origin: center bottom; transform: perspective(600px) rotateX(1deg) translateZ(-2px); } figshare-widget .fs-figshare-viewer .fs-document-display .fs-image-display.fs-next-page, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-image-display.fs-next-page { transform-origin: center top; transform: perspective(600px) rotateX(-1deg) translateZ(-2px); } figshare-widget .fs-figshare-viewer .fs-document-display .fs-image-display.fs-current-page, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-image-display.fs-current-page { opacity: 1; transform: perspective(600px) rotateX(0deg) translateZ(0px); } figshare-widget .fs-figshare-viewer .fs-document-display .fs-canvas-document-container, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-canvas-document-container { overflow: auto; position: absolute; } figshare-widget .fs-figshare-viewer .fs-document-display .fs-canvas-document-container .fs-page-wrapper, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-canvas-document-container .fs-page-wrapper { margin: 30px auto; border: 1px solid #ddd; box-shadow: 0 0 7px #ccc; position: relative; } figshare-widget .fs-figshare-viewer .fs-document-display .fs-canvas-document-container .fs-page-wrapper .fs-canvas-clone, figshare-overlay .fs-figshare-viewer .fs-document-display .fs-canvas-document-container .fs-page-wrapper .fs-canvas-clone { position: absolute; top: 0; left: 0; width: 100%; height: 100%; } figshare-widget .fs-figshare-viewer .fs-loading-layer, figshare-overlay .fs-figshare-viewer .fs-loading-layer { position: absolute; z-index: 4; left: 0; top: 0; width: 100%; height: 100%; background: #fff; } figshare-widget .fs-figshare-viewer .fs-text-layer, figshare-overlay .fs-figshare-viewer .fs-text-layer { position: absolute; left: 0; top: 0; right: 0; bottom: 0; overflow: hidden; opacity: 0.2; line-height: 1; } figshare-widget .fs-figshare-viewer .fs-text-layer > div, figshare-overlay .fs-figshare-viewer .fs-text-layer > div { color: transparent; position: absolute; white-space: pre; cursor: text; transform-origin: 0% 0%; } figshare-widget .fs-figshare-viewer .fs-text-layer .highlight, figshare-overlay .fs-figshare-viewer .fs-text-layer .highlight { margin: -1px; padding: 1px; border-radius: 4px; } figshare-widget .fs-figshare-viewer .fs-text-layer .highlight.begin, figshare-overlay .fs-figshare-viewer .fs-text-layer .highlight.begin { border-radius: 4px 0 0 4px; } figshare-widget .fs-figshare-viewer .fs-text-layer .highlight.end, figshare-overlay .fs-figshare-viewer .fs-text-layer .highlight.end { border-radius: 0 4px 4px 0; } figshare-widget .fs-figshare-viewer .fs-text-layer .highlight.middle, figshare-overlay .fs-figshare-viewer .fs-text-layer .highlight.middle { border-radius: 0; } figshare-widget .fs-figshare-viewer .fs-text-layer ::-moz-selection, figshare-overlay .fs-figshare-viewer .fs-text-layer ::-moz-selection { background: #3496fb; } figshare-widget .fs-figshare-viewer .fs-text-layer ::selection, figshare-overlay .fs-figshare-viewer .fs-text-layer ::selection { background: #3496fb; } figshare-widget .fs-figshare-viewer .fs-text-layer ::-moz-selection, figshare-overlay .fs-figshare-viewer .fs-text-layer ::-moz-selection { background: #3496fb; } figshare-widget .fs-figshare-viewer .fs-text-layer .endOfContent, figshare-overlay .fs-figshare-viewer .fs-text-layer .endOfContent { display: block; position: absolute; left: 0; top: 100%; right: 0; bottom: 0; z-index: -1; cursor: default; -webkit-user-select: none; -moz-user-select: none; -ms-user-select: none; user-select: none; } figshare-widget .fs-figshare-viewer .fs-text-layer .endOfContent.active, figshare-overlay .fs-figshare-viewer .fs-text-layer .endOfContent.active { top: 0; } figshare-widget .fs-figshare-viewer .fs-dataset-display, figshare-overlay .fs-figshare-viewer .fs-dataset-display { overflow: hidden; width: 100%; height: 100%; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-display, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-display { position: relative; overflow: auto; height: 100%; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper { width: 100%; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper table, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper table { white-space: normal; text-align: left; min-width: 100%; border-collapse: collapse; border-spacing: 0px; vertical-align: middle; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper th, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper th, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper td, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper td { border: 1px solid #ccc; padding: 0.4em 0.8em; vertical-align: middle; white-space: nowrap; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper th, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper th { background: #ddd; font-weight: 400; text-align: center; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper tbody th, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-sheet-wrapper tbody th { width: 60px; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel { position: relative; width: 100%; height: 34px; overflow: hidden; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper { position: relative; width: 100%; height: 34px; overflow: hidden; z-index: 0; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active { width: calc(100% - 80px); } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::after, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::after, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::before { z-index: 1; content: " "; opacity: 0; transition: opacity 0.5s; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::before { position: absolute; top: 0px; left: 0px; right: auto; display: block; width: 20px; height: 34px; background: linear-gradient(to right, #ffffff, rgba(255, 255, 255, 0)); } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::after, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active::after { position: absolute; top: 0px; left: auto; right: 0px; display: block; width: 20px; height: 34px; background: linear-gradient(to left, #ffffff, rgba(255, 255, 255, 0)); } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active.fs-fade-left::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active.fs-fade-left::before, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active.fs-fade-right::after, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-wrapper.fs-controls-active.fs-fade-right::after { opacity: 1; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-container, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-container { position: relative; top: 0px; display: inline-block; height: 34px; text-align: left; white-space: nowrap; z-index: 0; transition: left 0.5s; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-controls, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-controls { position: absolute; top: 0px; left: auto; right: 0px; width: 80px; height: 34px; text-align: center; background: #fff; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control { width: 21px; height: 34px; color: #ddd; line-height: 34px; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control.fs-active, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control.fs-active, figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control.fs-active, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control.fs-active { color: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-left-control::before { content: "\F108"; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control::before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel-right-control::before { content: "\F109"; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-carousel, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-carousel { border-top: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector { position: relative; display: inline-block; margin-right: -13px; padding: 0 20px; color: #fff; text-align: center; vertical-align: middle; height: 24px; width: auto; z-index: 0; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector:before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector:before { position: absolute; top: 0px; left: 0px; right: 0px; bottom: 0px; border-top: 24px solid #c74d5a; border-left: 13px solid transparent; border-right: 13px solid transparent; border-radius: 0 0 18px 18px; content: ""; z-index: -1; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector:after, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector:after { position: absolute; top: 0px; left: -1px; right: -1px; bottom: -1px; border-top: 25px solid #ddd; border-left: 13px solid transparent; border-right: 13px solid transparent; border-radius: 0 0px 18px 18px; content: ""; z-index: -2; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector:focus, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector.fs-active, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector.fs-active { color: #464646; font-weight: 700; } figshare-widget .fs-figshare-viewer .fs-dataset-display .fs-selector.fs-active:before, figshare-overlay .fs-figshare-viewer .fs-dataset-display .fs-selector.fs-active:before { border-top: 24px solid #fff; } figshare-widget .fs-figshare-viewer .leaflet-bar, figshare-overlay .fs-figshare-viewer .leaflet-bar { background-color: #eee; background-color: rgba(255, 255, 255, 0.4); border-radius: 4px; padding: 2px; box-shadow: none; } figshare-widget .fs-figshare-viewer .leaflet-bar a, figshare-overlay .fs-figshare-viewer .leaflet-bar a, figshare-widget .fs-figshare-viewer .leaflet-bar a:hover, figshare-overlay .fs-figshare-viewer .leaflet-bar a:hover { color: #fff; font-size: 1.14em; font-weight: 700; text-decoration: none; text-align: center; height: 1.375em; width: 1.375em; line-height: 1.4em; background-color: #7b98bc; background-color: rgba(0, 60, 136, 0.5); margin: 1px; border: none; } figshare-widget .fs-figshare-viewer a.leaflet-disabled, figshare-overlay .fs-figshare-viewer a.leaflet-disabled, figshare-widget .fs-figshare-viewer a.leaflet-disabled:hover, figshare-overlay .fs-figshare-viewer a.leaflet-disabled:hover { color: #000; background-color: rgba(255, 255, 255, 0.4) !important; } figshare-widget .fs-figshare-viewer .leaflet-bar a:hover, figshare-overlay .fs-figshare-viewer .leaflet-bar a:hover { background-color: #4c6079; background-color: rgba(0, 60, 136, 0.7); } figshare-widget .fs-figshare-viewer .leaflet-control-zoom-in, figshare-overlay .fs-figshare-viewer .leaflet-control-zoom-in { border-radius: 2px 2px 0 0 !important; } figshare-widget .fs-figshare-viewer .leaflet-control-zoom-out, figshare-overlay .fs-figshare-viewer .leaflet-control-zoom-out { border-radius: 0 0 2px 2px !important; } figshare-widget .fs-figshare-viewer .fs-info:before, figshare-overlay .fs-figshare-viewer .fs-info:before { content: "i"; } figshare-widget .fs-figshare-viewer .fs-info.fs-open:before, figshare-overlay .fs-figshare-viewer .fs-info.fs-open:before { content: ">"; } figshare-widget .fs-figshare-viewer .fs-info:focus, figshare-overlay .fs-figshare-viewer .fs-info:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .leaflet-control-attribution, figshare-overlay .fs-figshare-viewer .leaflet-control-attribution { left: -10px; top: -10px; padding: 7px 30px 7px 7px; border-radius: 4px; } figshare-widget .fs-figshare-viewer .fs-text-display, figshare-overlay .fs-figshare-viewer .fs-text-display { position: relative; } figshare-widget .fs-figshare-viewer .fs-text-display .fs-text-container, figshare-overlay .fs-figshare-viewer .fs-text-display .fs-text-container { height: 100%; overflow: auto; text-align: left; } figshare-widget .fs-figshare-viewer .fs-text-display pre, figshare-overlay .fs-figshare-viewer .fs-text-display pre { white-space: pre; font-size: 12px; line-height: 14px; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs, figshare-widget .fs-figshare-viewer .fs-text-display [class^="hljs-"], figshare-overlay .fs-figshare-viewer .fs-text-display [class^="hljs-"] { font-family: "Lucida Console", Monaco, monospace; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs { display: block; padding: 14px; color: #464646; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-comment, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-comment, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-quote, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-quote { color: #999; font-style: italic; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-keyword, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-keyword, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-selector-tag, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-selector-tag, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-subst, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-subst { color: #157009; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-number, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-number, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-literal, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-literal, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-variable, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-variable, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-template-variable, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-template-variable, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-tag .hljs-attr, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-tag .hljs-attr { color: #008080; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-string, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-string, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-doctag, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-doctag { color: #c74d5a; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-title, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-title, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-section, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-section, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-selector-id, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-selector-id { color: #c81a2b; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-subst, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-subst { font-weight: normal; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-type, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-type, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-class .hljs-title, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-class .hljs-title { color: #458; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-tag, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-tag, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-name, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-name, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-attribute, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-attribute { color: #000080; font-weight: normal; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-regexp, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-regexp, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-link, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-link { color: #A3CD3D; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-symbol, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-symbol, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-bullet, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-bullet { color: #990073; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-built_in, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-built_in, figshare-widget .fs-figshare-viewer .fs-text-display .hljs-builtin-name, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-builtin-name { color: #0086b3; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-meta, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-meta { color: #556471; font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-deletion, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-deletion { background: #ffdbdb; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-addition, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-addition { background: #edf3d7; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-emphasis, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-emphasis { font-style: italic; } figshare-widget .fs-figshare-viewer .fs-text-display .hljs-strong, figshare-overlay .fs-figshare-viewer .fs-text-display .hljs-strong { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-jupyter-display, figshare-overlay .fs-figshare-viewer .fs-jupyter-display { position: relative; } figshare-widget .fs-figshare-viewer .fs-jupyter-display .figshare-loader, figshare-overlay .fs-figshare-viewer .fs-jupyter-display .figshare-loader { overflow: hidden !important; } figshare-widget .fs-figshare-viewer .fs-viewer3d-display, figshare-overlay .fs-figshare-viewer .fs-viewer3d-display { position: relative; } figshare-widget .fs-figshare-viewer .fs-viewer3d-display canvas, figshare-overlay .fs-figshare-viewer .fs-viewer3d-display canvas { background: linear-gradient(to top, #202020 0%, #000 29%, #131313 29%, #000 55%); } figshare-widget .fs-figshare-viewer .fs-graph-display, figshare-overlay .fs-figshare-viewer .fs-graph-display { position: relative; } figshare-widget .fs-figshare-viewer .fs-graph-display canvas, figshare-overlay .fs-figshare-viewer .fs-graph-display canvas { background: radial-gradient(circle, #fff 50%, #ddd); } figshare-widget .fs-figshare-viewer .fs-fits-display .figshare-loader, figshare-overlay .fs-figshare-viewer .fs-fits-display .figshare-loader { overflow: hidden !important; } figshare-widget .fs-figshare-viewer .fs-fits-display .fs-canvas-container, figshare-overlay .fs-figshare-viewer .fs-fits-display .fs-canvas-container { background: rgba(0, 0, 0, 0.3); } figshare-widget .fs-figshare-viewer .fs-fits-display .fs-canvas-container canvas, figshare-overlay .fs-figshare-viewer .fs-fits-display .fs-canvas-container canvas { display: block; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-explore-container, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-explore-container { background-color: rgba(255, 255, 255, 0.1); } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info { position: absolute; left: 14px; top: 14px; background: rgba(48, 48, 48, 0.75); border-radius: 3px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-image-meta, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-image-meta { width: 230px; padding: 14px; color: #bbb; line-height: 21px; font-size: 12px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-image-meta .fv-fits-meta-separator, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-image-meta .fv-fits-meta-separator { height: 1em; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation { color: #fff; padding: 7px; font-size: 14px; line-height: 28px; display: flex; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button { width: 28px; height: 28px; line-height: 28px; background: transparent; color: inherit; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.next::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.next::after { content: "\F17B"; font-size: 14px; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.prev::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.prev::after { content: "\F179"; font-size: 14px; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta { margin-left: 12px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta::after { content: "\F11C"; font-size: 14px; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close { margin-left: auto; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close::after, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.meta.close::after { content: "\F134"; font-size: 14px; line-height: 28px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.disabled, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button.disabled { color: #bbb; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button:focus, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-nav-button:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-index, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-image-info .fv-fits-layer-navigation .fv-fits-layer-index { vertical-align: middle; font-size: 14px; margin: 0 7px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fs-histo-container, figshare-overlay .fs-figshare-viewer .fs-fits-display .fs-histo-container { position: absolute; right: 14px; bottom: 14px; background: rgba(0, 0, 0, 0.3); } figshare-widget .fs-figshare-viewer .fs-fits-display .fs-histo-container.disabled, figshare-overlay .fs-figshare-viewer .fs-fits-display .fs-histo-container.disabled { display: none; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-controls-container, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-controls-container { background-color: #f8f8f8; -webkit-user-select: none; -moz-user-select: none; -ms-user-select: none; user-select: none; line-height: 21px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-button, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-button { margin: 0 7px; font-size: 14px; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-button.disabled, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-button.disabled { color: #bbb; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close { position: relative; top: 3px; float: right; color: #bbb; } figshare-widget .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close::before, figshare-overlay .fs-figshare-viewer .fs-fits-display .fv-fits-histo-close::before { content: "\F112"; } figshare-widget .fs-figshare-viewer .fs-fits-display .fs-histo-canvas-container canvas, figshare-overlay .fs-figshare-viewer .fs-fits-display .fs-histo-canvas-container canvas { display: block; } figshare-widget .fs-figshare-viewer .fs-figshare-viewer, figshare-overlay .fs-figshare-viewer .fs-figshare-viewer { position: relative; text-align: left; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper { height: auto; width: 100%; position: relative; z-index: 2; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-molecule-controls-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-molecule-controls-wrap { width: 100%; height: 38px; display: flex; flex-flow: row; justify-content: flex-end; align-items: center; background: #f5f5f5; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal { margin: 0 14px 0 7px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-pagination-info, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-pagination-info { display: inline-block; width: 42px; text-align: right; font-size: 11px; font-weight: normal; margin-right: 7px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page { margin: 0 7px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before { vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page:focus, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page:focus, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page:focus, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-prev-page::before { content: "\F179"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-pagination.fs-horizontal .fs-next-page::before { content: "\F17B"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comment-list, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comment-list { max-width: 270px; height: 100%; display: inline-block; vertical-align: middle; position: relative; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger { position: relative; margin-right: 10px; padding-right: 5px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before { position: absolute; right: -9px; font-size: 5px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::after { top: 12px; content: "\F10A"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger::before { top: 20px; content: "\F107"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-trigger:focus, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-trigger:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-current-comment, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-current-comment { font-size: 13px; line-height: 38px; white-space: nowrap; max-width: 100%; overflow: hidden; text-overflow: ellipsis; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-menu-wrapper, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-menu-wrapper { position: absolute; right: -105px; bottom: 0; width: 0; height: 0; display: flex; justify-content: center; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu { position: absolute; top: 0; left: auto; right: 0; max-width: 385px; border: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::after, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::before { top: 0; right: 110px; border: solid transparent; content: " "; height: 0; width: 0; position: absolute; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::after { border-bottom-color: #fff; border-width: 7px; margin-right: -7px; margin-top: -14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu::before { border-bottom-color: #ddd; border-width: 8px; margin-right: -8px; margin-top: -16px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item { white-space: nowrap; overflow: hidden; text-overflow: ellipsis; font-size: 13px; line-height: 20px; border-bottom: 1px solid #ddd; padding: 7px 14px; color: #464646; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item:last-child, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item:last-child { border-bottom: 0; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item.fs-active, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item.fs-active { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item:hover, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item:hover, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item.fs-active, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-comments-menu .fs-drop-item.fs-active { background-color: #f8f8f8; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-items-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-items-wrap { max-height: 244px; min-width: 182px; background-color: #fff; overflow-y: auto; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile.fs-molecule-controls-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile.fs-molecule-controls-wrap { border-top: 1px solid #464646; justify-content: center; background: #000; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-pagination, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-pagination { display: none; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-trigger, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-trigger { color: #f8f8f8; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-menu-wrapper, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-menu-wrapper { right: 0; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu { position: fixed; top: 83px; left: 0; right: 0; bottom: 0; width: auto; height: auto; max-width: none; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu::before, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-comments-menu::after { right: 50%; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-current-comment, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-current-comment { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-drop-item, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-drop-item { padding: 21px 14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-items-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-mode-mobile .fs-items-wrap { max-height: 60%; width: 100%; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button { margin: 0 7px; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button:before { font-size: 16px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button[disabled], figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button[disabled] { color: #bbb; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button:focus, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-button:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-in:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-in:before { content: "\F1A1"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-out:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-out:before { content: "\F1A2"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-fit:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom .fs-zoom-fit:before { content: "\F138"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls { width: 100%; height: 38px; display: flex; flex-flow: row; justify-content: flex-end; align-items: center; background: #f5f5f5; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-image-navigation-control, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-image-navigation-control { font-size: 14px; line-height: 38px; color: #464646; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu { display: inline-block; position: relative; height: 38px; line-height: 38px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name { color: #464646; font-size: 14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name::after { content: "\F10B"; margin-left: 7px; font-size: 14px; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name span, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-fits-layer-name span { display: inline-block; max-width: 170px; text-overflow: ellipsis; white-space: nowrap; overflow: hidden; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper { right: 0; position: absolute; top: 100%; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu { position: relative; border: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap { background: white; z-index: 2; position: relative; width: 250px; font-size: 14px; max-height: 320px; overflow-y: auto; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item { display: flex; padding: 0 14px; border-top: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item:first-child, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item:first-child { border: none; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item .layer-count, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item .layer-count { font-weight: 13px; color: #464646; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item.fv-active, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .fv-drop-item.fv-active { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .layer-title, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .layer-title { text-overflow: ellipsis; white-space: nowrap; overflow: hidden; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .layer-count, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu .fv-items-wrap .layer-count { margin-left: 4px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper .fv-comments-menu::before { content: ""; z-index: 1; position: absolute; top: -5px; right: 6px; width: 8px; height: 8px; transform: rotate(45deg); background: white; border: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fs-zoom, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fs-zoom { display: inline; margin-left: 14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-button, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-button { margin: 0 7px; height: 100%; vertical-align: middle; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-button:focus, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-button:focus { box-shadow: 0 0 3px 0px #ffa500 inset, 0 0 3px 3px #ffa500; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layer-index, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-layer-index { margin-left: 28px; margin-right: 7px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-prev-layer-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-prev-layer-button:before { content: "\F179"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-next-layer-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-next-layer-button:before { content: "\F17B"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button { margin-left: 14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button:before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button:before { content: "\F147"; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button.disabled, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-histo-button.disabled { color: #bbb; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu { position: absolute; max-width: 461px; background-color: #fff; border: 1px solid #ddd; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::after, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::before { top: -16px; left: 50%; border: solid transparent; content: " "; height: 0; width: 0; position: absolute; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::after, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::after { border-bottom-color: #fff; border-width: 8px; margin-left: -8px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu::before { border-bottom-color: #ddd; border-width: 7px; margin-left: -7px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item { white-space: nowrap; overflow: hidden; text-overflow: ellipsis; line-height: 20px; border-bottom: 1px solid #ddd; padding: 7px 14px; color: #464646; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item:last-child, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item:last-child { border-bottom: 0; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item.fv-active, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item.fv-active { font-weight: bold; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item:hover, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item:hover, figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item.fv-active, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-menu .fv-drop-item.fv-active { background-color: #f8f8f8; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile { justify-content: space-between; padding: 0 14px; box-sizing: border-box; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-fits-layer-name, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-fits-layer-name { margin: 0; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-fits-layer-name span, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-fits-layer-name span { width: 75px; text-align: left; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-menu-wrapper.fv-fits-menu-wrapper { right: auto; left: -14px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-comments-menu::before, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-comments-menu::before { right: auto; left: 95px; } figshare-widget .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-comments-menu .fv-items-wrap, figshare-overlay .fs-figshare-viewer .fs-controls-wrapper .fv-fits-controls.mobile .fv-fits-layers-menu .fv-comments-menu .fv-items-wrap { width: 320px; } figshare-widget .fs-files-viewer-loader, figshare-overlay .fs-files-viewer-loader { width: 100%; height: 100%; } figshare-widget .fs-files-viewer-loader.fs-loading, figshare-overlay .fs-files-viewer-loader.fs-loading { display: none; } figshare-widget .fs-figshare-viewer, figshare-overlay .fs-figshare-viewer { width: 100%; height: 100%; position: relative; z-index: 0; } figshare-widget .fs-figshare-viewer .figshare-loader, figshare-overlay .fs-figshare-viewer .figshare-loader { width: 100%; height: 100%; } figshare-overlay { display: none; z-index: 200; z-index: 1002; } figshare-overlay.fs-active { display: block; position: fixed; top: 0; left: 0; bottom: 0; right: 0; } figshare-overlay .fs-overlay-content { position: relative; margin: 0 auto; z-index: 201; } figshare-overlay .fs-overlay-backdrop { position: absolute; width: 100%; height: 100%; display: block; overflow: auto; z-index: 200; background: rgba(0, 0, 0, 0.85); } figshare-overlay .figshare-loader { height: 100%; width: 100%; } figshare-overlay .figshare-loader::before { display: none; } figshare-overlay .figshare-loader .fs-figshare-loader-holder .fs-figshare-loader-message g[class^='group'] :local { -webkit-animation-name: opacityPulse; animation-name: opacityPulse; } .frontend-widgets-filesViewerGeneric-theme-module__primaryButton--wriES { position: relative; background: #62422b; border: 1px solid #ddd; color: #fff; } .frontend-widgets-filesViewerGeneric-theme-module__primaryButton--wriES:hover { background: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__primaryButton--wriES[disabled] { background: #bbb; } .frontend-widgets-filesViewerGeneric-theme-module__primaryIconButton--IP2jh { padding: 0; } .frontend-widgets-filesViewerGeneric-theme-module__secondaryButton--wc-Ev { position: relative; background: #fff; border: 1px solid #ddd; color: #62422b; } .frontend-widgets-filesViewerGeneric-theme-module__secondaryButton--wc-Ev:hover { color: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__secondaryButton--wc-Ev[disabled] { background: #eee; color: #999; } .frontend-widgets-filesViewerGeneric-theme-module__secondaryIconButton--oPU3V { padding: 0; } .frontend-widgets-filesViewerGeneric-theme-module__iconButton--LMkyj { position: relative; color: #62422b; } .frontend-widgets-filesViewerGeneric-theme-module__iconButton--LMkyj:hover { color: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_toggleListButton--S6tD6 { border-color: #62422b; } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_toggleListButton--S6tD6:hover { border-color: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_closeOverlayButton--P5vPZ { border-color: #62422b; } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_closeOverlayButton--P5vPZ:hover { border-color: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_downloadButton_desktop--mgmlM { } .frontend-widgets-filesViewerGeneric-theme-module__enlargedMode_downloadButton_mobile--Kk0iK { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_prevFileButton--fz5IU { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_nextFileButton--u2nbK { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_toggleListButton--JFSVF { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_enlargeButton--emsBo { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_shareButton--YrMKy { } .frontend-widgets-filesViewerGeneric-theme-module__inlineMode_downloadButton--5ykHk { } .frontend-widgets-filesViewerGeneric-theme-module__fileListScreen_viewButton--ksNmF { } .frontend-widgets-filesViewerGeneric-theme-module__fileListScreen_downloadButton--kejiB { } .frontend-widgets-filesViewerGeneric-theme-module__downloadScreen_downloadButton--cA8EI { } .frontend-widgets-filesViewerGeneric-theme-module__shareScreen_citeButton--EUh-k { } .frontend-widgets-filesViewerGeneric-theme-module__citationScreen_doiLink--N31Pd { } .frontend-widgets-filesViewerGeneric-theme-module__citationScreen_backButton--wVL72 { } .frontend-widgets-filesViewerGeneric-theme-module__citationSelector_triggerButton--qrTtd { } .frontend-widgets-filesViewerGeneric-theme-module__citationSelector_itemButton--uerg\+ { } .frontend-widgets-filesViewerGeneric-theme-module__citationSelector_backButton--O9FL- { } .frontend-widgets-filesViewerGeneric-theme-module__citationSelector_searchButton--vgJWb { } .frontend-widgets-filesViewerGeneric-theme-module__fileDescription--pASc1 .fs-toggle button { color: #62422b; } .frontend-widgets-filesViewerGeneric-theme-module__fileDescription--pASc1 .fs-toggle button:hover { color: #ac703d; } .frontend-widgets-filesViewerGeneric-theme-module__skipButton--DO9UR { } We Value Your Privacy! We and our partners are using technologies like cookies and process personal data like the IP-address or browser information in order to personalize the advertising that you see. This helps us to show you more relevant ads and improves your internet experience. We also use it in order to measure results or align our website content. Because we value your privacy, we are herewith asking for your permission to use these technologies. You can always change/withdraw your consent later. Necessary Cookies Store and/or access information on a device Personalised ads, ad and content measurement, audience insights and product development Accept All Settings Reject All Save Settings | Cookies | Privacy Policy | Terms & Conditions | Imprint powered by consentmanager.net Skip to Main Content Close Who We Serve Who We Serve Researchers Authors Reviewers Healthcare Professionals Patients & Their Supporters Librarians Health Sciences Industry Societies Agents & Distributors What We Offer What We Offer Subject Areas Journals Books & Series Collections Courses Podcasts Open Access What We Solve What We Solve Accessing Knowledge Presenting Knowledge Applying Knowledge About Us Publish with Us Publish with Us Publish Your Paper Calls for Papers Open Access Publishing Publication Services Partner Publications Publication Ethics Resources for You Resources for You Experience Karger The Waiting Room Embarrassing Problems Nephrology Viewpoints ISCN Online DermaCompass Search Dropdown Menu header search search input Search input auto suggest filter your search All Content All Journals Public Health Genomics Search /#MicrositeSearch /.navbar-search Advanced Search /.navbar-search-container (function () { var hfSiteUrl = document.getElementById('hfSiteURL'); var siteUrl = hfSiteUrl.value; var subdomainIndex = siteUrl.indexOf('/'); hfSiteUrl.value = location.host + (subdomainIndex >= 0 ? siteUrl.substring(subdomainIndex) : ''); })(); MOBILE SHOPPING CART ICON User Tools Dropdown DESKTOP SHOPPING CART ICON DESKTOP REGISTRATION Register DESKTOP INSTITUTIONS Institutional Accounts Saarländische Universitäts und Althoff Konsortium, Berlin Sign out of all accounts /.dropdown-menu /.dropdown-panel .dropdown-panel-institution /.site-theme-header-menu-item .desktop DESKTOP SIGN IN Login /.dropdown-panel /.site-theme-header_content /.site-theme-header- Toggle Menu Menu Content Articles Issues Early View About Details Contact Editorial Board Guidelines Open Access Advertising Journal Factsheet Submission /.navbar /.center-inner-row /.journal-header Skip Nav Destination Close navigation menu Article navigation Volume 18, Issue 4 July 2015 Abstract Introduction Materials and Methods Results Discussion Appendix Acknowledgments Disclosure Statement

Those who have doubted the ability of customers to understand results of personal genomic testing may have been selling them short, researchers at the University of Michigan School of Public Health have found. In one of the first large studies to measure customer comprehension of health-related genetic test reports from personal genomic testing companies, U-M researchers found that overall people were able to understand this information. "Our main research aim was to assess how well customers understood several hypothetical genetic test reports. We found, for the most part, they were able to correctly interpret the scenarios we presented," said Jenny Ostergren, lead author and doctoral candidate in health behavior and health education at the School of Public Health. Participants in the research were the customers of 23andMe Inc. and Pathway Genomics. Results of the research are reported in the current issue of Public Health Genomics. Since the first personal genetic testing company started in 2006, questions have been raised about sharing this kind of information with the public without someone to interpret it. Several leading governmental agencies have said that such information in the hands of the public could lead to psychological harms and misuse of health-care system resources. The Food and Drug Administration sent a warning letter in November 2013 to 23andMe raising concerns that misunderstandings of the tests could lead to patient noncompliance or mismanagement of medications. The genetic testing company stopped selling its health-related reports from personal genetic tests to customers in the U.S. shortly thereafter, and now only provides ancestry data. Earlier this year, the FDA approved one of the company's tests, which allows a healthy person to find out if he or she has a genetic variant that could lead to a serious disorder in offspring. In this study, the U-M-led team presented four hypothetical scenarios with sample genetic test reports to 1,030 customers of the two companies. The reports included genetic risk for Alzheimer's disease and type 2 diabetes, carrier screening results for more than 30 conditions, specific carrier screening results for phenylketonuria (PKU) and cystic fibrosis, and drug response results for a statin drug. Participants had high overall comprehension of the information presented, with an average score of 79 percent correct across scenarios, and scored extremely high on their understanding of statin drug response and carrier screening results, ranging from 81-to-97 percent. They were less certain about specific screening results for PKU, with scores from 64-to-75 percent, which the researchers said may be due to a lack of understanding about recessive traits. Participants also scored lower on questions about the diabetes scenario, which researchers believe was because the description of the hypothetical person as "obese" clouded the bigger picture of the results. "In general, people did fairly well across these scenarios so some of the concerns that people won't be able to handle the information on their own might be unfounded," said senior author Scott Roberts, U-M associate professor of health behavior and health education, who added, however, that the relatively high education level of the population taking the survey must be considered. "This was a highly selected group of people overall, so we can't really say how a more diverse group would fare," said Roberts, who also is director of the Genomics, Health and Society Program at the U-M Center for Bioethics and Social Sciences in Medicine. Overall, the customers of these services are predominantly white and have higher levels of education and income than the general population, Ostergren said. "We found those with better comprehension had high numeracy skills, more genetic knowledge and higher education overall. Older age was associated with lower comprehension," she said. "Because of this, there may not be a one-size-fits-all approach to communicating this kind of genetic test information. Tailoring the presentation of this information based on individual characteristics or preferences and the type of test results could potentially enhance comprehension of results."

10.1159/000431250

Biology

Using AI to push the boundaries of wildlife survey technologies

Zijing Wu et al, Deep learning enables satellite-based monitoring of large populations of terrestrial mammals across heterogeneous landscape, Nature Communications (2023). DOI: 10.1038/s41467-023-38901-y Journal information: Nature Communications

https://dx.doi.org/10.1038/s41467-023-38901-y

https://phys.org/news/2023-06-ai-boundaries-wildlife-survey-technologies.html

Abstract New satellite remote sensing and machine learning techniques offer untapped possibilities to monitor global biodiversity with unprecedented speed and precision. These efficiencies promise to reveal novel ecological insights at spatial scales which are germane to the management of populations and entire ecosystems. Here, we present a robust transferable deep learning pipeline to automatically locate and count large herds of migratory ungulates (wildebeest and zebra) in the Serengeti-Mara ecosystem using fine-resolution (38-50 cm) satellite imagery. The results achieve accurate detection of nearly 500,000 individuals across thousands of square kilometers and multiple habitat types, with an overall F1-score of 84.75% (Precision: 87.85%, Recall: 81.86%). This research demonstrates the capability of satellite remote sensing and machine learning techniques to automatically and accurately count very large populations of terrestrial mammals across a highly heterogeneous landscape. We also discuss the potential for satellite-derived species detections to advance basic understanding of animal behavior and ecology. Introduction The African continent has the greatest diversity and abundance of mammals in the world 1 . This status, however, is threatened by intensive land use changes driven by increasing natural resource extraction and infrastructure development 2 , 3 . Even in protected areas, Africa’s large mammal populations have declined by 59% in three decades 4 , and many are now categorized as endangered or threatened by the International Union for Conservation of Nature. Climate change promises to only accelerate these losses, underscoring the need for advanced monitoring techniques that can provide managers with information at a rate that keeps pace with local environmental changes 5 , 6 . Conventional methods for surveying large wildlife, especially in Africa, have relied on crewed aerial surveys for decades 7 , 8 , 9 , 10 , 11 . This approach has generated some of the longest-running ecological datasets in the world and formed the foundation of leading conservation strategies across the continent. However, crewed surveys introduce risks to human and wildlife and in many cases can only provide animal counts with coarse location precision. Moreover, all crewed aerial survey techniques are subject to biases arising from detection probability, observer experience and double counting 8 , 12 . Uncrewed aerial vehicles (UAVs) with imaging sensors offer a promising alternative to crewed surveys in some cases 13 , 14 , 15 , 16 , 17 , 18 . However, like crewed flights, UAVs are generally limited by fuel or battery life and, thus, are limited in scale and can be difficult to maintain in remote locations 19 . Moreover, UAVs can disturb wildlife when flown at low altitudes 20 , 21 , 22 , which has led to flight restrictions in some protected areas 23 . Recent advances in satellite technology have dramatically increased the feasibility of conducting uncrewed surveys in remote landscapes and at greater scales than UAVs are currently capable of. Many of the first applications of this technology focused on visualizing and analyzing easier-to-view environmental markers that, in certain contexts, provide insights to estimate population size (e.g., guano stains 24 , nests 25 , mounds and burrows 26 ). It took less than a few years, however, for the technology to accommodate manual counts at the scale of individual animals for species in unobscured contexts (e.g., polar bears 27 , albatrosses 28 , and Weddell seals 29 , 30 ). However, reliance on labor-intensive manual detection has restricted uptake by the conservation community, highlighting the need for automated techniques for processing fine-resolution satellite images. Machine learning and the associated sub-field of deep learning, have offered promising solutions to the challenge of conducting wildlife surveys from space. Over the past decade, deep learning has been a key driver of progress in science and engineering 31 . Such advancements have had a transformative impact on the field of computer vision, where the performance of some deep learning algorithms has achieved or surpassed human-level performance in many tasks 32 , 33 , 34 , 35 , 36 . At the same time, new collaborations between ecologists and computer scientists have provided several key advancements in automated animal detection from satellite imagery, including detection of the world’s largest marine and terrestrial vertebrates, such as whales 37 and elephants 38 , using object detection algorithms. However, the performance of current object detectors suffers from the small size of the objects in imagery 39 , 40 , 41 . The feasibility of successfully using object detection methods is dependent on the body size of the animal: mature whales have a body length of more than 20 m 42 , and African elephants are generally 3–5 m long 43 , both of which have more than eight pixels along the body length axis in submeter-resolution (e.g., 0.3–0.5 m) satellite imagery. A few studies have conducted automated surveys for smaller species with satellite images, such as for seals 44 and albatrosses 45 using pixel-based semantic segmentation algorithms. Image segmentation deep learning architectures such as U-Net 46 predict the class probability for every pixel, showing the potential to detect animals with a smaller size in satellite imagery. However, these early successes were limited to high-contrast species in homogeneous environments. The capability to reliably distinguish smaller animals (e.g., ≤9 pixels in size in satellite imagery, such as wildebeest, one of the African ungulate species) from complex backgrounds (e.g., mixed forest and savanna ecosystems) remains uninvestigated and continues to be a major question in satellite-based techniques for wildlife surveys 47 . Here, we address this shortcoming by presenting a robust framework for efficiently locating and counting wildebeest-sized animals with a body length of 1.5–2.5 m from submeter-resolution satellite imagery across a large, highly heterogeneous landscape. We do this by integrating a post-processing clustering module with a U-Net-based deep learning model, which uses high-precision pixel-based image segmentation to locate animals at the object level. We demonstrate the power of this framework by deploying it to locate and count the largest terrestrial mammal migration on the planet—the migration of white bearded wildebeest ( Connochaetes taurinus ) and plains zebra ( Equus quagga ) across the Serengeti-Mara ecosystem. Wildebeest have an estimated population of ~1.3 million individuals, making them the most numerous species in the ecosystem by an order of magnitude 48 , 49 . There are also over 250,000 zebras and other ungulate species that move seasonally across the system in tandem with wildebeest 48 . As a result, their annual migration drives multiple ecological processes that support the health of humans and wildlife across the region (i.e., nutrient cycling, trophic interactions, biomass removal and habitat recovery from over utilization 50 , 51 , 52 , 53 ). In addition, the spectacle of the great migration supports a robust tourism industry, which underpins regional economies across Kenya and Tanzania. However, with the migration subject to seasonality of rainfall and habitat preference, this iconic system is facing unprecedented threats from rapid climate and environmental change 54 , 55 , 56 , 57 . Thus, the ability to frequently and accurately assess the status of migratory ungulate populations is key to forming conservation policies that address current threats and promote ecosystem function. In addition to supporting conservation planning in East Africa, these methodological advances stand to inform basic scientific understanding of ecological patterns and processes, such as quantitatively describing the emergent properties of animal aggregations 58 , 59 and answering long-standing questions about the mechanisms that drive behavioral shifts from individuals to populations. Such insights are crucial for advancing the fields of functional ecology and collective behavior, yet the technological challenges associated with studying animal aggregations in the wild have hindered scientific understanding outside of a laboratory environment 60 . Here, we take a germinal step towards overcoming such challenges by presenting a method for locating and counting large groups of animals in fine-resolution satellite imagery. Results A U-Net-based ensemble learning model for wildebeest detection As a network designed for image segmentation tasks, U-Net allows precise pixel-level localization of a target class in an image 46 . However, it is not directly suitable for object detection applications. To address this issue, we present a U-Net-based detection pipeline that involves a post-processing module using a clustering method (Fig. 1 ). The pipeline is composed of three main blocks. In the first block, we subdivide the raw satellite image scenes into 336 by 336-pixel images (hereafter patches) as the input images for the model. The wildebeest in the input images are annotated as points, which are expanded to 3 by 3-pixel segments and are then converted to binary wildebeest/non-wildebeest image segmentation masks. In the second block, the satellite image patches and the corresponding masks of labeled wildebeest are fed into the U-Net model, which predicts the probability of wildebeest presence for each pixel. The U-Net model has a U-shaped symmetrical encoder-decoder structure that consists of a contracting path on the left, which extracts high-level features, an expanding path on the right that increases the resolution, and multiple levels of skip connections between two paths that allows for precise localization. To increase the robustness of the model, we adopt ensemble learning through a K -fold splitting method. The training dataset is split into ten folds, with nine folds used for training and the remaining fold used for validation. This ensemble block introduces variation in the training and validation datasets and achieves 10 individual base models. We then summarize the predictions by averaging the probability maps produced by these 10 base models. In the last post-processing block, we convert the pixel-wise prediction into wildebeest individuals through K -means clustering. The clumped wildebeest pixels were disaggregated by K -means clustering to separate individual wildebeest (Supplementary Fig. 1 ), which were used as the final outputs for evaluation at the individual level. Note that as wildebeest is the dominant ungulate species in the system and most animals we located and counted were wildebeest, we refer hereafter to the migratory ungulates detected by our model as wildebeest for the purpose of simplicity. Fig. 1: Model framework. The wildebeest detection pipeline consists of three main blocks: 1) The wildebeest are labeled in the satellite imagery and the masks are generated; 2) The satellite images and the masks are fed into the U-Net-based ensemble model for model training/validation and to produce the wildebeest probability maps; 3) The probability maps produced by the 10 base models are averaged to obtain the final predictions and the wildebeest individuals are detected using K -means clustering. The blue dots on example image of wildebeest labels represent manually annotated wildebeest labels. The red dots on example image of detected wildebeest represent wildebeest detected by the framework. In the U-Net architecture visualization, each box in gray color represents a multi-channel feature map layer. The gray box with dashed line represents copied feature map from the left part. Each arrow represents an operation. Satellite image © 2010 Maxar Technologies. Full size image We applied the pipeline to satellite images acquired over six years (August 2009, September 2010, August 2013, July 2015, August 2018, and October 2020) covering 2747 km 2 in the Serengeti-Mara ecosystem (Fig. 2 ). The images were captured by different satellite sensors with distinct spatial resolutions ranging from 38 cm to 50 cm, including GeoEye-1 (GE01), WorldView-2 (WV02) and WorldView-3 (WV03). Each individual wildebeest in the satellite imagery was represented by ~3-to-4 pixels in length and 1-to-3 pixels in width, with 1 or 2 relatively darker pixels in the center, including the shadow of the body (Fig. 3 ). The training dataset contained 1097 image patches captured from these six years, including 53,906 manually labeled wildebeest points across various environmental conditions. We incorporated labels created by four independent expert observers by majority voting. The details about the level of their agreement are presented in Supplementary Table 1 . During the labeling process, we used a set of reference satellite images acquired on different dates, but with the same background landscapes for cross-referencing to ensure the labels were moving animals and were not similar-looking static objects (e.g., termite mounds, small bushes). The acquisition dates and spatial resolutions of the reference images are presented in Supplementary Data 1 . During model training, the training dataset was split randomly into 10 folds, among which nine folds were used for training and the remaining one fold was used for validation. Fig. 2: Study area map. The satellite imagery used in this research cover mainly the Masai Mara National Reserve and the northernmost section of the Serengeti National Park (the area outlined in red). The wildebeest typically migrate over 1500 km on average every year (the purple dashed line). During June and August, the wildebeest migrate from the Serengeti plains in Tanzania into the Masai Mara National Reserve and then spread to the east crossing the Mara River in September. Then during November and December, they move south to the southern Serengeti. Image credit: EreborMountain/Shutterstock.com for the wildebeest art photo. Full size image Fig. 3: Labeling the wildebeest on the satellite image. a The reference satellite image that was used for cross-referencing while labeling the wildebeest. This example image was acquired on May 17th,2012. b The satellite image acquired on September 24th, 2010 for wildebeest labeling. c Wildebeest labels on B. The red points denote wildebeest annotations. The zoomed boxes are three examples of the wildebeest labels on the GE01 image with 44-cm resolution. Satellite image © 2010 Maxar Technologies. Full size image To evaluate model performance, we used a stratified random sampling method to select test sample plots across the images in each year to ensure their representativeness and independence from the training dataset. The strata are based on the number of animals in the image patches. The distribution of the number of animals per image is summarized in Supplementary Fig. 2 . In total, we selected 2700 test images containing 11,594 wildebeest individuals. Key information about the images used and the size of training and test dataset is summarized in Supplementary Table 2 . More details about the sampling method and data preparation process are described in the Methods section. We calculated the model performance for each year and also calculated the overall accuracy by combining all the test datasets. The accuracy (precision, recall, F1-score) was evaluated on a per-individual basis as demonstrated in Fig. 4 . The model achieved an overall F1-score of 84.75% with a precision of 87.85% and a recall of 81.86%. The model performed well in each year (Supplementary Table 3 ): all F1-scores were above 80% (between 80.40% and 91.70%). The precision across the six years varied between 82.68% and 97.80% and recall between 74.00% and 87.52% (Fig. 5a ). This indicates that the model has good generalization ability across varied image resolution (from 38 to 50 cm), despite the great temporal and spatial variation in landscape type, ecological conditions, and mode of image acquisition over different years. Fig. 4: Examples of model evaluation on individual wildebeest. In the Evaluation column, the predictions that match the ground references are True Positives (TP, red crosses), and those that do not match are False Positives (FP, blue crosses). Ground references that were not detected by the model are False Negatives (FN, yellow crosses). The examples are taken from the test set of 2009–2020, showing that the model avoids most of the background objects that have similar size and color to wildebeest objects, such as small bushes, shadows on the edges of ponds, and roads. Satellite image © 2009–2020 Maxar Technologies. Full size image Fig. 5: Model performance. a The wildebeest detection accuracy of the U-Net-based ensemble model for each of the six years and the whole dataset. Error bars represent mean values ± SD ( n = 5). b The Precision-Recall curve of the ensemble model and each base model. The red line (representing the ensemble model) lies above all other blue curves (representing the individual base models), indicating greater accuracy. Full size image To validate the advantage of using an ensemble model, we also compared the performance of the ensemble model with the individual base models. The original training dataset was split into 10 folds, nine of which were used for training and the remaining fold for validation, resulting in 10 models trained on various datasets. The predictions of the 10 models were averaged to obtain the final results. We assessed the performance of each individual model using the Precision-Recall curve and Area Under the Curve (AUC). The ensemble model achieved an AUC of 0.88, which is significantly higher than all other base models (Fig. 5b ). We also compared the F1-score: the F1-score of 10 base models on average is 78.22% (±0.86%), also lower than the F1-score of ensemble model (84.75%). A more detailed comparison is listed in Supplementary Table 4 . Model transferability To assess the temporal and spatial transferability of the model, we ran two tests: 1. Transferability of the model to a temporally different dataset: we selected the image from 2015 as an independent test dataset and trained the model with wildebeest labels from the other five years (2009, 2010, 2013, 2018, 2020). The 2015 dataset was an unseen image captured with a different sensor, with the finest spatial resolution (38 cm of WV03 versus 42–50 cm of GE01 and WV02). The model achieved high accuracy on this new dataset, with a precision of 90.77%, recall of 95.61%, and F1-score of 93.13%. Such high accuracy indicates the model can be transferred to a temporally different dataset without adding additional training samples and still demonstrate excellent performance. 2. Transferability of the model to a spatially different dataset: we selected the images from 2020 as an independent test dataset and trained the model with wildebeest labels from the other five years (2009, 2010, 2013, 2015, 2018). The coverage of the 2020 data is on the east side of Masai Mara National Reserve and Serengeti National Park, which is outside the coverage of the remaining datasets, and its spatial resolution is the coarsest (50 cm of WV02) of all years. The model achieved a 96.98% precision, showing that the model is able to avoid false positives without adding any new training samples for this new task with different landscapes and ecological conditions. The recall score is 60.65% (with F1-score of 74.63%), indicating the ability to detect all positives can still be improved by adding more samples from the 2020 dataset. Wildebeest detection and counting To detect and count migratory wildebeest within the area, we applied the U-Net-based ensemble model trained with full training datasets from all six years to the entire satellite imagery dataset that covered a large portion of the dry-season range of migratory wildebeest. Figure 6 shows examples of the detection across varied landscape characteristics including savanna, woodland and riverine forests. The detection results demonstrate the model’s robustness to variation in three dimensions: (1) variation between different satellite sensors, namely, various spatial resolutions over the six different years; (2) variation in the landscape context, such as river, woodland, bushland and grassland, with the risk of confusion with background objects such as termite mounds, small bushes and shadows caused by terrain, and (3) variation in the wildebeest aggregation patterns, such as scattered, linear and clustered. Further examples of detected wildebeest patterns across very large areas can be found in Supplementary Fig. 3 - 8 and Supplementary Data 2 . The method resulted in a sum count of 480,362 (ranging between 470,121 and 490,603) individual wildebeest (F1-score: 84.75 ± 0.18%) across the whole dataset (Table 1 ). See Fig. 7 for the location and coverage of the imagery of each year and Table 1 for the number of animals detected in each year. Fig. 6: Detecting wildebeest across different landscapes with variation in wildebeest spatial clustering patterns. The figures in the first column show the detected wildebeest (red circles). The second column is a zoom of the imagery covered by the white square in the first column. a Detected wildebeest in GeoEye-1 imagery acquired on August 11th, 2009. In the zoomed-in image, the wildebeest are crossing the road near a dry riverbed. b Detected wildebeest in GeoEye-1 imagery acquired on August 10th, 2013. Wildebeest herd in open grasslands. c Detected wildebeest in WorldView-3 imagery acquired on July 17th, 2015. The wildebeest prepare to cross the Mara River. d Detected wildebeest in GeoEye-1 imagery acquired on August 2, 2018. Herds of wildebeest avoid the closed woodlands. e Detected wildebeest in WorldView-2 imagery acquired on October 8th, 2020. The wildebeest herds move through open woodlands and grasslands. These examples also show the heterogeneity between the satellite images, inclusive of spectral variation and different levels of contrast between the wildebeest and the background. Satellite image © 2009–2020 Maxar Technologies. Full size image Table 1 The number of wildebeest detected and counted in six different years of satellite imagery Full size table Fig. 7: Spatial distribution of detected wildebeest from July to October in 2009-2020. The area outlined in red represents the study area, covering the Masai Mara National Reserve and the northernmost section of the Serengeti National Park.The area outlined in white indicates the corresponding area presented in Fig. 8 . The histogram shows the calculated wildebeest frequency distribution for each scene. a Spatial distribution hotspot map of wildebeest detected in July 2015. The image is located in the northernmost section of Serengeti National Park with the Mara River flowing through. The maximum wildebeest density is about 1500 per km 2 . b Spatial distribution hotspot map of wildebeest detected in August 2018. The image is located in the Mara Triangle inside the Masai Mara National Reserve, covering the border of Kenya and Tanzania. The wildebeest are near the border and the density peak is more than 4000 individuals per km 2 . c Spatial distribution hotspot map of wildebeest detected in August 2013. The image covers the Mara Triangle in the Masai Mara National Reserve and the northern section of the Serengeti National Park. The wildebeest are mostly distributed in the Serengeti National Park near the border and the density peak is about 4000 individuals per km 2 . d Spatial distribution hotspot map of wildebeest detected in August 2009. The image is located in the northwest corner of the Masai Mara National Reserve. The wildebeest density peak is about 6000 individuals per km 2 . e Spatial distribution hotspot map of wildebeest detected in September 2010. The image is located in the north Serengeti National Park with the Mara River flowing through. The wildebeest are mostly on the north side of the Mara River and the density peak is about 3000 per km 2 . f Spatial distribution hotspot map of wildebeest detected in October 2020. The images cover the east side of the Mara National Reserve and northeast Serengeti National Park. The wildebeest span sparsely across the Mara National Reserve and Serengeti National Park and the density peak is about 2000 per km 2 . The maximum wildebeest density displays a large difference in terms of months in the dry season. Satellite image © 2009–2020 Maxar Technologies. Full size image To further analyze the spatial distribution pattern of the migrating wildebeest in the Serengeti-Mara ecosystem, we calculated the wildebeest count per km 2 in each scene and plotted the resulting histogram (see Fig. 7a–f ). The maximum wildebeest density displays great variation across months in the dry season (July–October). Peaks in wildebeest density appear in August in the western Masai Mara National Reserve (more than 4000 to 6000 individual wildebeest per km 2 ). In September, the peak wildebeest density is ~3000 per km 2 , while in July and October, the maximum density is between 1500 and 2000 per km 2 . The spatially and temporally varied density is visualized in the hotspot maps in Fig. 7 . We also present the enlarged hotspot map in Fig. 8 . The high densities and dense clusters of wildebeest were observed in the three representative images from August (2009, 2013, 2018). Variation in this pattern is evident in the lower wildebeest densities observed in the representative image analyzed from September 2010 and the more scattered distribution observed spread out over a larger area in the October 2020 image. The distribution dynamics observed comply with the general wildebeest migration patterns shown in Fig. 2 . The wildebeest migrate to the north towards the Mara Triangle in July and August, and aggregate there for grazing before moving further southeast across the Masai Mara National Reserve in September, and spread south into the vast Serengeti National Park in October, as shown in the sparse distribution in the hotspot map. Fig. 8: Hotspot map and spatial density of wildebeest over time (from July to October, 2009 to 2020). In this figure, a subset of each timeframe was taken for display purposes and the hotspot map was produced for each timeframe with a cell size of 100 m and a radius of 500 m using Point Density tool in ArcGIS. The density of wildebeest varies from 0 to more than 10,000 wildebeest per km 2 , and it shows a clear spatial variation of wildebeest aggregation patterns in different months. Satellite image © 2009–2020 Maxar Technologies. Full size image Discussion The detection pipeline presented here demonstrates the potential for deep learning techniques to efficiently track fine-scale environmental changes through automated, satellite-based wildlife surveys. To create outputs that would have real-world utility to researchers and managers, we deployed our model at an especially large spatial scale (2747 km 2 ) and validated it on a dataset that varied in space, time, and resolution. This approach yielded highly accurate results (with an overall F1-score of 84.75%) and the largest training dataset ever published from a satellite-based wildlife survey (53,906 annotations). In addition to its size, the landscape diversity captured by this dataset will facilitate model transferability to applications in similar environmental contexts, such as future satellite-based wildebeest census surveys at the ecosystem scale. Although generalization of our model is inherently limited to wildebeest-like animals in open landscapes, the pipeline itself is generic and can be applied to other animal detection applications after retraining. Beyond providing a truly open-source and transferable method for satellite-based wildlife surveys, our approach holds extreme promise for scaling spatially to produce the first ever total counts of migratory ungulates in open landscapes. Such information is particularly important to the management of aggregating species like wildebeest because their heterogeneous and autocorrelated grouping patterns violate the assumptions of most statistical methods for estimating population abundance from survey data 61 . As a result, traditional methods are prone to systematic undercounts and high uncertainty 61 . An automated total count would eliminate the need for statistical inference and potentially produce a correction factor that could be used to reduce error in historic estimates through post-hoc analysis. While a total count would still assume near-perfect detection of animals, we note that this ideal may be achieved in open systems where biological cycles drive predictable periods of aggregation. For example, wildebeest could be censused while gathered to calve on the nutritious shortgrass plains of Serengeti, caribou could be censused while gathering to cross seasonal ice floes in the arctic, and white-eared kob could be imaged while concentrated in low-lying meadows along the margins of major watercourses during the dry season. A next valuable step in the science of enumerating large mammal populations using the proposed satellite-based method will be ground-truthing the predictions against both historical and contemporary estimates of population size derived using traditional methods (e.g., ground-based or aerial counts). For the present case of the wildebeest population, satellite-derived counts should be compared against the data collected every 2–3 years using aircraft surveys in the Serengeti National Park 7 , 62 . Comparisons can be conducted both at the transect level (with satellite image acquisition synced to the timing of aircraft transects—although noting that temporal alignment of surveys with suitable conditions for both survey types can be challenging) and at the whole population level via data extrapolation. In addition to facilitating total counts for multiple species, the ability to observe expansive herds of migratory ungulates from space presents an exciting opportunity for the study of the ecology of animal aggregations from an entirely novel perspective. For example, the spatially explicit point data produced by our model can be readily analyzed as an ecological point process 63 to facilitate the first-ever quantitative descriptions of wildebeest herding patterns in the wild. Such insights are crucial for answering key ecological questions about social and environmental drivers of animal behavior and identifying emergent biological patterns that scale from individuals to populations 63 . Likewise, a robust time series of satellite images may be used to extend previous work on the ecology of large-scale aggregation patterns of wildebeest across the landscape 64 . We demonstrate the potential for our pipeline to inform this approach by producing density plots from model outputs, which can then be mapped and analyzed within their native environmental context (Fig. 8 ). This ability to track the distribution of large animal aggregations over time is important for guiding adaptive management of mobile species and for deriving a systematic understanding of population-level responses to rapid environmental change. Another potentially promising application of the proposed method would be the detection of large mammal migrations that have not previously been documented. Despite the charisma of such fauna, the migrations can go uncharacterized and are infrequently discovered or rediscovered (e.g., the Burchell’s zebra migration in Nambia/Botswana; 65 white-eared kob in South Sudan 66 ). Given the advantages of surveying at large scales, satellite imaging techniques, coupled with GPS tracking of individual animals, could provide a powerful methodological combination for detecting or confirming such migrations. GPS tracking data could benefit the survey by giving prior information about the potential range, while regularly acquired satellite imagery can be used to identify the migration routes of large animal groups over time, as satellite imaging at high time frequency becomes possible. Such methods are also especially useful for detecting and studying wildlife migrations in remote or insecure regions 66 . Despite the clear potential for satellite-based wildlife surveys to advance both basic and applied research, this technology is still limited by the inherent challenge of distinguishing small objects from only a few pixels on satellite imagery. While the commonly used deep-learning based object detectors for animal detection are confined by the size of the object on the image 37 , 67 , 68 , our method addresses this challenge by utilizing a class of convolutional neural networks (specifically the U-Net model) designed for pixel-level segmentation, thus enabling detection of objects that occupy less than 9 pixels. This method uses ensemble learning to further increase the accuracy of individual U-Net models. By combining the clustering module, the ensemble model can separate multiple clustered animals and identify individual animals with high accuracy and efficiency. This is an advancement compared to previous studies, which had lower detection accuracy for similarly sized animals (e.g., seal detection with <50% accuracy 44 ), or focused on identifying large animals in homogeneous environments (e.g., whales 37 ). Nevertheless, the current limitation of satellite image resolution impacted our study by preventing distinction between wildebeest and other species of similar size, including domestic cattle ( Bos taurus ), topi ( Damaliscus korrigum ), Coke’s hartebeest ( Alcelaphus buselaphus cokii ), and eland ( Taurotragus oryx ). While we controlled for the most numerous species (e.g., cattle) by limiting collections to sites and seasons with minimal overlap, finer-resolution imagery (for example, <10 cm) will be required to discriminate these species. We also note that smaller-bodied species (e.g., gazelle) were not visible at the current resolution, but larger species (e.g., hippos and elephants) were successfully excluded by the model. Given these promising results, we are confident that pending technology will rise to meet the demand to resolve smaller species, as multiple satellite companies have already announced the arrival of breakthrough technologies that will make sub-daily, sub-50 cm imaging a reality. One limitation in satellite imaging wildlife currently is the cost of very-fine-resolution imagery. However, costs are falling as more companies are now offering sub-meter imaging capabilities from multiple constellations at lower prices. In addition, many satellite providers (e.g., Maxar, Airbus and Planet) are providing more opportunities for researchers to access sub-meter imagery at low or zero cost. As more fine-resolution constellations come online, we anticipate that satellite-based wildlife surveys will become increasingly affordable and accessible. We aim to capitalize on this technological moment by validating a data pipeline, which advances the scale and scope of current techniques to include medium-sized mammals in highly heterogeneous landscapes. While there are many applications for this pipeline, we wanted to demonstrate its potential to monitor animals across an area of unprecedented size by counting hundreds of thousands of wildebeest in the Serengeti-Mara ecosystem. When combined with anticipated advances in satellite imaging, the outputs of our model will improve the frequency and accuracy of population estimates for multiple species in open landscapes and produce novel datasets for investigations of animal behavior, ecosystem ecology, and global change biology. Methods Satellite imagery The satellite imagery used for wildebeest detection and counting includes nine multispectral images captured by three satellite sensors (GeoEye-1, WorldView-2, and WorldView-3) over six years in the Serengeti-Mara ecosystem. We selected these images from the archived very-fine-resolution satellite images acquired by the Maxar Worldview constellation, which can cover more than 3.8 million square kilometers per day and has a revisit rate of 1-2 times per day. The images we used mainly cover the Masai Mara National Reserve and the northernmost section of the Serengeti National Park (see Fig. 2 of the study area). The images cover 2747 km 2 within the delimited boundary. The spatial resolution varies from 38 to 50 cm (see Supplementary Table 2 of image resolution and date). Most of the acquired images were delivered as pan-sharpened products, while the WorldView-2 images in 2020 were pan-sharpened using the UNB-pansharp method 69 . The pre-processed satellite images have four bands: Red, Green, Blue and Near-Infrared. All the images are covered by cloud by less than 2%. In addition, another set of eight satellite images covering the same area as the images above, but acquired on different dates are used as a set of reference images for wildebeest labeling. Details of the input satellite images and the reference images are listed in Supplementary Data 1 . Labeling the wildebeest In the satellite imagery, we labeled the individual wildebeest as points in vector format. On the true color composite image, a wildebeest is a group of gray-brownish pixels with a dark black pixel commonly in the center representing the animal’s neck and spine with a black mane. Each wildebeest individual in the image was about 3 to 4 pixels in length and 1 to 3 pixels in width, with 1 or 2 relatively darker pixels in the center as shown in Fig. 3 . Therefore, for each wildebeest, we labeled one point at the center of this wildebeest segment, and then expanded the point to a polygon with a size of 3 by 3 pixels, such that the polygon covers most of the wildebeest pixels. The wildebeest labels were derived using majority voting from visual interpretation undertaken by four expert observers of the same satellite image, cross-referenced against another (reference) satellite image acquired in a different year. The purpose of using reference images was to distinguish between wildebeest and spectrally similar background objects, such as small bushes and the shadows of termite mounds, which are static in both images. Training and test dataset For each satellite image, we built a grid system with a cell size ranging from 150 m to 170 m, dependent on image resolution. Each grid covered 336 × 336 pixels, which was the size of the image patch for model training. The training and test datasets were sampled based on the cell units of the grid. In the training dataset, we selected a total of 1097 training grids, covering different types of landscapes and various wildebeest abundances across all six years. The training dataset contains 53,906 wildebeest, occupying 27.13 km 2 , which is 0.7% of the whole area. The test datasets were sampled using the proportionate stratified random sampling method on each image date, containing 2700 sample grids with 11,594 wildebeest. We adopted this method to guarantee the representativeness of the test dataset. The strata of the test dataset were based on the wildebeest density in the grids in accordance to the spatially imbalanced distribution of wildebeest, ensuring the test dataset contains sample grids with different levels of animal density. Therefore, preliminary information on wildebeest density was required. We first built an initial test dataset using a random sampling method and trained a model to achieve an acceptable detection performance on the initial test dataset. Then we applied the preliminary model to the whole imagery dataset to detect and count the wildebeest, which were used to estimate the wildebeest density in all the grid cells. The grid-level wildebeest density was used as the criteria to classify the grid cells into one of four categories (low density, medium density, high density and very high density) based on the mean and standard deviations. Supplementary Fig. 9 shows an example of the wildebeest density map in the year 2009 for sampling. Majority of the grids have low density of animals. We determined the test sample size as 100 or 200 test grid cells depending on the area covered by each image, and then selected a proportionate number of samples randomly within each category to build the final test dataset. For example, as there was a single image collected on 10 August 2009, 100 test samples were selected from it. Since there are two images on 13 August 2013, 200 test samples were chosen from them. For images collected on 08 October 2020, the area was much larger and the wildebeest density was rather low. As a result, we selected 1900 image grid cells for testing. The sample size for the year 2020 was relatively large to ensure the test datasets covered sufficient wildebeest-abundant image patches. In total, there were 2700 test grids for all six years, occupying 1.7% of the entire dataset. We manually labeled all the wildebeest in the test sample grids. Training the U-Net based ensemble model for wildebeest detection Before incorporating the training dataset into the model, we first pre-processed the images and labeled wildebeest to fit the requirements of the input data. The wildebeest polygon labels were rasterized into a small patch with 3 × 3 pixels to represent the wildebeest segments. The segments were then used to generate the binary masks, including the wildebeest pixels and non-wildebeest pixels. The masks have the same size as the corresponding satellite sensor gridded images. The gridded images and the binary masks were cropped into patches with 336 × 336 pixels. Then all data patches were augmented using horizontal flip, vertical flip, and 90° rotation to increase sample variation. These data augmentation techniques can help prevent overfitting and increase the generalization capability of the model on unseen data with unfamiliar patterns 70 . All the training image patches and the masks from the six different years were combined to train the U-Net deep learning model. The U-Net architecture is a type of convolutional neural network designed originally for biomedical image segmentation 46 , which has subsequently been applied widely in other applications, including remote sensing image segmentation. U-Net uses a U-shaped symmetrical encoder-decoder structure that consists of a contracting path on the left and an expanding path on the right 46 (Fig. 1 ). The contracting path encodes high-level contextual features through successive layers, which generates low-resolution, but high-dimensional feature maps. The expanding path decodes the information of these feature maps and up-samples the image to obtain the original resolution step-by-step. The up-sampled output is concatenated through skip connections with the corresponding feature map (with the same spatial resolution) in the contracting path on the left, thus, merging both sources of information to provide evidence for classification, and to support precise localization of the obtained semantic information. The last layer of the model maps the feature maps into the class number for each pixel in the original image using a sigmoid activation function, resulting in a probability map with a value ranging from 0 to 1 representing the wildebeest presence probability as the final output of the U-Net model. We employed the ensemble learning approach 71 , 72 , 73 to increase the generalization capability and robustness of the U-Net model. We split the training dataset into K folds ( K = 10 in this research), of which K -1 folds were used for training the U-Net model, and the remaining one was used for validation. Therefore, a total of K individual U-Net models were trained and validated with different subsets of the data. Then the K models were combined to construct the final ensemble model, where the probability predictions of the base models were first normalized to the scale of 0 to 1 using the standard min-max approach and then averaged to produce the final outputs as depicted in Fig. 1 . To address the imbalance between the wildebeest and non-wildebeest classes, we adopted a weighted loss function, namely, the Tversky loss function 74 , to measure the discrepancy between the predictions and ground references. The parameters of the Tversky loss, α and β , are the penalty weights for False Negatives (FN) and False Positives (FP), respectively, and the sum of α and β is 1 (Supplementary Equation ( 1) ). Considering that wildebeest detection from satellite images is a highly imbalanced problem, namely, the percentage of wildebeest pixels is less than 1% in the training imagery, the model tends to predict all the pixels into non-wildebeest pixels to achieve high overall accuracy. By increasing β , emphasis is added to minimize the number of misclassified wildebeest pixels. The parameter β was finely tuned over a range of values (0.01, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.99) to reach the optimal trade-off between FPs and FNs. We used the dataset of 2009 in a sensitivity analysis to evaluate how different settings of β influence the model performance and the optimal parameters used were α = 0.1 and β = 0.9 (Supplementary Table 5 ). The model was trained with the Adam optimizer using an initial learning rate of 0. 0001 75 . The learning rate was reduced by a factor of 0.33 when the loss on the validation set stopped improving after 20 epochs. The weights in the convolution layers were initialized by the He_normal kernel initializer 36 . The dropout rate 76 was set to 0 as preliminary experiments showed that a higher dropout rate did not increase significantly the model performance. The batch size was 12, and the model was trained for 120 epochs. The model generating the smallest loss on the validation dataset amongst all epochs was selected as the final model. The software was implemented using TensorFlow 77 2.1.0, and Python 3.7. The model was trained on Azure Virtual Machine with NVIDIA Tesla V100 GPU supported by Microsoft AI for Earth. We post-processed the outputs of the ensemble model to obtain precise wildebeest point predictions. The outputs of the base U-Net models were probability maps of wildebeest presence. The probability map of each base model was first rescaled into the range of 0 to 1 (if the maximum value is greater than 0.05) and then averaged to obtain the final probability map as the output of the ensemble model. Each pixel on the final probability map was then classified as either wildebeest or non-wildebeest using a threshold of 0.5 (Supplementary Fig. 10 ). We converted the raster results of wildebeest segments into points that represent individual wildebeest using K -means clustering. As such, the centroids of the segments were extracted and individual wildebeest were separated (Supplementary Fig. 1 ). The number of clusters in each segment was determined automatically by the ceiling division result of the number of pixels within the segment by the general wildebeest object size (namely, 9 pixels). Model evaluation We evaluated the accuracy of the U-Net-based wildebeest detection model based on the alignment between the predicted wildebeest points and the ground reference points. A small local searching region was considered while matching the points to compensate for a slight shift, considering that the wildebeest segments were not always perfect 3 × 3 squares and the extracted centroids of the ground reference and predicted segment may not be perfectly aligned, but still represent the same animal. In this way, the extracted wildebeest centroids can still represent the correct detection of wildebeest even if they deviate by one pixel away from the ground reference points. The radius of the searching region was set to be 0.71 m, which is equivalent to the actual length of the diagonal line of one 0.5 m-resolution pixel. Predicted points that could be matched with one of the closest ground reference points within the searching region were counted as True Positive predictions. Predicted points that could not be matched with any ground reference points within the searching region were treated as False Positives, and all the remaining ground reference points that were not matched with any predicted points were treated as False Negatives. To assess the overall performance of the model quantitatively, we utilized the following accuracy metrics: precision, recall and F1-score. Precision measures the accuracy of predicting wildebeest amongst all positive detections. It is calculated as the ratio between the number of True Positives and all detected positives. Recall measures how well the model performs at finding the actual true positives from all the ground reference points. It is the ratio between the number of detected True Positives and all existing ground reference positives. F1-score is the harmonic mean of precision and recall, which reflects the overall accuracy. The accuracy of each year was evaluated separately on the test dataset of each year, and the total accuracy obtained on all the test datasets was assessed as well. We repeated the model training and evaluation five times to obtain the uncertainty of the model accuracy. In addition to the above, we adopted the precision-recall curve and area under the curve (AUC) to compare the performance of the sub-models with the U-Net-based ensemble model. By applying different thresholds to the probability map, we calculated multiple pairs of precision and recall. For the threshold of 0 or 1, we set the paired precision and recall rates as (0, 1) and (1, 0), respectively. These precision-recall pairs were then added to the plot, and AUC was calculated using the composite trapezoidal rule. The value of AUC is between 0 and 1. A larger AUC indicates better model performance. To test the spatial and temporal transferability of the model, we ran two tests: (1) transferring the model to a temporally different dataset: we set aside the dataset in 2015 as an independent test dataset and trained the wildebeest detection model using only the data of the other five years (2009, 2010, 2013, 2018, 2020). The 2015 dataset is therefore an entirely new dataset obtained by a unique sensor with a different spatial resolution from others (38 cm of WV03 versus 42–50 cm of GE01 and WV02); (2) transferring the model to a spatially different dataset: we set aside the dataset in 2020 as an independent test dataset and trained the wildebeest detection model using only the data of the other five years (2009, 2010, 2013, 2015, 2018). The coverage of 2020 data is on the east side of the Masai Mara National Reserve and Serengeti National Park, which is outside the coverage of the remaining datasets, and its spatial resolution is the coarsest (50 cm of WV02) among all the years. In each of the scenarios, the model was trained with datasets of five years and transferred to another new year with unseen features, such as new spectral characteristics of a different year, new image resolution and new landscapes. The model transferability in these two tests was evaluated directly using the test dataset of the independent year (2015 or 2020). Detecting and counting the wildebeest After the U-Net-based ensemble model demonstrated a high accuracy using the test dataset, we applied the model to all the satellite imagery to detect all the wildebeest across the study area inside the Serengeti-Mara ecosystem. The images were cropped into patches to match the input size of the model, and the ensemble model outputs were converted using K -means clustering to obtain wildebeest point predictions. The detected wildebeest were then mapped across the study area. We counted the number of wildebeest points on each satellite image to obtain the population estimates. We repeated model training five times and calculated the count five times to obtain the associated modeling uncertainties (at a 95% confidence level) for each date. To explore the spatial distribution patterns of the migrating wildebeest on different dates, we generated a point density map with a cell size of 100 m and a radius of 500 m (Fig. 8 ) for each date. The point density map visualizes the density of wildebeest points within the neighborhood of each pixel, showing the spatial and temporal variation in wildebeest distribution. We also calculated the wildebeest count per km 2 and summarized the frequency of the density as a histogram in Fig. 7 . Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability The minimum set of segmentation mask samples that can be used to demonstrate the U-Net-based wildebeest detection framework generated in this study was deposited in the Github repository ( ). Samples of satellite images for model training and testing are available on a restricted basis due to data protection laws and access may be obtained by contacting the corresponding author upon reasonable request. The very-fine-resolution commercial satellite image data for wildebeest detection are protected under a NextView Imagery End User License Agreement and are not available as a result of data protection laws. The copyright remains with Maxar Technologies (formally DigitalGlobe), and redistribution is not possible. The detected wildebeest point data are available at: . Other data generated in this study to support the findings are provided in the Supplementary Information and Source Data File. Source data are provided with this paper. Code availability The wildebeest detection framework based on U-Net is publicly available at Github repository 78 ( ); support and more information are available from Z.W. ([email protected]).

In their research, associate professor from the NRS Department Tiejun Wang and his master's student Zijing Wu developed an AI-model to automatically locate and count large herds of migratory ungulates (wildebeest and zebra). They used their method in the Serengeti-Mara ecosystem using fine-resolution (38–50 cm) satellite imagery. They achieved accurate detection of nearly 500,000 individuals across thousands of square kilometers and multiple habitat types. The scientists at the Department of Natural Resources (ITC Faculty—University of Twente) recently published their results in the journal Nature Communications. The Great Wildebeest Migration is the largest terrestrial mammal migration on our planet. It drives multiple ecological processes that support the health of humans and wildlife across the region. However, due to climate and land cover/use change, this natural process is becoming compromised. Developing accurate, cost-effective monitoring methods has quickly become a vital necessity to protect wildebeests and the ecosystem. To address this issue, Tiejun Wang demonstrates, for the first time, the capability of satellite remote sensing and machine learning techniques to automatically and accurately count very large populations of wildebeest and zebras. Even across the highly heterogeneous landscape of their migration journey. Approximately 120.000 individual wildebeests are accurately detected by the algorithm in this large area. Each purple dot is one wildebeest. Credit: University of Twente "We are currently in the first year of implementing the post-2020 global biodiversity framework, which the UN adopted at the 15th Conference of Parties to the UN Convention on Biological Diversity. The UN also recently established the Global Sustainability Development Goals as well as the first round of risk assessments on biodiversity and ecosystem services through the Intergovernmental Platform on Biodiversity and Ecosystem Services," explains Wang. "The unprecedented biodiversity loss as well as the gaps in knowledge on key aspects of biodiversity (e.g., species populations) clarify the need to integrate biodiversity measurements on Earth from the ground and from space. Hence, the effort to combine bottom-up and top-down approaches for biodiversity monitoring has never been so critical," said Wang. New satellite remote sensing and machine learning techniques make it possible to monitor global biodiversity with speed and precision. These efficiencies promise to reveal novel ecological insights at spatial scales which are germane to the management of populations and entire ecosystems. Wang's research demonstrates, for the first time, the capability of satellite remote sensing and machine learning techniques to automatically and accurately count very large populations of terrestrial mammals across a highly heterogeneous landscape. This study yielded highly accurate results and the largest training dataset ever published from a satellite-based wildlife survey (53,906 annotations). Beyond providing a truly open-source and transferable method for satellite-based wildlife surveys, the approach is spatially scalable for the first-ever total counts of migratory ungulates in open landscapes. Observing expansive herds of migratory ungulates from space presents an entirely novel perspective for the study of the ecology of animal aggregations.

10.1038/s41467-023-38901-y

Earth

The revolt of the plants: The arctic melts when plants stop breathing

So-Won Park et al, The intensification of Arctic warming as a result of CO2 physiological forcing, Nature Communications (2020). DOI: 10.1038/s41467-020-15924-3 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-020-15924-3

https://phys.org/news/2020-05-revolt-arctic.html

Abstract Stomatal closure is one of the main physiological responses to increasing CO 2 concentration, which leads to a reduction in plant water loss. This response has the potential to trigger changes in the climate system by regulating surface energy budgets—a phenomenon known as CO 2 physiological forcing. However, its remote impacts on the Arctic climate system are unclear. Here we show that vegetation at high latitudes enhances the Arctic amplification via remote and time-delayed physiological forcing processes. Surface warming occurs at mid-to-high latitudes due to the physiological acclimation-induced reduction in evaporative cooling and resultant increase in sensible heat flux. This excessive surface heat energy is transported to the Arctic ocean and contributes to the sea ice loss, thereby enhancing Arctic warming. The surface warming in the Arctic is further amplified by local feedbacks, and consequently the contribution of physiological effects to Arctic warming represents about 10% of radiative forcing effects. Introduction The increase in atmospheric CO 2 concentration has an influence on plant physiology. Physiological responses to increasing CO 2 include changes in leaf area index (LAI) and stomatal conductance, and those affect the plant transpiration in opposite ways. First, one of the main physiological responses is the CO 2 fertilization effect—that is, an increase in the rate of photosynthesis 1 , 2 , 3 . This effect accounts for the largest contribution to the positive trends in the LAI detected by satellite data sets 4 and can also lead to an increase in plant transpiration, resulting in cooling effects 5 . Second, another plant response is a reduction in stomatal conductance. In other words, stomatal apertures open less widely under elevated CO 2 concentrations. These CO 2 -induced reductions in the stomatal conductance have been confirmed through experiments and from reconstruction data 2 , 3 , 6 , 7 . The stomatal closure resulting from elevated CO 2 levels can decrease the rate of transpiration by diminishing the amount of water loss from plants. This reduction in plant transpiration can lead to an increase of near-surface air temperature by decreasing the evaporative cooling effect and simultaneously increasing the sensible heat flux above the land surface 8 , 9 , 10 . This nonradiative effect from physiological acclimation is known as CO 2 physiological forcing 11 . Previous studies using model experiments have investigated how the physiological forcing will affect the future climate in vegetation-covered regions, through influences such as amplified heat extremes 12 , intensified zonal asymmetry of rainfall over tropical land 13 , drying over the Eastern Amazon 14 and Sahel greening 15 . This physiological effect has a potential to remotely alter the entire climate system through the redistribution of the surface energy and disturbance of hydrological cycle, but still, the remote impacts of physiological effect on the climate system are unclear especially in the Arctic region (north of 70°N). The Arctic is the region most sensitive to greenhouse warming and has experienced warming faster than the global average, a phenomenon known as Arctic amplification 16 . Many mechanisms have been suggested to explain the Arctic amplification including a role of diminishing sea ice 17 , 18 , seasonal storage and release of the absorbed shortwave (SW) radiation coupling with sea-ice loss 19 , 20 , 21 , enhanced downward longwave (LW) radiation due to an increase in water vapor and cloud fraction 22 , 23 , ocean biogeochemical feedback 24 , 25 , increased poleward energy transport 26 , 27 and other processes 28 , 29 . However, their relative contributions are still under debate and also many alternative mechanisms are under investigation. Here, we suggest that the CO 2 physiological forcing has a remote forcing on the Arctic climate and can intensify the Arctic amplification through the enhanced atmospheric poleward heat transport and the physical processes coupling with the Arctic sea-ice change. To examine the impacts of physiological acclimation under elevated CO 2 on the future climate system, we analyzed eight Earth system models (ESMs), which can simulate interactions between the physical climate system and the biogeochemical processes, from the Coupled Model Intercomparison Project Phase 5 (CMIP5) 30 (Supplementary Tables 1 and 2 ). In line with previous studies 12 , 13 , 31 , 32 , 33 , 34 , we respectively quantified the physiological forcing (Phy), which includes the CO 2 fertilization effect and the dependency of stomatal conductance on CO 2 , and CO 2 radiative forcing (Rad) (average CO 2 concentrations ~823 ppm) using carbon–climate feedback experiments (see the Methods section and Supplementary Table 3 ). Results Land surface warming by plant physiological effects Figure 1 shows changes in the annual mean evapotranspiration (ET), Bowen ratio (the ratio of sensible to latent heat fluxes), and near-surface air temperature resulting from the CO 2 physiological forcing. In contrast with the radiative effect inducing the increase in ET due to enhanced water-demand from the temperature rise (Supplementary Fig. 1 ), physiological effects cause a conspicuous and significant reduction in the annual mean ET in densely vegetated areas of the tropics and mid-to-high latitudes (Fig. 1a ) in line with previous studies 12 , 31 , 32 , 34 , 35 , 36 , 37 . In this idealized experiment for evaluating the CO 2 physiological forcing, the fertilization effect plays a role in increasing ET due to the resulting increased LAI, but the effect of stomatal closure works in the opposite way at the same time 5 , 10 , 37 . Therefore, this overall drop in ET suggests that the stomatal closure have a greater influence in controlling the total ET than the CO 2 fertilization, when only the physiological effects is considered under elevated CO 2 levels, in consistency with the argument in previous studies 12 , 31 , 34 , 37 . Fig. 1: Change in evapotranspiration, Bowen ratio and temperature resulting from CO 2 physiological forcing. Multimodel mean change of the annual mean evapotranspiration ( a ), Bowen ratio (sensible heat flux/latent heat flux) ( b ), and near-surface air temperature ( c ) resulting from CO 2 physiological forcing. Only significant values at the 95% confidence level based on a bootstrap method are shown. Full size image The physiological effects change the surface energy budgets by reducing the evaporative cooling and simultaneously increasing the sensible heat flux (Fig. 1b and Supplementary Table 4 ). These heat flux changes induce surface and near-surface air warming around regions where ET is significantly decreased. Interestingly, significant surface warming occurs in the Arctic Ocean under the influence of CO 2 physiological forcing, despite the fact that vegetation obviously does not exist in the Arctic Ocean (Fig. 1c ). Another interesting point is that a synergy effect, a nonlinear interaction of physiological forcing with the radiative forcing 15 , 37 , additionally contributes to the surface warming (see the Methods section and Supplementary Table 5 ). The magnitude of synergy effect in Arctic region is equivalent to ~24% of annual mean temperature change resulting from physiological forcing. These results imply that the global warming signal by the radiation forcing plays a role in amplifying the physiological effect through their interactions. Meanwhile, the physiological forcing excluding a synergy effect still induces the statistically significant Arctic warming, which confirms the consistency and robustness of our findings (Supplementary Fig. 2 ). Seasonal changes caused by the physiological effects (Fig. 2 and Supplementary Fig. 3 ) were further examined. While the variations of ET remain almost constant throughout the year in tropical regions, the reduction in ET at mid-to-high latitudes (40°–70°N) exhibits a strong seasonality. As a result, the changes in the surface energy fluxes and temperature in mid-to-high latitudes also show a strong seasonality. In summer (June–July–August; JJA) when photosynthesis is most active, the maximum decline in ET and the resulting strongest continental warming occur. Unlike this continental warming (40°–70°N), however, the maximum warming in the Arctic regions due to physiological effects, an increase of +0.99 K, occurs in winter with a time lag (Supplementary Table 6 ). The mechanisms of this remotely induced Arctic warming are discussed in the next section. Fig. 2: Impacts of the physiological forcing on the evapotranspiration and temperature. Zonally and monthly averaged change in the evapotranspiration ( a ) and surface air temperature ( b ). The shading represents the change resulting from CO 2 physiological forcing. The contouring represents the change resulting from CO 2 radiative forcing. The contour intervals for radiative forcing are 0.1 mm day –1 in ( a ) and 1.5 K in ( b ). Full size image Arctic warming remotely induced by CO 2 physiological forcing As illustrated in Fig. 1 , changes in plant physiology lead to a statistically significant temperature rise in the Arctic Ocean. The continental warming in JJA resulting from the physiological responses seems to be propagated to the polar region with time (Fig. 2b , shading), whereas this pattern was not observed in the CO 2 radiative forcing experiment (Fig. 2b , contour). The Arctic warming resulting from the physiological effects is most distinctive during the boreal winter (December–January–February; DJF). In addition, the magnitude of this Arctic warming in DJF is comparable with that of continental warming during JJA. It is important to understand how this continental warming resulting from the physiological effects can remotely cause the distinctive delayed warming in the Arctic Ocean. A previous study demonstrated that mid- and high-latitude forcing can remotely contribute to the Arctic warming through various physical processes 38 . In particular, the increase in atmospheric northward energy transport (NHT ATM ) due to the continental warming can be responsible for the delayed Arctic warming. It is evident that the northward energy transport significantly increases during the warm season from April to July (see the Methods section and Supplementary Fig. 4a ). Furthermore, this NHT ATM continuously increases during the whole period of simulations due to the intensified CO 2 physiological forcing with increasing CO 2 levels (Supplementary Fig. 4b ). These results imply that the NHT ATM plays a role in connecting the extratropical continental warming to the Arctic warming under an influence of the physiological effect, which shows the similarity with previous studies, suggesting that the high-latitude greening and mid-latitude afforestation can enhance the Arctic amplification through an increase in poleward energy transport 39 , 40 , 41 , 42 . The increase in NHT ATM is associated with sea-ice melting and the resultant newly open waters in the Arctic allow it to absorb more sunlight during the warm season (Supplementary Table 6 ). Most of this energy is released to the atmosphere through the longwave radiative flux, and sensible and latent heat flux in the Arctic Ocean during autumn and winter, thereby inducing the Arctic warming (Supplementary Table 6 ). These mechanisms, the seasonal storage and release of the absorbed shortwave radiation coupling with an Arctic sea-ice change, have already been proposed in previous studies to explain the Arctic amplification 19 , 20 , 21 . Nonetheless, our results suggest that the plant physiological forcing as well as radiative forcing can contribute to the Arctic amplification under elevated CO 2 levels. A previous study has shown that the mid-troposphere in the Arctic sensitively responds to the energy advection across the Arctic boundary 26 . The bottom-heavy warming profile has been attributed to increased upward turbulent heat fluxes by the loss of sea ice in previous studies 17 , 19 . The vertical structure of atmospheric warming shows that the mid-tropospheric warming first occurred with a large vertical extent in the Arctic region during summer, and then this warming was propagated to the lowermost region of the atmosphere with time (Supplementary Fig. 5 ). These results support our hypothesis that the remote and lagged effects of plant physiological acclimation can intensify Arctic warming through an enhancement of NHT ATM and the resulting Arctic sea-ice change. However, there is a large inter-model diversity in the magnitude of Arctic warming, which seems to be closely related to the strength of local feedback related to Arctic sea ice, but further research is needed to confirm this (Supplementary Figs. 6 and 7 ). In summary, the surface warming resulting from physiological effect enhances an atmospheric energy convergence into the Arctic basin and this increases the net SW absorption during the warm season in the Arctic Ocean by melting sea ice. Subsequent energy release to the atmosphere increases the air temperature and ice-free waters in the Arctic, thereby intensifying the Arctic amplification during the cold season. As a result, the CO 2 physiological forcing accounts for 27.7% of the continental warming in summer and 9.7% of the annual surface warming in Arctic region resulting from CO 2 radiative forcing (Fig. 3 ). These results emphasize that the contribution of the plant physiological effects to the Arctic warming is quite significant. Fig. 3: Ratio between changes of temperature in response to CO 2 physiological forcing and radiative forcing. Ratio of the change in the near-surface air temperature resulting from CO 2 physiological effects to that resulting from CO 2 radiative forcing ([CO 2 physiological forcing/CO 2 radiative forcing] × 100) in the continents (40°–70°N) and Arctic region (70°–90°N), respectively. Each bar shows the area-weighted average of multimodel ensemble. The black dots represent the individual results from ESMs. The error bar for each column indicates the range of the 95% confidence level on the basis of a bootstrap method. Full size image Intensified and continued surface warming by local feedback Besides the direct heating from the enhanced sensible heat flux, an increase in net shortwave absorption (4.58 W m −2 in JJA) additionally heats the air above the land surface in JJA (Supplementary Table 4 ). In this experimental design, the net SW absorption can be largely affected by these two factors: An increase in LAI resulting from CO 2 fertilization effect can alter the surface albedo and increase the net SW absorption, thereby contributing to the temperature rise. The decrease in cloud fractions caused by physiological acclimation-driven reduction of relative humidity 35 , 43 , 44 can also be a cause of surface warming because it enhances downward SW radiative flux 42 . From their relative contributions, we found that vegetation-cloud feedback has a dominant role in the increased net SW absorption during summer (Supplementary Fig. 8 ), thereby contributing the continental warming (40°–70°N) (Fig. 4 and Supplementary Fig. 9 ) particularly in summer (Supplementary Table 4 ). Furthermore, the relative magnitude of the vegetation-cloud feedback in ESMs seems to explain the inter-model diversity of the land surface warming (40°–70°N) in JJA ( r = −0.79, P = 0.02) (Supplementary Fig. 7 ). Specifically, two models, HadGEM2-ES and MPI-ESM-LR, show the greatest warming in JJA due to this greatest cloud effect despite the moderate reduction of ET (Supplementary Figs. 10 – 12 ). Fig. 4: Local feedback triggered by CO 2 physiological forcing. Multimodel mean change of the annual mean snow concentration ( a ), sea-ice concentration ( b ), total cloud fraction ( c ) and surface albedo ( d ). Only significant values at the 95% confidence level based on a bootstrap method are shown. e , f Annual cycle of change in the surface air temperature (red bar) resulting from CO 2 physiological effect with the ratio between change in the snow concentration (blue line) in response to physiological forcing and radiative forcing in continents (40°–70°N) ( e ). The same as in ( e ), but in the Arctic region (70°–90°N), with the ratio between change in the sea-ice concentration (blue line) in response to physiological forcing and radiative forcing ( f ). The error bars represent a range of the 95% confidence level on the basis of a bootstrap method. All of the values are area-weighted averages of eight ESMs, except for the snow concentration, which is an average of five ESMs, because GFDL-ESM2M, HadGEM2-ES, and IPSL-CM5A-LR do not provide the surface snow area fraction data. Full size image In contrast to the change of cloud cover over the continents (40°–70°N), the cloud formation is enhanced in the Arctic region especially during winter (Fig. 4 and Supplementary Table 6 ). This increased cloud fraction additionally intensifies the surface warming by decreasing the outgoing longwave radiation especially in non-summer season 45 , 46 (Supplementary Table 6 ). Although it is difficult to prove the causality in this experiment, it is conceived that this increase in cloud formation contributes to the Arctic sea-ice loss, which in turn causes the increase in water vapor from the newly opened Arctic waters, as proposed previously 21 , 46 . In summary, the cloud feedback in the Arctic can enhance the surface warming by increasing a downward LW radiation, and in turn, the enhanced surface warming can accelerate the sea-ice loss, thereby causing positive feedback during the cold season. Another local feedback might be triggered by physiological forcing over the continents (40°–70°N). As shown in Fig. 4 , a snow concentration and a surface albedo in high latitudes significantly decline in response to the CO 2 physiological forcing. The warming resulting from the physiological effects presumably melts snow and the resultant less-snow-covered surface absorbs more solar radiation (Supplementary Table 4 ). Furthermore, an increase of LAI from the fertilization effects and the land cover change in models with interactive vegetation might partially contribute to the surface warming by altering the surface albedo independently of a change in temperature and would melt the snow as noticed previously 47 , 48 (Supplementary Figs. 13 and 14 ). Consequently, this snow–albedo feedback may help enhance and maintain the land surface warming throughout the year, especially in high latitudes where the surface albedo is relatively high due to the high snow cover (Fig. 4 ). On the whole, our results suggest that the local feedbacks triggered by physiological effects might additionally contribute to the amplified and maintained surface warming in both continents and Arctic Ocean. Discussion So far, it has been shown based on a multimodel mean that distinctive Arctic warming occurs due to the physiological effects, but this conclusion can be model-dependent because the structure of models and the parameterization schemes are different from each other. The magnitudes and spatial patterns of change in ET and temperature are diverse and HadGEM2-ES seems to greatly contribute to the multimodel ensemble mean temperature change (Supplementary Figs. 10 and 11 ). Nevertheless, most models consistently simulate the reduction in ET, the resulting surface warming over the continents and enhanced Arctic warming as a result of physiological effect (Supplementary Fig. 6 ), which suggests that the results are not sensitive to a subsampling of the models. In addition, multimodel ensemble results excluding HadGEM2-ES are not much different with those including HadGEM2-ES and still statistically significant though the magnitude is a bit altered (Supplementary Fig. 15 ). These again attest to the robustness of our findings and also suggest that ensemble mean is not controlled by an outlier. This study shows that the physiological effects amplify Arctic warming by 9.7% compared with that from the radiative forcing. This surface warming in the Arctic region resulting from the physiological response might have the potential ramifications of future changes in the carbon and hydrological cycles by intensifying the interaction between the Arctic climate and Arctic biological system 47 , 48 . Considering the physiological effects of CO 2 might be helpful for understanding the inter-model diversity in future climate change. A previous study has reported that the stomatal conductance schemes in the current ESMs do not consider various plant water use strategy 49 , which can lead to the underestimation of the surface warming across Northern Eurasia 50 . This result raises a possibility that Arctic warming may be greater than that in the current projections. Furthermore, there are still the limitations of land surface models in simulating LAI and the albedo dynamics and the stomatal conductance schemes in ESMs are rather static and semi-empirical (see the Supplementary Notes 1 and 2 ). These factors make it hard to simulate the realistic plant behavior to elevated CO 2 levels and also increase the uncertainty in the quantification of climate change caused by the physiological forcing. These point to the need for improvement of land models’ schemes based on a fundamental understanding of the involved processes. Methods CMIP5 data and experimental design Eight ESMs (bcc-csm1-1, CanESM2, CESM1-BGC, GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MPI-ESM- LR, and NorESM1-ME) were used, which were coupled with the full carbon cycle and used in idealized experiments designed to assess carbon–climate feedback, from the CMIP5 archive 30 (see Supplementary Tables 1 and 2 ). These experiments were run for 140 years with 1% per year increase in atmospheric CO 2 concentration from preindustrial levels to quadrupling (285−1140 ppm) for both radiation and biogeochemistry (1pctCO2), radiation only (esmFdbk1) and biogeochemistry only (esmFixClim1) (see Supplementary Table 3 ). In GFDL-ESM2M, the atmospheric CO 2 levels were prescribed to increase from their initial mixing ratio level of 286.15 ppmv at a rate of 1% per year until year 70 (the point of doubling, 2 × CO 2 ) and thereafter CO 2 concentrations were kept at a constant for the remainder of the run. To quantify the CO 2 physiological forcing (Phy) (average CO 2 concentrations ~823 ppm), we calculated the difference between the final 70 years of two simulations data: Full CO 2 simulation (1pctCO2) that includes the fully interactive radiative, physiological, and fertilization effects in response to increasing CO 2 and radiation simulation (esmFdbk1) that includes only radiative effects in response to increasing CO 2 . Following the previous study 12 , we extracted the physiological forcing (Phy) from full CO 2 simulation rather than using the physiology simulation directly to evaluate CO 2 physiological forcing (Phy) relative to future CO 2 radiative forcing (Rad). Since a nonlinear interaction between CO 2 radiative forcing and physiological forcing exists in full CO 2 simulation, the physiological forcing (Phy), defined as 1pctCO2-esmFdbk1, includes this nonlinear interaction, or synergy effect, as well as the pure physiological forcing. We additionally assessed the physiological forcing in a different way by calculating the difference between the average of the final 70 years of physiology simulation (esmFixClim1) and averaged values of preindustrial control simulation (piControl) over the whole period to verify the robustness of our finding. Unlike the previous method, this alternative physiological forcing does not include an interaction between physiology and radiation. We evaluated a synergy effect by calculating the difference between the two physiological forcing with the different definition. We quantified the CO 2 radiative forcing (Rad) (average CO 2 concentrations ~823 ppm) by calculating the difference between the average of the final 70 years of radiation simulation (esmFdbk1) and averaged values of preindustrial control simulation (piControl) over the whole period. The multimodel ensemble was derived by re-gridding the outputs from ESMs to a common 1° × 1° grid, then averaging together. The bootstrap method was used to test the statistical significance of the difference between the simulations. For MME, eight values were randomly selected from eight ESMs with replacements, and then their average was computed. By repeating this process 1000 times, the confidence intervals were determined, and only significant values were shown to show the model agreement. For each individual model, we randomly selected 70 years with replacements from year 71 to 140, calculated their average and finally computed the confidence intervals by repeating this process 1000 times. Atmospheric northward energy transport calculation The atmospheric energy convergence into the Arctic basin for transient conditions was estimated using energy budgets and residual methods 51 , 52 . Following the framework in the previous studies 53 , 54 , the energy budget of an atmospheric column can be denoted as: $$\frac{{\partial E_{{\mathrm{ATM}}}}}{{\partial {{t}}}} = {\mathrm{{NHT}}}_{{\mathrm{ATM}}} + F_{{\mathrm{TOA}}} + F_{{\mathrm{SFC}}},$$ (1) where \({\textstyle{{\partial E_{{\mathrm{ATM}}}} \over {\partial {{t}}}}}\) is the time change of atmospheric energy storage (W m –2 ), \(F_{{\mathrm{TOA}}}\) is the sum of the net radiation budget at the top of atmosphere (W m –2 ), \(F_{{\mathrm{SFC}}}\) is the net surface energy flux (W m –2 ) and \({\mathrm{{NHT}}}_{{\mathrm{ATM}}}\) is the vertically integrated northward heat transport (W m –2 ). All terms are defined as positive when they increase the atmospheric energy, hence positive downward for the TOA net radiation, positive upward for the net surface energy flux and positive for northward heat transport. Based on Eq. ( 1 ), the transient vertically integrated atmospheric northward heat transport can be expressed as: $${\mathrm{{NHT}}}_{{\mathrm{ATM}}} = \frac{{\partial E_{{\mathrm{ATM}}}}}{{\partial {{t}}}} - F_{{\mathrm{TOA}}} - F_{{\mathrm{SFC}}}.$$ (2) The atmospheric energy storage is written as: $$E_{{\mathrm{ATM}}} = \frac{1}{g} \int \nolimits ^{p_{\mathrm{s}}}_0 {\left( {c_{\mathrm{p}}T + k + L_{\mathrm{v}}q + \Phi _{\mathrm{s}}} \right)} \,{\mathrm{d}}p,$$ (3) where \(p\) is pressure (Pa), \(p_{\mathrm{s}}\) is the reference surface pressure (hPa), \(g\) is gravitational acceleration (m s –2 ), \(c_{\mathrm{p}}\) is the specific heat of the atmosphere at constant pressure (J K –1 kg –1 ), \(T\) is temperature (K), \(k\) is the kinetic energy (J kg –1 ), \(L_{\mathrm{v}}\) is the latent heat of evaporation (J kg –1 ), \(q\) is the specific humidity (kg kg –1 ), and \(\Phi _{\mathrm{s}}\) is the surface geopotential which is not a function of pressure 54 . The contribution of kinetic energy, \(k\) , is ignored here due to its comparatively small magnitude 53 . The net radiation at the TOA, \(F_{{\mathrm{TOA}}}\) , is defined as: $$F_{{\mathrm{TOA}}} = F_{{\mathrm{SW}}} - F_{{\mathrm{LW}}},$$ (4) where \(F_{{\mathrm{SW}}}\) is the net shortwave (solar) and \(F_{{\mathrm{LW}}}\) is the longwave (thermal) radiation, both in W m –2 . The net surface energy budget at the surface, \(F_{{\mathrm{SFC}}}\) , is defined as: $$F_{{\mathrm{SFC}}} = {\mathrm{{SW}}}_{{\mathrm{SFC}}} + {\mathrm{{LW}}}_{{\mathrm{SFC}}} + Q_{\mathrm{H}} + Q_{\mathrm{E}},$$ (5) where \({\mathrm{{SW}}}_{{\mathrm{SFC}}}\) and \({\mathrm{{LW}}}_{{\mathrm{SFC}}}\) are the net surface shortwave and longwave surface radiative fluxes, and \(Q_{\mathrm{H}}\) and \(Q_{\mathrm{E}}\) are the net surface sensible and latent heat fluxes, all in W m –2 . Data availability All CMIP5 data 30 that support the findings of this study are publicly available on Earth System Grid Federation website: . Code availability Processed data, products, and code produced in this study are available from the corresponding authors upon reasonable request.

The vapor that plants emit when they breathe serves to lower land surface temperature, much like watering the yard on a hot day. Until now, the greenhouse effect has been blamed for the rise in global temperature. But an interesting study has shown that the Artic temperature rises when the moisture released by plants is reduced due to the increase of carbon dioxide (CO2 ) in the atmosphere. The joint research team led by Professor Jong-Seong Kug and doctoral candidate So Won Park of POSTECH's Division of Environmental Science and Engineering, and Researcher Jin-Soo Kim of the University of Zurich has confirmed that the increase in atmospheric CO2 concentration closes the pores (stomata) of plants in high-latitude areas and reduces their transpiration, which ultimately accelerates Arctic warming. The findings, which were studied through the Earth system models (ESM) simulations, were recently published in Nature Communications, an authoritative journal in science. Plants take in CO2 and emit oxygen through photosynthesis. During this process, the stomata of leaves open to absorb CO2 in the air and release moisture at the same time. However, when the CO2 concentration rises, plants can absorb enough CO2 without opening their stomata widely. If the stomata open narrowly, the amount of water vapor released also decreases. When this transpiration of plants declines, the land temperature rapidly rises under greenhouse warming. Recently, such a decrease in transpiration has been cited as one of the reasons for the surge in heat waves in the northern hemisphere. This response from the vegetation leads to the global climate change by controlling the exchange of energy between the surface and atmosphere, referred to as 'physiological forcing.' But so far, no study has confirmed the effects of physiological forcing on the Arctic climate system. The joint research team analyzed the EMS simulation and confirmed that the increase in CO2 leads to stomatal closure in land vegetation causing land warming, which in turn remotely speeds up Artic warming through atmospheric circulation and positive feedback in Earth systems process. In addition, a quantitative estimate of the stomatal closure's effect on Arctic warming due to increased CO2 showed that about 10% of the greenhouse effect is caused by this physiological forcing. Professor Jong-Seong Kug, who has studied Arctic warming in a variety of perspectives, commented, "The stomatal closure effect due to the increased CO2 levels is not fully counted in the future climate projection." He pointed out, "This means that Arctic warming can proceed much faster than currently forecast." He also warned that "the increase in CO2 is accelerating global warming not only through the greenhouse effect that we all knew of, but also by changing the physiological function of plants."

10.1038/s41467-020-15924-3

Earth

Biochar from agricultural waste products can adsorb contaminants in wastewater

Marlene C. Ndoun et al, Adsorption of pharmaceuticals from aqueous solutions using biochar derived from cotton gin waste and guayule bagasse, Biochar (2020). DOI: 10.1007/s42773-020-00070-2

http://dx.doi.org/10.1007/s42773-020-00070-2

https://phys.org/news/2020-11-biochar-agricultural-products-adsorb-contaminants.html

Abstract Biochars produced from cotton gin waste (CG) and guayule bagasse (GB) were characterized and explored as potential adsorbents for the removal of pharmaceuticals (sulfapyridine-SPY, docusate-DCT and erythromycin-ETM) from aqueous solution. An increase in biochar pyrolysis temperature from 350 ο C to 700 ο C led to an increase in pH, specific surface area, and surface hydrophobicity. The electronegative surface of all tested biochars indicated that non-Coulombic mechanisms were involved in adsorption of the anionic or uncharged pharmaceuticals under experimental conditions. The adsorption capacities of Sulfapyridine (SPY), Docusate (DCT) and Erythromycin (ETM) on biochar were influenced by the contact time and solution pH, as well as biochar specific surface area and functional groups. Adsorption of these pharmaceutical compounds was dominated by a complex interplay of three mechanisms: hydrophobic partitioning, hydrogen bonding and π–π electron donor–acceptor (EDA) interactions. Despite weaker π–π EDA interactions, reduced hydrophobicity of SPY − and increased electrostatic repulsion between anionic SPY − and the electronegative CG biochar surface at higher pH, the adsorption of SPY unexpectedly increased from 40% to 70% with an increase in pH from 7 to 10. Under alkaline conditions, adsorption was dominated by the formation of strong negative charge-assisted H-bonding between the sulfonamide moiety of SPY and surface carboxylic groups. There seemed to be no appreciable and consistent differences in the extent of DCT and ETM adsorption as the pH changed. Results suggest the CG and GB biochars could act as effective adsorbents for the removal of pharmaceuticals from reclaimed water prior to irrigation. High surface area biochars with physico-chemical properties (e.g., presence of functional groups, high cation and anion exchange capacities) conducive to strong interactions with polar-nonpolar functionality of pharmaceuticals could be used to achieve significant contaminant removal from water. Graphic Abstract Working on a manuscript? Avoid the common mistakes FormalPara Highlights Biochar was obtained from the carbonization of cotton gin waste and guayule bagasse at three temperatures Biochar shows considerable potential for the adsorption of SPY, DCT and ETM and can be used as a cheap, environmentally friendly adsorbent Hydrophobic partitioning, electrostatic interactions, hydrogen bonding and π-π electron donor–acceptor (EDA) interactions can simultaneously influence adsorption and determine the sorptive affinity of biochars for pharmaceuticals 1 Introduction Anthropogenic activities, including increasing urbanization, population growth and intensive agricultural activities have introduced a broad range of emerging contaminants (ECs) into the environment (Houtman 2010 ; Pal et al. 2010 ; Fenet 2012). ECs are defined as chemicals and microorganisms that have been detected in the environment and can potentially cause toxic effects in aquatic and human life at trace-level concentrations (ng- μg/L levels), but for which no water quality regulations exist (Ahmed et al. 2017a ; USGS, 2015). There are several classes of ECs including pesticides, industrial additives, flame retardants, endocrine disrupting compounds (EDCs) and pharmaceuticals and personal care products (PPCPSs). Among all classes of ECs, PPCPs are the most abundant in the environment and are often detected at elevated concentrations in surface and groundwaters (Daughton 2004 ) due to continuous human use. Additionally, wastewater treatment plants (WWTPs) are not effective at removing many of these chemicals, leading to the presence of various PPCPs in wastewater effluents (Gros et al. 2010 ; Sui et al. 2011 ). The World Health Organization (WHO) estimated that in 2025, two-third of the world’s population could be living in regions with limited access to water. Agriculture is the sector which requires the most water and is expected to be impacted by the shortage. To mitigate these impacts, treated wastewater or reclaimed water has become an important source of water for agricultural irrigation especially in arid regions of the world. Reclaimed water is wastewater from homes, offices, hospitals and industries that has undergone treatment to remove impurities such as nutrients and pathogens (DSWS 2011). Reclaimed water can be successfully used for irrigation because even after treatment, it still contains nitrogen and phosphorus that are essential to plant growth (Kinney et al. 2006 ). When treated wastewater is used for irrigation, contaminants such as pharmaceuticals may be introduced into crops from the soil through root uptake and translocation, leading to accumulation of these contaminants not only in the roots, but also in edible above-ground plant parts (Bartha et al. 2010 ). Erythromycin, a commonly used antibiotic, was found to accumulate over 5 months in soil irrigated with reclaimed water (Kinney et al. 2006 ). A study by Gibson et al. ( 2010 ) demonstrated that wastewater reused for irrigation contains pharmaceuticals such as ibuprofen (0.7–1.4 μg L −1 ), naproxen (7.2–13.5 μg L −1 ) and diclofenac (2.0–4.8 μg L −1 ). In addition, runoff from irrigated farmlands has been shown to impact surface water bodies and leach into groundwater. A study by Kolpin et al. ( 2002 ) showed that organic contaminants, including PPCPs, were detected in 80% of a network of 139 streams located downstream of urban networks and livestock production. Another study in New York on streams receiving wastewater discharge showed the presence of sulfamethoxazole and clindamycin at concentrations varying between 0.043 and 0.076 μg L −1 (Batt et al. 2006 ). Continuous accumulation and persistence of these chemicals in the environment can lead to ecotoxicological risks such as interference with endocrine systems of higher organisms, intersex characteristics in organisms such as fish, and microbiological resistance among bacterial populations (Belhaj et al. 2015 ). Continuing concerns regarding possible population-level impacts of pharmaceuticals from wastewater effluents has contributed to a search for sustainable and cheap technologies which will result in the effective removal of pharmaceuticals from reclaimed water. Recently, biochar has been explored as a potential material for the adsorption of pharmaceuticals from aqueous solutions (Ahmed et al. 2017 a; Mondal et al. 2016 ; Mostafapour et al. 2019 ; Sumalinog et al. 2018 ; Taheran et al. 2016 ; Yanyan et al. 2018 ). Biochar is a stable carbon (C)-rich, energy dense by-product synthesized through the pyrolysis of waste biomass in the absence of oxygen (Lehmann and Joseph 2009 ). Biochar has been employed as an adsorbent and it has the ability to compete with activated carbons (ACs) for the removal of contaminants from water due to its microporous structure, high C content, and specific surface area. Research on biochar as a potential filter media for urban stormwater runoff demonstrated that biochar filters effectively remove total suspended solids, heavy metals, nutrients, polycyclic aromatic hydrocarbons, and E. coli (Mohanty and Boehm 2014 ; Reddy et al. 2014 ). Bolster ( 2019 ) demonstrated that adding biochar to sand columns enhanced the removal of bacterial isolates E. coli and Salmonella, with sorption being the primary mechanisms for bacteria removal. Studies have also reported the potential use of biochar as an adsorbent for the treatment of agricultural wastewater effluents. Particular attention has been given to the removal of pesticides from water using biochar, with reported sorption coefficients as high as 1158 mg kg −1 for atrazine and 1066 mg kg −1 for simazine (Zheng et al. 2010 ). The cotton gin and guayule industries are viable sources of biomass for biochar. The production of textiles from cotton gin accounts for approximately 2.5 million metric tons of cotton gin waste being generated every year (Maglinao et al. 2015 ). A minimum 200 metric ton per day of guayule bagasse is discarded from the production of latex and biofuels from guayule (Sabaini et al. 2018 ). The enormous quantities of waste generated present several economic and environmental problems such as the cost associated with landfilling (e.g., tipping fees). Thus, the transformation of cotton gin waste and guayule bagasse into value-added products such as biochar for the treatment of wastewater used for irrigation warrants investigation. The overall aim of this research was to investigate the use of biochars derived from cotton gin waste and guayule bagasse as filter media for removal of pharmaceuticals known to persist in wastewater effluent used for irrigation. Herein we investigate the effect of biochar pyrolysis temperature on important adsorption-related properties (surface area, porosity, surface charge and functionality, pH). The biochars were then evaluated for their abilities to remove selected pharmaceuticals from aqueous solution using batch adsorption experiments. Mechanisms and kinetics governing the interaction between pharmaceuticals and biochars were elucidated. Based on our results, the utility of these biochars in sand filters for removal of pharmaceuticals from reclaimed water prior to irrigation of food crops is addressed. 2 Materials and methods 2.1 Target adsorbates Three pharmaceuticals (sulfapyridine, docusate and erythromycin) were purchased from Sigma–Aldrich (St. Louis, MO, USA). These pharmaceuticals were selected due to their frequency of occurrence in environmental systems as a result of widespread usage. Additionally, their pK a values and octanol water-partition coefficients (log K ow ) cover broad ranges. The physico-chemical properties of the pharmaceuticals are provided in Table 1 . Table 1 Physico-chemical properties and structures of pharmaceuticals Full size table 2.2 Adsorbents Biochars obtained from the pyrolysis of guayule ( Parthenium argentatum ) bagasse (GB) and cotton ( Gossypium L. ) gin (CG) waste were studied to compare their adsorption capacities for pharmaceuticals in batch adsorption experiments. The biochar samples were prepared according to Novak et al. ( 2012 ). All feedstocks were processed before pyrolysis through air-drying, grinding, and sieving to pass a 6 mm sieve. Between 0.5 and 1.5 kg of ground biomass were placed in a stainless-steel tray and pyrolyzed slowly at low heating rates (0.05–0.1 ο C) using a gas tight retort (Lindberg/MPH, Riverside, MI) at three different temperatures 350, 500 and 700 ο C for 2 h under a stream of N 2 gas. The resulting biochar samples are herein referred to as GB350, GB500, GB700, CG350, CG500 and CG700. The biochar samples were ground to pass a 0.5 mm sieve and stored in a desiccator to minimize water absorption. 2.3 Characterization of biochar The pH of the biochar samples was measured following a modified procedure by Angin ( 2013 ). Biochars were added to deionized (DI) water in a mass ratio of 1:20 (0.5 g of biochar + 10 mL of water). The mixture was shaken for 1 h using a mechanical shaker and the pH was measured. The Brunauer-Emmett-Teller (BET) surface areas (SAs) and pore volumes of the biochar were determined using the ASAP 2020 plus surface area and porosimetry system (Micrometrics, Norcross, GA) and the accompanying t-plot analysis software. The surface functional groups of the biochars were identified according to a modified procedure described by Kloss et al. (2012), using a Fourier transform infrared (FT-IR; Bruker IFS 66/S and Bruker Vertex V70) spectrometer equipped with a liquid nitrogen cooled mercury cadmium telluride (MCT) detector. Experiments were carried out in diffuse reflectance mode on a Praying Mantis diffuse reflectance accessory. Sample powder was placed in a 3 mm diameter 316 stainless steel sample cup assembly and a total of 500 scans were averaged per spectrum at a resolution of 4 cm −1 . The zeta potential values of the samples were measured using Malvern Zetasizer ZS and according to a modified procedure by Johnson et al. ( 1996 ). The biochar samples were ground and sieved to pass a 34 µm sieve. About 0.015 g of each biochar sample was added to 30 mL of DI water. Solution pH was adjusted using 0.05 M HCl or NaOH. Zeta potential was measured three times at each pH (150 scans each time), with the average values reported. 2.4 Batch adsorption experiments Batch adsorption experiments were conducted to determine the rate at which adsorption might reach equilibrium for the different biochars and pharmaceuticals. Stock solutions (200 mg L −1 ) of SPY, DCT and ETM were diluted with DI water to produce initial concentrations of 10 mg L −1 for each pharmaceutical compound. While concentrations are relatively high compared to typical wastewater effluent concentrations, the initial concentration in this study was chosen for batch adsorption experiments to ensure that concentrations following adsorption were above the limit of detection. Single batch adsorption experiments were conducted to determine the adsorption kinetics for pharmaceuticals using 125 mL polytetrafluoroethylene (PTFE)-lined® bottles containing 0.5 g of each biochar and 100 mL of solution containing 10 mg L −1 of each pharmaceutical. The mixtures were agitated at 200 rpm using a mechanical shaker at ambient laboratory conditions ( ≈ 23 ο C) and 10 mL aliquots (pharmaceutical solution + biochar) were collected after 5, 15, 30, 60, 120, 180, 240 min and 24 h contact times. Adsorption experiments were conducted in triplicates and PTFE-lined® bottles were covered with aluminum foil to minimize photodegradation. Control treatments were included to investigate potential contamination or non-adsorption losses. Collected samples were filtered through a 1 µm Whatman membrane filter and filtrates were analyzed by high-pressure liquid chromatography and mass spectrometry (HPLC–MS). The pharmaceutical removal efficiency and amount adsorbed (q t ; mg g −1 ) were calculated using Eqs. ( 1 ) and ( 2 ), respectively: $$\mathrm{Removal \, efficiency} \,\left(\%\right)= \frac{{C}_{0}-{C}_{t}}{{C}_{0}}\times 100$$ (1) $${q}_{t}=\frac{{C}_{0}-{C}_{t}}{m}\times V$$ (2) where C 0 is the initial concentration of pharmaceuticals in solution (mg L −1 ), C t is the concentration (mg L −1 ) at time t (5, 15, 30, 60, 120, 180, 240 min and 24 h), V is the volume of the solution (L) and m is the mass of the biochar (g). 2.5 Effect of solution pH The effect of solution pH on the adsorption of SPY, DCT and ETM was studied using CG700 biochar as the adsorbent. Batch adsorption experiments were conducted at the inherent solution pH ranging from 10–11 and at pH 7 to mimic the typically near-neutral conditions of wastewater effluents. The initial pH of the pharmaceutical solutions was adjusted to pH 7 by adding 0.1 M H 2 SO 4 . Mixtures of 0.5 g of CG700 biochar in 100 mL of 10 mg L −1 stock solution of SPY, DCT and ETM were agitated at 200 rpm on a mechanical shaker and then filtered at selected times between 5 min and 24 h. 2.6 Kinetic studies The kinetics of adsorption were analyzed using a pseudo-second-order (PSO) kinetic model, which is widely used for pollutant adsorption from aqueous solution (Ho 2006 ). Equation ( 3 ) shows the linearized form of the PSO kinetic rate equation: $$\frac{t}{{q}_{t}}=\frac{1}{{k}_{2}{q}_{e}^{2}}+t/{q}_{e}$$ (3) where q e and q t (mg g −1 ) are the amounts of pharmaceuticals adsorbed per unit of mass of biochar at equilibrium and at time t (min), respectively, and k 2 (g mg −1 min) is the rate constant of the PSO adsorption. From the PSO plot of t/q t versus t , the values of q e and k 2 were obtained from the slope and intercept, respectively. Model conformity was evaluated using the linear regression coefficients (R 2 ). 2.7 Adsorption isotherms Adsorption isotherms of SPY, DCT and ETM on CG700 biochar were performed at room temperature. CG700 biochar (0.5 g) was added to SPY, DCT and ETM solutions (100 mL) of varying initial concentrations (2, 10, 20, 40 and 50 mg L −1 ). Solutions were agitated for 24 h at 200 rpm to reach equilibrium and then filtered. The equilibrium data obtained from the study were fitted to the Langmuir and Freundlich isotherms. The linear forms of the Langmuir and Freundlich models (Goswami et al. 2011 ) are represented by Eqs. ( 4 ) and ( 5 ): $$\frac{1}{{q}_{e}}=\frac{1}{{q}_{m}}+\frac{1}{{K}_{L}{q}_{m}{C}_{e}}$$ (4) $$\mathrm{log}{q}_{e}=\mathrm{log}{K}_{f}+\frac{1}{n}\mathrm{log}{C}_{e}$$ (5) where C e (mg L −1 ) is the equilibrium pharmaceutical concentration, q e and q m are the equilibrium and maximum adsorption capacity, respectively (mg g −1 ), K L is the Langmuir adsorption equilibrium constant in L mg −1 , K f (mg g −1 ) is the Freundlich constant related to adsorption capacity and n is a measure of adsorption intensity. 3 Results and discussion 3.1 Biochar characterization 3.1.1 Specific surface area (SA) and pH All biochar samples were alkaline and biochar pH increased with increasing pyrolysis temperature (Table 2 ). The basic nature of the biochar is due to the transformation of C into ash during pyrolysis and alkali salts begin to separate from the organic matrix, increasing the pH (Cao and Harris 2010 ). Porous structure (BET surface area and pore volume) of the studied biochar samples are summarized in Table 2 . Table 2 BET surface area, pore size and pH of the biochars Full size table The SAs of the cotton gin waste biochars were found to be low, but increased as the pyrolysis temperature increased to 700 °C. Numerous studies have documented an increasing surface area of biochars with increasing pyrolysis temperatures (Ding et al. 2014 ; de Caprariis et al. 2017 ; Goswami et al. 2016; Kloss et al. 2012 ; Uchimiya et al. 2011 ). As the pyrolysis temperature reached 700 °C, the BET SAs and number of micropores for the biochars significantly increased resulting from the removal of volatile matter that was either inside or blocking the micropores (Guedidi et al. 2017). Biochars obtained from guayule bagasse did not exhibit adequate SA and porous structure characteristics. This may be attributed to the softening, melting, fusing and carbonization which likely resulted in the pores of the biochar being partially blocked. This would prevent the access of the absorption gas to the pores and, therefore, lead to lower surface areas and pore volumes (Fu et al. 2011 ). The SAs of the biochars were generally lower than values for biochars produced from various feedstocks used in other studies, although several biochars are reported to have values less than 10 m 2 g −1 . Uchimiya et al. ( 2011 ) reported SAs of biochar obtained from the pyrolysis of cottonseed hull at 350 °C and 500 °C to be 4.7 and 0.0 m 2 g −1 , respectively. Such low SAs do not preclude substantial adsorption of organic pollutants. Cao and Harris ( 2010 ) reported nearly 77% removal of atrazine (1.5 mg g −1 adsorption capacity) by dairy manure-derived biochar pyrolyzed at 200 °C with a SA of 2.7 m 2 g −1 . 3.1.2 Zeta potential Figure 1 shows the pH-dependent zeta potential of the biochars. The measured zeta potential for CG and GB was negative for all pH conditions tested. Increasing pH resulted in increasing negative zeta potential for all of the biochars. A similar pH-dependent trend has been observed for pine wood biochars (Essandoh et al. 2015 ; Taheran et al. 2016 ). Given the nature of the pH dependence (Fig. 1 ), it was not possible to identify a pH point of zero charge (pHzpc) for the biochars. However, the relevance to the current research is that CG and GB biochar samples carry a net negative charge under all pH conditions used. Fig. 1 Zeta potential-pH curves of CG and GB biochar samples Full size image 3.1.3 FT-IR The FT-IR spectra of the biochar samples were used to characterize the surface functional groups. As shown in Fig. 2 a,b, all spectra exhibit the OH, C–H, C=O and C=C, bond stretching at 3400, 2850, 1750, 1600 cm −1 , respectively. The peaks at about 3400 and 3550 cm −1 correspond to vibrations of OH groups and were still present in the biochar prepared at 700 °C, but were dramatically straightened at 350 and 500 °C. The peak at 1770 cm −1 is due to the C=O stretching vibrations of the carbonyls (aldehyde, ketones esters, carboxylic acids) both unconjugated and conjugated with aromatic rings (Uchimiya et al. 2011 ). Fig. 2 (a) FT-IR spectra of CG350, CG500 and CG700. (b) FT-IR spectra of GB350, GB500 and GB700 Full size image Carboxyl groups contribute to negative surface charge at circumneutral pH (Sect. 3.1.2 ) which promotes electrostatic adsorption of cations in aqueous solution. The absorbance peaks between 1400 and 1500 cm −1 represent C=C stretching vibrations indicative of alkanes and aromatics (Inyang et al. 2010 ). The C–O stretching (1350 cm −1 ) occurred due to the presence of primary, secondary and tertiary alcohols, phenols, ethers and esters. The peak at wavenumber near 870 cm −1 reflects the C–H bending vibration in β-glycosidic linkage (Krishnan and Haridas 2008 ). This also indicates the presence of adjacent aromatic hydrogen on the biochar surface. The absorbance peak at 2850–2960 cm −1 indicates the presence of an alkyl C–H and the intensity of this group decreased as temperatures increased from 350 °C to 500 °C and then to 700 °C (Fig. 2 a, b). The 2850–2960 cm −1 peak area is strongly correlated to the hydrophobicity of the biochars (Kinney et al. 2012 ). From these results, it can be suggested that the increase of this functional group results from conversion of functional groups in a low oxidation state to those in a high oxidation state by means of heat treatment. The decrease of these functional groups at 700 °C was attributed to the release of these groups or conversion to other functionalities. 3.2 Adsorption kinetics and mechanism Adsorption is a process governed by several mechanisms operating simultaneously and it is often difficult to precisely identify the role of each mechanism in a specific system. For interpreting adsorption behavior, it is convenient to consider the overall free energy for the adsorption reaction, Δ G ads , as a combination of terms representing various adsorption mechanisms: $${\triangle G}_{{\rm ads}}= {\triangle G}_{{\rm elect}}+{\triangle G}_{{\rm hydro}}+{{\triangle G}}_{{{\rm H}}{-}{{\rm bond}}}+{{\triangle G}}_{\pi-\pi{\rm EDA} }+\dots$$ (6) where Δ G elect is the electrostatic adsorption term, Δ G hydro accounts for removal from solution due to hydrophobic interaction, Δ G H-bond accounts for adsorption due to hydrogen bonding and Δ G π–π EDA accounts for electron-donor–acceptor interactions. The adsorption behaviors of DCT, SPY and ETM using CG and GB biochars are interpreted based on the interplay of these different mechanisms. 3.2.1 Sulfapyridine (SPY) adsorption The time-dependent removal of SPY from solution using the six biochars is shown in Fig. 3 a, b. The data indicates that the highest removal of SPY by biochar was observed with the CG700 (70% removal), followed by CG350 (50% removal) and CG500 (15% removal) (Fig. 3 a). Removal was correlated to the surface areas of the adsorbents that follow the same order CG700 > CG350 > CG500 (Table 2 ). The pyrolysis at 700 °C increased the surface area eightfold compared to the surface area at 350 °C and this was accompanied by an increase of approximately 20% in the extent of adsorption. Fig. 3 Removal of Sulfapyridine by ( a ) CG700, CG500 and CG350 and ( b ) GB700, GB500 and GB350 (Red errors bars < 10%) Full size image Binding of the SPY to the biochars is initially explained by the surface charge of the biochars and the properties of SPY − ; log K ow (0.35) and acidity constants (pK a1 = 2.30; pK a2 = 8.43). The ionic character of SPY varies greatly with pH, as reflected by the acidity constants. At the inherent solution pH which varied between 8.3–11.2 over a 24 h contact time using CG700 and CG350, SPY exists mainly in its neutral and anionic forms (pH > pK a1 and pK a2 ; SPY − ) and hydrophobic interactions caused by solvation of apolar molecular surfaces between SPY and the negatively charged biochar surfaces (pH > pHzpc) resulted in SPY adsorption (Yao et al. 2017 ). However, the amount of SPY adsorbed by CG700 and CG350 (Fig. 3 a) is greater than what would be expected due to hydrophobic interactions alone, because SPY has a log K ow value of 0.35. This suggests that other mechanisms are responsible for the removal of SPY from solution. An additional binding mechanism for SPY removal is the formation of negative charge-assisted H-bonds (CAHB) between the anionic SPY and the O-containing functional groups of the biochars. CAHB occurred in response to the elevation in pH as the contact time increased to 24 h. The increase in pH is attributed to the release of OH − during the proton exchange between SPY − and water molecules, which is followed by interaction of the SPY molecules with the O-functional groups present on the biochar surface leading to SPY adsorption (SPY − + biochar → SPY o ≡ biochar; Teixidó et al. 2011 ). Furthermore, the sorption of SPY by CG700, CG500 and CG350 can be explained by π–π electron-donor acceptor (EDA) interactions. SPY can act as a π-electron acceptor due to the presence of the amino functional group and N and/or O -hetero-aromatic rings (Ahmed et al. 2017 a, b ). CG700, CG500 and CG350 biochars enriched with C=C, OH, C=O groups act as strong electron donors. Both of these factors may have facilitated π–π EDA interactions between CG700, CG500 and CG350 and SPY resulting in the removal from SPY from solution. Yao et al. ( 2017 ) recently proposed that the adsorption of SPY and sulfamethoxazole (SMX) onto biochars derived from anaerobically digested bagasse was due to π–π EDA interactions between graphitic regions of biochars and the sulfonamide group in SMX and SPY. Ji et al. ( 2009 ) also reported that the adsorption of SMX and SPY to graphite and carbon nanotubes is enabled by π–π EDA interactions involving the heterocyclic rings of the antibiotics. The low removal by CG500 (14% after 24 h) compared to CG700 and CG350 might be attributed to the inherent pH of the solution. The pH varied between 8.79 and 9.30 from the beginning of the experiment to the final time. This is lower than the inherent solution pH using CG700 and CG350 (pH between 10.5–11.2 and 9.9–11.1, respectively). Consequently, with the pH being slightly above the pK a2 of SPY (8.4), a significant fraction of SPY exists as neutral species which do not participate in negative charge-assisted H-bonding. It is therefore hypothesized that adsorption of SPY by CG500 occurred primarily via two mechanisms (hydrophobic interaction and π–π EDA interactions) instead of the three mechanisms attributed to the removal using CG700 and CG350 (hydrophobic interaction, CAHB and π–π EDA interactions). Differences in adsorption of SPY between CG and GB are correlated to the different physico-chemical characteristics of these biochars, which are predominantly controlled by the inherent molecular configuration of the plant-based biomass feedstock. GB700, GB500 and GB350 are characterized by more O-containing functional groups (Fig. 2 b) compared to CG biochars, which renders GB biochars more hydrophilic and limits the potential for hydrophobic interaction. This is explained by the fact that during adsorption, the oxygen groups on the biochar surfaces usually act as the primary adsorption center. Water molecules show a greater affinity for surface oxygen groups on the biochar via hydrogen bonding compared to the more hydrophobic SPY molecules. Water molecules are therefore adsorbed onto the GB biochars surfaces and act as polarized secondary adsorption centers, promoting further water-molecule adsorption and cluster formation. These clusters form an envelope extending beyond the localized adsorption centers, reducing the accessibility of SPY molecules to the solid particles (Zheng et al. 2013). Moreover, water molecules strongly compete for adsorption sites with SPY on the functionalized biochar surface. As a result, the potential for hydrophobic interaction between SPY molecules and the GB biochars was strongly inhibited (Wu and Pendleton 2001 ). One additional factor is that low pore volume and specific surface areas of the GB700 (5.92 m 2 g −1 ) GB500 (0.06 m 2 g −1 ) and GB350 (0.00 m 2 g −1 ) compared to the CG700 (16.33 m 2 g −1 ), made the active sites less available for SPY adsorption, thereby resulting in minimal removal (Fig. 3 b). 3.2.2 Docusate (DCT) adsorption All tested biochars exhibited some ability to remove aqueous DCT (Fig. 4 a, b). The removal of DCT reached 98% using CG700, followed by 85% and 79% using CG500 and CG350, respectively (Fig. 4 a). The biochar surfaces have a net negative charge (Fig. 1 ) and DCT, containing a strongly acidic sulfonate group (Table 1 ), is anionic over a wide pH range. For DCT to be adsorbed, it is likely that hydrophobic interactions are involved. Fig. 4 Removal of Docusate by ( a ) CG700, CG500 and CG350 and ( b ) GB700, GB500 and GB350 (Red errors bars < 10%) Full size image The hydrophobic nature of the CG700, CG500 and CG350 biochars (Table 2 ), compared to GB biochars, coupled with the relatively high log K ow (5.24) for DCT, aided in its adsorption. DCT has a greater tendency to withdraw from the aqueous phase compared to the other selected pharmaceuticals, facilitating hydrophobic interaction between the hydrophobic moiety of the surfactant DCT and the hydrophobic regions of the biochar surface. In addition, hydrophobic interactions can occur between the hydrophobic moiety of previously adsorbed DCT and other DCT molecules in solution, resulting in multilayer adsorption (Brown et al. 1998 ). Moreover, even the GB biochars showed more than 50% removal after 24 h (Fig. 4 b). Adsorption using GB700, GB500 and GB350 reached 51%, 53% and 66%, respectively. However, GB biochars showed the least removal, likely due to the presence of the additional O-containing functional group. The predominance of O-containing functional groups on the GB biochar causes a reduction in the hydrophobic character of the carbon surface, which, in turn, militates against the development of hydrophobic interactions between GB700, GB500 and GB350 and DCT. This resulted in lower DCT removal compared to the CG biochars. The lower temperature biochar, GB350, showed the greatest removal of DCT compared to GB700. Besides the parent feedstock used to produce biochar, the pyrolysis temperature can also influence the surface area and natural organic matter (NOM) content of biochar. This will, in turn, affect the removal of DCT. Biochars made at lower pyrolysis temperatures contain higher NOM content and can sorb organic compounds through the mechanism of partitioning into the organic phase in contrast to hydrophobic interactions (Kupryianchyk et al. 2016 ). Thus, lower temperature GB350 biochar could have higher DCT adsorption potential due to its higher NOM content. The higher sorption of DCT onto lower temperature biochars suggests that surface functional groups on the biochars and NOM may play a more important role in interactions between DCT and biochar than other factors, such as specific SA. No research addressing the adsorption of DCT onto biochar could be found; therefore, further investigation of binding mechanisms is warranted. 3.2.3 Erythromycin (ETM) adsorption The ETM adsorption profiles onto CG700, CG500 and CG350 biochars show 74%, 44% and 37% removal, respectively (Fig. 5 a). Using the GB biochars, removal reached 53%, 64% and 50% for GB700, GB500 and GB350, respectively (Fig. 5 b). Since ETM has a pK a value of 8.88 and the inherent experimental pH of the solution varied between 10.2 and 11.4 when using the biochars, ETM existed predominantly in its anionic form. Since the biochar surface also carries a net negative charge (Fig. 1 ), electrostatic attraction is precluded. Consequently, adsorption is in response to hydrophobic interactions induced by van der Waals forces arising between the hydrophobic ETM molecules (log K ow = 3.06) and the negatively charged graphemic planes of hydrophobic biochars (Sun et al. 2009 ). Fig. 5 Removal of Erythromycin by ( a ) CG700, CG500 and CG350 and ( b ) GB700, GB500 and GB350 (Red errors bars < 10%) Full size image The adsorption of ETM onto microporous hydrophobic beads studied by Sun et al. ( 2009 ) showed that increasing the temperature and pH resulted in decreased K d , indicating that more ETM molecules were present in the solution and that adsorption occurred due to van der Waals forces. In addition, the adsorption of ETM onto the biochars is increased by rapid diffusion of ETM molecules from the solution into the porous structures of the biochars (Mostafapour et al. 2019 ). Moreover, ETM has the ability to form hydrogen bonds between its hydroxyl moieties and the (C=O) and (C=C) group present on the biochars surfaces. The adsorption of ETM follows the order CG700 > GB500 > GB700 > GB350 > CG500 > CG350. Apart from the CG700, the more hydrophilic GB biochars removed ETM more than the CG biochars. These results indicate that the more O-containing functional groups present on the GB biochar increased the formation of H-bonds, thereby leading to greater removal. Moreover, this suggests that adsorption might be dominated by the formation of H-bonds and not the availability of active surface sites. 3.3 Modeling kinetics and adsorption isotherms 3.3.1 Kinetics modeling Quantifying the rate of pharmaceutical removal is necessary for modeling adsorption and determining the contact time needed to achieve a desired amount of pharmaceutical removal in water treatment processes. In this study, both PFO and PSO kinetic models were employed to model the experimental data but only the PSO results are shown (Table 3 ), since the R 2 values for the PFO model are low. Table 3 Kinetic parameters of pseudo-second-order model for the adsorption of the selected pharmaceuticals, SPY, DCT and ETM Full size table From the results, it can be deduced that the PSO model could be used to explain the adsorption processes onto almost all of the biochars because of the high R 2 values (> 0.8), with the exception of a few cases. It is also observed from Table 3 that the experimental adsorption capacity ( q e (exp) ) value is very close to the model-calculated adsorption capacity ( q e (cal) ) for SPY, DCT and ETM, which is consistent with the high correlation of the adsorption of pharmaceuticals onto biochars to the PSO model. The better fit of the experimental data by the PSO model implies that the adsorption of SPY, DCT and ETM onto CG and GB biochars was a rate-limited process controlled by chemical adsorption involving sharing or exchange of electrons between pharmaceuticals and biochars (Qiu et al. 2009 ). The results for the PSO model fitting for the adsorption of the SPY onto GB700, CG500 and GB350 biochars are less favorable ( R 2 < 0.9) compared to the adsorption of SPY, DCT and ETM onto CG biochars ( R 2 > 0.9). This less favorable fit may be attributed to the irregular variation of the adsorption results characterized by very low removal, followed by high removal efficiencies and then no removal, leading to a flat horizontal line showing 0% additional removal after 24 h. The adsorption kinetics do not plateau, and therefore determining the values of R 2 and q e (cal) becomes difficult. No kinetic studies could be found on adsorption of SPY, DCT and ETM onto cotton gin waste and guayule bagasse biochars. Nonetheless, other researchers have reported that the PSO model is useful for describing the adsorption of pharmaceuticals onto different adsorbents. The sorption of sulfonamides by functionalized biochar followed the PSO chemisorption kinetic model (Ahmed et al. 2017 b). Reguyal et al. ( 2017 ) found that the removal of sulfamethoxazole (SMX) and sulfamethazine (SMT) by magnetized pine saw dust biochar followed the PSO model. The adsorption of ETM by carbon nanotubes was better explained by the PSO model ( R 2 = 0.995) compared to the PFO with an R 2 value of 0.892 (Mostafapour et al. 2019 ). Therefore, our results are consistent with other previous findings. 3.3.2 Adsorption isotherms The adsorption isotherm represents the relationship between the mass of pharmaceuticals adsorbed per unit weight of biochar and liquid-phase equilibrium concentration of the pharmaceuticals. These isotherms provide important design data for adsorption systems (Lata et al. 2007 ). When SPY, DCT and ETM concentrations in the aqueous solutions were increased from 2 to 50 mg L −1 , adsorptive uptake of the CG700 also increased. Table 4 shows the different isotherm parameters and their corresponding values. It is seen from Table 4 that the Langmuir isotherm model (Fig. 6 a) exhibited a better fit (i.e., a higher R 2 ) to the adsorption data than the Freundlich isotherm model (Fig. 6 b). The data obtained from the Langmuir isotherm model produces a straight line fitted with higher R 2 of 0.962, 0.966 and 0.989 for SPY, DCT and ETM, respectively, and this clearly suggests that the Langmuir isotherm validates the experimental data for the adsorption of pharmaceuticals onto CG700 biochar. The maximum SPY, DCT and ETM adsorption capacities ( q m ) were 1.221, 19.685 and 17.123 mg g −1 , respectively. The value q m for DCT suggests a greater affinity between DCT molecules and CG700 compared to ETM and SPY molecules. These results are in accordance with the results from the adsorption of the pharmaceuticals showing 98% DCT removal after 24 h (Sect. 3.2.2 ). The adsorption data fitting the Langmuir isotherm suggests that there is uniform binding energy on the surface of the adsorbent and negligible sorbate-sorbate interaction which, in turn, facilitates physical monolayer adsorption (Gong et al. 2008 ). Table 4 Related parameters Langmuir and Freundlich isotherm model for erythromycin and docusate adsorption on to CG700 Full size table Fig. 6 Adsorption isotherms of ETM, DCT and SPY (secondary axis) removal by CG700 ( a ) Langmuir and ( b ) Freundlich isotherm Full size image The suitability of the Freundlich model for SPY, DCT and ETM is indicated by R 2 values of 0.909, 0.905 and 0.947, respectively. The Freundlich constants ( K f ) for SPY, DCT and ETM are 0.531, 2.957 and 0.337 mg g −1 , respectively, and the n values lie between 1 and 10 signifying favorable adsorption by CG700. There is a stronger affinity between DCT and CG700 which is demonstrated by the larger K f value for DCT compared to ETM and SPY. The Freundlich isotherm model suggests that adsorption of these pharmaceuticals onto the surface of CG700 is considered to be a multi-layer, chemisorption process in which the amount of pharmaceuticals absorbed per unit mass of the CG700 increases gradually and is not restricted to the formation of the monolayer (Chung et al. 2015 ). The data fitting both isotherm models ( R 2 > 0.9) indicates that the adsorption of pharmaceuticals on CG700 biochar is not restricted to physical monolayer adsorption and that chemical interactions may be involved. Other studies (Wang et al. 2010 ; Liu et al. 2012 ) have reached similar conclusions. Caution is warranted in choosing one isotherm model over another to explain the adsorption mechanism, given the fact that linear and nonlinear models give different correlation coefficients and this leads to difficulties in explaining the adsorption mechanism as it relates to isotherms (Foo and Hameed 2010 ). However, once the isotherm parameters are determined, these parameters can be used as constants and experimental conditions such as initial pharmaceutical concentrations and biochar mass required to achieved desired removal efficiencies could be estimated prior to the actual adsorption process. This is particularly useful when designing columns for scaling the adsorption process (Essandoh et al. 2015 ). 3.4 Effect of solution pH Solution pH affects both the ionization of the pharmaceuticals and the surface charge of the biochars which, in turn, influences the different mechanisms for pharmaceutical adsorption onto biochar. The removal of SPY was approximately 70% at pH 10–11 but was significantly reduced (40%) at pH 7 (Fig. 7 a). These results differ from other previous studies, with several reporting a decrease in SPY adsorption with a rise in pH due to increased electrostatic repulsion between the anionic SPY − and the negatively charged biochar surface (Ji et al. 2009 ; Xie et al. 2014 ; Yao et al. 2017 ). Additionally, anionic SPY − present at higher pH is more hydrophilic than the neutral form (present at lower pH), which causes a decrease in hydrophobic interactions (Huang et al. 2017 ). At pH 7, the solution is between the pK a1 (2.30) and pK a2 (8.43) of SPY, hence the neutral SPY° species dominate and act as π-electron acceptors. These π-electron acceptors facilitated strong π–π EDA interactions between SPY and CG700 and this was the predominant adsorption mechanism at pH 7 (Ahmed et al. 2017b ). Fig. 7 Removal of ( a ) Sulfapyridine, ( b ) Docusate and ( c ) Erythromycin on CG700 at pH 7 and 10 (Red errors bars < 10%) Full size image In contrast, at a pH of 10, the anionic SPY − species principally exists in solution and an increase in adsorption was observed (70% removal after 24 h). However, the increase in SPY − removal with increase in pH is attributed to the formation of negative-charge assisted hydrogen bonds (CAHB). As demonstrated by Teixidó et al. ( 2011 ), this mechanism proceeds through the adsorption of negative molecules by the release of − OH to proton exchange with water (SPY − + H 2 O → SPY° + OH − ), followed by the formation of exceptionally strong H-bonds between the neutral molecules and the carboxylate functional group present on the CG700 biochar surface (e.g., [RSO 2 N(R') …H…O 2 C-surf] − ). Zheng et al. ( 2013 ) reported similar results from the adsorption of sulfamethoxazole (a member of the sulfonamide family with similar properties to SPY) using biochar, where removal still occurred at alkaline pH due to the formation of CAHB. The effect of pH on the adsorption of DCT and ETM is shown in Fig. 7 b, c, respectively. There were no consistent and pronounced differences between the removal of DCT and ETM using CG700 at pH 7 and at pH 10. DCT and ETM are negatively charged at pH 7 and the proportion of negatively charged species increases as the pH rises. Likewise, CG700 surface becomes increasingly more negative as the pH becomes more alkaline. Thus, electrostatic repulsion between negatively charged DCT and ETM and the biochar should be reduced at pH 7 relative to pH 10. However, π–π electron donor acceptor interactions between the π-electrons of the pharmaceuticals and the π-electrons in the aromatic ring of the CG700 exists throughout the entire pH range. Similarly, diffusion, hydrogen bonding and hydrophobic interaction between the biochar and the pharmaceutical are still dominant mechanisms and this results in the similar equilibrium adsorption amount after 24 h even at different pH conditions. 3.5 Kinetics of adsorption at different pH The kinetics for the adsorption of the pharmaceuticals onto CG700 biochar was evaluated at pH 7 and 10 and the parameters are shown in Table 5 . The experimental data were fitted to the PSO model and high correlation coefficients were observed ( R 2 > 0.8) with excellent linearity. Additionally, the calculated and experimental q e values were similar for both pH conditions. The excellent fit of the data to the PSO models indicates chemisorption may be the rate limiting step at different pH values, where electrons sharing through hydrogen bonding, hydrophobic interactions and π–π EDA interactions occur by valence forces between the pharmaceuticals and CG700 biochar (Qiu et al. 2009 ). Moreover, it was seen that the PSO rate constant k 2 is lower at pH 7 than at pH 10 for SPY, DCT and ETM indicating that adsorption at pH 7 required a higher amount of biochar than at pH 10 to achieve the same adsorption efficiency (Ferreira et al. 2015 ). Table 5 Kinetic parameters of pseudo-second-order models for the adsorption of SPY, DCT and ETM at pH 7 and 10 using CG700 biochar Full size table 4 Conclusion Biochars produced from the pyrolysis of cotton gin waste and guayule bagasse exhibited significant capacity to remove pharmaceuticals from aqueous solution. Removal is a strong function of the solution pH and the mechanisms involved are hydrophobic interactions, hydrogen bonding and π–π electron donor acceptor interactions. Adsorption data fit the PSO model, indicating that adsorption was dominated by chemisorption through electron sharing or transfer. These findings demonstrate the potential for biochar to serve as a low-cost additional treatment for reducing pharmaceuticals in treated wastewater prior to beneficial reuse in a wastewater irrigation system. The surface properties of biochar can vary depending on the biochar feedstock, with some functional groups more effective at reducing some pharmaceuticals compared to others. Given that most pharmaceuticals are weak acids or bases that are moderately hydrophobic (log K ow ~ 0–4), biochar materials that have intermediate degrees of hydrophobicity will likely be most effective in enhancing the removal of pharmaceuticals commonly found in wastewater effluent.

Biochar—a charcoal-like substance made primarily from agricultural waste products—holds promise for removing emerging contaminants such as pharmaceuticals from treated wastewater. That's the conclusion of a team of researchers that conducted a novel study that evaluated and compared the ability of biochar derived from two common leftover agricultural materials—cotton gin waste and guayule bagasse—to adsorb three common pharmaceutical compounds from an aqueous solution. In adsorption, one material, like a pharmaceutical compound, sticks to the surface of another, like the solid biochar particle. Conversely, in absorption, one material is taken internally into another; for example, a sponge absorbs water. Guayule, a shrub that grows in the arid Southwest, provided the waste for one of the biochars tested in the research. More properly called Parthenium argentatum, it has been cultivated as a source of rubber and latex. The plant is chopped to the ground and its branches mashed up to extract the latex. The dry, pulpy, fibrous residue that remains after stalks are crushed to extract the latex is called bagasse. The results are important, according to researcher Herschel Elliott, Penn State professor of agricultural and biological engineering, College of Agricultural Sciences, because they demonstrate the potential for biochar made from plentiful agricultural wastes—that otherwise must be disposed of—to serve as a low-cost additional treatment for reducing contaminants in treated wastewater used for irrigation. "Most sewage treatment plants are currently not equipped to remove emerging contaminants such as pharmaceuticals, and if those toxic compounds can be removed by biochars, then wastewater can be recycled in irrigation systems," he said. "That beneficial reuse is critical in regions such as the U.S. Southwest, where a lack of water hinders crop production." The pharmaceutical compounds used in the study to test whether the biochars would adsorb them from aqueous solution were: sulfapyridine, an antibacterial medication no longer prescribed for treatment of infections in humans but commonly used in veterinary medicine; docusate, widely used in medicines as a laxative and stool softener; and erythromycin, an antibiotic used to treat infections and acne. The results, published today (Nov. 16) in Biochar, suggest biochars made from agricultural waste materials could act as effective adsorbents to remove pharmaceuticals from reclaimed water prior to irrigation. However, the biochar derived from cotton gin waste was much more efficient. In the research, it adsorbed 98% of the docusate, 74% of the erythromycin and 70% of the sulfapyridine in aqueous solution. By comparison, the biochar derived from guayule bagasse adsorbed 50% of the docusate, 50% of the erythromycin and just 5% of the sulfapyridine. The research revealed that a temperature increase, from about 650 to about 1,300 degrees F in the oxygen-free pyrolysis process used to convert the agricultural waste materials to biochars, resulted in a greatly enhanced capacity to adsorb the pharmaceutical compounds. "The most innovative part about the research was the use of the guayule bagasse because there have been no previous studies on using that material to produce biochar for the removal of emerging contaminants," said lead researcher Marlene Ndoun, a doctoral student in Penn State's Department of Agricultural and Biological Engineering. "Same for cotton gin waste—research has been done on potential ways to remove other contaminants, but this is the first study to use cotton gin waste specifically to remove pharmaceuticals from water." For Ndoun, the research is more than theoretical. She said she wants to scale up the technology and make a difference in the world. Because cotton gin waste is widely available, even in the poorest regions, she believes it holds promise as a source of biochar to decontaminate water. "I am originally from Cameroon, and the reason I'm even here is because I'm looking for ways to filter water in resource-limited communities, such as where I grew up," she said. "We think if this could be scaled up, it would be ideal for use in countries in sub-Saharan Africa, where people don't have access to sophisticated equipment to purify their water." The next step, Ndoun explained, would be to develop a mixture of biochar material capable of adsorbing a wide range of contaminants from water. "Beyond removing emerging contaminants such as pharmaceuticals, I am interested in blending biochar materials so that we have low-cost filters able to remove the typical contaminants we find in water, such as bacteria and organic matter," said Ndoun.

10.1007/s42773-020-00070-2

Medicine

Researchers use MRI to show brain changes, differences in children with ADHD

Weiyan Yin et al, Altered neural flexibility in children with attention-deficit/hyperactivity disorder, Molecular Psychiatry (2022). DOI: 10.1038/s41380-022-01706-4 Journal information: Molecular Psychiatry

https://dx.doi.org/10.1038/s41380-022-01706-4

https://medicalxpress.com/news/2022-07-mri-brain-differences-children-adhd.html

Abstract Attention-deficit/hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders of childhood, and is often characterized by altered executive functioning. Executive function has been found to be supported by flexibility in dynamic brain reconfiguration. Thus, we applied multilayer community detection to resting-state fMRI data in 180 children with ADHD and 180 typically developing children (TDC) to identify alterations in dynamic brain reconfiguration in children with ADHD. We specifically evaluated MR derived neural flexibility, which is thought to underlie cognitive flexibility, or the ability to selectively switch between mental processes. Significantly decreased neural flexibility was observed in the ADHD group at both the whole brain (raw p = 0.0005) and sub-network levels ( p < 0.05, FDR corrected), particularly for the default mode network, attention-related networks, executive function-related networks, and primary networks. Furthermore, the subjects with ADHD who received medication exhibited significantly increased neural flexibility ( p = 0.025, FDR corrected) when compared to subjects with ADHD who were medication naïve, and their neural flexibility was not statistically different from the TDC group ( p = 0.74, FDR corrected). Finally, regional neural flexibility was capable of differentiating ADHD from TDC (Accuracy: 77% for tenfold cross-validation, 74.46% for independent test) and of predicting ADHD severity using clinical measures of symptom severity ( R 2 : 0.2794 for tenfold cross-validation, 0.156 for independent test). In conclusion, the present study found that neural flexibility is altered in children with ADHD and demonstrated the potential clinical utility of neural flexibility to identify children with ADHD, as well as to monitor treatment responses and disease severity. Introduction Attention-deficit/hyperactivity disorder (ADHD) is one of the most prevalent psychiatric disorders in children, affecting 3–5% of children worldwide [ 1 , 2 ]. ADHD is characterized by developmentally inappropriate symptoms of inattention, impulsivity, and hyperactivity. Children with ADHD exhibit difficulties controlling their behaviors and attention, which affects their academic performance and social functioning [ 3 ]. Importantly, these symptoms often persist into adulthood [ 4 ]. Current ADHD diagnosis relies largely on behavioral assessments after symptoms onset. In addition, clinical diagnostic approaches could be subjective and thus may influence diagnostic accuracy. Thus, approaches capable of early and objective diagnosis of ADHD may provide an opportunity of early intervention to potentially minimize its long-term sequelae [ 5 , 6 , 7 ]. The development of machine learning methods has provided the opportunity to solve these concerns. By integrating neuroimaging data and machine learning methods, it is possible to differentiate subjects with ADHD from typically developing children (TDC) and, critically, to predict clinical outcomes [ 8 ]. A large number of neuroimaging studies have observed that symptoms of ADHD are driven by atypical brain network organization and impaired functional connectivity (FC) [ 9 , 10 , 11 , 12 ]. Disrupted whole brain and sub-network FC [ 9 , 13 ], altered small-world topology, higher local and lower global efficiency [ 14 ], and, finally, reduced segregation between default mode network (DMN) and task-relevant networks [ 15 , 16 ] have been reported in ADHD. However, one of the common assumptions of the aforementioned studies was that the brain is temporally stable during the entire imaging acquisition period (5–8 min). Recent evidence indicates that subjects are likely to engage in several types of mental activities during a resting period of imaging acquisition [ 17 , 18 ], which could result in altered functional brain network organization throughout the course of the scan [ 19 , 20 ]. More importantly, several lines of evidence have reported that brain dynamics are relevant for complex cognitive processes [ 21 , 22 , 23 ] and are related to psychiatric and neurologic disease [ 24 , 25 ]. Thus, measuring “dynamic” brain features through estimation of time-related variations across multiple short time windows has gained substantial interest [ 24 ]. Notably, altered dynamic brain states, distorted quasi-periodic patterns of brain activity, and changes in temporal variability of FC have been observed in ADHD [ 26 , 27 , 28 , 29 , 30 ]. Cognitive flexibility is a critical aspect of human cognition [ 31 ] and has been reported to be a biomarker for brain disorders [ 32 , 33 , 34 , 35 ]. Children with ADHD have reduced cognitive flexibility as compared to TDC (higher switch costs and slower reaction time) [ 36 , 37 , 38 ]. Recent studies have suggested that brain dynamics are important features underlying cognitive flexibility [ 21 , 23 , 39 , 40 , 41 ]. Specifically, neural flexibility, calculated as the frequency at which brain regions change their allegiance from one functional module to another during fMRI acquisition, has recently been proposed [ 42 , 43 ]. Neural flexibility not only potentially links to cognitive flexibility, but also has been reported to predict learning outcomes and executive functions in healthy subjects [ 42 , 44 ]. Therefore, neural flexibility might be a useful metric to reflect impaired cognitive flexibility in ADHD subjects. In this study, we aimed to determine whether neural flexibility can serve as a biomarker to differentiate children with ADHD from TDC and whether it is associated with ADHD severity. Specifically, we implemented machine learning methods on neural flexibility estimates to: (1) distinguish children with ADHD from TDC; and (2) to assess symptom severity in children with ADHD. We hypothesized that children with ADHD would exhibit lower neural flexibility than that of TDC and that it would successfully differentiate groups and predict symptom severity. Furthermore, although pharmacological treatments can ameliorate the core symptoms of ADHD and improve subjects’ future functional outcomes [ 45 ], only a few studies have investigated how medication use impacts functional network organization in ADHD [ 46 ]. Therefore, we additionally hypothesized that children with ADHD who were on medication would show a “recovery” of neural flexibility toward that observed in TDC. Methods We used two sites from a publicly available, multi-site ADHD dataset, the ADHD-200 study [ 47 ]: Peking University (PKU) and New York University (NYU), with 236 and 192 subjects, respectively. The experimental protocols were approved by the local Internal Review Board. Written informed consent was obtained from all participants. The PKU and NYU study cohorts are the two largest samples and have balanced numbers of ADHD and TDC subjects. The ADHD Rating Scale IV [ 48 ] and Conner’s Parent Rating Scale-Revised, Long version (CPRS-LV) [ 49 ] were used by PKU and NYU respectively to clinically assess the severity of ADHD. Exclusion criteria included left-handedness, IQ below 80, no ADHD rating scale, loss of consciousness due to head trauma, neurological illness, schizophrenia, affective disorder, pervasive development disorder, or substance abuse. In addition, subjects who failed to pass quality control of image preprocessing were also excluded from the analysis, including no full brain coverage, failed tissue segmentation, failed image registration, and excess motion (mean FD > 0.3 mm, maximal head motion of more than 5 mm or 5 degrees). Children with ADHD were included whether or not they were currently taking stimulant medication. In main analyses, all children with ADHD were included in a single group. We additionally investigated the effect of stimulant medication use on neural flexibility by separating children with ADHD into medicated and unmedicated groups. For medicated group, psychostimulant medications were withheld 24–48 h prior to scanning. RsfMRI data were preprocessed using FSL [ 50 ], which included discarding the first 10 volumes, slice-timing correction, motion correction with the mcflirt function of FSL [ 50 ], spatial smoothing (6 mm full-width at half-maximum), bandpass filtering (0.01 Hz–0.08 Hz), global mean/white matter/cerebrospinal fluid (CSF) signal regression, 24 head motion parameters regression and wavelet denoising [ 51 , 52 ]. The time series lengths varied among subjects and imaging sites. To minimize biases contributed by the varying lenghts of time series data, the total time series length was kept at 225 for PKU subjects and 165 for NYU subjects. For each subject, T1-weighted images were first segmented into three tissue types, including gray and white matter and CSF. The tissue segmentation images were then normalized to a standard template using the advanced normalization tools (ANTs) [ 53 ]. After preprocessing, a 5 mm sphere around the coordinates defined by Power264 atlas [ 54 ], was deformed back to the rsfMRI space to extract the mean time series of each region of interest (ROI). Specifically, Power264 atlas parcellates the brain into 264 regions and 14 functional systems, including sensorimotor hand (SH), sensorimotor mouth (SM), auditory (AUD), visual (VIS), cingulo-opercular (CO), frontoparietal (FP), default model (DMN), memory retrieval (MEM), salience (SAL), subcortical (SUB), ventral attention (VA), dorsal attention (DA), cerebellar (CB), as well as an uncertain system (UC). A sliding window approach with a window width of 30 volumes and an increment of 1 volume was employed for the time series data. Pearson’s correlations were calculated for each pair of the 264 ROIs in each window, a p value for each correlation coefficient was estimated using the MATLAB function corrcoef, and only connections significantly different from zero were retained ( p < 0.05, FDR corrected for all 34,716 connections). A multilayer network was constructed by connecting each node to itself in adjacent time windows (Fig. 1 ). Dynamic community detection based on multilayer modularity was performed on the weighted multilayer network using the Generalized Louvain method [ 55 , 56 ]. The Generalized Louvain algorithm outputs a community assignment for each node in each time window. Using the community assignment results, neural flexibility was calculated [ 42 , 57 ] as: $$f_i = \frac{{n_i}}{N}$$ where N is the total number of possible community changes and n i is the number of times node i changes its community label. To account for pseudo-randomness of the community detection algorithm, the Generalized Louvain method was repeated 100 times and mean values of all community-based measures were taken. Whole brain and network-level neural flexibility were calculated as the mean neural flexibility of all nodes of the entire brain and within a predefined canonical network from the Power264 atlas, respectively. Fig. 1: Illustration of a multilayer network. Top panel: in the multilayer network representation of temporal data, each node is connected to itself in adjacent contiguous windows. Next, each node is assigned to a functional community, represented by different colors. Bottom panel: representative correlation matrices of each sliding window. Full size image Linear regression was applied to evaluate statistical differences between groups (e.g., ADHD vs TDC, unmedicated ADHD vs medicated ADHD, unmedicated ADHD vs TDC, medicated ADHD vs TDC). Age, sex, mean FD, and imaging site were included as covariates. Statistical significance was considered as p < 0.05. False discovery rate (FDR) correction was performed for multiple comparisons. The extreme gradient boosting (XGBoost) algorithm was used to develop two models: a classification model to differentiate subjects with ADHD from TDC, and a regression model to predict ADHD severity of individuals. Optimally predictive combinations of region-wise neural flexibility were identified based on their ranked nodal importance for the classification and regression models, respectively. Specifically, we evaluated the model performance of the top N ( \({{{{{{{\mathrm{N}}}}}}}} \in [1,264]\) ) regions based on their importance scores using tenfold cross-validation (CV), which we conducted ten times. Through this process a set of brain regions yielding the best performance of accuracy for the classification model and R 2 for the regression model was determined. Tuning parameters of the final models were provided in Table S2 . Additional details and analyses are provided in Supplementary Information . Results After data preprocessing and quality control, a total of 180 ADHD subjects and 180 TDC were included in the statistical analyses. Demographic and clinical information, as well as motion parameter estimates, are summarized in Table 1 . Table 1 Demographic, clinical and motion information for TDC and ADHD subjects. Full size table Neural flexibility Significantly decreased whole brain neural flexibility was observed in subjects with ADHD as compared to TDC (raw \(p = 0.0005\) ; Fig. 2a ). Consistent with this finding, we observed that modules were significantly more stable in subjects with ADHD (Fig. S1 ). To further examine if the observed decrease in neural flexibility in ADHD was driven by specific functional brain systems or a was general feature of the whole brain, network-level neural flexibility was compared between ADHD and TDC subjects (Fig. 2b ). Compared to TDC, subjects with ADHD exhibited significantly decreased neural flexibility in all but the CO and CB networks ( p values < 0.05, FDR corrected for 14 networks). Additional region-level analysis revealed similar patterns (Fig. S2 ). Together, these results indicate that subjects with ADHD exhibited reduced neural flexibility spanning across multiple functional networks encompassing both higher order and basic cognitive systems. Fig. 2: Alteration of neural flexibility in ADHD. a A boxplot shows significantly decreased whole brain neural flexibility in subjects with ADHD as compared to TDC. b Comparisons of neural flexibility of different functional networks. Black asterisks indicate significant differences after FDR correction. SH sensorimotor hand, SM sensorimotor mouth, CO cingulo-opercular, AUD auditory, DMN default mode, MEM memory retrieval, VIS visual, FP frontoparietal, SAL salience, SUB subcortical, VA ventral attention, DA dorsal attention, CB cerebellar, UC uncertain system. Statistical significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001. Full size image Medication influence A total of 46 subjects with ADHD received medication. Thus, we evaluated the influence of medication on neural flexibility in the ADHD group (Fig. 3 ). We found that whole brain neural flexibility was significantly higher in the medicated ADHD group than the unmedicated group ( \({{{{p}}}} = 0.025\) , FDR corrected for three comparisons). Meanwhile, no statistical differences were observed between the medicated ADHD group and the TDC group ( \({{{{p}}}} = 0.74\) , FDR corrected for three comparisons). Moreover, ADHD subjects in the unmedicated group exhibited significantly decreased neural flexibility when compared to TDC ( \(p = 0.012\) , FDR corrected for three comparisons). Network-level and region-level analyses demonstrate these findings are largely consistent across brain networks and regions (Fig. S4 , Table S6 ). Together, these results indicate treatment with medication has significant impact on the “recovery” of neural flexibility toward that observed in TDC subjects. Fig. 3: Influence of medication on neural flexibility in ADHD. A boxplot shows significantly increased whole brain neural flexibility in subjects with ADHD who are receiving treatment with medication as compared to unmedicated ADHD subjects (corrected p = 0.025), significantly decreased whole brain neural flexibility in unmedicated ADHD subjects as compared to the TDC group (corrected \(p = 0.012\) ), no significant difference in whole brain neural flexibility between medicated ADHD subjects and subjects in the TDC group (corrected \(p = 0.74\) ). Black asterisks indicate significant differences after FDR correction. Statistical significance levels: * p < 0.05. Full size image Differentiating ADHD from TDC We hypothesized that neural flexibility could serve as a biomarker to differentiate children with ADHD from TDC. Due to the smaller number of female subjects with ADHD, the observed sex differences in neural flexibility (Fig. S5 ), and the observed “recovery” of neural flexibility for subjects with ADHD on medication, we excluded both female and medicated ADHD subjects from training of the machine learning model. We performed ten-times tenfold CV, on the PKU dataset (TDC/ADHD: 65/51). The final model was then applied to the NYU dataset (TDC/ADHD: 31/16) as an independent test. When including the neural flexibility of all 264 brain regions, an accuracy of 54.98% (sensitivity: 42.43%; specificity: 64.66%; and AUC: 56.46%) was achieved using the PKU dataset (Fig. 4a ). Using the ranked importance scores, an optimal set of brain regions whose neural flexibility could better differentiate subjects with ADHD from TDC was identified (Fig. 4b and Fig. S6 ). When limiting the model to brain regions with the highest 24 importance scores, an accuracy of 77% (sensitivity: 72.13%; specificity: 80.78%; and AUC: 84.32%) was achieved (Fig. 4a, b ). These 24 regions spanned 8 functional systems (VIS: 6 nodes; UC: 5 nodes; DMN and FPN: 4 nodes each; SAL and SH: 2 nodes each; SUB: 1 node) (Table S3 ). Lastly, we applied the PKU trained classification model with the 24 most predictive brain regions to the NYU dataset, and achieved an accuracy of 74.46% (sensitivity = 62.5%, specificity = 80.64% and AUC = 67.94%), demonstrating the robustness of the proposed approaches for differentiating ADHD from TDC. Fig. 4: Successful prediction of ADHD status and severity using neural flexibility. a The accuracy, sensitivity, specificity, and AUC when using all 264 ROIs, top 24 ROIs, and independent testing using the NYU dataset with the top 24 ROIs, respectively for the ADHD classification model, and b the spatial distribution of the most predictive 24 regions using ranked importance scores. c The R 2 scores when using all 264 ROIs, top 28 ROIs, and independent testing using the NYU dataset with the top 28 ROIs, respectively for the ADHD severity regression model, and d scatter plots comparing the representative neural flexibility-based severity score using tenfold cross validation and clinically obtained ADHD severity score using the top 28 regions for PKU dataset (left) and independent testing using the NYU dataset (right). e The spatial distribution of the most predictive 28 regions using ranked importance scores. Full size image Neural flexibility-based ADHD score We next evaluated the performance of the regression model by comparing the neural flexibility-based ADHD severity score and the clinically obtained ADHD severity score. Using neural flexibility of all 264 brain regions and the PKU dataset, the average R 2 score from tenfold CV with ten repetitions was 0.079. In contrast, an average R 2 of 0.2794 was achieved using the features selected from the brain regions with the highest 28 importance scores (Fig. 4c, d and Fig. S7 ). These 28 regions spanned 11 functional systems (VIS: 8 nodes; DMN: 4 nodes; FPN: 3 nodes; SH, SAL, SUB, DA and CB: 2 nodes each; VA, CO and UC: 1 node each). The spatial distribution of these regions is provided in Fig. 4e and details of these regions are summarized in Table S4 . Applying this regression model and the selected 28 regions to the NYU dataset yielded an R 2 of 0.156. Discussion Extending research in ADHD indicating disruptions in dynamic FC [ 26 , 27 , 28 , 29 , 30 ], here we examined the alterations of neural flexibility in children with ADHD by employing a sliding window approach to estimate multilayer networks. Overall, we found that subjects with ADHD exhibited significantly decreased neural flexibility and altered dynamic modular structure (Fig. S1 ). These findings suggest that functional modules are less segregated in subjects with ADHD than in TDC, consistent with the previously reported impaired segregation of the default network and task-positive networks in ADHD [ 15 ]. Since neural flexibility has been associated with learning and executive functions [ 42 , 44 , 57 ], this decreased neural flexibility may underlie the compromised performance in the domain of executive function observed in ADHD. ADHD leads to a system-wise neural flexibility reconfiguration Decreased neural flexibility was observed in both higher order networks and primary networks. Our results suggest a system-wide dynamic reconfiguration in ADHD rather than a disruption limited to specific sub-systems. Indeed, recent neurobiological models of ADHD have favored multi-network explanations, and the observed differences in functional organization as compared to TDC are widely distributed [ 13 , 15 , 58 ]. Furthermore, though dynamic network reconfiguration has been less investigated in ADHD, existing research is consistent with our observation of a system-wise reduction in neural flexibility. For instance, Rolls et al reported decreased temporal variability of the FC [ 30 ]. Another study reported that children with ADHD spent more time in a hyperconnected state as compared to TDC [ 26 ]. Duffy et al. further demonstrated that ADHD children with reduced temporal variability is related to higher commission errors using a go/no-go task, indicating the impaired executive functions of ADHD subjects [ 16 ]. Meanwhile, hyperconnectivity patterns were also widely reported in ADHD subjects [ 30 , 58 , 59 ], which would potentially “lock” regions together, constrain regional module transition frequency, and reduce neural flexibility. Nevertheless, recent adult studies using the multilayer framework reported that adults with ADHD had higher flexibility and lower integration coefficient than of control subjects [ 60 , 61 ], opposite to our findings. While several potential factors may explain the observed discrepancies, one of the most plausible reasons may be the difference in age across the studies (current cohort: mean age 11.65 years; adult cohorts: mean age 32 years). It has been widely documented that higher-order brain functions follow a protracted developmental timeline, well into adolescence or early adulthood [ 21 , 62 , 63 ]. Therefore, the observed discrepancies may reflect the complex developmental processes associated with ADHD. Future studies executing a direct comparison between different age groups and/or utilizing a longitudinal design are warranted. Medication effects In this study, a total of 46 subjects were treated with stimulant medication. By comparing them to the unmedicated ADHD children, we found that medication use led to increased neural flexibility that was no longer significantly different from that observed in TDC. Since psychostimulant medications were withheld 24–48 h prior to scanning, the observed “recovery” of neural flexibility in the medication group may reflect the long-term benefit of stimulant medication to brain function. Consistent with our findings, previous structural and functional MR imaging studies also suggested that ADHD subjects who received stimulant treatment were more similar to TDC than unmedicated subjects with ADHD [ 46 , 64 ]. These results suggest that neural flexibility is a sensitive metric revealing the alterations of intrinsic brain function in response to stimulant medication. Neural flexibility-based prediction Machine learning methods have been increasingly employed to differentiate ADHD patients from TDC and predict clinical outcomes [ 8 , 65 ]. Using the open access ‘ADHD-200’ dataset, a number of prediction models have been developed, with a range of accuracy from 55 to 90% [ 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 ]. Although, previous studies have reported promising prediction performance, concerns haven been raised about their methodological robustness [ 8 ]. Specifically, after reviewing 69 studies using neuroimaging features to predict ADHD diagnosis, Pulini et al. indicated that high classification accuracy appears to be inflated by circular analysis and small sample size and that many studies lack independent validation [ 78 ]. To mitigate these concerns, we combined tenfold CV and independent testing procedures in this study. First, ten repetitions of tenfold CV were applied. Since partitioning the dataset into tenfolds could yield random effects that may influence prediction performance, performing multiple repetitions should minimize this effect. Second, our models were validated using an independent dataset from NYU. Considering the experimental differences between these two datasets (imaging protocol, PKU: eye open/closed, NYU: eye closed, PKU: ADHD Rating Scale IV, and NYU: CPRS-LV), which could greatly influence the consistency of the data, our models still yield acceptable prediction performance. Finally, aforementioned studies largely focused on accurate classification of ADHD. To date, only a few studies have been conducted to predict ADHD symptom severity [ 79 ]. In this study, a regression model was developed by identifying patterns of neural flexibility that are predictive of clinically obtained ADHD severity score ( \(R^2\) : 0.2794 for tenfold CV, 0.156 for independent test). Currently, clinical ADHD diagnosis mainly relies on behavioral assessments after symptoms onset and have potential rater bias during implementation. Our findings could potentially inform efforts at earlier detection for vulnerable youth. Core regions of prediction Using the ranked regional importance scores yielded by the XGBoost algorithm, we identified 24 brain regions that yielded the highest classification accuracy and 28 core regions that yielded the highest \({{{{{{{\mathrm{R}}}}}}}}^2\) for predicting ADHD severity. We refer to these brain regions as the core regions for classification (CR_c) and regression (CR_r) models. We found that models including only the core regions outperformed models including all regions. These findings indicate that using all brain regions likely include noise, inevitably leading to a negative impact on the performance of the models. As a result, the use of the importance scores yielded by the XGBoost offers a method to potentially remove noise while preserving the features that are important for model training. Among the detected 24 CR_c and 28 CR_r, 11 regions (VIS: 4 nodes; DMN: 3 nodes; FPN: 2 nodes; SUB and UC: 1 node each) were consistently observed, suggesting that they may play key roles for delineating ADHD from TDC as well as for predicting ADHD severity. Indeed, using these 11 regions, comparable performance to the CR_c and CR_r models was achieved from ten times tenfold cross validation within the PKU dataset ( \({{{{{{{\mathrm{Accuracy}}}}}}}} = 71.43{{{{{{{\mathrm{\% }}}}}}}}\) , \({{{{{{{\mathrm{R}}}}}}}}^2 = 0.23\) ), as well as from the independent testing set ( \({{{{{{{\mathrm{Accuracy}}}}}}}} = 70.21{{{{{{{\mathrm{\% }}}}}}}}\) , \({{{{{{{\mathrm{R}}}}}}}}^2 = 0.11\) ). We also identified 13 regions specific to classification accuracy and 17 regions specific to regression accuracy. These regions were mainly located in DMN, FP, SAL, VIS, and attention networks. This indicates that while there is some overlap, there are also important differences between brain regions capable of distinguishing TDC and ADHD and predicting severity of ADHD symptoms. Limitations There are several limitations of this study. First, as publicly available datasets, the imaging parameters and the rating systems of IQ and ADHD severity differ between PKU and NYU cohorts. Nevertheless, our prediction models were generalizable across the two cohorts. Second, we only considered males in the machine learning models due to the limited female subjects with ADHD. Third, we opted not to adjust IQ since the ADHD subjects tends to exhibit low IQ and controlling this variable can provide counterintuitive estimates of the effects of interest [ 11 ]. Finally, current study was performed with the limited sample size, the use of a cross-sectional design, and the use of data with short acquisition time. A longitudinal and prospective study with a larger sample size, longer scanning time and detailed medication information will be needed to further confirm our findings. Conclusion In conclusion, we investigated the dynamic functional network reconfiguration in ADHD. Significantly decreased neural flexibility was observed in children with ADHD spanning multiple brain functional networks, supporting the multi-network explanations of ADHD. Using the XGBoost approach, core regions critically important for differentiating ADHD from TDC, as well as predicting ADHD severity were reported. Finally, we were able to successfully classify group membership and predict ADHD severity using an independent testing dataset, demonstrating the robustness of these approaches. Our study demonstrated the potential clinical utility of neural flexibility to diagnose children with ADHD and monitor disease severity.

Multitasking is not just an office skill. It's key to functioning as a human, and it involves something called cognitive flexibility—the ability to smoothly switch between mental processes. UNC scientists conducted a study to image the neural activity analogues to cognitive flexibility and discover differences in the brain activity of children with ADHD and those without. Their findings, in the journal Molecular Psychiatry, could help doctors diagnose children with ADHD and monitor the severity of the condition and treatment effectiveness. Some people are more cognitively flexible than others. It's just the luck of the genetic draw in some ways, though we can improve our cognitive flexibility once we realize we're being inflexible. Think of it like this: we're cognitively flexible when we can start dinner, let the onions simmer, text a friend, return to making dinner without scorching the onions, and then finish dinner while also carrying on a conversation with your spouse. We're also cognitively flexible when we switch communication styles while talking to a friend and then a daughter and then a coworker, or when we solve problems creatively, say, when you realize you don't have onions to make the dinner you want, so you need a new plan. It's part of our executive function, which includes accessing memories and exhibiting self control. Poor executive function is a hallmark of ADHD in children and adults. When we're cognitively inflexible, we can't focus on some of the tasks, we pick up the phone and scroll social media without thinking, forgetting what we're doing while making dinner. In adults but especially in children, such cognitive inflexibility can wreak havoc with an individual's ability to learn and accomplish tasks. UNC scientists led by senior author Weili Lin, Ph.D., director of the UNC Biomedical Research Imaging Center (BRIC), wanted to find out what's happening throughout the brain when executive function, particularly cognitive flexibility, is off line. Lin and colleagues used functional magnetic resonance imaging (fMRI) to study the neural flexibility of 180 children diagnosed with ADHD and 180 typically developing children. "We observed significantly decreased neural flexibility in the ADHD group at both the whole brain and sub-network levels," said Lin, the Dixie Boney Soo Distinguished Professor of Neurological Medicine in the UNC Department of Radiology, "particularly for the default mode network, attention-related networks, executive function-related networks, and primary networks of the brain involved in sensory, motor and visual processing." The researchers also found that children with ADHD who received medication exhibited significantly increased neural flexibility compared to children with ADHD who were not taking medication. Children on medication displayed neural flexibility that was not statistically different from the group of traditionally developing children. Lastly, the researchers found that they could use fMRI to discover neural flexibility differences across entire brain regions between children with ADHA and the traditionally developing children. "And we were able to predict ADHD severity using clinical measures of symptom severity," Lin said. "We think our study demonstrate the potential clinical utility of neural flexibility to identify children with ADHD, as well as to monitor treatment responses and the severity of the condition in individual children." Other authors of the paper are first author Weiyan Yin, Tengfei Li, Jessica Cohen, Hongtu Zhu, Ziliang Zhu—all of UNC-Chapel Hill—and Peter Mucha, formerly of UNC and now at Dartmouth University.

10.1038/s41380-022-01706-4

Earth

Global warming threatens the existence of an Arctic oasis

Sofia Ribeiro et al, Vulnerability of the North Water ecosystem to climate change, Nature Communications (2021). DOI: 10.1038/s41467-021-24742-0 Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-021-24742-0

https://phys.org/news/2021-09-global-threatens-arctic-oasis.html

Abstract High Arctic ecosystems and Indigenous livelihoods are tightly linked and exposed to climate change, yet assessing their sensitivity requires a long-term perspective. Here, we assess the vulnerability of the North Water polynya, a unique seaice ecosystem that sustains the world’s northernmost Inuit communities and several keystone Arctic species. We reconstruct mid-to-late Holocene changes in sea ice, marine primary production, and little auk colony dynamics through multi-proxy analysis of marine and lake sediment cores. Our results suggest a productive ecosystem by 4400–4200 cal yrs b2k coincident with the arrival of the first humans in Greenland. Climate forcing during the late Holocene, leading to periods of polynya instability and marine productivity decline, is strikingly coeval with the human abandonment of Greenland from c. 2200–1200 cal yrs b2k. Our long-term perspective highlights the future decline of the North Water ecosystem, due to climate warming and changing sea-ice conditions, as an important climate change risk. Introduction The impact of changing sea-ice conditions on the productivity of resources that sustain Indigenous livelihoods in the Arctic has been identified by the IPCC as an emerging climate change risk, stemming from the triangular-intersection of an exposure, a hazard and a vulnerability 1 In this context, the well-documented millennial-scale dependence of local communities on the North Water (NOW) represents an exposure 2 , 3 , defined by the IPCC as the presence of e.g. people, livelihoods or ecosystems in settings that could be adversely affected 1 The NOW or Pikialasorsuaq (‘the great upwelling’ in Greenlandic) is the largest and most productive polynya in the northern hemisphere, an annually recurring ice-free area in northern Baffin Bay (Fig. 1 ). The ice-free waters of the NOW allow for an enhanced and unusually early phytoplankton bloom lasting for two-four months 4 The polynya ecosystem sustains keystone Arctic species, including Arctic cod, seabirds, and marine mammals such as narwhal, beluga, walrus, and polar bear 5 , which all serve to underpin the hunting and fishing economies of Inuit communities in the region 3 , 6 The World Conservation Union (IUCN) has identified the NOW as one of the most ecologically significant marine areas in the Arctic and proposed it as a UNESCO Natural Marine World Heritage Site due to its Outstanding Universal Value 7 . Fig. 1: Location and configuration of the North Water polynya. a Marine and lacustrine core sites, main ocean currents, prehistoric human migration routes, and present-day distribution of little auk colonies. The extent of the polynya is defined as in ref. 17 . b Example of late-June polynya configuration with a stable ice arch c . Example of late-June configuration in the absence of the Kane Basin ice arch. Satellite images from NASA EOSDIS. Several Inuit communities rely directly on the polynya resources today, including Qaanaaq, Siorapaluk, Qeqertat and Savissivik in Northwest Greenland, and Grise Fiord in Nunavut. Background map figures were created using Ocean Data View 73 . Full size image The seasonal formation and physical conditions of the NOW rely primarily on the consolidation of an ice arch or bridge across the southern Kane Basin during winter, which blocks the inflow of multiyear sea ice from the Arctic Ocean (Fig. 1 ). When the ice arch is stable, newly formed sea ice south of the arch is continuously removed by northerly winds and ocean currents, which also promotes a deep mixed layer during winter and high nutrient availability for the spring phytoplankton bloom 4 Entrainment of deeper and warmer Atlantic-derived water masses further limits sea ice growth and provides nutrients that sustain high primary productivity rates 4 . Diatoms, including open-water, marginal ice zone, and sea-ice (sympagic) taxa, are the main primary producers in the NOW (e.g. ref. 8 ) and although grazing pressure may vary, a significant fraction of the diatom production in the euphotic zone reaches the seafloor sediments either as intact cells and spores or empty frustules (after grazing or lysis) 9 . Following nutrient exhaustion, species belonging to the diatom genus Chaetoceros can produce large amounts of highly silicified, fast-sinking resting spores 10 that generally preserve well in the sediments and reflect changes in productivity levels over time. The NOW supports >80% of the global breeding population of little auk, which is the most abundant seabird in the North Atlantic 11 . The little auk is tightly linked to the copepod Calanus hyperboreus , on which the chicks are raised, and >60 million birds depend on the unique availability of this prey item in the NOW 12 . By transporting vast quantities of marine-derived nutrients (MDN) from sea to land in the form of guano, little auks have transformed extensive parts of the NOW coastal landscapes into green oases 13 , 14 . At little auk colonies, temporal changes in the MDN flux in sediments can be used as a proxy for changes in bird numbers and, by inference, NOW productivity over time 15 . Due to its resource richness in one of Earth’s most inhospitable environments, and its strategic location at the southern edge of the narrowest point between Canada and Greenland, the NOW region represents the gateway of prehistoric migrations into Greenland and has been the stage for cultural transitions since the first humans crossed the Nares Strait c. 4400 yrs b2k 2 , 16 . The NOW is of critical significance to Inuit communities today 6 , 17 , and transformative shifts in the ecosystem, due to changing sea-ice conditions, represent an emerging hazard, in this context referring to a climate-related trend with negative impacts 1 . Ice-arch dynamics in Nares Strait have shown a tendency towards instability and earlier spring collapses, linked to changes in sea-ice regime and wind forcing 18 , 19 , 20 . This raises concerns as to whether the NOW ecosystem will persevere in a warming climate. Evaluating risks and defining climate adaptation measures for this unique ecosystem requires an understanding of past ecological and societal responses, but this knowledge is lacking due to a paucity of long-term records. Here, we explore the third side of the NOW climate change risk triangle: vulnerability (propensity or predisposition to be adversely affected 1 ). We applied a retrospective approach, reconstructing long-term trends in sea ice, primary (diatom) production and little auk colony dynamics based on two sediment core records (one marine and one lacustrine) spanning the last c. 4000 and 6000 years, respectively (Figs. 2 , 3 and Supplementary Table 1 ). To reconstruct marine primary production, we quantified changes in the sedimentary fluxes of diatoms (diatom valves and Chaetoceros resting spores). As proxies of past sea-ice dynamics, we used IP 25 , a source-specific molecular biomarker produced by sympagic diatoms 21 , 22 , and its related compound HBI III (Triene), produced by certain pelagic diatoms thriving in the cold waters of the marginal ice zone 23 . Combined, these source-specific biomarkers track polynya activity and stability over time. To infer the presence and relative abundance of little auks, we analysed δ 15 N, cadmium to titanium ratios (Cd:Ti), the fluxes of cholesterol and ß-sitosterol, and changes in diatom assemblage composition in a sediment core from a lake within the catchment of a large little auk colony (Fig. 1 ). Fig. 2: Marine sediment core. a Computerised tomography scan image of the Casq1 core. b CT number. Denser areas appear whiter in the CT scan image. c Total organic carbon (TOC) percentage weight. d TOC-normalised concentrations of the sea ice biomarker IP 25. e TOC-normalised concentrations of HBI III (Triene). f Sedimentation rates. g Modelled median age-depth relationships constructed in BACON for CASQ1 and CASQ1 BC (insert). Error bars and dashed lines represent 95% confidence intervals (for dates see Supplementary Table 1 ). The grey bar indicates the stratigraphic interval covered only by the box-core record. Full size image Fig. 3: Lake sediment core. a Core photograph showing laminations and shift from organic-poor to organic-rich sediments at c 150 cm core-depth. b Percentage (weight) of organic material loss on ignition (LOI). c Carbon to Nitrogen ratio (C:N). d Log of Cl counts based on X-ray fluorescence (XRF) scanning. e Log Cl:Ti XRF data. f Sedimentation rates. g Modelled median age-depth relationship constructed in BACON. Error bars and dashed lines represent modelled 95% confidence intervals (for dates see Supplementary Table 1 ). Full size image Results and discussion Geochronology of sediment records and depositional environments The marine sediment record consists of a gravity (543 cm long) and a box core (40 cm long) retrieved from a site centrally located in Smith Sound at 692 m water depth, south of the southernmost ice arch location (Fig. 1 , see ‘Methods’ for details). The combined record shows continuous marine sedimentation spanning the past ca. 4000 years, with sedimentation rates varying from 0.09 to 0.27 cm y −1 in the gravity core and 0.4–0.67 cm y −1 in the box core, and total organic carbon contents ranging 1.5–2% (Fig. 2 and Supplementary Fig. 1 , Supplementary Table 1 ). The lacustrine sediment core (177 cm long) was retrieved from a lake at Annikitsoq on the Cape York Peninsula at 34 m water depth and spans the last ca. 6000 years (Figs. 1 , 3 and Supplementary Table 1 ). The lake currently lies about 1 km from the edge of the Greenland Ice Sheet and receives some inflow from streams originating there (Supplementary Fig. 16 ). As such, there is an input of water with lower nutrient contents and lower δ 15 N values than the inflow from the large little auk colony located right beside and above the lake. This lake is currently profoundly impacted by the birds with a pH of 5 and Chlorophyll-a concentration of 22.7 µg l −1 , low and high respectively for a High Arctic lake. The lake sediment core is marked by a sharp transition at 150 cm, which dates to 4400–4200 cal yrs b2k (Fig. 3 ). Sediment characteristics at the base of the core reflect the local geology 24 , which comprises high-grade crystalline rocks (Kap York Meta-igneous Complex) with glacial silty-clays and low organic content (Loss on Ignition 20%, Fig. 3 ). After the transition, there is a 4-fold increase in sediment accumulation rates, the colour changes markedly, and organic contents become extremely high for a high Arctic lake (up to 80%) (Fig. 3 ). Diatom analyses of the lake record, supported by XRF elemental data, provide independent evidence that the sediment record covers a period of continuous freshwater sedimentation (Fig. 3 and Supplementary Figs. 2 , 3 ). Prior to the transition at c. 4200 cal yrs b2k, the lake diatom assemblages were dominated by Fragilarioid species ( Staurosira construens and Stauroforma exiguiformis ). Fragilarioids are common in Arctic lakes and ponds and are typically found in oligotrophic and circum-neutral to somewhat alkaline environments with prolonged ice cover and hence a short growing season 25 . After 4200 cal yrs b2k, there is a marked change in the lake diatom assemblages indicating a decrease in lake pH, which is seen as the dominance of acidophilous taxa (within the genera Psammothidium and Eunotia ) and an overall decrease in planktic species, which are adversely affected by lower pH 26 (Supplementary Fig. 3 ). The acidification of the lake is a result of the marked peat accumulation in the catchment area after the arrival of little auk (see also ‘Methods’). The lake diatom assemblages show no clear signs of eutrophication despite the increased supply of marine-derived nutrients by the birds, which is likely due to the overriding effect of acidification on species composition. Significance of the North Water for the human settlement of Greenland The lake indicators record the arrival of little auks at the colony site between 4400 and 4200 cal yrs b2k (Fig. 4 ), corresponding to the marked transition in the core, and consistent with data from nearby terrestrial peat deposits 15 . Bird colony influence on the lake appears to be relatively stable from c. 4200 and 2700 cal yrs b2k (Figs. 4 and 5 ). The high diatom fluxes in the marine record indicate a productive polynya that would support bird colony expansion during this interval (Fig. 4 ). The high IP 25 fluxes demonstrate an active polynya with the recurrent formation of seasonal sea ice, while the minimal HBI III fluxes indicate reduced influence of marginal ice conditions at the core site (Fig. 4 ). Combined, the diatom and sea-ice biomarker records indicate prolonged open-water conditions consistent with an active and stable polynya. The arrival of little auk is coeval with the earliest documented human migrations into Greenland. Fig. 4: Holocene changes at the North Water as evidenced by marine and lake multi-proxy records. The blocks shaded in grey indicate periods of polynya instability. The main phases of human prehistory in Northwest Greenland are represented by bars. Stars on the individual proxy time -series indicate points of significant change based on generalised additive model (GAM) statistics (see ‘Methods’ and Supplementary Figs. 7 – 15 ), and dashed vertical lines denote significant changes in more than one proxy at the same time. a Marine primary production as indicated by the fluxes of Chaetoceros resting spores. b Marine primary production as indicated by fluxes of diatoms (excluding Chaetoceros resting spores). c Fluxes of the sea ice biomarker IP 25 . d Fluxes of the ice-marginal zone indicator HBI III (z triene). e Principal component 1 for the marine record proxies (Supplementary Fig. 4 and Supplementary Table 2 ). f Principal component 1 for the lake record proxies (Supplementary Figs. 5 , 6 and Supplementary Table 2 ). g Principal Components 1 and 2 for the lake diatom assemblages (Supplementary Fig. 3 ). h Changes in Cadmium (Cd) to Titanium (Ti) ratios tracing the level of Cd-enrichment of the lake sediments by bird guano. The dark line shows a LOESS smoothed curve. i δ 15 . j A combination of total cholesterol flux, and the ratio of cholesterol to cholesterol plus β-sitosterol (greyscale) indicates relative bird colony size. RWP Roman Warm Period, DA Dark Ages cold period, MCA Medieval Climate Anomaly, LIA Little Ice Age, LD Late Dorset. Full size image Fig. 5: Evolution of the North Water ecosystem and cultural transitions in Greenland. a A stable and highly productive polynya is inferred from our records after 4400–4200 cal yrs b2k, coincident with the arrival of the first humans in Greenland and the first appearance and expansion of little auks in the area. b From 2700 to 800 cal yrs b2k, the polynya is unstable and reduced in extent, particularly after 2200 cal yrs b2k. This period spans a void in the human settlement of Greenland from c . 2200–1200 yrs b2k and absence/low abundance of little auks. c From c. 800 cal yrs b2k, a stable but low productive polynya is inferred and little auk colonies recover. During this time, there is a replacement of Late Dorset groups by the Thule Culture, the direct ancestors of modern Inuit. d Predicted disappearance of the polynya following the current trajectory of Arctic warming and sea-ice decline. BC Baffin Current, WGC West Greenland Current. Changes in WGC influence in the polynya region are based on ref. 49 . Triangles represent drift ice. Shades of green represent the interpreted late-spring relative extent and productivity of the polynya (darker green corresponds to a more productive polynya and vice-versa). Background map figures were created using Ocean Data View 73 . Full size image According to archaeological and genetic evidence, the first humans to settle in Greenland migrated from Siberia via Alaska and Canada and, on arrival, followed two distinct routes, determined by the presence of the Greenland Ice Sheet 2 , 16 . This gave rise to the Independence I culture in Northeast Greenland 27 and the Saaqqaq culture on the West coast 28 , 29 —both Pre-Inuit cultures. We show that, at this time, the NOW was a stable feature that would have presented both a reliable ice bridge for crossing from Canada to Greenland and a bountiful ecosystem (Figs. 4 and 5 ). Polynya instability and periodic human abandonment of Greenland During c. 2700–2200 cal yrs b2k, the lake record indicates a significant decline in the abundance of little auks, culminating in a short-lived abandonment of the colony c. 2300 cal yrs b2k (Fig. 4 f–i). This decline is marked by falling δ 15 N and cholesterol values, and a reduction of moderately to strongly acidophilous diatom species ( Psammothidium marginulatum , Psammothidium helveticum , and Eunotia species) 30 , 31 (Supplementary Fig. 3 ). In the marine record, primary production remains high until c. 2200 cal yrs b2k after when it drops significantly (Fig. 4 ). The fact that the lake record indicators show an earlier decline than the marine record indicators, is presumably linked to the different locations of coring sites relative to the spatial extent of the polynya. The marine coring site lies at the centre of the polynya, whereas the lake site is located close to its present-day southern edge, and therefore any polynya contractions would be recorded somewhat earlier at this location (Fig. 1 ). Although dating uncertainties cannot be ruled out as an alternative explanation, this idea is supported by peat core studies from further north, demonstrating that bird colonies spread northwards around 2800 and 2200 cal yrs b2K 15 . Both records indicate a sustained period of polynya contraction and instability between 2700/2200–800 cal yrs b2k (Fig. 4 ). This period encompasses a long-term void in the human prehistory of Greenland, spanning both the Roman Warm Period and the Dark Ages Cold Period (Figs. 4 and 5 ). Archaeological studies point to Greenland being uninhabited for about a millennium from the disappearance of the Greenland Dorset in West Greenland at c. 2200 cal yrs b2k until the arrival of the late Dorset Culture by c. 1200 cal yrs b2k 32 (Fig. 5 ). The Late Dorset inhabited Northwest Greenland during the Medieval Climate Anomaly 33 (Figs. 4 and 5 ). Between c. 800 and 100 cal yrs b2k, atmospheric cooling intensified, reaching temperatures up to 3 °C degrees cooler than at the time of human and bird arrival c. 4400–4200 cal yrs b2k, consistent with cooler sea-surface temperatures in the North Atlantic region during the Little Ice Age (refs. 34 , 35 ; Fig. 6 ). A return to more stable conditions is suggested by the sea-ice biomarkers and, while marine diatom fluxes remain relatively low (Fig. 4 ), the little auk colony appears to recover after c. 800 cal yrs b2k (Fig. 4 ), reflected by rising δ 15 N and a higher ratio of cholesterol to cholesterol plus β-sitosterol, along with a return to lake diatom assemblages dominated by acidophilous species (Supplementary Fig. 3 ). This period is marked by another significant cultural transition: the arrival and rapid expansion of the Thule Culture, the direct ancestors of modern Inuit (refs. 16 , 34 ; Fig. 5 ). Fig. 6: Climate variability in the Arctic and North Atlantic region during the late Holocene from selected high-resolution records. a Air temperature anomalies from the Agassiz ice core record from 52 . b Arctic Oscillation (AO) reconstruction from a Kara Sea record as presented in ref. 45 . c North Atlantic Oscillation (NAO) reconstruction from a southwest Greenland lake record as presented in ref. 44 . d Atlantic multidecadal variability (AMV) as reconstructed from a lake record from Ellesmere Island 39 . RWP Roman Warm Period, DA Dark Ages Cold Period, MCA Medieval Climate Anomaly, LIA Little Ice Age, MoW Modern Warming. Grey bars indicate periods of polynya instability as inferred by our lake and marine records. The darker grey bar overlaps with the period of human abandonment of Greenland from c. 2200–1200 cal yrs b2k. Full size image Climate forcing of polynya dynamics Polynya dynamics in the North Water are strongly linked to sea ice conditions (sea ice thickness, concentration and motion) and prevailing winds, and thus influenced by local and regional climate forcing. The inception of recurrent ice arches in the Nares Strait region is estimated to have occurred at the mid-to-late Holocene transition, following atmospheric cooling and extensive sea ice formation in the Canadian Arctic and Greenland 35 , 36 . The North Atlantic Oscillation (NAO) and the Arctic Oscillation (AO) are the two dominant (and closely related) modes of atmospheric variability for the mid-latitudes of the North Atlantic region and the entire Northern Hemisphere, respectively 37 , 38 . The NAO/AO have shown periodicities on multidecadal timescales that are linked with sea-surface temperature changes in the North Atlantic (Atlantic Multidecadal Variability (AMV)), in turn influencing Arctic sea-ice variability 39 , 40 . Sea ice concentration and motion in the Arctic Ocean region are also highly impacted by the Dipole Anomaly (DA) pattern—the second mode of winter variability - which promotes sea ice transport from the western to the eastern Arctic Ocean and export towards the North Atlantic through Fram Strait, and appears to be particularly important during warmer periods 41 , 42 , 43 . From c. 4400 to 2700 cal yrs b2k, the polynya remained relatively stable and productive, and paleo-records point towards predominantly weakly positive NAO and low magnitude variations in the AO during this period (refs. 44 , 45 , Fig. 6 ). In contrast, the period of polynya instability c. 2700/2200–800 cal yrs b2k encompasses episodes of abrupt climate anomalies in the North Atlantic region, characterised by enhanced Arctic water influence in the subpolar North Atlantic 46 . Ocean-atmosphere climate shifts after 2700 yrs b2k have been linked to a strengthening of the latitudinal temperature gradient and southward migration of the Polar Front 46 , 47 , coincident with cooling and freshening of bottom waters off the West Greenland coast from 2600 to 1900 cal yrs b2k 48 The period of polynya instability spans two warm(er) intervals, the Roman Warm Period and the Medieval Climate Anomaly. Marine records from offshore Greenland indicate ocean warming, particularly during the Roman Warm Period (c. 2000–1600 cal yrs b2k), linked to an increased contribution of Atlantic-sourced waters to the West Greenland Current 48 , 49 . Predominantly positive modes of the AO (and NAO) are inferred for the period of human abandonment of Greenland c. 2200–1200 cal yrs b2k and coeval with our demonstrated decline in NOW productivity (Figs. 4 and 6 ). Several episodes of strong positive Arctic Oscillation (AO) anomalies of magnitudes unprecedented in the Holocene are indicated by Kara Sea-sourced ice-rafted iron grains preserved in sediment cores from the Alaskan coast (ref. 45 , Fig. 6 ). Positive winter AO conditions lead to cyclonic wind activity promoting sea ice motion and export out of the Arctic Ocean 50 . Such conditions can have a negative impact on ice arch formation and stability, and lead to an increase in sea ice drift in the NOW region, as demonstrated by satellite data (Fig. 7 ). While the AO affects sea ice conditions in the NOW, the Dipole Anomaly has been determinant in promoting sea ice loss from the Arctic Ocean via the Transpolar Drift system in recent decades 41 . This partly explains why negative AO excursions in recent years have also led to record low Arctic sea ice volumes, contributing to amplifying ice loss and warming in the region (ref. 51 , Fig. 7 ). Fig. 7: Trends in atmospheric and sea ice conditions in the North Water region since the mid-twentieth century. a Anomaly map showing the difference in May–September sea ice concentration following the extreme AO + anomaly year of 1989 compared to the extreme AO-anomaly year of 1996 and location of the weather station at Pituffik (Thule Air Base). Extreme AO + conditions resulted in 5% sea-ice cover reduction in the ice arch area and an increase in drift ice in the polynya area compared to AO- conditions. b Air temperature anomaly for May–September at Pituffik. c Winter AO index. d . Average sea ice concentrations during May–September for the two regions of interest based on satellite observations (see Methods for details); e , f Fluxes of the sea ice biomarkers IP 25 and HBI III, respectively, in the box-core marine sediment record. The grey bar denotes a regime shift in sea ice (declining %) and air temperatures (positive anomalies) after 1998. Full size image The future of the North Water in a warming climate Our long-term perspective on the evolution of the NOW underscores the tight coupling of this ecosystem to climate forcing and highlights its vulnerability to climate change. The concordance of past periods of decreased marine productivity with human abandonment of Greenland suggests that the future collapse of the NOW ecosystem is a legitimate climate change risk. Multiple lines of evidence suggest increasing polynya instability over the last two decades. Remote sensing data indicate earlier ice arch break-up, leading to sea ice export from the Arctic Ocean into the Baffin Bay via the Nares Strait 18 , 19 , which allows for increased late spring/summer ice drift and melt in the NOW region as ice is not held upstream of the arch (Fig. 7 ). This is reflected in our marine record as increased fluxes of both sea ice biomarkers (Fig. 7 ). Present-day air temperatures in the High Arctic are unprecedented in the history of the NOW 52 , and anthropogenic warming is expected to exacerbate Arctic sea ice thinning and loss, detrimental to ice arch stability and polynya formation. On our present climate trajectory, the NOW will likely cease to exist as a globally unique ice-bounded open-water ecosystem, and a winter refuge for keystone High Arctic species 53 . The vulnerability of the NOW is a clear example of the emerging climate change risk associated with changing sea ice conditions on the productivity of indigenously harvested resources 1 . The unprecedented speed of the changes already underway should provide an impetus to ensure that climate change does not outpace the adaptive capacity and threaten the anticipatory knowledge of High Arctic Inuit communities, who are also facing social and economic challenges. This is a priority highlighted by the Inuit Circumpolar Council’s Pikialasorsuaq Commission, which recommends that the governments of Greenland and Canada support ecosystem monitoring and conservation of living resources, and work with indigenous organisations and local communities to implement a management regime for the polynya, and the creation of an Indigenous Protected Area (IPA) 17 . Methods Marine sediment record The Calypso Square gravity core AMD15-CASQ1 (77°15.035′ N, 74°25.500′ W, 692 m water depth) and accompanying box core (BC; same location) were retrieved aboard the CCGS Amundsen during the ArcticNet 2015 Leg 4a expedition in 2015, in accordance with relevant permits and local laws. The CASQ corer recovered a sequence 543 cm long, while the box core was 40 cm long. Sediment material from these cores is stored at the Geological Survey of Denmark and Greenland and available upon reasonable request to the first and corresponding author (SRI). Computed Tomography (CT) scanning of the core was performed using a Siemens SOMATOM Definition AS + 128 at the Institut National de la Recherche Scientifique (INRS), Quebec, Canada. The tomograms were converted into digital DICOM format using a standard Hounsfield scale (HU scale) from −1024 to 3071, where −1024 corresponds to the density of air, 0 to the density of water and 2500 to the density of calcite. The age control on the marine sediment record was provided by 11 accelerator mass spectrometry (AMS) radiocarbon dates on mollusc shells (Supplementary. Table 1 ) at the Keck Carbon Cycle AMS Facility, University of California, Irvine, US, and 210 Pb/ 137 Cs measurements conducted on 20 samples at the Gamma Dating Center, Copenhagen University, Denmark. In the box core, the content of unsupported 210 Pb showed a clear exponential decline with depth (Supplementary Fig. 1 ). A clear 137 Cs peak was not detected, but the 210 Pb-based chronology dates the earliest sample with 137 Cs to 1969 ± 2 years, which is close to the expected date, 1963, for the global 137 Cs peak induced by nuclear weapons testing in the atmosphere. This, and the very uniform exponential decline in unsupported 210 Pb with depth, gives confidence in the calculated chronology. A mixed age-depth model, using both 210 Pb and 14 C dates, was constructed using BACON, an open-source package of ‘R’ 54 . This Bayesian accumulation model code allows for greater flexibility in sedimentation rates between dated intervals than traditional linear age-depth models 54 . The AMS radiocarbon dates were calibrated with the Marine13 IntCal13 55 , and the regional marine reservoir offset was estimated based on existing 14 C data from marine specimens collected before the mid-1950s. Distinct regional offset values have been proposed for Arctic Canada, but do not include the Smith Sound region 56 . Existing data from NW Greenland show local reservoir correction (ΔR) values ranging from -40 years in the Inglefield Fjord to +320 years in Ellesmere Island (the latter consistent with the proposed 335 ± 85 years for the Canadian Arctic Archipelago 56 ). However, these samples have been retrieved from shallow sites (<40 m water depth), which are unlikely to reflect influence from the West Greenland Current. Data from the only deeper site in the NOW region are based on measurements of the mollusc Astarte montagui and indicate an offset of 140 ± 60 years 57 . This value is consistent with measurements from the Disko Bay region, also under the influence of the West Greenland current 58 . We have therefore chosen to use an offset value of 140 ± 60 years. Ages are reported in calibrated years before 2000 (cal yrs b2k) and year CE (for the box-core record presented in Fig. 7 ). TOC measurements were carried out at the Geological Survey of Denmark and Greenland, at 5 cm intervals in the CASQ core and at 2 cm intervals in the box-core. Dried sediment samples (~0.5 g) were powdered (<250 micron) and subjected to Rock-Eval type bulk flow pyrolysis using a HAWK instrument (Wildcat Technologies, Texas). Sets of one control-standard (in-house standard) and one blank were run every 10 samples to ensure instrument stability. Sea ice biomarkers A small number of common pan-Arctic diatoms belonging to the Haslea and Pleurosigma genera are known to produce the sea ice biomarker IP 25 —a mono-unsaturated highly branched isoprenoid (HBI) alkene, biosynthesized in the sea-ice matrix 21 , 22 and deposited in marine sediments following ice melt. Due to its source-specificity and good preservation potential in marine sediments, IP 25 constitutes direct evidence for seasonal sea ice. In marine settings, high sedimentary IP 25 content generally reflects increasing seasonal sea ice concentrations, whereas the absence of IP 25 can either indicate perennial sea-ice cover or open-water 23 . As such, downcore changes in sedimentary IP 25 fluxes from a given location can be interpreted to reflect temporal fluctuations in sea ice conditions. A related lipid biomarker HBI III is produced by diatoms blooming in the often ice-loaded and relatively fresher and cooler surface waters typical of the marginal ice zone (e.g. ref. 59 ). It has recently been shown that the relative abundances of IP 25 and other HBIs remain essentially unaltered in trophic food webs and faecal pellets 60 . These findings suggest that source HBI distributions remain unaltered following grazing, which implies that changes in grazing efficiency do not have a significant impact on the sedimentary signature of HBIs. Prior to analytical treatment, an internal standard (7-hexylnonadecane) was added to 0.5 g of freeze-dried and homogenised sediment. Total lipids were ultrasonically extracted (×3) using a mixture of dichloromethane (DCM: CH 2 Cl 2 ) and methanol (MeOH) (2:1, v/v). Extracts were pooled together, and the solvent was removed by evaporation under a slow stream of nitrogen. The total extract was subsequently resuspended in hexane and purified through open column chromatography (SiO 2 ). Hydrocarbons (including IP 25 and HBI III) were eluted using hexane (8 mL). Procedural blanks and standard sediments were analysed every 15 samples. Hydrocarbon fractions were analysed using an Agilent 7890 gas chromatograph (GC) fitted with 30 m fused silica Agilent J&C GC columns (0.25 mm i.d. and 0.25 µm phase thickness) and coupled to an Agilent 5975 C Series Mass Selective Detector (MSD). The following oven temperature programme was used: 40–300 °C at 10 °C min −1 , followed by an isothermal interval at 300 °C for 10 min. The data were collected using Chemstation and analysed using the MassHunter quantification software. IP 25 and HBI III were identified on the basis of retention time and comparison of mass spectra with authenticated standards. Abundances were obtained by comparison between individual GC–MS responses relative to those of the internal standard. Biomarker data presented here are reported as fluxes to account for changes in sedimentation rate. Changes in marine primary production Diatom fluxes were used to infer changes in marine primary production. For diatom quantification, sediment samples were treated with hydrogen peroxide (H 2 O 2 , 30%) and hydrochloric acid (HCl, 10%) to remove the organic material and carbonate, respectively. Residues were then rinsed several times with distilled water. A known volume of the final residue, homogenised in suspension, was added to a coverslip. Once the samples were completely dried, microscopy slides were mounted in Naphrax® for observation. Quantification of diatom valves was done using an optical microscope (Olympus BX43) with phase contrast optics at a magnification of 1000x. Concentrations were calculated based on the surface area of the slide that was analysed. Fluxes were calculated by combining diatom concentrations (ind. g −1 ) with mass accumulation rates (g cm −2 yr −1 ). Lake sediment record Sediment core NOW25c was collected from a lake at Annikitsoq on the Cape York Peninsula (76°2.100′ N, 67°36.540 W, 8.1 m a.s.l.) on July 30th 2015. The sediment core was recovered from 34 m water depth using a highly portable piston corer specially adapted for remote location use. The 177cm-long core was kept upright and drained of water, the sediment surface was secured by packing the core top with a rigid foam block (Oasis), and the core was kept cool and dark before and during transport from Greenland to Denmark. Fieldwork was conducted in accordance with local laws and permits. Sediment material is deposited at Aarhus University and available upon reasonable request from the last author (TAD). The lake was not stratified at the time of sampling (July 30–August 2 2015) and was partially ice-covered on the day of arrival. The surface waters were 4 °C and oxygen saturation was over 100% all the way to the lakebed. The age control of NOW25c was attained by 10 accelerator mass spectrometry (AMS) radiocarbon dates at Aarhus AMS Centre (AARAMS), Aarhus University, 9 on terrestrial moss remains and 1 on humic extraction of a bulk sample (Supplementary Table 1 ). The radiocarbon ages of the samples were converted into calendar years using the IntCal13 calibration curve 55 . The age model was calculated using the R routine BACON 54 . Ages reported and used in the figures are median modelled ages converted to calibrated years before 2000 (cal yrs b2k). The Loss on Ignition (LOI) technique 61 was used to determine the organic matter content at a 1 cm resolution. The sediment was dried to calculate water content and then heated to 550 °C for two hours and reweighed to calculate the percentage organic matter. The lake sediment core was split along its length then placed in an ITRAX core scanner to obtain high-resolution pictures and measure micro-XRF. The XRF scans were made at the Aarhus University core scanning facility with a molybdenum tube set at 30 kV and 30 mA with a dwell time of 4 s. Prior to analysis, the sediment surface was flattened and covered with a 4 µm ultralene film. A step size of 0.1 mm was selected to capture possible elemental variations even in small laminations. Count readings for less abundant elements, such as Ti, maybe too low with a 4 s dwell time, so counts were summed to at least 1 mm for analysis and presentation. Tracking the presence, absence, and relative abundance of little auks We used a combination of δ 15 N (stable isotope of nitrogen), the ratio of Cadmium (Cd) and Titanium (Ti) concentrations in the lake sediments, concentrations and fluxes of cholesterol and β-sitosterol and diatom assemblage composition changes to assess the presence/absence and relative size of the adjacent little auk colony through time. δ 15 N differs markedly between marine and freshwater systems and has been shown to provide an unequivocal signal of marine-derived nutrients (MDN) 13 . Cd is also more concentrated in the marine system and thus can be used to trace seabird influence 62 . Cd is abundant in the seabird excrement, but it may also be present in the lake’s catchment and so the ratio Cd:Ti is used, as Ti is a proxy of catchment input to the lake. The development of soil, peat and permafrost in the catchment stores C, N and perhaps to a lesser extent Cd, reducing the quantities delivered to the lake and affecting isotope fractionation. The extremely high values of δ 15 N and cholesterol at the time of bird arrival likely reflect the absence of soil in the catchment and the unimpeded flow of bird-derived compounds into the lake. In the event that the inputs of C, N and Cd are reduced due to reduced supply as bird numbers fall, the catchment has the potential to act as a source. Thus, whilst the isotopes of N give a clear signal of bird arrival, they have the potential to be less reliable in tracking absence as a result of their storage and subsequent release. This may depend on the residence time of the lake and the degree of flushing by non-bird impacted waters. The combination of total cholesterol flux and the ratio of cholesterol to cholesterol plus β-sitosterol has been used to identify marine bird influence in freshwater systems in the Arctic (e.g. ref. 63 ). Marine zooplankton, upon which the little auks feed, are rich in cholesterol but contain virtually no β-sitosterol. In contrast, terrestrial and freshwater primary producers contain a high proportion of β-sitosterol 64 . We found very high cholesterol concentrations (1497 µg g −1 ) and low β-sitosterol concentrations (15 µg g −1 ) in little auk excrement, contrasting with goose excrement collected in the same region (16 µg g −1 of cholesterol and 159 µg g −1 of β-sitosterol), a species which largely grazes on terrestrial vegetation. Thus, when seabirds are abundant, cholesterol values in the lake sediments can be expected to be high, and the ratio of cholesterol to cholesterol plus β-sitosterol should also be high. In the NOW region, the ratio is 10-fold higher in little auk excrement (0.99) compared with goose excrement (0.09). This ratio of cholesterol to cholesterol plus β-sitosterol has been previously used as an index of seabird influence, with a value of around 0.6 indicating that the majority of the cholesterol originates from seabirds 64 . This index was derived from studies of fulmar colonies in relatively oligotrophic systems, which is not the case at Annikitsoq, where the lake is situated in extremely eutrophic systems (due to the little auk colony) with lush vegetation in the catchment and freshwater algae thriving in the water column, both of which are sources of β-sitosterol. Here, the arrival of little auks, and with them large quantities of MDN, transformed nutrient availability, prompting a period of exceptionally high terrestrial and aquatic biological productivity. This is demonstrated by the fact that 2 m of peat accumulated in the catchment adjacent to the lake in just 1000 years, following little auk colonisation 65 , and the extremely high LOI values recorded in the lake (75% instead of <10% as expected for a High Arctic lake). Therefore, in addition to the large input of marine-derived cholesterol, there was also the input of cholesterol of terrestrial and freshwater origin, which has a higher proportion of β-sitosterol. Thus, a slight drop in bird input combined with increased terrestrial production, as nutrient levels remain sufficient, would result in a lower ratio, even when birds may still be present. Thus, whilst the ratio is still a useful indicator (especially when <0.4), we have not used a cut-off value signalling the dominance of seabird influence but instead simply present the index value over time. Bird absence is most likely when both cholesterol and the ratio of cholesterol to cholesterol plus β-sitosterol are low. Sterol analysis Sterol analysis followed standard protocols 66 . Specifically, Androstanol (0.1 mg mL −1 ) was added as an internal standard to each sample of approximately 0.5 g of dried, homogenised sediment. Lipid compounds were extracted with solvents (DCM:MeOH, 3:1) using Microwave-Assisted Extraction, saponified and separated into neutral and acid fractions using aminopropyl SPE columns. The neutral fraction of each sample was then separated using silica gel column chromatography. The sterol fraction was trimethylsilylated using N,O- bis (trimethylsilyl)trifluoroacetamide (BSTFA)/trimethylchlorosilane (TMCS) (99:1 v/v) and heated at 70 °C overnight. Excess BSTFA-TMCS was removed by drying gently under nitrogen. Samples were dissolved in 50–100 µl of ethyl acetate prior to gas chromatography-flame ionisation detection (GC-FID) and gas chromatography–mass spectrometry (GC–MS) analysis. GC–MS analyses were performed on an Agilent 7890B GC injector (280 °C) linked to an Agilent 5977B MSD in full scan mode (50–600 amu s −1 ). Separation was performed on an Agilent fused silica capillary column (HP-5, 60 m, 0.25 mm ID, 0.25 um df) with Helium as a carrier gas. Sterol derivatives were analysed using the following temperature programme: 50 °C (held for 2 min) to 200 °C at 10 °C min −1 , then to 300 °C at 4 °C min −1 and held for 20 min. GC–MS peaks were identified through comparisons with known mass spectra (NIST08) and standards where possible. Analytes were quantified based on internal standards. For 13 of the samples, lipids were extracted from the dry sediment using a chloroform:methanol 2:1 mixture and then sonicated for 10 min, after which 0.75 mL of distilled water was added. Lipids were fractionated into neutral lipids (NLs; including sterols), glycolipids, and phospholipids (PLs) using a Bond Elut (0.5 mg) silica cartridge. First, the resin of the cartridges was conditioned using 5 mL of chloroform. Subsequently, the total lipids (1 mL) were applied to the resin, rinsed using chloroform, and then the NLs (including sterols) were collected under vacuum using 10 mL of chloroform. Sterols from the NL fraction were silylated with BSTFA, TMCS, and pyridine at 70 °C for 1 h. Trimethylsilyl (TMS) derivatives of sterols were analysed with GC–MS (Shimadzu) and GC-FID (Shimadzu) equipped with a Phenomenex (USA) ZB-5 Guardian column (30 m × 0.25 mm × 0.25 μm). Cholesterol and β-sitosterol were identified using characteristic ions of GC–MS runs 64 and quantified with GC-FID using authentic standard solutions of plant sterol mixture from Larodan (including 53% β-sitosterol, 7% stigmasterol, 26% campesterol, 13% brassicasterol), and cholesterol from Sigma-Aldrich. The recovery percentage of the sterol samples was calculated using 5-α-cholestane (Sigma-Aldrich) as an internal standard. Stable isotope analyses For stable isotope analysis, samples were taken at 2 cm intervals, freeze-dried for 48 h and ground into fine powder. The samples were packed into tin cups and analysed at UC Davis Stable Isotope Facilities, California, USA. Here, carbon ( 13 C) and nitrogen ( 15 N) isotope analyses were conducted using an elemental analyser and a continuous flow isotope ratio mass spectrometer (IRMS). Specifically, an Elementar Vario EL Cube (Elementar Analysensysteme GmbH, Hanau, Germany) interfaced with an Isoprime VisION IRMS (Elementar UK Ltd, Cheadle, UK). Samples were combusted at 1080 °C in a reactor packed with chromium oxide and silvered copper oxide. After combustion, a reduction reactor trap removed oxides and a helium carrier then flowed through a water trap (magnesium perchlorate and phosphorous pentoxide). CO 2 was held in an adsorption trap until the N 2 peak was analysed; after which the CO 2 was released by heating to the IRMS. Reference materials included: IAEA-600, USGS-40, USGS-41, USGS-42, USGS-43, USGS-61, USGS-64, and USGS-65. A sample’s isotope ratio is expressed relative to a reference gas peak analysed with each sample. These provisional values are finalised by correcting the values for the entire batch based on the known values of the included laboratory reference materials. The long-term standard deviation was 0.2 per mil for 13 C and 0.3 per mil for 15 N. The delta values are expressed relative to international standards VPDB (Vienna Pee Dee Belemnite) and Air for carbon and nitrogen, respectively. Diatom analyses of the lake record Diatom analyses were carried out on 77 samples at a resolution of 1–3 cm, covering the entire lake record. Sediment samples were cleaned using H 2 O 2 and HCl, and permanent slides were prepared using Naphrax®. A minimum number of 400 valves were counted per sample, and relative species abundances were calculated as percentages of the total counts in each sample. Modern sea ice concentration in the NOW region We assessed mean sea-ice concentration during the months May–September (MJJAS) in two regions of interest—NOW and Ice Arch (Fig. 7 ). We used the gridded sea ice concentration and extent product based on satellite observations during 1979–2015, available from the NOAA/NSIDC Climate Data Record of Passive Microwave Sea Ice Concentration 67 . Numerical methods Monte Carlo simulations and principal components analysis To derive principal components of the marine and lake records, we first performed Monte Carlo simulations of the constituent time series from each record (Supplementary Figs. 4 – 6 ). We performed 10,000 simulations of both the marine and lake records that acknowledged both measurement and dating uncertainties of the constituent time series. The 10,000 Monte Carlo simulations yield 2D frequency histograms depicting the probability density of a given measurement at a given time for each time series. We took the median (50th percentile) as the ‘most likely’ time series for each variable and used the 5th and 95th percentiles to define a 90% confidence envelope around this ‘most likely’ time series. For the marine record, each of the three constituent time series analysed (Diatom flux, Chaetoceros spores flux, and HBI III) were assumed to have a measurement uncertainty of ±7.5% (5–10%) and the depth-dependent dating uncertainty shown in Fig. 2 , assumed to represent 2-sigma uncertainty. For the lake record, the four constituent time series were assumed to have measurement uncertainties of 2% for Cd:Ti, 0.5 per mil for δ 15 N, and 0.001 mg cm −2 yr for fractional sterol, with depth-dependent dating uncertainty (as shown in Fig. 3 ) assumed to represent 3-sigma uncertainty. To ensure that sterol was not over-represented as two independent variables in the subsequent principal component analysis, we employed fractional sterol, which is sterol flux times sterol ratio. Similarly, for the fourth constituent series, lake diatoms, we used a weighed sum to combine the two principle components (resulting from analyses carried out on centred and standardised, square root transformed relative abundance data) into a single index. This weighed sum is proportional to the per cent variance explained by each principal component (34% for PC1 and 27% for PC2) and assumed an uncertainty equivalent to 10%. We performed separate principal component analyses on the median time series of the three constituent marine time series and the four constituent lake time series and only considered the resulting first principal component. We propagated the Monte Carlo uncertainty envelopes as fractional mean uncertainties bounding the first principal components of both the marine and lake records (Supplementary Table 2 ). In the marine record, Diatoms and Chaetoceros spores both load positively on the first principal component (PC1 loadings of 0.63 and 0.62, respectively), while HBI III loads negatively (PC1 loading of −0.47). We interpret this as suggesting that the first marine principal component is positively correlated with primary production. In the lake record, all-time series (Cd:Ti, δ 15 N, sterols and diatoms) load positively on the first principal component (Supplementary Table 2 ). We interpret this as suggesting that the first lake principal component is positively correlated with little auk abundance. We do not consider principal components beyond the first but acknowledge that they may also contain climate and/or population signals. When plotting both the first marine and lake principal components (Fig. 4 ), we included the respective Monte Carlo uncertainty envelopes, which have been propagated as the fractional mean uncertainties of the constituent time series underlying each principal component. The Y-envelope corresponds to uncertainty in principal component magnitude, while the X-envelope corresponds to uncertainty in time. The temporal uncertainties we discuss are associated with these latter horizontal uncertainty envelopes. General additive models for detection of significant changes To identify points of significant change in our proxy data over time, we used general additive models (GAM), and took a flexible approach to the degree of smoothing employed in modelling the response of both the marine and lake record proxies. Where possible, the degree of smoothing was estimated using restricted maximum likelihood (REML), and the model contained a continuous-time first-order autoregressive process (CAR(1)) to account for temporal autocorrelation. This solution was optimal for the marine record proxies Diatoms, Chaetoceros spores, IP 25 and HBI III, and for δ 15 N and the diatom PC1 and PC2 axes scores from the lake record. However, for Cd:Ti and cholesterol from the lake record, this approach produced an over-smoothed model, and so in these cases, we used a generalised cross validation (GCV) approach, which optimises predictive accuracy and allows for heteroscedasticity in the data. After the model was fitted (Supplementary Figs. 7 – 15 ), posterior simulation was run with 20 random draws from the posterior distributions of the fitted GAMs to reflect the degree of certainty of the model at a given time (Supplementary Figs. 7 – 15 ), and to allow for the calculation of confidence intervals. Finally, first derivatives (black lines) and associated 95% confidence intervals were estimated for the GAM trend of each proxy. Using this approach, a significant change is demonstrated when the confidence interval of the first derivative does not include 0 (Supplementary Figs. 7 – 15 ) 68 . GAM analyses were carried out in R version 4.0.0 69 , using the package mgcv 70 , 71 and additional code outlined in ref. 68 . Data availability Proxy data are available at the open data repository of the Geological Survey of Denmark and Greenland (GEUS Dataverse ) 72 . The historical air temperature data from Pituffik is available from the Danish Meteorological Institute (dmi.dk), and Arctic Oscillation index data from NOAA/CPC. Code availability Code used for numerical analyses in this study is available at the open data repository (GEUS Dataverse ) 72 of the Geological Survey of Denmark and Greenland.

The University of Helsinki's Environmental Change Research Unit (ECRU) took part in an international study investigating the millennia-long history of the most important oasis in the Arctic and the potential effects of climate change on its future. The North Water Polynya is an area of year-round open water located between northwest Greenland and Ellesmere Island, Canada, in northern Baffin Bay, which is otherwise covered by sea ice roughly eight months of the year. The area is known as an Arctic oasis, and one of the main migration routes of Greenland's original population runs just north of the area. In the study, microfossils and chemical biomarkers preserved in marine and lake sediments were analyzed as keys to the past, exposing historical variation in the North Water Polynya in the past 6,000 years. The polynya's high rate of primary production, for which, in marine environments, diatoms and other microalgae are responsible, maintains a diverse and unique ecosystem that serves as a safe haven for a range of species in Arctic conditions, which are otherwise harsh. Keystone Arctic species, such as the polar bear, the walrus and the narwhal, also thrive there. For the indigenous populations reliant on hunting and fishing, this area, the largest polynya in the northern hemisphere, has been a lifeline. According to the study, the polynya was stable and its primary production was high roughly 4,400–4,200 years ago, at the time when people arrived in Greenland from Canada over the frozen Nares Strait. A millennium of instability and new heat records However, the polynya's stability has varied over the last millennia: during the warmer climate periods 2,200–1,200 years ago, the area was unstable and its productivity fell drastically. When primary production rates are low, significant reductions are seen in the populations of organisms in the upper levels of the food web, such as zooplankton, fish and marine mammals. "According to archaeological finds, there were no inhabitants in the area during this period. It's a mystery that can potentially be explained, in light of the research findings, by conditions that were unfavorable to people reliant on hunting and fishing," says researcher Kaarina Weckström from the Environmental Change Research Unit, University of Helsinki. The researchers point out that air temperatures have never reached the current level in northwest Greenland in the 6,000-year-long period of the polynya's history studied. Global warming and reduction in sea ice caused by human activity have led to the polynya's instability. The area is maintained by favorable ocean currents and winds, and particularly by an ice bridge located north of the polynya, which prevents drift ice in the Arctic Ocean traveling further south. It is the annual formation of this natural block that the warming of the climate is now threatening. "This area, the Arctic's most important oasis, is likely to disappear if temperatures continue to rise as forecast. It would be important to at least slow climate change down, in order for Arctic indigenous peoples to have some kind of a chance to adapt to their future living conditions. Then again, as the history of the polynya suggests, if we can reduce greenhouse gas emissions and mitigate the rising air temperature, both Arctic sea ice and the polynya can be restored," Weckström sums up.

10.1038/s41467-021-24742-0

Nano

Nanotubes self-organize and wiggle: evolution of a non-equilibrium system demonstrates maximum entropy production

Scientific Reports, www.nature.com/srep/2015/15020 … /full/srep08323.html Journal information: Scientific Reports

http://www.nature.com/srep/2015/150209/srep08323/full/srep08323.html

https://phys.org/news/2015-02-nanotubes-self-organize-wiggle-evolution-non-equilibrium.html

Abstract While behavior of equilibrium systems is well understood, evolution of nonequilibrium ones is much less clear. Yet, many researches have suggested that the principle of the maximum entropy production is of key importance in complex systems away from equilibrium. Here, we present a quantitative study of large ensembles of carbon nanotubes suspended in a non-conducting non-polar fluid subject to a strong electric field. Being driven out of equilibrium, the suspension spontaneously organizes into an electrically conducting state under a wide range of parameters. Such self-assembly allows the Joule heating and, therefore, the entropy production in the fluid, to be maximized. Curiously, we find that emerging self-assembled structures can start to wiggle. The wiggling takes place only until the entropy production in the suspension reaches its maximum, at which time the wiggling stops and the structure becomes quasi-stable. Thus, we provide strong evidence that maximum entropy production principle plays an essential role in the evolution of self-organizing systems far from equilibrium. Introduction Classical thermodynamics offers elaborated theoretical instruments to study behavior of systems that are in or near equilibrium. But often objects and processes which we deal with in everyday life, such as living biological organisms or weather events, are far from equilibrium. Therefore, the search for principia governing nonequilibrium systems is of great interest. The notion of entropy in equilibrium states and its production in nonequilibrium processes not only form the basis of modern thermodynamics and statistical physics, but have also been at the core of various philosophical discussions concerned with the evolution of the world, the course of time, etc. There is a general agreement on how the entropy of a nonequilibrium system should evolve, postulated by the second law of thermodynamics. At the same time, the question how entropy production changes with time is still extensively discussed. So-called the maximum entropy production principle (MEPP) 1 , 2 , 3 was proposed in the middle of 20 th century 4 , 5 , 6 , 7 and may be viewed as the natural generalization of the Clausius-Boltzmann-Gibbs formulation of the second law. According to MEPP, a nonequilibrium system evolves such as to maximize its entropy production under present constraints. Proponents of this principle demonstrated that it can be used as a powerful tool to solve various environmental 8 , 9 , 10 , biological 11 , 12 and physical problems 13 , 14 , 15 , 16 . It is interesting to notice that there are publications attempting to disprove MEPP 17 , 18 , 19 , 20 , 21 , 22 . However, the most recent review 23 argues that all these claims seemingly disproving MEPP are based on experiments, which do not belong to the sphere of applicability of the principle. Frequently, thermodynamic systems driven out of equilibrium have a tendency for symmetry breaking. This includes pattern formation 24 , spontaneous self-assembly of particles either dispersed in colloids 25 , 26 , 27 , suspended at the air-fluid 28 , 29 or fluid-fluid interfaces 30 . One of the best-known examples of symmetry breaking is the emergence of Rayleigh–Bénard (RB) convection cells 31 in a layer of oil placed on a hot plate. Ordering that is present in the RB cells reduces the oil entropy. At the same time, convective flows associated with the RB structures increase the heat transfer rate, leading to a greater entropy production of the entire system, which includes the hot plate and the environment. It is unclear under which conditions nonequilibrium systems develop a new order. Yet, it is plausible to hypothesize that MEPP might be one of the main guiding principles for not only the evolution of biological species, but their appearance on Earth due to the strongly nonequilibrium energy distribution created by the Sun 32 , 33 . Nonequilibrium conditions occur because the Sun supplies photons with energies of many thousands of Kelvins, which is much larger than the temperature of the surface of the Earth. Taking into account these facts, we believe that investigations of the MEPP applicability for self-organizing systems is a very promising direction of research, from the viewpoint of both the substantiation of the MEPP and for better understanding and predicting behavioral and evolutionary trends in nonequilibrium systems. In this work, we investigate possible evolution paths of an electrorheological (ER) fluid 34 exposed to a strong electric field (E-field) towards an attractor state. Our setup provides means to study nonequilibrium processes quantitatively . In particular, it enables precise measurements of generated Joule heating (the precision is limited by a very small and uncertain fraction of internal work done by the voltage source that is used to create order in the system of nanotubes). As a result, entropy production in the fluid, which is directly proportional to the Joule heating, can be correlated with the process of self-assembly of carbon nanotube (CNT) chains occurring under the effect of the applied E-field 35 , 36 , 37 , 38 , 39 . We demonstrate that in such complex systems as ER fluids the state with maximum entropy production acts as an attractor. Results When the ER fluid is driven far from equilibrium by a strong E-field, it may exhibit a spontaneous collective behavior, such as emergence of avalanches and chain formation. Previous studies were suggestive that these processes lead to maximization of the entropy production rate 40 . Here we use a different suspension of conducting particles and observe that the ER fluid evolves towards quasi-stable states characterized by the maximum (or near-maximum) rate of entropy increase in a wide range of excitation parameters. The evolution usually proceeds in two phases: an “avalanche” phase followed by a “stable” phase, which onsets when the entropy generation reaches its pre-determined maximum. We discovered that close to the end of the avalanche phase self-assembled dissipative structures can start to wiggle, i.e. exhibit quasi-periodic motion and deformation (for the video containing details see the supplementary online material ). Due to a visual resemblance to a living organism, we term such states as a “bug”. Interestingly, the bug, which takes energy from the biased electrodes, halts its motion as soon as the entropy production in the ER fluid reaches the maximum. Experiment details The fluid is comprised of electrically conducting either of single- (SES Research, outer diameter < 2 nm , length 5 – 15 μm ) or multi-wall carbon nanotubes (Alfa Aesar, outer diameter 3 – 24 nm , length 0.5 – 5 μm ) suspended in an insulating non-polar solvent (toluene). The field is generated between a pair of metallic electrodes immersed in the ER fluid and connected to a voltage source in series with a commercial resistor. The measurement setup involves a voltage source connected in series with a resistor, ammeter and two cylindrical stainless steel electrodes submerged into the sample ( Fig. 1a ). The diameter of the electrodes is 0.7 mm , the spacing between them is 10 mm and the depth of immersion into the fluid is over 10 mm . The nanotube concentration is varied between 0.05 and 0.2 mg/ml . This is significantly below the percolation threshold, i.e. the current in the circuit is negligibly small when the nanotubes are not aligned with each other. However, all the concentrations we use allow a nontrivial self-assembly: when the tubes form chains, the electric current, I = I ( t ), can reach and even exceed U/2R s , where R s is the series resistance and U is the applied voltage. Since the current depends on the presence of continuous nanotube chains and electro-convection, we can use it as a measure of the order in the suspension. Figure 1 Consecutive snapshots of the sample illustrating the formation of nanotube chains. The distance between electrodes is 1 cm, applied voltage is 400 V and the series resistor is 100 MOhm . Panel (a) demonstrates the photograph of the ER fluid before the voltage is applied and the schematic of the experimental setup. The following photographs are taken after (b) t = 45 s , (c) t = 90 s and (d) t = 1500 s of interaction with E-field. Full size image Prior to each experiment, the nanotubes in the fluid are dispersed by means of sonication, ensuring that the initial resistance of the suspension, R f ( t = 0 ), exceeds R s by at least 3 orders of magnitude. The time t is measured from the moment when the voltage is turned on. During the measurement, the value of the series resistor is fixed, while the resistance of the fluid is changing owing to the formation of CNT chains. We use a precision electrometer (Keithley 6517B) to set the voltage and to measure the current in the circuit. The power dissipated in the fluid, P f , can be deduced from the equation P f ( t ) = I ( t )[ U-I ( t ) R s ], where I ( t ) R s is the voltage drop on the resistor. Chain growth Snapshots of the system evolution under the influence of a strong electric field are demonstrated in Figure 1 . Before the voltage is turned on, the carbon nanotubes are homogeneously dispersed in the fluid, providing its uniform grey coloring ( Fig. 1a ). Soon after the bias is applied, we observe a turbulent motion of the fluid. This motion is caused by the electro-convection, or the “shuttling” effect 40 , see Fig. 1b , when the electric charge is transported from one electrode to the other by small but visible clusters of charged particles. In parallel with the charge transfer process, chains of nanotubes begin to grow. Characteristically, the growth starts from the positive electrode (left electrode in Fig. 1 ) and proceeds towards the negative one, which is grounded in our setup. The process of self-organization is not always monotonic, i.e. the chains can get destroyed, what represents an avalanche . This evolution stage is called the “avalanche” phase due to the fact that they frequently, but not necessary, occur here. The reasons for the avalanches have been clarified in the model presented in Ref. 40 . At a certain value of the power dissipated in the fluid, the system transitions to a stable state, which is characterized by termination of a turbulent motion and formation of a steady connection between the electrodes ( Fig. 1c ). We term this stage of the system evolution a “stable” phase. The stable phase begins at time t 0 and continues as long as the experiment goes on. During that phase, the chain breaking events can no longer be detected visually. Quite the contrary, more and more nanotubes are attracted to the electrodes, making the chain network larger and denser, while the rest of the fluid becomes clearer ( Fig. 1d ). As will be discussed later, the stabilization coincides in time with the moment t 0 when the power dissipated in the fluid reaches its highest possible value, determined by the applied voltage and the value of the series resistor. Now we want to comment on the linearity of the process. Before the maximum entropy production is achieved, the process is nonlinear. Chains of nanotubes are formed under the effect of strong electric field and get destroyed by fluctuations, gravity, Coulomb forces, and, possibly, by some local Joule heating inside electrically conducting chains. When the maximum of the entropy production is achieved, we notice that the cloud formed by nanotubes and attached to the electrodes is rather stable. Therefore, small variations of the bias voltage do not change the resistance of the structure in this regime. Hence, the charge flow is approximately a linear function of the applied electric field. On the other hand, if the voltage is reduced significantly, the gravity and fluctuations cause a certain amount of degradation of the self-assembled cloud of nanotubes. In case of strong variations of the applied voltage the response is nonlinear. Generally speaking, when we emphasize that the applied electric field is strong we mean that it is sufficient to change the organization of the nanotubes and thus it can change the resistance of the fluid. The self-assembly in nonequilibrium systems is inherently nonlinear process. Discussion The chain formation mechanism can be explained by means of simple dipole-dipole interaction. Indeed, the applied E-field polarizes nanotubes, making them act as electric dipoles. Since a dipole moment aligns along the field, all nanotubes tend to be in parallel with each other. Such orientation of dipoles creates an attractive force between them, when the positively charged end of one nanotube connects to the negatively charged end of a neighboring nanotube. Thus, chains of nanotubes start to grow. Our setup has an important property that we want to emphasize: Joule heating within the fluid, generated by the current flowing through CNT chains, has a maximum 40 : P max = U 2 /4R s . This maximum is achieved only when the nanotubes organize in such a way that their collective resistance matches the series resistance, i.e. when R f = R s . The above condition could be thought of as a manifestation of the maximum power transfer theorem. However, it should be stressed that this theorem does not predict any particular direction of the evolution for self-organized systems included into the circuit, as is the case in our experiments. Thus, it cannot be used to predict that the power generated in the self-assembled fluid should achieve its maximum in the process of evolution initiated by the applied strong electric field. The temperature increase of the ER fluid, Δ T , depends on the applied voltage, duration of the experiment and on the suspension's resistance, mass and specific heat. Simple estimates show that even if the system was thermally isolated, the maximum Δ T for the characteristic parameters of our experiment should not exceed 0.15 K . It constitutes only 0.05% of the fluid temperature. In the real situation there is a heat flow from the fluid to an environment and therefore Δ T is even smaller. To confirm this estimate, we have measured the temperature of the fluid during the self-assembly process and detected no temperature increase to an accuracy of 0.2 K . Therefore, we take T f ≈ T = constant , where T and T f is the ambient and the fluid temperature, correspondingly. The entropy production in the fluid is calculated as the power dissipated in the fluid divided by the environment temperature, i.e. dS f /dt = P f /T . Thus, MEPP in this system corresponds to the condition of the maximum power dissipated in the fluid, which is achieved when R f = R s . Let us now turn to Figure 2 , which demonstrates how the power dissipated in the fluid changes under various experimental conditions. In general, we distinguish two evolution scenarios. When the system evolves according to the first scenario, which we call the “ complete evolution ”, P f reaches P max . At the initial stage of such evolution and before P max achieved, the average power increase may either proceed smoothly (curve S1 in Fig. 2a ) or be accompanied by spontaneous avalanches (curve S2 in Fig. 2a ). The general trend is that overall the dissipation tends to increase. Figure 2 Normalized power dissipation in the fluid as a function of time, t . Concentration of nanotubes is 0.075 g/l . (a) Complete evolution: curve S1 (black), R s = 10 MΩ , U = 75 V ; curve S2, R s = 10 MΩ , U = 325 V (blue). The time t 0 is the time when the maximum possible dissipated power is achieved. (b) Incomplete evolution: curve U1, R s = 100 MΩ , U = 5 V ; curve U2, R s = 2 MΩ , U = 300 V . Full size image From multiple visual observations we identify that the moment at which P f matches P max coincides with the beginning of the stable phase. During the system evolution in the stable phase, dissipation changes slowly over time and P f stays on the order of P max for the remaining and the longest part of the experiment (note the logarithmic scale in Fig. 2 ). Similarly to Ref. 40 , no avalanches are observed in the stable phase of a complete evolution. Departure of P f from P max at t > t 0 corresponds to a gradual decrease of R f below R s , which is caused by the thickening of the CNT chains and is not a result of their destruction. When the system evolves according to the second scenario, termed “ incomplete evolution ”, P f does not reach P max ( Fig. 2b ) and CNT precipitate. In the early phase, the chains may start to form, which together with the electro-convection leads to an increasing P f . Later on, this growth slows down and the system eventually enters a stagnation phase. At the next, the final phase, the system “dies”: nanotubes precipitate and the dissipated power drops to a value many orders of magnitude below P max . There are three feasible reasons for the incomplete evolution. First, when an applied E-field is too weak (curve U1 in Fig. 2b ), the “destructive” forces, such as gravity and thermal vibration of nanotubes, are able to overcome the attractive dipole-dipole forces, which results in uncorrelated diffusion of the nanotubes and, later, to their slow, but irreversible precipitation. Incomplete evolution can also be caused by a too strong electric field (curve U2 in Fig. 2b ), which creates intense turbulent flows in the fluid, precluding formation of stable chains. Moreover, when the first chains, connecting two electrodes, are formed, strong E-fields induce high local electric currents heating up these chains and eventually burning them out. And finally, at low concentrations of CNT, the system is not able to build a stable network of chains that matches the series resistance. Despite the number of factors impeding the formation of CNT chains, the complete evolution takes place in a surprisingly wide range of experimental parameters. To study the ability of the system to adjust its resistance to reach P max , we performed a number of experiments in which we fixed the applied voltage U and varied the series resistance R s . The results of these measurements are plotted in Figure 3 . As one can see, R f tends towards R s with time. This finding is especially remarkable since the series resistance is varied across three orders of magnitude, whereas all other parameters, namely the concentration of nanotubes and the applied voltage, are kept the same. Figure 3 The resistance of nanotube chains formed in the fluid at different series resistances. For each R s four values of R f are shown, measured at different times t . Blue diamonds represent the longest evolution time. Applied voltage is 150 V , concentration of nanotubes is 0.07 5 g/l . The blue dashed line corresponds to R f = R s , t 0 is the time when P ( t 0 )/ P max = 1 . Full size image The tendency of the system to maintain dissipation that is close to its maximum is demonstrated in Figure 4 . Here, the probability density function of the dissipated power averaged over 10 measurements is plotted for several moments of time during a complete evolution. In the beginning of the avalanche phase, i.e. at t = 0.1t 0 , relatively low Joule dissipation values have the highest probability. Later on, the distribution shifts towards the states with higher dissipation level and at t = 0.5t 0 we already observe three distinct peaks. During the stable phase of the evolution, the probability peak that corresponds to the maximum dissipation continues to grow, while the amplitudes of all other peaks diminish. This result supports our conclusion that throughout the evolution the system tends to self-organize in such a way that the dissipation of the injected energy proceeds most efficiently. Therefore, the entropy production in the fluid proceeds at the maximum rate for a given set of constraints (applied voltage and the series resistor). Once the maximum is achieved, the evolution processes slow down, giving rise to the peak in the vicinity of < P f /P max > = 1 . Note that if we consider the entire system that includes the fluid and the resistor, then the maximum of the entropy production is never achieved, since it would correspond to R f = 0 . Figure 4 Probability density function of the normalized power dissipated in the fluid. These curves are calculated using the measurements of the type shown in Fig. 2 . The probability distributions are calculated for the following time intervals: 0 < t < 0.1t 0 , 0 < t < 0.5t 0 , 0 < t < 1.0t 0 and 0 < t < 2.0t 0 . The total ensemble consists of 438820 measured points, corresponding to ten independent evolution curves, in which R s and U had different values. Full size image Occasionally, towards the end of the avalanche phase, nanotubes form a dissipative pattern that exhibits quasi-periodic collective motion and deformation. Figure 5 shows an example of such a dissipative structure (CNT bug) that self-assembled around one of the electrodes (see video in Supplementary Information ). A few chains of nanotubes form thick CNT “arms” that protrude out of this bug and extend towards the opposite electrode ( Fig. 5a, 2s ). Their motion forces the cloud to displace, causing its slight deformation. When the arms touch the electrode, they remain in contact with it for a short period of time ( Fig. 5a, 3s ) and then retract ( Fig. 5a, 9s ). Subsequently, the process repeats again ( Fig. 5a ). This wiggling of the dissipative structure is surprising since the applied voltage is constant in time. The extension and retraction of several CNT arms is almost synchronous. The motion of the CNT bug is repeated and can be characterized as quasi-periodic, up until the stabilization transition when the system enters the stable evolution phase. Figure 5 Motion of the CNT “bug” under the influence of dc electric field. Panel (a) illustrates positions of the “arms” of the bug when they are retracted ( t = T ), approaching ( t = T+2 s ), touching ( t = T+3 s ) and again retracted ( t = T+9 s ) from the right electrode. Here T~2500 s . The movie showing the motion of the selfassembled wiggling structure is contained in Supplementary Information online. Panel (b) demonstrates the time dependence of the normalized power dissipated by the bug. Inset: a segment of the time dependence of the normalized power, corresponding to the phase when the bug is fully developed and exhibits the arm motion. At t~4500 s the stable phase begins. Full size image In the course of the CNT bug formation, the initial rapid growth of the dissipated power is accompanied by relatively large avalanches ( Fig. 5b ). Once the body and the arms of the CNT bug have formed, the dependence P f /P max ( t ) starts to jitter near the maximum, but does not actually reach it (inset in Fig. 5b ). This jittering correlates with the arms' motion (see Fig. 5a ), namely the power increases when they touch the other electrode (right electrode in Fig. 5a ) and decreases as soon as they retract. After approximately an hour of this wiggling behavior, a transformation to a state with the maximum Joule heating in the nanotube structure takes place. At that moment the center of mass of the cloud of nanotubes shifts to the central position between the electrodes. The connection of the cloud to the left electrode remains stable and a similar stable connection with the right electrode gets established. The bug halts its motion and P f remains near P max for as long as the experiment is continued. As time goes by, the center of mass of the cloud slowly shifts closer to the bottom of the container, apparently due to gravitational forces. Yet, the connection of the cloud to both electrodes remains stable and the dissipated power continues to be at its maximum. To sum up, we have studied the behavior of the electrorheological fluid under the influence of a strong electric field. Two types of the evolution have been observed: complete, max( P f ( t )) = P max and incomplete, max( P f ( t )) < P max . In the course of complete evolution, the total resistance of self-assembled carbon nanotube structure tends to a value which maximizes energy dissipation in the structure. This fact reflects the realization of the maximum entropy production principle. Moreover, the self-assembled dissipative structures can start to wiggle, thus resembling a self-created living organism. We find that the system is dynamic if its resistance is greater than the series resistor and it becomes static when the impedance matching condition, R f = R s , is achieved. Our conclusions apply to systems subjected to external forces and gradients, which are sufficiently strong so that the behavior is nonlinear and the self-assembly cannot be prevented by external or internal perturbations. On the other hand, the applied forces and/or gradients should not be too strong. Otherwise the generated heating might destroy self-assembled patterns.

The second law of thermodynamics tells us that all systems evolve toward a state of maximum entropy, wherein all energy is dissipated as heat, and no available energy remains to do work. Since the mid-20th century, research has pointed to an extension of the second law for nonequilibrium systems: the Maximum Entropy Production Principle (MEPP) states that a system away from equilibrium evolves in such a way as to maximize entropy production, given present constraints. Now, physicists Alexey Bezryadin, Alfred Hubler, and Andrey Belkin from the University of Illinois at Urbana-Champaign, have demonstrated the emergence of self-organized structures that drive the evolution of a non-equilibrium system to a state of maximum entropy production. The authors suggest MEPP underlies the evolution of the artificial system's self-organization, in the same way that it underlies the evolution of ordered systems (biological life) on Earth. The team's results are published in Nature Publishing Group's online journal Scientific Reports. MEPP may have profound implications for our understanding of the evolution of biological life on Earth and of the underlying rules that govern the behavior and evolution of all nonequilibrium systems. Life emerged on Earth from the strongly nonequilibrium energy distribution created by the Sun's hot photons striking a cooler planet. Plants evolved to capture high energy photons and produce heat, generating entropy. Then animals evolved to eat plants increasing the dissipation of heat energy and maximizing entropy production. In their experiment, the researchers suspended a large number of carbon nanotubes in a non-conducting non-polar fluid and drove the system out of equilibrium by applying a strong electric field. Once electrically charged, the system evolved toward maximum entropy through two distinct intermediate states, with the spontaneous emergence of self-assembled conducting nanotube chains. In the first state, the "avalanche" regime, the conductive chains aligned themselves according to the polarity of the applied voltage, allowing the system to carry current and thus to dissipate heat and produce entropy. The chains appeared to sprout appendages as nanotubes aligned themselves so as to adjoin adjacent parallel chains, effectively increasing entropy production. But frequently, this self-organization was destroyed through avalanches triggered by the heating and charging that emanates from the emerging electric current streams. Carbon nanotube cloud becomes "alive" in strong electric field. This experiment was done by A. Belkin and A. Bezryadin in collaboration with A. Hubler. Credit: University of Illinois at Urbana-Champaign "The avalanches were apparent in the changes of the electric current over time," said Bezryadin. Following avalanches, the chains with their appendages "wiggled," resembling a living thing, similar to an insect. "Toward the final stages of this regime, the appendages were not destroyed during the avalanches, but rather retracted until the avalanche ended, then reformed their connection. So it was obvious that the avalanches correspond to the 'feeding cycle' of the 'nanotube inset'," comments Bezryadin. In the second relatively stable stage of evolution, the entropy production rate reached maximum or near maximum. This state is quasi-stable in that there were no destructive avalanches. The study points to a possible classification scheme for evolutionary stages and a criterium for the point at which evolution of the system is irreversible—wherein entropy production in the self-organizing subsystem reaches its maximum possible value. Further experimentation on a larger scale is necessary to affirm these underlying principals, but if they hold true, they will prove a great advantage in predicting behavioral and evolutionary trends in nonequilibrium systems. The authors draw an analogy between the evolution of intelligent life forms on Earth and the emergence of the wiggling bugs in their experiment. The researchers note that further quantitative studies are needed to round out this comparison. In particular, they would need to demonstrate that their "wiggling bugs" can multiply, which would require the experiment be reproduced on a significantly larger scale. Such a study, if successful, would have implications for the eventual development of technologies that feature self-organized artificial intelligence, an idea explored elsewhere by co-author Alfred Hubler, funded by the Defense Advanced Research Projects Agency. "The general trend of the evolution of biological systems seems to be this: more advanced life forms tend to dissipate more energy by broadening their access to various forms of stored energy," Bezryadin proposes. "Thus a common underlying principle can be suggested between our self-organized clouds of nanotubes, which generate more and more heat by reducing their electrical resistance and thus allow more current to flow, and the biological systems which look for new means to find food, either through biological adaptation or by inventing more technologies. "Extended sources of food allow biological forms to further grow, multiply, consume more food and thus produce more heat and generate entropy. It seems reasonable to say that real life organisms are still far from the absolute maximum of the entropy production rate. In both cases, there are 'avalanches' or 'extinction events', which set back this evolution. Only if all free energy given by the Sun is consumed, by building a Dyson sphere for example, and converted into heat then a definitely stable phase of the evolution can be expected." "Intelligence, as far as we know, is inseparable from life," he adds. "Thus, to achieve artificial life or artificial intelligence, our recommendation would be to study systems which are far from equilibrium, with many degrees of freedom—many building blocks—so that they can self-organize and participate in some evolution. The entropy production criterium appears to be the guiding principle of the evolution efficiency."

www.nature.com/srep/2015/15020 … /full/srep08323.html

Physics

Artificial solid fog material creates pleasant laser light

Fabian Schütt et al. Conversionless efficient and broadband laser light diffusers for high brightness illumination applications, Nature Communications (2020). DOI: 10.1038/s41467-020-14875-z Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-020-14875-z

https://phys.org/news/2020-03-artificial-solid-fog-material-pleasant.html

Abstract Laser diodes are efficient light sources. However, state-of-the-art laser diode-based lighting systems rely on light-converting inorganic phosphor materials, which strongly limit the efficiency and lifetime, as well as achievable light output due to energy losses, saturation, thermal degradation, and low irradiance levels. Here, we demonstrate a macroscopically expanded, three-dimensional diffuser composed of interconnected hollow hexagonal boron nitride microtubes with nanoscopic wall-thickness, acting as an artificial solid fog, capable of withstanding ~10 times the irradiance level of remote phosphors. In contrast to phosphors, no light conversion is required as the diffuser relies solely on strong broadband (full visible range) lossless multiple light scattering events, enabled by a highly porous (>99.99%) non-absorbing nanoarchitecture, resulting in efficiencies of ~98%. This can unleash the potential of lasers for high-brightness lighting applications, such as automotive headlights, projection technology or lighting for large spaces. Introduction Solid-state lighting (SSL) is defined as light emitted by solid-state electroluminescence 1 . Its current power efficiency, i.e., the optical output power of the SSL device per unit input electrical power 2 , is ~70% and there is no fundamental physical reason why efficiencies well beyond 70% could not be reached 2 , 3 , 4 . SSL is thus expected to replace all conventional light sources by 2035 5 , including halogen, xenon, incandescent, and fluorescent lamps 4 , 6 , 7 , 8 . At present, light emitting diodes (LEDs) are the most efficient devices for white-light generation 2 , 3 , 6 . Their adoption is predicted to achieve a 75% reduction of energy consumption for lighting by 2035 5 in the US alone, which would result in a total energy saving of 6.75 × 10 16 TJ (equivalent to nearly $630 billion in avoided energy costs) and thus drastically reduce greenhouse emission worldwide 5 . However, the so-called “efficiency droop” still limits the operation of LEDs to very low input power densities, with current densities ~0.01 kA cm −2 2 , 9 . Consequently, for a higher light output the physical size of a LED has to be increased. In contrast to LEDs, laser diodes (LDs) can be operated at much higher current densities (>10 kA cm −2 ), with peak efficiencies close to that of LEDs 2 . This results in a higher light output per unit area, e.g., a 0.1 mm 2 LD source can produce the same amount of light as a 1 cm 2 LED. Hence, the target to generate more photons at high-power densities (kW cm −2 ) and decrease the cost per lumen can only be satisfied by using LDs 2 , 4 , 8 . State-of-the-art LD-based lighting devices exploit a blue LD pumping, e.g., a yellow-light emitting phosphor, resulting in white light (Fig. 1a ) 2 , 4 . However, the performance of such systems is strongly limited by the properties of the phosphor. The efficiency of state-of-the-art light emitting phosphors, such as doped yttrium-aluminum-garnet, is mainly determined by two types of energy losses, the stokes shift (~80% efficiency) and the photoluminescence quantum yield (~90%) 10 . Both these loss mechanisms scale with temperature 10 (e.g., as a result of illumination) and therefore phosphor luminescence suffers from saturation 10 , aging 11 , and thermal quenching 10 , limiting the irradiance to ~5 kW cm −2 and thus the overall light output. Even though new concepts such as glass encapsulation 12 , 13 , phosphor monoliths 14 , or composite ceramic phosphors 15 , 16 , 17 can increase the irradiance level up to ~10–20 kW cm −2 , the true potential of lasers for high-brightness lighting applications, with possible light outputs of several MW cm −2 , still remains unemployed. Fig. 1: Schematics of laser-based lighting concepts. a White-light generation by employing a remote phosphor that converts a part of the blue laser light into yellow light resulting in white light. b White-light generation based on an artificial solid fog in combination with an R+G+B laser system. A macroscopically expanded porous (>99.99%) network of interconnected and hollow hBN microtubes with nanoscopic wall thickness is used to convert directed laser light into an isotropic high-brightness white light source, exploiting multiple light scattering. c Schematic comparison of efficiencies of both systems. In the case of remote phosphor, light conversion results in a strong efficiency reduction, whereas the negligible absorption and conversionless light-scattering properties of the hBN foam allow for almost zero losses in light intensity. Full size image Here, we demonstrate a tunable, disordered, cubic centimetre-sized ceramic nanoarchitecture as an efficient (> 98%) broadband (> 450–640 nm) diffuser, that in combination with a RGB (red-green-blue) laser system (Fig. 1b ), is an alternative to the conventional used phosphors with a single laser (Fig. 1a ). The diffuser withstands ~10 times the irradiance level achievable by state-of-the-art phosphors, enabling a lighting system whose efficiency is mainly determined by that of the LDs used, due to the lack of any conversion effects (Fig. 1c ). The concept is based on a highly porous (> 99.99), macroscopic, and translucent network of randomly arranged and interconnected hexagonal boron nitride (hBN) hollow microtubes, that we call Aero-BN. The material acts like an artificial solid fog, but with a defined hierarchical internal structure - a combination of well separated feature sizes greater than, equal to, as well as below the magnitude of the impinging wavelength. The Aero-BN diffuser enables an isotropic light distribution from a multitude of coherent laser light sources at the same time, while simultaneously reducing speckle contrast to values well below the detection limit of the human eye (< 4%) 18 . Especially the latter is a strict requirement for LD-based lighting, that is not met by today’s commercially available diffuser systems (Supplementary Note 1 and Supplementary Table 1 ). Even though the current state of LD technology - with laser efficiencies < 20% for green 19 and < 40% for blue 2 , 19 - is still limiting the application of LD-based lighting systems, fast progress in the development of more efficient laser diodes is expected in the near future 2 , 3 , 4 , 8 . Therefore, the development of new optical components, such as the Aero-BN discussed here, is a necessity, indicating a way to unlock the full potential of LDs for high-brightness illumination, such as needed in projector technology, automotive headlights, large room illumination, and sports lighting. Results Light diffuser based on interconnected hBN microtubes The laser light diffuser is based on a macroscopically (> mm 3 ) expanded nanoarchitecture consisting of interconnected nanoscopic hBN films (thickness < 25 nm) in the form of hollow tubes, see Fig. 2 . hBN has a large band gap of up to 6.5 eV 20 , ensuring low (< 1%) absorption coefficients in the visible light regime. Optical transmission up to 99% at 250–900 nm was reported for thin (1–2 nm) hBN films 21 . Our synthesis process (Supplementary Fig. 1 ) is based on a ceramic template material (Supplementary Fig. 2 ) 22 , which offers, in contrast to the common Ni templates 23 used for the synthesis of hBN and graphene foams, fabrication flexibility, as the template can be tailored 22 in its density, microstructure (e.g., pore size and pore interconnectivity) as well as geometry. It consists of randomly distributed, interconnected ZnO microrods, with large (up to 100 μm) voids and porosities up to 98% 22 . The synthesis of the final BN network involves a one-step transformation of the ZnO microrod structure in which a thin (< 25 nm) hBN layer is formed by a chemical vapor deposition (CVD) process, while the ZnO template is simultaneously removed (Supplementary Figs. 3 and Supplementary Note 2 ). The final semitransparent Aero-BN (porosity > 99.99%) microtube network is shown in Fig. 2a . Calculations indicate that the specific surface area of the hBN foams is in the order of 900 m 2 g −1 (see Supplementary Note 3 ). Energy dispersive X-ray spectroscopy (EDX, Supplementary Fig. 5 ) show that the ZnO template is completely removed during CVD. The process results in a disordered 24 macroscopic network, Fig. 2b , consisting of interconnected hollow hBN microtubes, with individual features varying in well-defined sizes and dimensions. The as-synthesized hollow hBN microtubes have an average length ~25 µm, and their diameter, d tube , is between 300 and 3000 nm, depending on the geometry of the used ZnO microrods (Fig. 2c) as shown in Fig. 2d–f . Thus, d tube is of the same order of magnitude as the wavelength of visible light. The hBN CVD process results in wall-thicknesses d wall < 25 nm. This is much smaller than the wavelength of visible light, promoting light-matter interactions that are dominated by Rayleigh-type scattering 25 . As for Fig. 2f , the hBN microtube walls consist of randomly arranged, interconnected hBN nanoplates (see also Supplementary Fig. 6 ). The average distance between the individual microtubes, d MT , is several μm, larger than the visible light wavelength. The resulting Aero-BN network architecture leads thus to an optical system with microscopic (optical) density fluctuations (volumes filled with air and with hBN microtubes) throughout the macroscopic structure, as indicated in Fig. 2b . The CVD process is similar to that used to prepare macroscopically expanded nano architectures based on interconnected ZnO microrod networks 26 , 27 , 28 , with the main difference that the hBN is grown here by using the sacrificial ceramic template. Fig. 2: Artificial solid fog. a Photographs of Aero-BN. A thin (< 25 nm) hBN layer is grown by CVD using macroscopically expanded templates of tetrapodal ZnO microparticles. The hBN layer encloses the entire template structure, while it is simultaneously removed by hydrogen etching, resulting in a free-standing, low density (< 1 mg cm −3 ) network, consisting of interconnected hollow hBN microtubes. b The structure resembles an artificial solid fog, i.e., a highly optically disordered (completely randomised) photonic system with a combination of feature sizes greater than, equal to, or well below the impinging light wavelength. c Representative scanning electron microscopy (SEM) micrographs of the ZnO template consisting of interconnected microrods. d – f SEM micrographs of the resulting Aero-BN after CVD. The microtubes have an average length ~25 μm, diameter between 300 and 3000 nm, and < 25 nm wall thickness. Full size image Figure 3a shows Raman spectra ( λ = 514 nm, 1.32 mW) of Aero-BN (blue curve) and ZnO (red curve). The Aero-BN spectrum has a characteristic single peak centred ~1366 cm − 1 29 , 30 , 31 . The ZnO spectrum shows several resonances. The sharp peak ~439 cm − 1 indicates the crystal quality of the sample 32 . The peak at ~335 cm −1 is assigned to the difference between \(E_2^{{\mathrm{high}}}\) and \(E_2^{{\mathrm{low}}}\) [ \(E_2^{{\mathrm{high}}}\) − \(E_2^{{\mathrm{low}}}\) ], which corresponds to the high and low longitudinal optical branches of ZnO, while the feature at 384 cm −1 is assigned to A 1 (TO) mode 33 . In addition, the black curve shows a peak at ~584 cm − 1 attributed to E 1 (LO) mode. The broad, intense peak at 1158 cm − 1 , which is found between the doubled frequencies measured for the A 1 (LO) and E 1 (LO) modes, contains contributions of 2A 1 (LO) and 2E 1 (LO) modes at the Γ point of the Brillouin zone, and possibly also of 2LO scattering 33 . The weaker peak ~1105 cm −1 can be attributed to 2LO at H and K point of the Brillouin zone 33 . However, no peaks of ZnO are observed in the Aero-BN spectrum, consistent with the removal of the sacrificial ZnO template. Fig. 3: Raman and EELS characterisation of the Aero-BN network. a Raman spectra of Aero-BN structure (blue) and ZnO template (red). b Low-loss EELS spectra of bulk-hBN 37 (dash-point), Aero-BN (solid), and double layer hBN 37 (dashed). The positions and shapes of the π-plasmon at ~6 eV match. The positions of the σ-plasmon ~15 eV match, while shape and relative intensity differ slightly, whereas no peak ~26 eV (bulk-BN) is seen. Spectra are normalized from the onset of the π-plasmon to its apex. Full size image Transmission electron microscopy (TEM) investigations reveal, that the atomic structure of Aero-BN resembles that of hBN nanotubes (see Supplementary Note 4 and Supplementary Fig. 7 ) 34 . Furthermore, high-resolution (HR) micrographs show the existence of numerous point and triangle defects, potentially advantageous for catalytic applications 35 (Supplementary Fig. 8 ). The wall thickness of the BN microtubes is determined via the electron energy-loss spectroscopy (EELS) log-ratio method 36 (Supplementary Fig. 9 and discussion) to be 4–25 nm. The EEL spectra in the plasmon region up to 40 eV are shown in Fig. 3b and compared with those of double hBN layer 37 . The positions of the π-plasmon at 6 eV and the σ-plasmon at 15 eV match the double hBN layer reference 37 . This confirms the nanoscale thickness of Aero-BN microtube walls, as bulk hBN shows σ-plasmon resonance peaking ~26 eV 37 . Optical absorption measurements with an integrating sphere (Supplementary Fig. 10 ) performed on a macroscopic Aero-BN sample ( ρ Aero-BN ~ 0.68 mg cm −3 ) give absorption ~4.04, 0.85, and 0.11% for blue (450 nm), green (520 nm), and red (638 nm) laser lights, respectively. The slightly larger absorption at 450 nm might be caused by traces of ZnO, however the amount is too low to affect the measurements critically, as the measured absorption is consistent with that of 1–2 nm thick hBN structures 21 . The low absorbtion in combination with the structural feature sizes greater than, equal to, as well as below the magnitude of the impinging wavelength, results in a disordered system 24 , in which the light transport properties are determined by multiple light scattering. Light-scattering characteristics In order to analyze the light-scattering properties and to determine the underlying mechanisms we fabricate Aero-BN with different densities ρ Aero-BN (0.17–0.68 mg cm −3 ) by changing the initial template density ρ T between 0.3 and 1.2 g cm −3 (see Supplementary Fig. 11 ). This enables us to tune and control the internal light-scattering properties, key to build the envisaged laser light diffuser. For example, a template density ρ T ~300 mg cm −3 results in ρ Aero-BN as low as ~0.17 mg cm −3 (equal to a porosity > 99.99%), lower than that of other reported macroscopically expanded BN architectures 34 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 (see Supplementary Table 2 ). The light-scattering properties are demonstrated by illuminating an Aero-BN sample from one side with a focused laser. Figure 4a shows a photograph (perpendicular to the laser axis) of a low density ( ρ Aero-BN of ~0.17 mg cm −3 ) Aero-BN illuminated at 520 nm. The resultant frontal photograph of the same sample illuminated in the centre of a semitransparent glass bulb, Fig. 4b , shows that most of the incident laser beam is transmitted through the material. Figure 4c, d display the corresponding photographs of a sample with a higher initial ρ Aero-BN ~0.68 mg cm −3, for 520 nm illumination at 100 mW. As shown in Fig. 4d , a nearly homogeneous, isotropic light distribution, with no visible transmitted primary beam, is seen at the semitransparent glass bulb screen. The corresponding intensity plots, Supplementary Fig. 12 , obtained from Fig. 4a (highlighted areas) indicate that the intensity of the incident laser beam decreases linearly through the sample along x and y directions. This can be described by a system in which the scattering mean free-path l * is much larger than the sample dimensions 46 , resulting in an overall low scattering efficiency (most light is transmitted rather than scattered), i.e., ratio of scattered to transmitted light, and a dominating ballistic transmission. For the ZnO microrod template used to prepare our Aero-BN, with the same microstructure (microrods instead of hollow microtubes), a pronounced visible backscattering is observed (Supplementary Figs. 13 – 15 ), indicating the fundamental role of the hollow tubular geometry with multiple feature sizes. A more detailed discussion on the influence of different ceramic microstructural arrangements can be found in Supplementary Note 5 and Supplementary Fig. 16 . Fig. 4: Light-scattering characterisation. a Photograph of ρ Aero-BN ~0.17 mg cm −3 illuminated with 100 mW (spot size ~1 mm) at 520 nm. b Photograph of the same sample shown in a when illuminated in the centre of a semitransparent glass bulb (front view). c , d Photograph of ρ Aero-BN ~0.68 mg cm −3 illuminated with 100 mW at 520 nm, and resultant light scattering imaged using a semitransparent glass bulb (front view). e Angular photocurrent dependence for Aero-BN with different ρ Aero-BN compared with an interconnected microrod structure (t-ZnO; ρ T ~300 mg cm −3 ) for 520 nm at 100 mW. The photodiode is polar rotated over the sample, as illustrated in the schematics. The graphs represent the photocurrent produced by scattered light only. f Corresponding normalised photocurrent with respect to the azimuthal rotation of the photodiode. For details of measurements see Supplementary Figs. 17 – 20 . Full size image The detailed light distribution produced by illuminating Aero-BN samples is investigated with a photo-goniometer 47 (a photodiode movable around the illuminated specimen on a spherical surface, see Supplementary Fig. 17 ) to characterise the broadband light-scattering properties as a function of the angle (azimuthal and polar rotation) at 450, 520, and 638 nm, Fig. 4e, f . We also consider a network of interconnected ZnO microrods 48 , 49 as a comparison. A perfect 3D light diffuser exhibits angle independent (isotropic) emission over the complete angular range, so that the light is uniformly emitted in all directions. Figure 4e, f show plots of both azimuthal and polar rotations, extracted from the polar plots of the goniometer measurements of the laser illuminated ZnO and Aero-BN networks (see Supplementary Figs. 18 – 20 ). In both cases, the photodiode is pivoted, while the sample and the LD are stationary. These graphs provide quantitative data for the amount of scattered, reflected, and transmitted light. Figure 4e depicts the photocurrent for different samples as a function of polar angle. This represents the photocurrent produced by scattered light only, while no reflected and transmitted light reaches the detector. In contrast, Fig. 4f shows the normalised photocurrent as a function of azimuthal angle. In this case, the photocurrent detected for 90° < γ < 180° is caused by scattering only. The value at γ = 90° represents transmission ( T ) and forward scattering. For γ > 180° the photocurrent is a result of reflection and scattering. As depicted in Fig. 4e , the ZnO network with ρ ZnO ~300 mg cm −3 shows only a small but homogeneous photocurrent (~1.5 µA) caused mainly by back-scattered light. Thus, nearly no light is transmitted through the structure, Fig. 4f . Aero-BN, on the other hand, shows a much stronger emission and more uniform light distribution of the laser beam. The measured photocurrent caused by azimuthally scattered light from the Aero-BN (Fig. 4e ) is ~2.5–4 µA and 6–7 µA for ρ Aero-BN ~0.17 and ~0.68 mg cm −3 , respectively. Thus, for the higher density sample, the amount of scattered light is ~4.6 times higher than for the ZnO network, even though the density of the network structure is reduced by a factor of ~440. As illustrated in Fig. 4f , the amount of light transmitted ( γ = 90°) through ρ Aero-BN ~0.68 mg cm −3 is ~4 times higher than the reflected (and scattered) light ( γ > 180°). The ratio between transmitted and scattered light (ideal value of 1 for an isotropic diffuser) decreases with increasing network density. For ρ Aero-BN ~0.17 mg cm −3 this is ~200, whereas it is ~3.7 for ρ Aero-BN ~0.68 mg cm −3 . A value of 1 might be achieved by increasing ρ Aero-BN further. The average scattering intensity S is calculated by averaging the photocurrent intensities of the polar plots for 105° < γ < 170° and 100° < δ < 230°. The relative deviation of S with respect to T ( γ = 90°) is illustrated in Supplementary Fig. 21 as a function of the optical areal density, i.e., the density times the sample length, for three wavelengths. By increasing the optical areal density ( ρ Aero-BN × L ), the ratio ( T-S ) /S by over three orders of magnitude, irrespective of the wavelength. Light with a shorter wavelength is scattered more effectively, as for Rayleigh scattering 50 , 51 (see Supplementary Fig. 22 ). Further details on the light-scattering properties are in Supplementary Note 6 and Supplementary Figs. 23 – 27 , showing that the multiple light scattering observed in Aero-BN is a result of the combination of negligible absorption losses and a control of density of scattering centres over several orders of magnitude. Furthermore, we show independent tunability of the density in all three dimensions (given the almost null equivalent Poisson’s ratio of such low-density foam materials 52 , see also Supplementary Fig. 24). This enables control of light diffusion and a nearly constant density of photons close the surface, with at most a linear decay in one dimension. Speckle contrast reduction The scattering behaviour also enables us to use Aero-BN for laser illumination without recognizable speckle patterns, thus solving one of the main challenges of using LDs as a light source 18 , 53 , 54 . Speckle is the result of interference of light beams with the same frequency, but different phase and amplitude, resulting in a wave with random amplitude variations 55 . The most promising approach to avoid speckle is to use an optical downstream component that superimposes multiple speckle patterns at once 18 , 53 , 56 , so that on average no pattern is visible to the human eye, for an exposure time ~1/60 s 57 . In our Aero-BN, the primary laser beam is scattered multiple times. Thereby it is split into a large number of independent beams, causing multiple overlapping speckle patterns. This reduces the objective speckle contrast χ (i.e., the mean intensity of the speckle pattern divided by the standard deviation of the intensity) down to ~2%, lower than that for the human eye (4%) 18 . Figure 5 plots the objective speckle pattern for different wavelengths as a function of material and Aero-BN density ( ρ Aero-BN ). For high-density Aero-BN ( ρ Aero-BN ~0.68 mg cm −3 ) the speckle contrast is lowest, with minimal values ~2.96%, 1.52%, and 5.52% for 450, 520, and 638 nm (each at 100 mW), respectively. Therefore, nearly no speckle can be observed by the human eye. Even lower speckle contrast could be achieved by using higher ρ Aero-BN. Our Aero-BN outperforms commercially available plate diffusers like DG10-220 ( Thorlabs ) in terms of speckle contrast, since these have >5 times higher speckle contrasts (16%). It also surpasses that of the interconnected ZnO microrod networks, as no pure light diffusion can be reached there (see also Supplementary Fig. 15 ), due to missing Rayleigh-type scattering centres at the nanoscale. The lower speckle contrast for Aero-BN at lower wavelengths is a direct effect of the wavelength dependence ( λ −4 ) of Rayleigh scattering 50 , 51 . Due to continuous beam splitting by multiple light-scattering processes, the low speckle contrast might also be related to small (< 50 nm) thermally activated movements of the hollow microtubes with wall thicknesses < 25 nm, resulting in a time-varying speckle pattern (see also Supplementary Note 7 and Supplementary Table 4 ). This is similar to the speckle contrast reduction obtained by using colloidal dispersions, with values as low as 3% due to particles Brownian motion 18 . In comparison with other methods to reduce speckle contrast, e.g., by random lasing 58 , using small moving diffusers 53 , rotating ground glass diffusers 59 or nonmoving Hadamard matrix diffusers 60 our approach does not require complex micromechanical devices, making it easier to use and less prone to failure (see also Supplementary Note 1 and Supplementary Table 1 ). Fig. 5: Speckle contrast reduction. Objective speckle pattern at 450, 520, and 638 nm for two Aero-BN samples with high ( ρ Aero-BN ~0.68 mg cm −3 ) and low ( ρ Aero-BN ~0.17 mg cm −3 ) density, a porous ZnO microrod network ( ρ T ~300 mg cm −3 ) and a commercial plate diffuser. Values for speckle contrast are in %. Full size image Tunable RGB laser light illumination The viability of our Aero-BN in combination with an RGB laser system as an illumination source, as an alternative to remote phosphors, is demonstrated by illuminating the ρ Aero-BN ~ 0.68 mg cm −3 sample at different laser intensities under a translucent glass sphere screen. The resulting images are presented in Fig. 6a together with the respective International Commission on Illumination (CIE) colour space values marked in the colour map of Fig. 6b . An all-primary RGB laser wavelength mixing approach, i.e., a combination of three (red, green, and blue) or even four (red, yellow, green, and blue) laser wavelengths is known to outperform the efficiency of any other known white-light source 2 , 3 , 4 , 8 . Furthermore, the possible colour gamut (i.e., the subset of colours which can be accurately represented) of such a system is on par to that of LEDs or LCDs 61 . By tuning the individual intensities of our RGB laser source, all colours in the resultant RGB triangle (Fig. 6b ) can be produced. For the maximum intensity of all lasers, white light is produced, close to the CIE standard white illuminant D65 62 . The corresponding photographs of ρ Aero-BN ~0.68 mg cm −3 illuminated at 450, 520, and 638 nm are in Fig. 6c , together with a photograph of the same sample illuminated with all wavelengths at once, resulting in a diffuse white-light illumination. Thus, our Aero-BN is an ideal broadband diffuser (see also Supplementary Movie 1 ) and can be used to fabricate tunable RGB laser light sources with a large colour gamut, depending only on characteristics of the actual laser system used, rather than on light conversion effects such as in the case of remote phosphors. Fig. 6: Colour mixing. a Light distribution of a high ρ Aero-BN sample illuminated in the middle of a translucent glass bulb under different intensities for the each wavelength (450, 520, and 638 nm). White light is produced when all lasers (R + G + B) are at the maximum power (100 mW). b CIE colour map with marked values for the pictures in a . The value obtained for mixing R + G + B is close to the CIE standard white illuminant D65 62 (yellow circle). c Photographs of a sample with ρ Aero-BN ~0.68 mg cm −3 illuminated at 450, 520, and 638 nm (100 mW each, 1 mm spot size), respectively, as well as the resultant white light produced if all lasers are used at once. The arrows mark the direction of the incident laser beam. Full size image Laser damage threshold To demonstrate that Aero-BN can overcome the irradiance levels of state-of-the-art phosphors needed for high-brightness illumination applications, such as automotive headlights or projectors, we characterised its thermal decomposition and laser damage threshold. Thermogravimetric analysis (TGA) under nitrogen atmosphere indicates nearly no change in weight (±2 wt% up to 1000 °C). In an oxygen-containing atmosphere (nitrogen/oxygen ~1/4) the material is stable up to 700 °C, where the formation of B 2 O 3 starts 63 (Supplementary Fig. 28 ). The chemical reaction also confirms the presence of hBN over other crystalline forms of BN such as wurtzite boron nitride (wBN) and cubic boron nitride (cBN) 64 . To determine the laser damage threshold we use a focused (spot diameter ~8.4 µm) high-power (3 W) continuous wave laser at 450 nm. The threshold is determined by moving the focused laser beam over an individual tube and simultaneously recording the microtube with a charge-coupled device (CCD) camera. After each passage, the laser power is increased until the laser induces morphological damage (e.g., microtube destruction, see Supplementary Fig. 29 ). However, even at the highest power (~650 kW cm −2 ) the Aero-BN network remains intact, whereas a commercially available state-of-the-art phosphor shows degradation at ~80 kW cm −2 (see Supplementary Fig. 30 ). In contrast to Aero-BN, the phosphor actively converts the incident laser light into energy, which leads to increased heat accumulation. To achieve even higher power densities we use a highly focused pulsed laser (spot diameter ~1 µm) at 355 nm, with 100-Hz repetition rate ~7-ns pulse duration (see Supplementary Fig. 29 ). In this configuration the Aero-BN shows a high laser damage threshold ~430 MW cm −2 (~2.65 J cm −2 ), ~10 times higher than commercially available phosphor (see Supplementary Fig. 31 ) 12 , 13 , 17 . This is directly related to the microscopic structure of the Aero-BN. The nanoscopic wall thickness leads to high transmittance of the individual tubes, meaning that only a small portion of the laser light interacts with a single tube. Furthermore, the low hBN absorption in the visible spectrum 21 , implies that a minimal amount of energy is transformed into heat. The high heat conductivity (~400 W m −1 K −1 ) 23 of hBN helps to quickly transport thermal energy away from the illuminated spot 65 . The high porosity (> 99.99%), the small wall thickness < 25 nm, as well as the micrometre-sized voids enable efficient heat management, similar to that reported for other foam-like nanostructures, such graphene foams 66 , 67 , since heat can be easily transported to the surrounding air. Furthermore, the volumetric heat capacity of our Aero-BN foam is comparable with that of the surrounding air, as both have similar densities ( \(\rho _{{\mathrm{air}}}\sim 1.2\,{\mathrm{mg}}\,{\mathrm{cm}}^{ - 3};0.17\,{\mathrm{mg}}\,{\mathrm{cm}}^{ - 3}\,<\,\rho _{{\mathrm{Aero}} - {\mathrm{BN}}}\,<\,1\,{\mathrm{mg}}\,{\mathrm{cm}}^{ - 3}\) ). Therefore, the damage threshold is only an estimate for the lower destruction limit. The macroscopic destruction threshold of Aero-BN is potentially much higher when a macroscopic laser beam is used, not focused to such a small spot. Discussion We demonstrated a concept for high-brightness and broadband laser illumination based on a diffuser consisting of a network of interconnected hollow hBN microtubes, overcoming the problems associated with inorganic phosphor materials. Their structurally disordered arrangement, combined with the nanoscopic wall thickness, and low absorption are key to enable homogeneous light diffusion through the cm 3 -sized networks and promote light-scattering properties suitable for laser illumination applications. Our Aero-BN has efficient Rayleigh-type scattering centres arranged in thinly spread and controlled manner, resulting in non-exponential light diffusion. By controlling the density of the aero-material system, we are able to adjust the light diffusion so that multiple scattering events result in an almost homogeneous, isotropic light illumination. For Aero-BN densities ~0.68 mg cm −3 the speckle contrast is well below the perception threshold of the human eye. The highly porous structure, together with the low absorption in the visible range, as well as the low volumetric heat capacity and high heat conductivity, enable an efficient heat management. We achieve laser irradiance levels ~10 times higher than commercially available remote phosphors, unleashing the full potential of laser diodes for high-brightness illumination. Being based on multiple light scattering, rather than on light conversion effects, the broadband properties of our diffuser enable an all-primary RGB laser approach for white-light generation and full-colour range mixing with a large colour gamut 8 , without efficiency reduction, thereby overcoming the problems associated with state-of-the-art remote phosphors (see also Supplementary Note 8 ). With the expected increase in LD efficiencies in the near future, our concept paves the way to design a new generation of highly effective light sources. Methods Fabrication of highly porous ZnO networks The t-ZnO ceramic networks are produced by a flame transport synthesis technique 68 . Zinc powder with a grain size ~1–10 µm is mixed with polyvinyl butyral in a mass ratio of 1:2. The mixture is then heated in a muffle furnace at 60 °C min −1 to 900 °C for 30 min. After that a loose powder of ZnO tetrapods is obtained, then pressed into pellets (e.g., height ~10 mm, diameter ~12 mm) with a density ~0.3 g cm −3 . Reheating the pellets for 5 h at 1150 °C leads to junctions between the tetrapods and an interconnected network. Fabrication of Aero-BN In Supplementary Fig. 3 the computer-controlled CVD setup for the fabrication of the Aero-BN is illustrated. The highly porous (up to 98%) ZnO ceramic template is placed in the middle of a quartz tube furnace in a ceramic crucible. Next to that (~1 cm), a crucible filled with B 2 O 3 is placed into the furnace. The reactor is flushed with Ar and the pressure adjusted to 30 mbar. The Ar flow is then adjusted to 30 sccm and the temperature is increased to 910 °C (heating rate ~20 °C min −1 ). Urea is used as a nitrogen source, located in an evaporator which is connected to the quartz tube furnace as illustrated in Supplementary Fig. 3 . When the quartz tube furnace reaches 910 °C the evaporator for urea is switched on. By heating to 170 °C at 30 mbar NH 3 forms 69 , which decomposes to N and H 2 in the reaction zone of the reactor 70 . At the process temperature (910 °C) N and B react at the surface of the ceramic template, forming a thin (<25 nm) hBN layer. Simultaneously the ZnO template is etched by hydrogen. After 3 h the urea evaporator and the quartz tube furnace are switched off. When the reactor reaches 30 °C the Ar flow is switched off and the sample is removed. A detailed discussion of the reaction is in Supplementary Note 2 . Characterisation The morphologies of the different structures are investigated by SEM (Zeiss Supra 55VP) equipped with an EDX detector. Aero-BN is analysed by a FEI Tecnai F30 G2 STwin TEM (300 kV acceleration voltage, cs-coefficient 1.15 mm) and a FEI Titan G2 60-300 TEM equipped with a monochromator. Macroscopic aggregates of Aero-BN are tapped with TEM grids in order to transfer some tetrapods or single fragments onto the grid, minimising the breaking rate for the Aero-BN network. Unfolded BN sheets are also analysed by HRTEM to visualise atomic scale defects. The electronic structure is investigated by HR-EELS with a GIF Quantum/Enfina energy analyser. TGA measurements are performed using a TA Instruments Q50 under nitrogen and nitrogen/oxygen (1/4) at a scan rate of 10 °C min −1 from 25 to 1000 °C. Raman spectroscopy is done with a Renishaw 1000 InVia micro-spectrometre at 514.5 nm for the ZnO template and a Witec Instruments Alpha300 RA at 532 nm for the Aero-BN sample. Reflectivity calculations Reflectivity calculations as a function of wall thickness for a hollow hBN microtube and ZnO microrods as a function of diameter for different wavelengths, respectively, follow those in ref. 71 . Refractive indexes of 1.8 20 and 2.1 72 are used for hBN and ZnO. The mean reflectivity is derived by averaging that for incident beam angles of 0–180° (step size of 1°). For each angle the unpolarised and polarised reflectivity is derived. This procedure is repeated for different hBN wall thicknesses as well as ZnO microrod diameters. Light-scattering measurements A photodiode (FDS1010, Thorlabs) is rotated around the sample with an angular step ~5° at a distance ~15 cm, using a photogoniometer. From one side the cylindrical samples are illuminated with an RGB laser (RTI OEM 300 mW RGB Modul, LaserWorld). The sample is positioned so that the laser beam illuminates it in the middle. The spot size is adjusted by a lens to ~1 mm. Each laser has a maximum output power of ~100 mW. Absorption measurements Absorption measurements are performed using an integrating sphere (Opsytec) with an inner diameter of 200 mm, coated with a reflective BaSO 4 thin film. The illumination intensity is measured by connecting to it a radiometer (RM-22, Opsytec). The sample is mounted on a thin (diameter of 3 mm) Al slab in the centre of the sphere. Through an opening of 2 mm, the laser is focused on the sample. The absorption is calculated as the ratio of the luminous flux measured by the radiometer with and without sample. This is integrated for at least 20 s. Transmission measurements Transmission measurements are performed using the same integrating sphere used for absorption. The sample is placed in front of a 2 mm opening of the sphere. The laser is adjusted to be in the same axis as the opening of the sphere and focused on the sample. The transmission is calculated as the ratio of the measured luminous flux with and without sample. For measurements as a function of compression, the sample is clamped between two highly reflective (>99%) plates to ensure as little light absorption as possible by the surrounding (clamping) material. The sample is compressed step by step using a high-precision screw. After each compression, a transmission measurement is performed as described before, using an integration time of at least 20 s. This is increased to 60 s for small fluxes. Laser damage threshold The sample is moved using a xy -translation stage, such that the laser beam directly hits an individual nanostructure, e.g. a microtube. The laser focus is adjusted using the back-scattered signal of the laser spot, tuned towards its highest intensity by moving the translation stage in z -direction. The laser signal is then filtered on a video camera by using a notch filter, while only the microscope image is monitored. The laser power is increased stepwise until the first morphological changes of the nanostructures become evident in the microscope (white light) image. The corresponding laser intensity (on the sample) defines the destruction threshold of the investigated materials. The commercially used phosphor is a Intematix CL830R45XT. Speckle pattern photography and contrast Objective speckle patterns (i.e. the intensity pattern produced by the interference of a set of wavefronts) are obtained by illuminating with a focused laser beam with 100 mW at 450, 520, and 638 nm. The objective speckle pattern forms on a sheet of white paper at 90° with respect to the incoming laser. The distance between sample and speckle pattern is ~40 cm. The pattern is photographed using a CCD camera (Nikon D300) equipped with a lens with 120 mm focal length. The camera is positioned slightly over the sample, to avoid any light being directly scattered into the lens. The aperture of the lens is used at maximum of f/4 to take as much light in as possible. Since speckle patterns are time dependent 56 , the exposure time is important. We use 1/60 s, close to the detection limit of the human eye 57 . To avoid any overexposure of the CCD chip we use a camera sensitivity (ISO) of 800. For weakly scattering samples, this might lead to a dark speckle pattern. However, this has no influence on the speckle contrast, whereas an overexposure would result in wrong calculations. The photographs are taken at a maximum resolution of 2848 × 4288 pixels. The speckle contrast of the resultant photographs is calculated by using the Gatan Microscopy Suit. A representative quadratic area (several cm 2 ) is chosen. The colour is converted into a black and white representation. From these images the mean intensity Φ as well as the standard deviation σ is calculated using the above mentioned software. The speckle contrast χ is than calculated as follows 56 : $$\chi\,=\,\frac{\sigma }{\Phi }.$$ FEM simulations The FEM model developed to compute the network variation of projected porous areal density and Poissonʼs ratio under monoaxial compression (see Supplementary Note 6), consist of a periodic supercell ~71 × 83 × 46 μm 3 ( x , y , z ) containing nine tetrapods mutually interconnected. We consider an average geometry of the tetrapod with d air,1 = 1.67 µm, d air,2 = 1.00 µm, t wall = 4 nm, and r = 27 or 38 µm 73 to simulate networks with high or low densities, (respectively ρ Aero-BN = 0.367 mg cm −3 and ρ Aero-BN = 0.178 mg cm −3 ), similar to the ones tested in the experiments. Tetrapods are built associating the arm extremities and the central joint of the tetrapods to the vertexes and centroid of a regular tetrahedron, respectively. The tube walls are modelled with thin shell elements with selective-reduced integration, and the spurious modes effects are controlled. Monoaxial compression tests are reproduced with periodic boundary conditions along the lateral faces of the supercell ( x and y directions) while the two horizontal rigid surfaces act to apply the monoaxial load on the network along z (displacement controlled, 0.25 μm ms −1 ). Contact between tetrapods and within elements of the same tetrapods are implemented to prevent mutual and self-penetration. The density of the supercell is monitored along the simulations. To measure the evolution of the projected porous area, images of lateral view of the network ( xz and yz planes) are extracted from simulations at a constant time sampling. The normalised projected porous area (Ω/Ω 0 ) is measured via a graphics software (paint.net) by selecting the void area in the lateral projection fo the network (“magic wand” tool) and computing the corresponding number of pixels (ratio of the current vs. initial value). Data availability The data that support the findings of this study are available from the corresponding authors upon request.

With a porosity of 99.99 %, it consists practically only of air, making it one of the lightest materials in the world: Aerobornitride is the name of the material developed by an international research team led by Kiel University. The scientists assume that they have thereby created a central basis for bringing laser light into a broad application range. Based on a boron-nitrogen compound, they developed a special three-dimensional nanostructure that scatters light very strongly and hardly absorbs it. Irradiated with a laser, the material emits uniform lighting, which, depending on the type of laser, is much more efficient and powerful than LED light. Thus, lamps for car headlights, projectors or room lighting with laser light could become smaller and brighter in the future. The research team presents their results in the current issue of the renowned journal Nature Communications, which was published today. More light in the smallest space In research and industry, laser light has long been considered the "next generation" of light sources that could even exceed the efficiency of LEDs (light-emitting diode). "For very bright or a lot of light, you need a large number of LEDs and thus space. But the same amount of light could also be obtained with a single laser diode that is one-thousandth smaller," Dr. Fabian Schütt emphasizes the potential. The materials scientist from the working group "Functional Nanomaterials" at Kiel University is the first author of the study, which involves other researchers from Germany, England, Italy, Denmark and South Korea. Powerful small light sources allow numerous applications. The first test applications, such as in car headlights, are already available, but laser lamps have not yet become widely accepted. On the one hand, this is due to the intense, directed light of the laser diodes. On the other hand, the light consists of only one wavelength, so it is monochromatic. This leads to an unpleasant flickering when a laser beam hits a surface and is reflected there. Bornitride, on which the new light material is based, is also called "white graphene" because of its similar atomic structure. Credit: Julia Siekmann, CAU Porous structure scatters the light extremely strongly "Previous developments to laser light normally work with phosphors. However, they produce a relatively cold light, are not stable in the long term and are not very efficient," says Professor Rainer Adelung, head of the working group. The research team in Kiel is taking a different approach: They developed a highly scattering nanostructure of hexagonal boron nitride, also known as "white graphene," which absorbs almost no light. The structure consists of a filigree network of countless fine hollow microtubes. When a laser beam hits these, it is extremely scattered inside the network structure, creating a homogeneous light source. "Our material acts more or less like an artificial fog that produces a uniform, pleasant light output," explains Schütt. The strong scattering also contributes to the fact that the disturbing flickering is no longer visible to the human eye. The nanostructure not only ensures that the material withstands the intense laser light, but can also scatter different wavelengths. Red, green and blue laser light can be mixed in order to create specific color effects in addition to normal white—for example, for use in innovative room lighting. Here, extremely lightweight laser diodes could lead to completely new design concepts in the future. "However, in order to compete with LEDs in the future, the efficiency of laser diodes must be improved as well," says Schütt. The research team is now looking for industrial partners to take the step from the laboratory to application. Within the fine network of hollow tubes measuring only a few micrometers in size incident laser beams are so strongly scattered that a homogeneous white light is produced. Credit: Kiel University Due to its inner structure, the material can scatter different wavelengths, i.e. green, red and blue laser light. Credit: Fabian Schütt Wide range of applications for aeromaterials Meanwhile the researchers from Kiel can use their method to develop highly porous nanostructures for different materials, besides boron nitride also graphene or graphite. In this way, more and more new, lightweight materials, so-called "aeromaterials," are created, which allow particularly innovative applications. For example, the scientists are currently doing research in collaboration with companies and other universities to develop self-cleaning air filters for aircraft.

10.1038/s41467-020-14875-z

Nano

Scientists have a blast with aluminum nanoparticles

Jennifer L. Gottfried et al. Improving the Explosive Performance of Aluminum Nanoparticles with Aluminum Iodate Hexahydrate (AIH), Scientific Reports (2018). DOI: 10.1038/s41598-018-26390-9 Journal information: Scientific Reports

http://dx.doi.org/10.1038/s41598-018-26390-9

https://phys.org/news/2018-06-scientists-blast-aluminum-nanoparticles.html

Abstract A new synthesis approach for aluminum particles enables an aluminum core to be passivated by an oxidizing salt: aluminum iodate hexahydrate (AIH). Transmission electron microscopy (TEM) images show that AIH replaces the Al 2 O 3 passivation layer on Al particles that limits Al oxidation. The new core-shell particle reactivity was characterized using laser-induced air shock from energetic materials (LASEM) and results for two different Al-AIH core-shell samples that vary in the AIH concentration demonstrate their potential use for explosive enhancement on both fast (detonation velocity) and slow (blast effects) timescales. Estimates of the detonation velocity for TNT-AIH composites suggest an enhancement of up to 30% may be achievable over pure TNT detonation velocities. Replacement of Al 2 O 3 with AIH allows Al to react on similar timescales as detonation waves. The AIH mixtures tested here have relatively low concentrations of AIH (15 wt. % and 6 wt. %) compared to previously reported samples (57.8 wt. %) and still increase TNT performance by up to 30%. Further optimization of AIH synthesis could result in additional increases in explosive performance. Introduction The effect of aluminum (Al) additives on the laser-induced plasma chemistry of the military explosive cyclotrimethylenetrinitramine (RDX) was previously demonstrated using time-resolved emission spectroscopy 1 ; increasing the Al content resulted in an increase in plasma temperature (due to the exothermic formation of AlO and Al 2 O 3 ) and an increase in C 2 /soot formation because of the oxygen-scavenging Al reactions. Exothermic reactions were subsequently shown to increase the laser-induced shock wave velocity and/or the laser-induced deflagration reaction following pulsed laser excitation of energetic materials 2 . The strong correlation between the measured laser-induced shock velocities produced by laser ablation of energetic materials and the reported detonation velocities from large-scale detonation tests provides a laboratory-scale method for estimating the fast (i.e., microsecond-timescale) energy release from milligram quantities of energetic materials 3 . The laser-induced air shock from energetic materials (LASEM) technique has recently been used to estimate the detonation velocities of novel energetic materials 4 , 5 and conventional military explosives doped with Al or boron (B) additives 6 . Slower energy release such as that from combustion reactions in air results in strong laser-induced deflagration reactions on the millisecond timescale 7 ; typically, energetic materials such as trinitrotoluene (TNT) that make good propellants have significantly higher deflagration intensities than more powerful military explosives such as RDX that produce faster detonation waves (and thus, faster laser-induced shock velocities). Therefore, TNT is a good baseline explosive to evaluate the performance of metal particle additives that typically enhance late-time blast effects but inhibit detonation velocities. Because of its high heat of oxidation, Al powder (typically 20% by weight) is commonly added to TNT to enhance the late-time blast effects of the explosion; tritonal is a typical Al-TNT formulation 8 . Since only a fraction of the available chemical energy is released from conventional metal particle formulations on the microsecond-timescale (resulting in lower detonation velocities), significant research efforts have been devoted to decreasing the reaction time of the metal additive in order to enhance the explosive performance 8 . One approach recently introduced involves chemically altering the oxidation shell of an Al particle that is initially composed of an amorphous Al 2 O 3 outer shell and crystalline Al core. When Al particles are exposed to oxygen during synthesis, an amorphous Al 2 O 3 passivation layer forms around Al particles. The Al 2 O 3 passivation layer can act as a heat sink and oxygen diffusion barrier that slows Al oxidation reactions. Capellos et al . 9 and Baker et al . 10 developed an eigenvalue detonation model to determine the effects of the Al 2 O 3 passivation layer on Al oxidation when Al powders are used as additives in explosive mixtures. Their results show that the Al 2 O 3 passivation layer surrounding Al particles delays the onset of Al oxidation such that Al oxidation does not occur in the detonation wave 9 , 10 . Reducing the limiting effects of the Al 2 O 3 passivation layer could significantly decrease the ignition delay of Al particles and significantly enhance explosive performance. The motivation for this work is to explore chemical synthesis routes that alter the Al 2 O 3 shell toward greater reactivity at microsecond timescales and are aligned with the chemical reactions behind the detonation wave. When exposed to highly acidic solutions (i.e., pH < 4), Al 2 O 3 becomes soluble and dissolves in solution 11 . The solubility of Al 2 O 3 in acidic solutions has been utilized in the synthesis of Al energetic materials to remove the Al 2 O 3 passivation layer surrounding Al particles and replace Al 2 O 3 with a reactive salt 12 , 13 . In Smith et al . 12 , 13 , it was shown that aluminum iodate hexahydrate (AIH) replaces the Al 2 O 3 passivation layer when Al particles are added to concentrated solutions of iodic acid. The Al 0 -AIH core-shell particles significantly increased reactivity with flame speeds measured as high as 3200 m/s 14 . These high flame speeds suggest the Al-AIH particles have great potential to exceed the reactive timescale limitations of Al particles when combined with explosives such as TNT. The objective of this study is to examine the energy release from novel Al 0 -AIH samples that vary slightly in AIH concentration from ~6% AIH to 15% AIH (i.e., AIH6 and AIH15) then compare to the reactions of conventional micron-sized Al particles, alumina (Al 2 O 3 ), and a physical mixture of Al/I 2 O 5 . This analysis is further extended to composites of TNT with each of these additives: micron-sized Al, Al 2 O 3 , Al/I 2 O 5 , AIH6 and AIH15. The goal is to understand AIH contribution to detonation and/or deflagration reactions using the laboratory-scale LASEM technique. Experimental Aluminum iodate hexahydrate synthesis and characterization The mixing procedure used here is similar to the procedure used previously 12 , 13 , 14 . AIH mixtures were synthesized by first dissolving commercially available I 2 O 5 , supplied by Sigma Aldrich (St. Louis, MO), in distilled water prior to mixing 80 nm average diameter Al powder supplied by Novacentrix (Austin, TX). Both AIH6 and AIH15 were mixed to an equivalence ratio (ER) of 0.9 and the water to Al ratio for AIH6 and AIH15 was 1:1 and 2:1, respectively. The XRD data for each sample was collected on a Rigaku Ultima III powder diffractometer operated in continuous θ-2θ mode from 15–60° 2θ with parallel beam geometry. The step size was 0.02° with a collection time of 2°/min. The MDI Jade V9.1.1 software provides both qualitative and quantitative data analysis. The uncertainty in the XRD measurements from peak fitting and variation between samples is calculated to be less than 6.9% for all concentrations. The skeletal density of AIH6 and AIH15 were also measured using an AccuPyc II 1340 pycnometer from Micromeritics (Norcross, GA). For each sample, 15 volume measurements were run with a nitrogen flow of 0.005 psig/min. Samples were weighed prior to volume measurements and density was calculated from weight and volume measurements. The standard deviation from pycnometer measurements is less than 0.025 g/cm 3 . All of these measured densities are above 90% of the theoretical maximum density. The oxygen balance ( OB ) is calculated in terms of 100 grams of material to determine the percent of oxygen excess or deficient for 100 grams of a compound. $$OB \% =\frac{-1600}{{M}_{c}}\ast (2X+\frac{Y}{2}+\alpha M-Z)$$ (1) In Eq. ( 1 ), M c is the molecular weight of the compound, X is moles of carbon (i.e., 0 for this mixture since carbon is not present), Y is moles of hydrogen, αM is moles of metallic oxide, and Z is the moles of oxygen. The AIH6 and AIH15 samples were further examined using the transmission electron microscopy (TEM) technique. The TEM specimens were prepared using high purity ethanol (Decon Laboratories, Inc.) as the medium via the nanoparticle suspension technique 15 . The samples were studied in a JEOL 2100FX TEM operated at 200 keV (JEOL USA, Inc.). The overall field of view varied from 10 to 100 µm 2 under TEM diffraction contrast imaging conditions with an average emission current approximately 120 µA. Fourier transform (FFT) patterns were acquired to reflect the crystallinity of the sample from corresponding TEM images. The surface chemistry of the particles was further analyzed using a Physical Electronics VersaProbe II Ultra X-ray Photospectroscopy system equipped with a hemispherical analyzer and a take-off angle of 45 degrees. The sample was irradiated by a focused beam of monochromatic Al Kα X-rays. Differential scanning calorimetry (DSC) for one AIH sample was performed using a Netzsch Model STA 449 to measure the heat flow as a function of temperature and time. Only one sample was analyzed because the analysis resulted in thermal runaway and significant damage to the instrumentation. For the DSC analysis, a 7 mg sample of AIH6 was loaded into an alumina crucible with no lid and placed in the diagnostic. The sample was heated in an argon environment at 10 K/min (KPM). Temperature calibrations for the instrument were performed using melting of a set of metal standards resulting in a temperature accuracy of ±1 °C. Laser-Induced Air Shock from Energetic Materials (LASEM) Experiments The LASEM experimental setup has been described previously 2 , 3 . Briefly, a 6-ns pulsed Nd:YAG laser (Quantel Brilliant b, 1064 nm, 900 mJ) is focused just below the sample surface with a 10 cm lens. The laser ablates the sample material into the air above the sample surface, atomizing/ionizing a fraction of the ejected material and forming a high-temperature (>10,000 K) microplasma. Because of the properties of the laser (pulse duration and wavelength), most of the energy of the laser pulse is absorbed by the plasma, which shields the sample surface. During the plasma lifetime (10’s of μs), the ablated material is highly excited, resulting in strong atomic and molecular emission spectra. The plasma rapidly cools following the laser pulse, and recombination reactions occur which result in the formation of molecular species and, if the reactions are exothermic, increase the plasma temperature. While even inert samples produce strong laser-induced shock waves due to the energy deposited by the laser in the plasma, the chemical reactions of energetic materials further increase the plasma temperature and resulting shock wave velocity. The expansion of the shock wave into the air above the sample is visualized using a typical Z-type schlieren imaging setup (10.8 cm diameter mirrors, 114 cm focal length) illuminated by a 200 W Hg-Xe arc lamp. A high-speed color camera (Photron SA5) with a zoom lens (Nikon Nikkor 24–85 mm f/2.8-4D IF) records the shock wave expansion at 84,000 frames-per-second (64 × 648 pixels, 1.0 μs shutter). The measured shock wave positions are then used to plot the laser-induced shock wave velocity as a function of time. The resulting data is fit to a polynomial; the y-intercept of the fit is the characteristic shock velocity for the sample under the given experimental conditions and is used to compare the fast energy release of the materials. For energetic materials, the measured characteristic shock velocity can be used to estimate the detonation velocity of the sample (at the theoretical maximum density) using the previously determined calibration fit 3 . Additional diagnostics for the LASEM setup used in this study included a CCD spectrometer (Ocean Optics USB4000, 200–890 nm, 100-ms integration time) to investigate the chemical reactions of the laser excited material and an infrared-sensitive photodiode (New Focus Model 2053, 900–1700 nm, 10 2 gain) to measure the extent (i.e., peak emission intensity and duration) of the combustion reactions. An exhaust outlet was located next to the laser ablation region to remove scattered particulates and product gases. As with previous LASEM experiments, the sample substrate was double-sided tape on a glass microscope slide. Residues of varying thicknesses (8–63 μg/mm 2 ) were prepared on multiple slides for each sample (except for Al 2 O 3 ) by pressing the sample firmly into the tape with a metal spatula (to confine the material in the laser focus and enhance the laser-material interaction). At least 20 laser shots were acquired for each sample. The sample slides were weighed before and after each laser shot with a balance accurate to 1 μg in order to determine the amount of material removed with each laser shot (via ablation, chemical reaction into gaseous products, or ejection off the sample surface). Pure metal samples investigated included micron-sized Al (Sigma-Aldrich, 27.5 μm mean particle size), reduction-grade alumina (Al 2 O 3 , NIST SRM 699), Al/I 2 O 5 (80 nm Al, 0.9 equivalence ratio), AIH6, and AIH15. Prior to the LASEM experiments, the AIH6 sample vial contained iodine vapor, indicating some sample degradation had occurred (Fig. S1 ). The AIH15 vial remained iodine-free for approximately one week following the LASEM data collection. The metals were added to TNT (provided by colleagues at the U.S. Army Research Laboratory) in composite mixtures that were approximately 20% metal by weight. Results AIH Characterization Table 1 shows the density and composition of AIH6 and AIH15 determined by Powder XRD analysis. OB is calculated from concentration data using Eq. 1 . The composition of a similar mixture (AIH ER 0.9) from previous work is shown for comparison. The mixing procedure used here is similar to the procedure used previously 12 , 13 , 14 ; however, the final composition of the AIH6 and AIH15 are significantly different as shown in Table 1 . The concentrations of Al°, AIH and iodic acid from AIH6, AIH15 and another AIH sample mixed at an initial ER of 0.9 discussed in Smith et al . 13 , 14 are included in Table 1 for comparison. The only difference between the AIH samples in Table 1 is the initial water to Al ratio used during mixing. The differences in concentrations of AIH between the three samples shown in Table 1 indicate that there are many unknown variables involved in AIH synthesis that need to be studied further, including drying time and solution pH. Table 1 AIH-Al mixture composition. Full size table Figure 1 illustrates the representative nature of the resultant Al-AIH nanoparticles through high resolution TEM (HRTEM) images. Figure 1a shows that the AIH layer is exhibited as nodules protruding from the Al surface and enhancing the roughness. Figure 1b shows a large Al-AIH nanoparticle and its single crystallinity was exemplified by the lattice fringes shown as parallel thin straight lines in the TEM image and the corresponding diffraction spots in the FFT pattern. Figure 1c is an image at 800 kx magnification to highlight the single crystallinity of the Al-AIH nanoparticle and the different diffraction contrast due to the presence of the AIH phase. Figure 1 TEM images showing the nature of Al-AIH nanoparticles: ( a ) the rough surface with protruding nodules, ( b ) and ( c ) the single crystallinity exemplified by the distinct lattice fringes showing as parallel thin straight lines and the different strain contrast due to the AIH layer in 300 kx and 800 kx magnifications, respectively. Full size image The presence of iodine on the AIH6 and AIH15 samples is further confirmed via XPS analysis. Shown in Fig. 2a,b are the binding states of these samples displaying all signature iodine peaks at binding energies of ~ 49 eV, 121 eV, 186 eV, 619 eV, 631 eV, 875 eV, and 931 eV, along with Auger signals, which do not exist in uncoated Al particles (Fig. 2c ). Figure 2 XPS surveys confirming the presence of iodine on the surface of the ( a ) AIH6, ( b ) AIH15 particles, in comparison to ( c ) untreated Al particles. Full size image Laser-Induced Shock Waves Shot-to-shot variations in the laser-material interaction are well-known to influence ablation mass and laser-induced plasma properties 16 , 17 , 18 , particularly for residue samples 19 . Minimizing these variations involves control of the experimental parameters, obtaining sufficient data points for statistical averaging, and/or analytical methods to compensate for signal fluctuations 20 , 21 . In this work, the first two methods were used to decrease the shot-to-shot variations as much as possible – however, it was not possible to completely overcome the stochastic nature of the laser-material interaction. Moreover, quantifying the shot-to-shot variations is challenging because of all the different processes involved in the laser-material interaction. For example, although we can measure the amount of material removed after each laser shot, it is not possible to identify where the material went, i.e., whether it was ablated into the plasma where it participated in the chemical reactions, or it was ejected off the sample slide and subsequently reacted in either the plasma region or (later) the deflagration zone, or it was ejected off the sample slide away from the interaction region. Of the material ablated into the plasma, only some fraction of it is completely atomized. The largest amount of material is ejected off the sample slide during the first several laser shots, as the shock wave disperses the loose material on the slide (Fig. S2 ). Table 2 shows the wt.% of metal, residue thickness, average mass removal per laser shot, and number of laser shots for each sample Typically, the first several laser shots removed the most material from the slide (up to 10× the mass of subsequent shots) since the shock wave disperses the loose material on the slide. Table 2 Sample details and number of laser shots acquired. Full size table For most samples, the amount of material removed from the substrate reaches a steady state around 100 μg per shot; in general, the deflagrating samples (TNT composites) result in more material removal per shot (Table 2 ). For a few samples (e.g., AIH15, TNT), a jump in the amount of mass removed at later shot numbers results from a pile of material created by a previous shock wave igniting. We investigated the effect of the amount of material removed per laser shot on the measured laser-induced shock velocities (Fig. 3 ). Although some of the materials showed a very slight correlation between the mass removed per shot and the measured shock velocity, in general the amount of material removed from the slide did not significantly influence the characteristic shock velocity. Since the largest sources of error in the shock velocity measurement are the shot-to-shot variations in the laser-material interaction and the measurement errors in the shock wave position (and subsequent y-intercept determination), the slight effect of residue thickness is within the error bars of the LASEM measurement. This is because the amount of material in the laser focus is relatively fixed by the defined focal length and the amount of material held in place by the tape. Any excess material is ejected off the sample surface and reacts (if at all) on the millisecond timescale, not on the microsecond timescale that determines the shock velocity. Figure 3 Laser-induced shock velocities for Al, TNT, and TNT + AIH15 as a function of the amount of material removed from the sample slide per laser shot. Full size image Figure 4 shows the average characteristic laser-induced shock velocities measured for the pure metal additives (top) and TNT composite (bottom) samples. The Al 2 O 3 has the lowest energy release because relatively little Al is free to oxidize in air on the microsecond timescale. While the physical Al/I 2 O 5 mixture released more energy than the micron-sized Al, the AIH samples released the most energy on the microsecond timescale, with the AIH15 sample (which showed no visible signs of degradation) producing the highest shock velocity. It is not clear that the measured laser-induced shock velocities for the AIH samples would directly correlate to detonation velocities (as with the primarily organic military explosives); however, the significant increase in measured laser-induced shock velocities for the AIH samples compared to the other metal additives suggest that they release their energy on a much faster timescale relevant to explosive events. Figure 4 Laser-induced shock velocities for the metal additives (top) and TNT composites (bottom). Full size image To test this hypothesis, the laser-induced shock velocities for TNT composites with the additives were also measured (bottom of Fig. 4 ). As previously shown, the micron-sized Al decreases the laser-induced shock velocity (and in large-scale tests, the detonation velocity), in part due to the formation of solid reaction products 6 . The Al 2 O 3 additive also decreases the shock velocity, although to a somewhat lessor extent since fewer solid reaction products form during the first few microseconds of the high-temperature reactions to slow the shock wave (compared to micron-sized Al). The biggest decrease in shock velocity was observed for the TNT-Al/I 2 O 5 composite. In contrast, the TNT-AIH samples resulted in a significant increase in laser-induced shock velocities. Using the calibration fit determined for military explosives 3 , the estimated detonation velocities for the TNT-AIH6 and TNT-AIH15 composite materials are 8.69 ± 0.24 km/s and 9.10 ± 0.26 km/s, respectively (compared to an estimated TNT detonation velocity 3 of 7.03 ± 0.12 km/s). Snapshots from the high-speed videos of the laser excitation of the pure and composite samples are shown in Figs 5 and 6 , respectively. The brightness and contrast of some of the later frames have been adjusted to improve visualization of the shock wave, and the top of the images have been cropped. The metal additives produce brighter plasmas as a result of the extensive aluminum-related emission features while energetic materials such as TNT produce less intense plasmas, as previously observed 2 , 6 . The reduced plasma emission from the TNT-AIH composites (compared to TNT with the other Al-containing additives) is indicative of increased energy release; the reduced emission is likely a result of the formation of exothermic reaction products which do not emit in the visible wavelength region (unlike neutral or ionic Al). Figure 5 Snapshots from videos of laser excited ( a ) micron-sized Al, ( b ) Al 2 O 3 , ( c ) Al/I 2 O 5 , ( d ) AIH6, ( e ) AIH15, and ( f ) TNT. Full size image Figure 6 Snapshots from videos of laser excited ( a ) TNT + Al, ( b ) TNT + Al 2 O 3 , ( c ) TNT + Al/I 2 O 5 , ( d ) TNT + AIH6, and ( e ) TNT + AIH15. Full size image Laser-Induced Deflagrations While the pure micron-sized Al sample resulted in a significant combustion cloud above the sample surface following laser excitation (lasting approximately 40 milliseconds on average, Fig. S3 ), neither the Al 2 O 3 nor the Al/I 2 O 5 mixture resulted in significant reaction following the decay of the laser-induced plasma. The combustion of the Al particles ejected into the air above the sample surface was similar to that previously observed during the laser-induced deflagration of energetic materials 7 . Although the Al 2 O 3 and Al/I 2 O 5 residues were not as thick as the micron-sized Al because of the way the material spread on the tape, shots with similar amounts of material removal (~100 μg per shot) could be compared directly to account for the differences in excess material; even with comparable masses of material ejected above the sample surface, of these three samples only the micron-Al sample resulted in late-time combustion emission. Although no cloud of combusting particles was produced in the air directly above the focused laser position following laser excitation, the AIH samples were more sensitive to the burning metal particles produced by the laser ignition than the other Al-containing additives investigated here. For both the pure AIH6 and AIH15 samples, the reaction following laser excitation propagated to other areas on the sample slide for two of the laser shots (out of 22 and 24, respectively), as shown in Fig. S4 . In each case, additional material on the sample slide ignited several milliseconds after the laser pulse. The ignition of adjacent material only occurred for the first two laser shots on the affected sample slides, suggesting that the reaction involved excess material (subsequently ejected off the sample slide by successive laser-induced shock waves). The ignited material was located several centimeters distant from the location of the focused laser pulse and was most likely ignited by burning metal particles ejected from the initial reaction zone. No reaction propagation to subsequent areas of the sample slide was observed for the micron-sized Al, Al 2 O 3 , or Al/I 2 O 5 samples, even when excess material was present, suggesting that the AIH samples are more sensitive than the other Al-containing materials. The intensity and duration of the laser-induced deflagration increases with the amount of material ejected into the air above the substrate (for those materials that deflagrate). Thicker residues generally result in more material ejected per laser shot and thus stronger deflagration events on the millisecond timescale (Fig. 7 ). The average laser-induced deflagration emission intensities (measured in units of optical power by the photodiode) for the TNT composites (Fig. S5 ) show that the TNT-Al formulation produces the most intense deflagration events, followed by pure TNT, as previously observed 6 . The TNT-Al 2 O 3 and TNT-AIH composites produced comparatively weaker deflagrations, with the TNT-Al/I 2 O 5 producing the weakest deflagration. For comparison to the average deflagration intensities over all laser shots (Fig. S5 ), limited averages of the deflagration intensities were calculated using only the laser shots where 1.0 ± 0.2 mg of material was removed from the slide (Fig. 8 ). Although the measured mass removal amount is not necessarily indicative of how much material was available in the deflagration zone, limiting the number of averaged shots in this way could partially correct for the amount of material ejected from the sample surface. As shown in Fig. 8a , the limited averages suggest that when similar amounts of material are available to react with the heated air above the sample surface, the AIH samples produce the strongest deflagration reactions. The differences between the average and limited average deflagration intensities could therefore reflect differences in the adhesion of the material to the tape, and not just the material’s potential for extended combustion reactions in air. Unlike with the pure AIH samples, no reaction propagation to adjacent spots on the sample slide was observed for the TNT-AIH mixtures (although, in general, significantly more material is ejected from the sample slide, Table 2 ). Figure 7 Integrated deflagration emission for Al 2 O 3 , TNT, TNT-Al and TNT-AIH15 as a function of the amount of material removed from the sample slide per laser shot. Full size image Figure 8 Average time-resolved emission from the laser-induced ( a ) deflagration and ( b ) microsecond-timescale combustion of TNT composites (limited averages from shots with 1.0 ± 0.2 mg of material removed). Full size image The emission spectra of the aluminum additives have similar features (Fig. S6 ). Strong emission features due to Al, AlO, CN and Na (a common impurity) saturated the spectrometer for some samples. The most surprising feature is the strong CN molecular bands, since the metal additives likely do not contain much carbon. Although carbon-related features from the tape used to adhere the residue to the substrate typically are not very strong in the resulting emission spectra, the presence of the oxygen-scavenging Al may enhance the formation of CN through recombination of the available carbon with nitrogen from the air rather than oxygen (to form CO, CO 2 ) 1 . Unfortunately, the strong AlO molecular emission bands obscure the C 2 emission features (512 and 516 nm) which could help confirm this hypothesis (since a lack of oxygen available for reaction with the C would result in increased C 2 emission 1 ). Among the aluminum-containing additives, Al/I 2 O 5 has the strongest AlO bands, while micron-sized Al has the strongest CN and H emission features. Al 2 O 3 has the most Na contamination, along with Li and K. AIH6 has Mg and Ca contaminants. While the Al emission features are significant for all Al-containing materials, atomic I features are comparatively weak and do not appear in the spectra. No evidence of possible reaction intermediates such as the iodine monoxide radical (IO; 427.2, 449.3, and 484.9 nm), iodine oxide (OIO; 549.3 nm), or I 2 (500 nm continuum) 22 was observed, although any emission features present were likely obscured by the strong AlO bands. The TNT and TNT composite material spectra (Fig. S7 ) have similar emission features as the pure metal additives, along with a broad emission feature in the visible region due to grey body radiation from the deflagration of the material following laser excitation. Discussion Microsecond-Timescale Reactions Figures 3 and 4 show that AIH increases the laser-induced shock velocities of TNT. The estimated detonation velocity of the TNT-AIH15 was 9.10 ± 0.26 km/s, indicating AIH15 may increase the detonation velocity of TNT by up to 30%. The increase in estimated detonation velocity in TNT-AIH15 mixtures suggests that the reaction time of AIH is on a fast enough time scale (μs) that significant amounts of gaseous products (I 2 , O 2 , aluminum iodates species, etc.) could be formed in the chemical reaction zone behind the detonation wave. While similar products are formed from the thermite reaction of Al with I 2 O 5 , they are not produced on a fast-enough timescale to increase the shock velocity, as evidenced by the decrease in laser-induced shock velocity for the TNT-Al/I 2 O 5 composites. In Capellos et al . 9 and Baker et al . 10 , it was shown that the Al 2 O 3 passivation layer limits Al reactions and slows the detonation velocity of aluminized explosives. Figure 8b shows that as the laser induced plasma is rapidly cooling (i.e., as the plasma emission decreases within the first 50 μs), extended emission from combustion of the TNT-Al/I 2 O 5 composites continues for approximately 500 μs. These reactions (beyond approximately 10 μs) are too slow to contribute to the laser induced shock velocity. The increase in laser-induced shock velocity of TNT-AIH15 mixtures suggests that the limiting effects of the Al 2 O 3 passivation layer are reduced in AIH formulations. In Smith et al . 13 , 14 , it was shown that during synthesis of AIH, the Al 2 O 3 passivation layer is dissolved and replaced with AIH. The increase in estimated detonation velocity seen in TNT-AIH15 mixtures suggests that when the Al 2 O 3 passivation layer is replaced with AIH, the Al-AIH reaction occurs on a timescale relevant to detonation. In addition, Fig. 8b also shows peaks in the combustion emission near 180 and 300 μs for AIH15 and AIH6, respectively. These combustion reactions are likely ignited by the plasma and, unlike the subsequent deflagration reactions (on the millisecond timescale, Fig. 8a ), are not self-sustaining. In Smith et al . 14 , it was also proposed that AIH reactions occur when the hydrate layer in AIH is removed. In Cradwick et al . 11 , the evaporation of the hydrate layer in AIH is reported to occur at 135 °C. To test the hypothesis that Al oxidation in AIH mixtures occur when the hydrate layer is removed, a DSC experiment was performed to examine heat flow under equilibrium conditions. Figure 9 shows the DSC heat flow of AIH when heated at 10 °C/min in an argon atmosphere. The heat flow plot shows an endotherm at 140 °C followed by a runaway exothermic reaction, confirming that AIH reactions are initiated by removal of the hydration layer in AIH. The heat flow ends between 150 °C and 160° because at this point, the thermal runaway of the AIH reaction damaged the DSC thermocouple to a point where measurements could not be taken. The temperature required to evaporate the hydrate layer in AIH is exceeded within the first nanosecond of the laser-material interaction as the laser-induced plasma is formed. Removal of the hydrate layer to initiate the reaction may contribute to the AIH samples reacting on times scales relevant to a detonation. Figure 9 Heat flow of AIH6 heated at 10 °C/min in an argon atmosphere. Scale bar corresponds with 0.4 mW/mg. Full size image It is important to note that for this measurement, 7 mg of AIH mixture was used and the runaway reaction destroyed the sample and reference crucible, the platinum thermocouple, and ~6 cm of the thermocouple support. Use caution when preforming DSC experiments with AIH mixtures. Millisecond-Timescale Reactions We have shown that removal of the hydration layer causes AIH to react on microsecond timescales, producing gaseous products formed in the reaction zone behind the laser-induced shock wave (Fig. 4 ). Figure 8a shows that TNT-AIH not only reacts on microsecond timescales, but continues to deflagrate on a millisecond timescale. The initial spike in emission intensity shown in the photodiode traces is due to the laser-induced plasma emission; ignition of the material ejected into the air above the sample occurs following the passage of the laser-induced shock wave – for the TNT composites, this is sufficient to initiate self-sustaining laser-induced deflagration reactions (i.e., fast combustion) on the millisecond timescale. The TNT-Al 2 O 3 composite deflagration intensity peaked almost 10 ms later than the other composites, possibly indicative of the time required to free the Al content for further oxidation in air. The strong deflagrations of the TNT-AIH composites demonstrate additional energy release through combustion reactions on the millisecond timescale (compared to the other TNT composites with comparable amounts of ejected material). These results suggest that, in addition to improving the detonation performance, AIH could also potentially enhance the middle- and late-time blast effects (i.e., slow energy release). Mechanisms for Energy Release Figure 4 shows that AIH reacts on a microsecond timescale and Fig. 8a shows that deflagration continues on a millisecond timescale. Reactions on a microsecond timescale suggest two possible mechanisms for AIH contributing to the TNT shock velocity. First, the plasma temperature could be increased for the TNT-AIH composites because more exothermic reactions are occurring. While the temperature of a laser-induced plasma formed on an inert substrate rapidly decays following cessation of the laser pulse (reaching just a few thousand Kelvin within several microseconds), exothermic reactions from ablated material can contribute additional energy to the plasma, raising the temperature and increasing the shock velocity beyond that observed using the same laser energy on an inert material. Significant increases in plasma temperature with the addition of Al nanoparticles to RDX have been previously observed 1 . The larger temperature differential between the plasma and the surrounding air would produce a faster shock. Second, Fig. 9 suggests that AIH produces gas products when the hydration layer is removed (Eq. 2 ) and corresponds with early stages of the reaction. $$({[Al{({H}_{2}O)}_{6}]}^{3+}{(I{O}_{3})}_{3}^{-}{(HI{O}_{3})}_{2})\to {(A{l}^{3+}{(I{O}_{3})}_{3}^{-}{(HI{O}_{3})}_{2})}_{s}+{[{({H}_{2}O)}_{6}]}_{v}$$ (2) Water vapor from dehydration production may contribute to accelerating the shock wave on the microsecond time scale. When the hydration layer is removed, one possible intermediate reaction could include the formation of iodine and oxygen, shown in Eq. 3 . $$2{(A{l}^{3+}{(I{O}_{3})}_{3}^{-})}_{s}\to A{l}_{2}{O}_{3}+{(\frac{15}{2}{O}_{2})}_{v}+{(6{I}_{2})}_{v}$$ (3) Oxygen and iodine gas products from Eq. 3 may also contribute to accelerating the shock wave on the microsecond time scale. Only radical (or ionized) species are shown in Eq. 3 . The stable product species (e.g., HIO 3 from Eq. 1 ) that are not shown in Eq. 3 could enhance deflagration. The enhancements in deflagration that occur after AIH has reacted behind the laser-induced shock wave could be a result of intermediate species reacting with oxygen from the air and/or HIO 3 and HI 3 O 8 reacting with the remaining Al°. In both Fig. S5 and Fig. 8a , the deflagration intensity for the TNT-AIH mixtures is greater than the TNT-I 2 O 5 mixtures. If deflagration of the AIH mixtures is a result of the remaining HIO 3 and HI 3 O 8 reacting with Al°, similar deflagration intensities would be expected between the TNT-AIH mixtures and the TNT-Al/I 2 O 5 mixtures; however, the Al 2 O 3 shell in the TNT-Al/I 2 O 5 mixtures is still intact. Since the deflagration intensities for the TNT-AIH mixtures are greater than the TNT-Al/I 2 O 5 mixtures, dissolution of the Al 2 O 3 shell not only allows reactions to occur on the microsecond timescale, but also results in more complete Al° oxidation during deflagration. This is supported by comparing the deflagration intensities of the TNT-AIH6 and TNT-AIH15 mixtures. The iodic acid concentration (both HIO 3 and HI 3 O 8 ) for the AIH6 mixture is 10.0 wt. % greater than the AIH15 mixture and the deflagration intensity (Fig. 8a ) for AIH6 is greater than for AIH15. Higher concentrations of HIO 3 and HI 3 O 8 result in higher intensity deflagrations for the AIH6 mixture compared to the AIH15 mixture. Higher concentrations of AIH in AIH15 produce more vapor products in the intermediate reactions and result in higher laser-induced shock wave velocities. Figure 4 shows that replacement of the Al 2 O 3 passivation layer with AIH can significantly increase explosive performance. From Table 1 , the AIH15 (15.0 wt. % AIH) mixture has a significantly greater concentration of AIH compared to AIH6 (5.8 wt. %). The variables that affect the final concentrations of AIH mixtures are still largely unknown; however, it was shown in Smith et al . 14 that reactivity from AIH mixtures is directly related to concentration of AIH. In comparison to previous samples tested 13 , 14 (with AIH concentration as high as 78%), only a fraction of the Al 2 O 3 shell was replaced with AIH in AIH6 (5.8 wt. %) and AIH15 (15.0 wt. %) and as much as a 30% increase in estimated detonation velocity was seen in TNT-AIH mixtures. Higher concentrations of AIH in the AIH15 sample may account for the faster laser-induced shock velocities for TNT-AIH15 compared to TNT-AIH6. In addition, the AIH6 sample showed visible signs of degradation at the time of testing, while the AIH15 sample did not. In general, AIH is over-oxidized and has a hydrate layer separating the iodate oxidizer from Al° fuel that dehydrates at 140 °C. When the hydrate layer is evaporated, excess oxygen from AIH reacts with Al° without the diffusion limitations of the Al 2 O 3 passivation layer. The improved laser-induced shock velocities in Fig. 4 show that Al particles coated with AIH react on timescales relevant to detonation waves. Further optimization of AIH synthesis to tailor AIH mixtures could result in additional increases in explosive performance. Conclusions Aluminum particles have been synthesized with an AIH coating. The crystalline AIH replaces the Al 2 O 3 shell encapsulating an Al core particle and is observed using TEM analysis as well as detected in powder XRD analysis. Two samples were synthesized, the first contained 5.8% AIH on Al (i.e., AIH6) and the second contained 15% AIH on Al (i.e., AIH15). The energy release of both samples was examined via LASEM testing. The technique enables analysis of small quantities (~mg) of material in order to assess the feasibility of larger scale-up efforts that evaluate detonation performance. The LASEM results for the AlH samples demonstrated their potential use for explosive enhancement on both fast (detonation velocity) and slow (blast effects) timescales. Estimates of the detonation velocity for TNT-AIH composites suggest an enhancement of up to 30% may be achievable over pure TNT detonation velocities. The laser-induced deflagrations of the TNT-AIH composites also suggest the potential for blast enhancement effects. The mechanisms for AIH-Al coated particles contributing to the early-time reactions may include: enhanced reaction of the under-oxidized TNT (OB of −74%) with the over-oxidized AIH and the chemical structure of AIH, which includes a hydrate layer separating the iodate oxidizer from Al 0 fuel that dehydrates at low temperatures, 140 °C. In this way, AIH reacts with Al 0 without the diffusion limitations of the Al 2 O 3 passivation layer. Gas production via dehydration and exothermic reactions may also increase the temperature of the plasma therefore increasing the shock velocity. Unreacted Al remaining after the passage of the shock front may then provide the enhanced blast effects at later times without the hindrance of the Al 2 O 3 passivation layer inherent in Al particles. While the interpretation of these preliminary results are complicated by the complexity of the laser-material interaction and subsequent shot-to-shot variations, we believe these results suggest larger-scale detonation tests are worth pursuing, assuming the composite materials pass the prerequisite safety and compatibility testing prior to scale-up (including sensitivity testing). The degradation of the AIH samples under atmospheric conditions must also be sufficiently addressed for viability in military applications. Data Availability Statement Data is available in the supplementary information section.

Army scientists proved a decades-old prediction that mixing TNT and novel aluminum nanoparticles can significantly enhance energetic performance. This explosive discovery is expected to extend the reach of U.S. Army firepower in battle. Researchers from the U.S. Army Research Laboratory and Texas Tech University demonstrated up to 30-percent enhancement in the detonation velocity of the explosive TNT by adding novel aluminum nanoparticles in which the native alumina shell has been replaced with an oxidizing salt called AIH, or aluminum iodate hexahydrate. The structure of the AIH-coated aluminum nanoparticles was revealed for the very first time through high resolution transmission electron (TEM) microscopy performed by ARL's Dr. Chi-Chin Wu, a materials researcher who leads the plasma research for the lab's Energetic Materials Science Branch in the Lethality Division of Weapons and Materials Research Directorate. Wu said this revolutionary research offers the potential for the exploitation of aluminum and potentially other metallic nanoparticles in explosive formulations to extend the range and destructive power of Army weapons systems, a key objective of the Army's "Long Range Precision Fires" modernization priority. "We believe these results show tremendous promise for enhancing the detonation performance of conventional military explosives with aluminum nanoparticles for the first time," said ARL's Dr. Jennifer Gottfried, a physical chemist who collaborated on the research. Single nanoparticle extracted out from a view of native aluminum particles at 150,000 magnification. The image highlights the amorphous oxide shell surrounding the crystalline core. Credit: ARL "It is very exciting to advance science to a point where we can harness more chemical energy from metal particles at faster timescales. This is an exciting time for transforming energy generation technology," said Dr. Michelle L. Pantoya, the J. W. Wright Regents Chair in Mechanical Engineering and Professor at Texas Tech University. Details of this breakthrough work are described in the team's May 28 published paper "Improving the Explosive Performance of Aluminum Nanoparticles with Aluminum Iodate Hexahydrate (AIH)" by Jennifer L. Gottfried, Dylan K. Smith, Chi-Chin Wu, and Michelle L. Pantoya in the high-impact journal Scientific Reports. The team found that the crystalline aluminum core was effectively protected against unwanted oxidation by the AIH shell, which appears as protruding nodules on the aluminum surface. The enhanced reactivity due to this unique morphological feature and novel core-shell structure was demonstrated by laser-induced air shock from energetic materials experiments, an innovative laboratory-scale energetic testing method developed by Gottfried. This technique involves impacting the sample with a high-energy, focused laser pulse to violently break apart the explosive molecules. The interaction of the laser with the material forms a laser-induced plasma and produces a shock wave that expands into the surrounding air. The energy released from an explosive sample can then be experimentally determined by measuring the laser-induced shock velocity with a high-speed camera. It was predicted decades ago that aluminum nanoparticles have the potential to enhance the energetic performance of explosives and propellants because of their high energy content and potential for rapid burning. This is because they have exceptionally large surface areas compared to their total volume and a very large heat of reaction. However, the surface of the aluminum nanoparticles is naturally oxidized in air to form a thick alumina shell, typically 20% by weight, which not only lowers the energy content of the nanoparticles by reducing the amount of active aluminum, it also slows the rate of energy release because it acts as a barrier to the reaction of the aluminum with the explosive. Therefore, replacing the oxide shell, as successfully achieved by TTU, can significantly improve the explosive performance. An AIH-salt crystal found at 400,000 magnification. The background is the carbon support film on the specimen grid. Credit: U.S. Army These preliminary joint efforts have also led to a formal research collaboration under an ARL Director's Research Award, the fiscal 2018 External Collaboration Initiative between Wu and TTU. After publishing two papers in high-impact scientific journals in the past year, the team is poised to pursue additional energetics research with aluminum nanoparticles by working with the U.S. Army Research, Development and Engineering Command at Picatinny Arsenal, New Jersey, and the Air Force Research Laboratory.

10.1038/s41598-018-26390-9

Medicine

Serotonin and confidence underlie patience in new study

Katsuhiko Miyazaki et al, Reward probability and timing uncertainty alter the effect of dorsal raphe serotonin neurons on patience, Nature Communications (2018). DOI: 10.1038/s41467-018-04496-y Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-018-04496-y

https://medicalxpress.com/news/2018-06-serotonin-confidence-underlie-patience.html

Abstract Recent experiments have shown that optogenetic activation of serotonin neurons in the dorsal raphe nucleus (DRN) in mice enhances patience in waiting for future rewards. Here, we show that serotonin effect in promoting waiting is maximized by both high probability and high timing uncertainty of reward. Optogenetic activation of serotonergic neurons prolongs waiting time in no-reward trials in a task with 75% food reward probability, but not with 50 or 25% reward probabilities. Serotonin effect in promoting waiting increases when the timing of reward presentation becomes unpredictable. To coherently explain the experimental data, we propose a Bayesian decision model of waiting that assumes that serotonin neuron activation increases the prior probability or subjective confidence of reward delivery. The present data and modeling point to the possibility of a generalized role of serotonin in resolving trade-offs, not only between immediate and delayed rewards, but also between sensory evidence and subjective confidence. Introduction The neuromodulator, serotonin, is extensively involved in behavioral, affective, and cognitive functions of the brain. Chemical and electrode recordings from the dorsal raphe nucleus (DRN) have shown that the activity of serotonin neurons increases when animals perform tasks requiring them to wait for delayed rewards 1 , 2 , 3 . Local pharmacological inhibition of DRN serotonin neural activity in rats impairs their patience in waiting for delayed rewards 4 . We recently used transgenic mice that express the channelrhodopsin-2 (ChR2) variant C128S in serotonin neurons 5 , 6 and showed that their selective activation in the DRN enhances the patience of mice waiting for both a conditioned reinforcer tone and a food reward 7 . A recent study also confirmed that optogenetic activation of DRN serotonin neurons enhances patience in waiting 8 . These results established a causal relationship between serotonin neural activation and patience in waiting for future rewards. We therefore questioned whether activation of serotonin neurons always promotes waiting for delayed reward or whether its effect depends on the subject’s reward prediction. In our previous optogenetic study, serotonergic activation prolonged waiting time by ~30% before the mice eventually gave up waiting 7 . Serotonin neuron activation was most effective at the time when mice decided whether to continue waiting 7 . These results suggest that cognitive status, such as the anticipation of future rewards, modulates the promotion of patience by serotonin. In the current study, we tested whether the probability, amount, and timing uncertainty of future rewards affects promotion of patience by serotonin neuron activation. We find that serotonin effect in promoting waiting is maximized by both high-reward probability (RP) and high-reward timing uncertainty. We further propose a Bayesian decision model of waiting, which assumes serotonin neuron activation increases the prior RP to reproduce the major features of the experimental results. The model reproduces the more prominent effect of serotonin with reward timing uncertainty because the likelihood function for reward delivery has a longer tail in time. The present data and modeling suggest that serotonin neuron activation enhance patience in waiting for future rewards by increasing subjective confidence of future goals. Results Serotonin effect on waiting depends on reward probability Mice (seven transgenic mice and five wild-type (WT) mice) were trained to perform a sequential tone-food waiting task that required them to wait for a delayed tone (conditioned reinforcer) at a tone site and then to wait for delayed food (primary reward) at a reward site (Fig. 1a, b ). In experiment 1, to examine whether the predicted probability and amount of reward affect the promotion of patience by serotonin neuron activation, we prepared six combinations of RP (75, 50, and 25%) and reward amount (1, 2, and 3 food pellets) (Supplementary Fig. 1 ). Fig. 1 Schematic of the sequential tone-food waiting task. a Diagram of the test in which optogenetic stimulation was applied during the reward-delay period (experiments 1 and 2). b Time sequence of serotonin activation trials and serotonin no-activation trials. In serotonin activation trials, 0.8 s of blue light was delivered at the onset of the reward delay. In serotonin no-activation trials, 0.8 s of yellow light was applied at the onset of the reward delay. In each trial, 1 s of yellow light was used at the onset of food presentation or at the reward wait error. Blue and yellow bars denote blue and yellow light stimulation, respectively. Brown- and red-shaded regions denote tone- and reward-delay periods, respectively. Orange-shaded regions denote duration of tone presentation. c Locations of optical fibers in the DRN. Light blue bars in the DRN represent tracks of implanted optical fibers. Coronal drawings were adapted from ref. 48 with permission Full size image In the experiment, during which 75% of the nose pokes for 3 s were rewarded with one food pellet (Supplementary Fig. 1a ), waiting time in the 25% of trials with no reward (i.e., omission) was significantly longer with serotonin neuron activation (7.89 ± 0.08 s, mean ± s.e.m.) than without activation (6.95 ± 0.09 s; t (5) = 24.05, P = 2.32 × 10 −6 , n = 6 mice, paired t -test) (Figs. 2 a and 3a ; Supplementary Fig. 2 ). The effect was significantly seen in each of the six mice tested ( P < 0.022, Mann–Whitney U- test) (Supplementary Fig. 3 ). We confirmed, in five WT mice, that waiting time in the blue light trials (7.36 ± 0.31 s) was not significantly different from that in the yellow light trials (7.35 ± 0.32 s; t (4) = 0.33, P = 0.76, n = 5 mice, paired t -test). In the 75% one-pellet test, we analyzed control group (WT) data with ChR2-expressing group (ChR2) in a two-way analysis of variance (ANOVA). There was a significant main effect of light (two levels within-subject factors; yellow and blue, F (1,9) = 366.83, P < 10 −6 ) but no significant main effect of group (two levels between-subject factors; ChR2 and WT, F (1,9) = 0.062, P = 0.81). There was a significant main effect of interaction (light × group, F (1,9) = 353.14, P < 10 −6 ). There was a significant simple main effect of light in ChR2 ( F (1,9) = 791.90, P < 10 −6 ) but no significant simple main effect of light in WT ( F (1,9) = 0.06, P = 0.81) (Fig. 3a ). When the reward was increased to two pellets, waiting times for omission trials became significantly longer both without serotonin neuron activation (7.84 ± 0.12 s, t (4) = 7.45, P = 0.0017, n = 5 mice, paired t -test) and with (8.89 ± 0.11 s, t (4) = 5.42, P = 0.0056, n = 5 mice, paired t -test) (Fig. 2a, b ; Supplementary Figs. 2 and 4a, b ). Again, waiting time with such activation was significantly longer than that without ( t (4) = 14.74, P = 1.23 × 10 −4 , n = 5 mice, paired t -test) (Fig. 3b ). Fig. 2 Optogenetic activation of DRN serotonin neurons enhances waiting in 75% reward tests, but not in 25 or 50% reward tests. a Distribution of waiting time during omission trials in the 75% one-pellet test. b Distribution of waiting time during omission trials in the 75% two-pellet test. c Distribution of waiting time during omission trials in the 25% one-pellet test. d Distribution of waiting time during omission trials in the 25% three-pellet test. e Distribution of waiting time during omission trials in the 50% one-pellet test. f Distribution of waiting time during omission trials in the 50% three-pellet test. Orange circles illustrate the timing and numbers of food pellets presented in rewarded trials. White circles denote omission trials Full size image Fig. 3 Average waiting time during omission trials in the 75, 50, and 25% reward tests. a Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 75% one-pellet test for individual ChR2-expressing (blue thin lines) and WT (green thin lines) mice and for population of ChR2-expressing (blue line) and WT (green line) mice. b Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 75% two-pellet test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). c Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 25% one-pellet test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). d Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 25% three-pellet test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). e Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 50% one-pellet test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). f Average waiting time in serotonin no-activation (yellow) and activation (blue) during the 50% three-pellet test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). *** P < 0.001 by paired t -test. Error bars represent the s.e.m. In some case, the error bars are too small to be visible. n.s. not significant Full size image In contrast, when the probability of reward delivery was reduced to 25% (Supplementary Fig. 1b ), waiting time in omission trials with serotonin neuron activation (5.67 ± 0.16 s) was not significantly different from that without (5.69 ± 0.18 s; t (5) = 0.89, P = 0.41, n = 6 mice, paired t -test) (Figs. 2 c and 3c ; Supplementary Fig. 2 ). To examine whether the ineffectiveness of serotonin neuron activation was due to a lower expected reward value, we performed a test with a 25% reward of three pellets, in which the expected reward value was equated with that of a 75% reward of one pellet. Waiting time without serotonin neuron activation in the 25% three-pellet test (6.20 ± 0.18 s) was significantly longer than that in the 25% one-pellet test ( t (5) = 11.79, P = 7.74 × 10 −5 , n = 6 mice, paired t -test), but significantly shorter than in the 75% one-pellet test ( t (5) = 5.33, P = 0.0031, n = 6 mice, paired t -test) (Fig. 2c, d ; Supplementary Figs 2 and 4c, d ). However, even with a higher expected reward value, waiting time in omission trials in the 25% three-pellet test with serotonin neuron activation was not significantly different from that without serotonin neuron activation (6.19 ± 0.16 s; t (5) = 0.24, P = 0.82, n = 6 mice, paired t -test) (Figs. 2 d and 3d ; Supplementary Fig. 2 ). These results show that the increased reward value in the 25% reward tests prolongs waiting time, but does not modulate the effect of serotonin in promoting waiting time. To further examine whether the uncertainty of reward delivery affects the promotion of patience by serotonin, we introduced tests with a 50% RP, at which the uncertainty is maximized (Supplementary Fig. 1c ). In both the one-pellet and three-pellet tests, serotonin neuron activation did not prolong waiting time in omission trials compared with the trials without serotonin neuron activation (one-pellet test, 6.19 ± 0.15 s, with activation, 6.04 ± 0.16 s, without activation, t (2) = 3.36, P = 0.078, n = 3 mice, paired t -test; three-pellet test, 6.85 ± 0.30 s, with activation, 6.60 ± 0.21 s, without activation, t (2) = 3.44, P = 0.075, n = 3 mice, paired t -test) (Figs. 2 e, f and 3e, f ; Supplementary Fig. 2 ). In the 50% three-pellet test, the expected reward value (1.5 pellets per trial) was equal to that in the 75% two-pellet test. Waiting time in omission trials without serotonin neuron activation in the 50% three-pellet test was significantly longer than those in the 50% one-pellet test ( t (2) = 6.89, P = 0.020, n = 3 mice, paired t -test), but significantly shorter than those in the 75% two-pellet test ( t (2) = 4.86, P = 0.039, n = 3 mice, paired t -test) (Supplementary Figs 2 and 4e, f ). These results show that the uncertainty of reward acquisition does not facilitate waiting or the effect of serotonin neuron activation on waiting. To quantify the effectiveness of serotonin neuron activation at promoting waiting time during omission trials, we calculated waiting time ratio (waiting time with serotonin neuron activation/waiting time without serotonin neuron activation) for each test (Fig. 4 ) and performed Scheirer–Ray–Hare test with the RP and the expected reward value as explanatory variables. There was a significant main effect of the RP (three level; 75, 50, and 25%, H (2) = 112.38, P < 10 −6 ) but no significant main effect of the expected reward value (four levels; 0.25, 0.5, 0.75, and 1.5, expected pellets (EPs) per trial, H (3) = 0.11, P = 0.99). Fig. 4 The role of serotonin in promoting patience is modulated by reward probability, but not by reward value. Waiting time ratios: 75% one-pellet test (1.13 ± 0.01, n = 57 tests from 6 mice), 75% two-pellet test (1.14 ± 0.02, n = 30 tests from 5 mice), 25% one-pellet test (1.00 ± 0.01, n = 38 tests from 6 mice), 25% three-pellet test (1.00 ± 0.01, n = 41 tests from 6 mice), 50% one-pellet test (1.03 ± 0.01, n = 21 tests from 3 mice), and 50% three-pellet test (1.04 ± 0.02, n = 19 tests from 3 mice). Waiting time ratios were significantly larger in 75% reward tests compared with tests having the same expected reward value. ** P < 0.01, *** P < 0.001 by post hoc Bonferroni correction. n.s. not significant. Error bars represent the s.e.m. Full size image In addition, we performed analysis based on a linear mixed model, taking mouse identity (MI) as a random effect. This approach is based on a plausible assumption that the baseline waiting time ratio may be different among mice. The result of likelihood ratio test between the model with effects of RP and EP and the model without these covariates supports the former model ( χ 2 (5) = 121.00, P < 10 −6 ). Further, using the obtained model, we tested difference of mean waiting time ratios between different levels of RP and EP. Difference of means between RP 75 and 25% ( Z = 9.02, P < 10 −6 ), and between 75 and 50% ( Z = 5.07, P < 10 −6 ) were significant, while the remainder of difference of means were not significant (Supplementary Table 1 ). Subsequently, we tested variability of waiting time ratio among mice. We compared the obtained mixed model with the model including fixed effects of RP and EP, but not a random effect of MI. To accurately evaluate likelihood ratio of two models, we generated 1000 new samples of waiting time ratios by means of a parametric bootstrap method. The variability of waiting time ratio among mice was not significant ( P = 0.553). Lastly, we went for more detailed analysis on differences of waiting time ratios for specific combinations of RP and EP. Note that in this analysis, we did not distinguish between mice because such differences are not significant. In each RP, reward value change did not significantly influence the waiting time ratio (75% one-pellet vs. 75% two-pellet, P = 1.00; 50% one-pellet vs. 50% three-pellet, P = 1.00; 25% one-pellet vs. 25% three-pellet, P = 1.00, post hoc Bonferroni correction) (Fig. 4 ). This result was all seen in each of the tested mice (for 75% reward, P > 0.55, n = 5 mice; for 50% reward, P > 0.53, n = 3 mice; for 25% reward, P > 0.20, n = 6 mice, Mann–Whitney U -test) (Supplementary Fig. 5 ). When we directly compared tests with different RP and same expected reward value, the waiting time ratios were significantly larger in 75% reward tests compared with same expected reward value tests (75% one-pellet vs. 25% three-pellet, P < 10 −6 ; 75% two-pellet vs. 50% three-pellet, P = 0.0039, post hoc Bonferroni correction) (Fig. 4 ). These results show that serotonin’s effect on promoting waiting depends on the probability of delivery, but not the expected value, of future reward. Reward timing uncertainty alters serotonin effect on waiting In our previous study, the waiting time ratio was >1.3 7 , whereas in experiment 1 of the current study, the waiting time ratio was ~1.1 with a 75% probability of reward. A major difference between the previous and current studies was the variability of reward delays. In our previous study, in the 75% reward trials, the reward delay was randomly set to 3, 6, or 9 s, whereas it was a constant 3 s in the current study. Thus, we hypothesized that serotonin promotes waiting more effectively when mice cannot predict the timing of the reward delivery (timing uncertainty). In experiment 2, we prepared three reward-delay conditions with a 75% RP: (i) fixed 6 s (D6 test) (Supplementary Fig. 6a ); (ii) randomly set to 4, 6, or 8 s (D4-6-8 test) (Supplementary Fig. 6b ); and (iii) randomly set to 2, 6, or 10 s (D2-6-10 test) (Supplementary Fig. 6c ). In all three tests, waiting time for omission trials with serotonin neuron activation was significantly longer than that without serotonin neuron activation (D6 test, 12.23 ± 0.20 s vs. 11.00 ± 0.23 s, t (5) = 20.35, P = 5.30 × 10 −6 , n = 6 mice; D4-6-8 test, 14.48 ± 0.25 s vs. 12.26 ± 0.17 s, t (5) = 20.16, P = 5.55 × 10 −6 , n = 6 mice; D2-6-10 test, 18.05 ± 0.79 s, vs. 13.51 ± 0.51 s, t (5) = 13.75, P = 3.65 × 10 −5 , n = 6 mice, paired t -test) (Figs. 5 a–c and 6a–c ; Supplementary Fig. 7 ). These results were significantly seen in each of the six mice tested (D6 test, P < 0.043; D4-6-8 test, P < 0.0014; D2-6-10 test, P < 4.19 × 10 −6 , Mann–Whitney U -test) (Supplementary Fig. 8 ). For WT mice ( n = 5), we confirmed that the waiting time in the blue light trials was not significantly different from that in the yellow light trials in both D6 and D2-6-10 tests (D6 test, 11.62 ± 0.66 s vs. 11.66 ± 0.63 s, t (4) = 0.90, P = 0.42; D2-6-10 test, 14.61 ± 0.59 s, vs. 14.66 ± 0.70 s, t (4) = 0.39, P = 0.72, paired t -test) (Fig. 6a, c ). In D6 and D2-6-10 test, we analyzed WT group data with ChR2 group in a two-way ANOVA. There was a significant main effect of light (two levels within-subject factors; yellow and blue, D6 test, F (1,9) = 226.75, P < 10 −6 ; D2-6-10 test, F (1,9) = 139.82, P < 10 −6 ) but no significant main effect of group (two levels between-subject factors; ChR2 and WT, D6 test, F (1,9) = 0.0028, P = 0.96; D2-6-10 test, F (1,9) = 1.92, P = 0.20). There was a significant main effect of interaction (light × group, D6 test, F (1,9) = 259.83, P < 10 −6 ; D2-6-10 test, F (1,9) = 145.60, P < 10 −6 ). There was a significant simple main effect of light in ChR2 (D6 test, F (1,9) = 534.62, P < 10 −6 ; D2-6-10 test, F (1,9) = 313.89, P < 10 −6 ) but no significant simple main effect of light in WT (D6 test, F (1,9) = 0.52, P = 0.49; D2-6-10 test, F (1,9) = 0.03, P = 0.87) (Fig. 6a, c ). Fig. 5 Optogenetic activation of DRN serotonin neurons enhances waiting for temporally uncertain rewards. a Distribution of waiting time during omission trials in the D6 test. b Distribution of waiting time during omission trials in the D4-6-8 test. c Distribution of waiting time during omission trials in the D2-6-10 test. d Distribution of waiting time during omission trials in the D10 test. Orange circles illustrate the timing and number of food pellets presented in rewarded trials. White circles denote omission trials Full size image Fig. 6 The role of serotonin in promoting patience for future rewards with uncertain timing. a Average waiting time in serotonin no-activation (yellow) and activation (blue) during the D6 test for individual ChR2-expressing (blue thin lines) and WT (green thin lines) mice and for population of ChR2-expressing (blue line) and WT (green line) mice. b Average waiting time in serotonin no-activation (yellow) and activation (blue) during the D4-6-8 test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). c Average waiting time in serotonin no-activation (yellow) and activation (blue) during the D2-6-10 test for individual ChR2-expressing (blue thin lines) and WT (green thin lines) mice and for population of ChR2-expressing (blue line) and WT (green line) mice. d Average waiting time in serotonin no-activation (yellow) and activation (blue) during the D10 test for individual ChR2-expressing mice (gray lines) and for population of mice (blue line). *** P < 0.001 by paired t -test. Error bars represent the s.e.m. In some case, the error bars are too small to be visible. e Waiting time ratios in the 75% one-pellet tests in which food pellets were delivered with uncertain timing. The waiting time ratio in the D2-6-10 test was the largest among the five tests. The waiting time ratio in the D6 test was not significantly different from the waiting time ratios in the D3 test and in the D10 test. *** P < 0.001 by post hoc Bonferroni correction. Error bars present the s.e.m. n.s. not significant Full size image Among the three delay conditions, the waiting time ratio was largest in the D2-6-10 test (D6 test, 1.12 ± 0.01, n = 47 tests; D4-6-8 test, 1.19 ± 0.01 s, n = 50 tests; D2-6-10 test, 1.34 ± 0.02 s, n = 54 tests) ( H (4) = 110.22, P < 10 −6 , Kruskal–Wallis test; P = 8.60 × 10 −4 for D6 vs. D4-6-8, P < 10 −6 for D6 vs. D2-6-10, post hoc Bonferroni correction) (Fig. 6e ). In each of the six mice tested, the waiting time ratio was the largest in the D2-6-10 test ( P < 0.015, Mann–Whitney U -test) (Supplementary Fig. 9 ). These results show that serotonin promotes waiting more effectively when mice cannot predict the timing of the reward delivery. Next, we examined whether the increased waiting time ratio in the D2-6-10 test was due to the introduction of the longest delay (10 s). We introduced a D10 test, in which reward delay was fixed at 10 s with a 75% probability. In the D10 test, waiting time for omission trials with serotonin neuron activation (19.40 ± 0.59 s) was significantly longer than that without serotonin neuron activation (17.55 ± 0.56 s, t (3) = 13.75, P = 8.32 × 10 −4 , n = 4 mice, paired t -test) (Figs. 5 d and 6d ; Supplementary Fig. 7 ). With regard to waiting time ratio, we performed analysis based on a linear mixed model, taking MI as a random effect. The result of likelihood ratio test between the model with effects of reward-delay condition and the model without this covariate supports the former model ( χ 2 (4) = 133.04, P < 10 −6 ). Further, using the obtained model, we tested difference of mean waiting time ratios between different levels of reward-delay conditions. The mean waiting time ratio of D2-6-10 test was significantly larger than the remainder of the time delay conditions ( Z = 11.0, P < 10 −6 for D3 test; Z = 11.3, P < 10 −6 for D6 test; Z = 10.5, P < 10 −6 for D10 test; Z = 7.91, P < 10 −6 for D4-6-8 test). Also, the mean waiting time ratio of D4-6-8 test was significantly large than D6 and D10 tests ( Z = 3.35, P = 8.07 × 10 −4 ; Z = 3.50, P = 4.64 × 10 −4 , respectively). The remainder of differences were not significant (Supplementary Table 2 ). Subsequently, we tested variability of waiting time ratio among mice. We compared the obtained mixed model with the model including a fixed effect of reward-delay condition, but not a random effect of MI. To evaluate likelihood ratio of two models, we generated 1000 new samples of waiting time ratios by means of a parametric bootstrap method. The variability of waiting time ratio among mice was not significant ( P = 0.602). The waiting time ratio in the D6 test was not significantly different from the waiting time ratio in the 75% one-pellet test with a 3 s delay in experiment 1 (D3 test) ( P = 1.00, post hoc Bonferroni correction) (Fig. 6e ). The waiting time ratio in the D10 test (1.11 ± 0.01, n = 34 tests) was not significantly different from the waiting time ratios in the D6 test of experiment 2 ( P = 1.00, post hoc Bonferroni correction) and in the D3 test of experiment 1 ( P = 1.00, post hoc Bonferroni correction) (Fig. 6e ). These results show that timing uncertainty, but not the longest waiting time for future rewards, is critical for enhancing serotonin’s effect at increasing waiting times. Bayesian decision model of waiting Can these effects of serotonin on waiting, depending on the RP and timing uncertainty, be explained in a coherent way? Here we consider the possibility that serotonin signals the prior probability of reward delivery in a Bayesian model of repeated decisions to wait or to quit. In this model, the subject has an internal model of the timing of reward delivery and infers whether the current trial is a reward trial or a no-reward trial. As time goes by without a reward delivery, the likelihood of its being a reward trial diminishes (Fig. 7a , top panel). The posterior probability of a reward follows the same time course scaled by the prior probability for a reward trial (Fig. 7a , middle panel). The expected reward for waiting goes down accordingly and the subject quits waiting as the expected reward for waiting becomes close to that for quitting (zero). The distribution of the time of quitting shifts later as the prior probability of a reward trial increases (Fig. 7a , bottom panel). Fig. 7 A Bayesian decision making model for waiting reproduces features of effects of reward probability and timing uncertainty on promotion of patience by serotonin. a Top panel: the model assumes that the subject has a probabilistic model of reward delivery timing (magenta line), which is assumed to be Gaussian with μ = 3 s and σ = 2 s in this example. As the time passes without reward delivery, the likelihood of a reward trial diminishes according to the cumulative density function (green line). Middle panel: the posterior probability for a reward trial goes down along with the likelihood, but persists longer if the prior probability for a reward trial is higher. Bottom panel: the timing of quitting is shifted later with a higher prior probability (Methods). b We assume that dorsal raphe serotonin neuron stimulation causes an overestimation of the prior probability when the reward probability is higher ( p ′ = p + p 2 − p 3 in this example). The yellow and blue lines show the time of quitting without and with increased prior probability, respectively. The effect of serotonin neuron stimulation is largest with a reward probability p = 0.75 (top panel; μ = 3 s and σ = 2 s). c With a larger uncertainty σ of reward timing, the waiting time distribution shifts later and the effect of serotonin neuron stimulation (increase of prior probability from 0.75 to 0.95 in this example) increases. A shift in the average reward timing (bottom panel; μ = 10 s and σ = 3 s) does not cause a large increase in waiting time with serotonin neuron stimulation Full size image If we assume that dorsal raphe serotonin neuron stimulation causes an increase in the estimate of the prior probability when the RP is high, the effect on the waiting time distribution with different RPs (Fig. 2 ) can be reproduced (Fig. 7b ). As the uncertainty of reward timing increases, the likelihood of a reward trial has a longer tail in the time axis. Accordingly, the same increase in the prior probability causes a larger shift in waiting time distribution (Fig. 7c ). This effect approximates the differential effects of serotonin neuron stimulation with different timing uncertainty (Fig. 5 ). Discussion Through a series of studies, we revealed a causal relationship between dorsal raphe serotonin neuron activation and patience to wait for future rewards 1 , 2 , 4 , 7 . Previous recording studies have shown that DRN neural activity is correlated with levels of behavioral arousal 9 , rhythmic motor outputs 10 , salient sensory stimuli 11 , 12 , 13 , 14 , conditioned cues 13 , 14 , 15 , 16 , 17 , rewards 2 , 13 , 15 , 16 , 17 , reward values and expectation 15 , 16 , 17 , punishments 17 , 18 , waiting for delayed rewards 2 , and reward omission 13 . Classically, putative serotonin neurons have been identified by broad spikes, slow regular firing, and suppression of 5-HT 1A receptor antagonist 2 , 19 , 20 . However, it has been difficult to precisely identify serotonergic neurons using these criteria 21 , 22 , 23 , 24 . Response diversity in the DRN may reflect non-selective recording of both serotonin and non-serotonin neurons. Using ontogenetic tagging, recent recording studies have demonstrated that serotonin neurons respond to conditioned cure 25 , 26 , reward 3 , 26 , punishment 26 , average reward rate 26 , and waiting 3 . This response diversity may reflect anatomical, neurochemical, and electrophysiological heterogeneity of serotonergic neurons in the DRN 27 . Nevertheless, 79% of classically identified putative serotonergic neurons 2 and 90% of optogenetically identified serotonergic neurons 3 were tonically activated during waiting for delayed rewards, suggesting that regulating waiting behavior for delayed rewards is a principal function of the serotonin system. In the current study, we found that optogenetic activation of dorsal raphe serotonergic neurons was not always sufficient to enhance waiting for future rewards. In experiment 1, we found that in the 75% reward test, but not in the 25 or 50% reward tests, optogenetic serotonin activation promoted waiting. These results suggest that a high expectation or confidence in future rewards is necessary for serotonin neural activation to promote waiting and that the interaction of increased serotonin release and the cognitive state of the subject is crucial. Our finding that serotonin neuron activation did not enhance waiting time in the 25 and 50% reward tests also showed that under our stimulation parameters, optogenetic serotonin activation itself did not induce a reinforcing effect to cause prolonged nose poking at the reward site 7 , 8 , 25 , 28 , 29 . In experiment 2, we found that the effect of serotonin neuron activation on promoting patience was modulated by the variability of timing of reward presentation. Serotonin neuron activation enhanced waiting more effectively when the mice could not predict the timing of the delivery of highly certain rewards. This effect, most prominently observed in D2-6-10 condition, did not simply depend on the average or maximal waiting time because the average waiting time was the same among the D6, D4-6-8, and D2-6-10 conditions and the maximal waiting time was the same between the D10 and D2-6-10 conditions (Fig. 6e ). When the timing of reward delivery becomes variable, it becomes more difficult to reject the possibility that the reward may still come. The resulting lower confidence in no reward, or higher subjective probability of reward delivery, might be a reason for the stronger effect of serotonin in facilitating reward-directed behavior. How does serotonin neuron activation promote patience in waiting? A possible explanation is that serotonin affects the perception of time, such that the same physical time is perceived to be shorter with serotonin neuron stimulation 30 . However, our previous experiment showed that serotonin neuron stimulation during an early phase of waiting does not affect waiting time 7 , which is inconsistent with the time perception hypothesis. We previously hypothesized that serotonin controls the temporal discounting parameter in the model-free reinforcement learning framework 31 . While this hypothesis was consistent with many of the recording and manipulation experiments 2 , 4 , 7 , 32 , the effects depending on the RP and timing uncertainty are difficult to explain in terms of a simple temporal discounting paradigm. Thus, we considered a Bayesian model in which serotonin neuron stimulation affects the prior probability for the present trial to be a reward trial. Our simulation results (Fig. 7 ) reproduced the critical features of the shifts in waiting time distribution depending on RP and timing uncertainty. The present model is based on several arbitrary assumptions, namely, the internal model of reward timing distribution is Gaussian while the experimental setting is multi-modal, serotonin neuron stimulation causes overestimation of RP especially when the RP is high, and the choice of some free parameters. Nevertheless, this model is consistent with the effect of serotonin on emotional bias toward positive outcomes 33 and a recent report that serotonergic neuron activity keeps track of average reward rate 26 , and further points to the possibility of a generalized role of serotonin in arbitrating the trade-off between (negative) sensory evidence and (positive) subjective belief. Selective serotonin reuptake inhibitors (SSRIs) are widely used to treat psychiatric disorders, especially depression, by increasing the serotonergic tone in the whole brain 34 , 35 . However, remission rate is 36.8% for citalopram treatment alone 36 . Psychological treatment, such as cognitive behavioral therapy combined with antidepressant therapy, is associated with a higher improvement rate than drug treatment alone 37 . Our finding that activation of serotonin neurons alone is not enough and that it requires a subject’s confidence in a positive outcome (i.e., high probability for a future reward) to promote a goal-directed behavior, may explain the combined effect of SSRI treatment and cognitive therapies, which often removes patients’ negative biases in future outcomes. The effect of cognitive behavioral therapy is gradual, such that subjects cannot predict a specific time till recovery. Our results in experiment 2 suggest that augmentation of serotonergic tone by SSRI treatment is most effective for enhancing patience for a gradual recovery, and could prevent patients from dropping out. Therefore, SSRI treatment and cognitive behavioral therapy may produce mutually positive effects to realize synergistic therapy. A recent study showed that inactivation of the orbitofrontal cortex (OFC) disrupts waiting-based confidence reports without affecting decision accuracy 38 . Previous recording studies have also revealed that OFC neurons encode predictions of reward outcomes 39 , 40 . Optogenetic serotonin activation modulates reward anticipatory responses of OFC neurons 41 . These results suggest that the OFC may produce causal signals for waiting with serotonin neural activation 42 . Optogenetic stimulation of the terminal sites to which DRN serotonin neurons project will clarify the sites where serotonin contributes to enhance patience 43 . Recent rabies virus tracing strategies have yielded a comprehensive map of afferent inputs to serotonin neurons 44 , 45 , 46 . The combination of serotonergic neural recording with optogenetic manipulation of their afferent inputs will allow us to dissect the afferent inputs, local circuits, and cellular auto-regulatory mechanisms that shape activities of serotonin neurons 47 . These techniques should also allow us to reveal the brain’s algorithm for regulation of patience 31 . Methods Animals All experimental procedures were performed in accordance with guidelines established by the Okinawa Institute of Science and Technology Experimental Animal Committee. Serotonin neuron-specific ChR2(C128S)-expressing mice were produced by crossing Tph2-tTA mice with tetO-ChR2(C128S)-EYFP knock-in mice 5 , 6 . Seven male bigenic and five male WT mice, aged >4 months at the beginning of the behavioral training period, were used in the study. Animals were housed with one mouse per cage at 24 °C on a 12:12 h light:dark cycle (lights on 07:00–19:00 h). Seven bigenic (one for experiment 1 only, one for experiment 2 only, five for both experiments 1 and 2) and five WT animals contributed to the data reported here. Training and test sessions were conducted during the light period 5 days per week. Mice were deprived of food in their home cage and received their daily food ration during the experimental sessions only (~2–3 g per day). Food was freely available during the weekend and removed >15 h before the experimental sessions started. Water was freely available in the home cage. Surgery After mice had mastered the sequential tone-food waiting task, they were anesthetized with equithesin (3 ml/kg, i.p.), and an optical fiber (400 μm diameter, 0.48 NA, 4 mm length, Doric Lenses) was stereotaxically implanted above the DRN (from bregma: posterior, −4.6 mm; lateral, 0 mm; ventral, −2.6 mm). The optical fiber was fixed to the skull and anchored with dental acrylic and stainless steel screws. Animals were housed individually after surgery and were allowed at least 1 week to recover. Reconstruction of optical stimulation sites Mice were deeply anesthetized with 100 mg/kg sodium pentobarbital i.p. and were then perfused with 0.9% NaCl, followed by 10% formalin. Their brains were removed and stored in 10% formalin for a minimum of 24 h before being sliced into 60 mm coronal sections. Cresyl violet staining was used to help verify placements of optical fiber tracks (Fig. 1c ). Behavioral apparatus and training A free operant task that we designated as a sequential tone-food waiting task was used. Mice were individually trained and tested in an operant-conditioning box (Med-Associates) measuring 21.6 cm × 17.8 cm × 12.7 cm. The box could be illuminated with a single 2.8 W house light located in the top center of the rear wall. One speaker was positioned in the top right side of the rear wall. Three 2.5 cm square apertures were positioned 2 cm above the floor. The rear stainless steel wall of the chamber contained one aperture defined as the tone site. On the front wall, two apertures defined as the food sites were positioned 7 cm apart. Both apertures on the front wall were connected to a food pellet dispenser that delivered a food pellet (20 mg) to these apertures. In all experiments, only the right food site was used, and the left aperture was covered with an opaque window to prevent nose poking. An infrared photo-beam crossed the entrances of all of the apertures to detect nose poke responses positioned at a depth of 0.5 and 1 cm from the bottom of the aperture. The operant box was illuminated by a house light and was enclosed in a sound-attenuating chamber equipped with a ventilation fan. When the mouse poked its nose through the apertures in the back and front walls, the control infrared photo-beam was interrupted to detect the mouse’s responses. The tone site nose poke induced an 8 kHz tone (0.5 s, 85 dB) from the speaker. At the food site, a small food pellet (20 mg) was delivered into the aperture through the food dispenser. All experimental data were recorded with an EPSON personal computer that was connected to the operant box via an interface using MED-PC IV software (Med-Associates). The beginning of the sequential tone-food waiting task was signaled by turning on the house light, and termination was indicated by turning off the house light. The behavioral instrumental response in this task was for the mouse to hold its nose in a fixed posture in either the tone site aperture while waiting for the conditioned reinforcer tone or the reward site aperture while waiting for the food reward. This task required the mice to perform alternate visits and nose pokes to the tone site and the reward site. The mouse initiated a trial by nose poking in a fixed posture to achieve continuous interruption of the photo-beam at the tone site during a delay period until the tone was presented, signaling that a food reward was available at the reward site. After the tone was presented, the mouse was required to continue nose poking at the reward site during another delay period until the reward was delivered. The delay period that preceded the tone was called the tone delay and that which preceded the food was termed the reward delay. During the initial training period, the tone delay and the reward delay were fixed at 0.2 s. Two types of error were present in this task: the tone wait error and the reward wait error. The tone wait error and the reward wait error occurred when the mouse failed to wait for the tone and the food, respectively, during the delay period, by keeping its nose in a fixed posture. After the tone wait error, the mouse could restart the trial until it succeeded in waiting for the tone. A trial ended when the mouse received the food or a food wait error. During a trial, the tone wait error could occur multiple times. By contrast, the reward wait error could only occur one time. Occurrences of tone and reward wait errors were not signaled. Mice could start the next trial at any time after food consumption or after making a reward wait error. Mice were trained daily for a period of 2 h. In 2 weeks or less, mice learned the sequential tone-food waiting task. In vivo optical stimulation during the task During the test session, an external optical fiber (400 μm diameter, 0.48 NA, Doric Lenses) was coupled to the implanted optical fiber with a zirconia sleeve. The optical fiber was connected to an optic swivel (Doric Lenses) that allowed unrestricted in vivo illumination. The optic swivel was connected to 470 nm blue and 590 nm yellow LEDs (470 nm: 35 mW, 590 nm: 10 mW, Doric Lenses) to generate the blue and yellow light pulses through the optical fiber (960 μm diameter, 0.48 NA, Doric Lenses). Blue and yellow light power intensities at the tip of the optical fiber, as measured by the power meter, were 1.2–2.8 mW and 1.4–1.8 mW, respectively. The LED was controlled by the transistor-transistor-logic pulses generated by a MED-PC IV. Experiment 1: effect of reward probability and reward value To examine whether reward prediction modulates the effect of serotonin on patience during waiting, we prepared six tests in which the RP and the reward amount were changed (75% reward one-pellet, 75% reward two-pellet, 25% reward one-pellet, 25% reward three-pellet, 50% reward one-pellet, and 50% reward three-pellet tests) (Supplementary Fig. 1 ). The tone and reward delays were fixed at 0.3 and 3 s, respectively. One test of experiment 1 lasted 3000 s or until the mouse completed 40 trials. The tones in the 75% one-pellet, 75% two-pellet, 25% one-pellet, 25% three-pellet, 50% one-pellet, and 50% three-pellet tests were set at 8 kHz (0.5 s), white noise (0.5 s), 2 kHz (0.25 s) followed by 7 kHz (0.25 s), click (0.5 s), 7 kHz (0.25 s) followed by 2 kHz (0.25 s), and 2.5 kHz (0.5 s), respectively. Removing the nose for >500 ms before the end of the reward-delay period caused a reward wait error, in which no reward was presented. The trials in which serotonin neurons were or were not optogenetically stimulated were named serotonin activation trials or serotonin no-activation trials, respectively (Supplementary Fig. 1 ). For serotonin activation trials, 0.8 s of blue light was randomly applied for half of the trials at the onset of the nose poke to the reward site following the tone presentation. For serotonin no-activation trials, 0.8 s of yellow light were applied for half of the trials at the onset of the nose poke to the reward site following tone presentation. One trial was ended by applying 1 s of yellow light at the onset of food presentation or the reward wait error (Supplementary Fig. 1 ). We executed 75, 25, and 50% reward tests separately. The sequence of 75, 25, and 50% tests was changed for each mouse. During the 75% reward test, 1 or 2 days were used for training in the one-pellet and two-pellet tests and then the recording sessions were started. Each mouse experienced both the one-pellet and two-pellet tests at least once per day. During recording sessions, the order of the one-pellet and two-pellet tests was counterbalanced by daily recording. During both the 25 and 50% reward tests, 1 or 2 days were used for training in the one-pellet and three-pellet tests and then recording sessions were started. Each mouse experienced both one-pellet and three-pellet tests at least once per day. During the recording sessions, the order of the one-pellet and three-pellet tests was counterbalanced by daily recording. Experiment 2: effect of reward timing uncertainty To examine whether the timing of presentation of an expected reward influences promotion of patience by serotonin, we prepared four delayed reward tests with 75% RP, in which the timing of reward delivery was changed: (i) the reward delay was fixed at 6 s (D6 test) (Supplementary Fig. 6a ); (ii) the reward delay was randomly set to 4, 6, or 8 s (D4-6-8 test) (Supplementary Fig. 6b ); (iii) the reward delay was randomly set to 2, 6, or 10 s (D2-6-10 test) (Supplementary Fig. 6c ); and (iv) the reward delay was fixed at 10 s (D10 test). One test of experiment 2 lasted 3000 s or until the mouse completed 40 trials. The tone was 0.5 s at 8 kHz and was fixed through four reward-delay conditions. Removing the nose for >500 ms before the end of the reward-delay period caused a reward wait error, in which no reward was presented. Light stimulation patterns during the serotonin activation and serotonin no-activation trials were the same as in experiment 1. In the D4-6-8 and D2-6-10 tests, the eight trial patterns (two light conditions multiplied by four delay lengths) were randomly selected without repetition until all items were selected, and then this selection was repeated five times. In the D6 and D10 tests, eight trials (three fixed delay with serotonin activation, one omission with serotonin activation, three fixed delay without serotonin activation, and one omission without serotonin activation) were randomly selected without repetition until all items were selected, and then this selection was repeated five times. We executed the D6, D4-6-8, D2-6-10, and D10 test sessions in this order. In each reward-delay test session, the first day was a training session followed by 3 or 4 days of recording sessions. The 1-day recording sessions consisted of at least one reward-delay test. For two mice, D4-6-8 and D6 test sessions were further executed in this order after D2-6-10 test session (one mouse) or D10 test session (one mouse). Since in both D6 and D4-6-8 test sessions, waiting time in omission trials did not differ significantly between first and second sessions, data from first and second sessions were merged for analysis (in the D6 test, P > 0.10 with serotonin activation, P > 0.10 without serotonin activation, Mann–Whitney U -test; in the D4-6-8 test, P > 0.79 with serotonin activation, P > 0.13 without serotonin activation, Mann–Whitney U -test). Data analysis No statistical tests were used to determine sample size, but our sample sizes were similar to those employed in our previous study 7 . To examine how serotonin neuron activation promotes waiting for delayed rewards, we focused on waiting time during omission trials. To quantify effectiveness of serotonin neuron activation at promoting waiting time during omission trials, we calculated the waiting time ratio (waiting time with serotonin neuron activation/waiting time without serotonin neuron activation) for each test. Statistically significant differences (waiting time or waiting time ratio) between two groups were assessed by Mann–Whitney U -test. To compare waiting time in serotonin activation and in serotonin no-activation by within animal averages, we used paired t -test. For analysis of ChR2-expressing group (ChR2) data and control group (WT) data, two-way ANOVA using light effect (two levels; yellow and blue) as within-subject factors and group effect (two levels; ChR2 and WT) as between-subject factors were used. The normality of data for paired t -test and two-way ANOVA were assessed by Shapiro–Wilk test. We have checked a homogeneity of variance of the waiting time ratio data in experiments 1 and 2. Since data did not satisfy homogeneity of variance in both experiment, non-parametric statistical tests were used. To examine the main effect of RP (three level; 75, 50, and 25%) and that of expected reward value (four levels; 0.25, 0.5, 0.75 and 1.5 EPs per trial) on promoting waiting time, Scheirer–Ray–Hare test, which is non-parametric method equivalent to two-way ANOVA, followed by the Bonferroni correction for multiple comparisons was used for analysis of the waiting time ratio. A linear mixed model analysis was performed, taking the waiting time ratio ( Y ) as a dependent variable, RP, and EP as independent variables with fixed effect, and MI as an independent variable with random effect. We fitted the model to data using R package {lme4} with the formula Y = RP + EP + (1|MI). To test difference of means, we used Z -value instead of t -value because the degree of freedom of t -value is not readily available for an unbalanced mixed model. Further, to test whether variance of mice is zero, it is not appropriate to use a χ 2 -test because the null hypothesis is located in the end of domain of variance. As a bail-out method, we used a parametric bootstrap. Kruskal–wallis test followed by Bonferroni correction for multiple comparisons was used for analysis of the waiting time ratio in experiment 2. In Bonferroni correction for multiple comparisons, P -values of pairwise Mann–Whitney U -tests were multiplied by m , where m was the number of pairwise Mann–Whitney U -tests. Statistically significant differences were achieved when P -value × m < 0.05. m was 15 and 10 in Scheirer–Ray–Hare test and Kruskal–wallis test, respectively. Data collection and analysis were not performed blind during the experiment, and no randomization was used. In a very small number of omission trials, mice removed the nose from the reward site within 1.5 s (in the 75% one-pellet test, 2 for serotonin activation trial and 2 for serotonin no-activation trial; in 50% three-pellet test, 3 for serotonin activation trial, and 4 for serotonin no-activation trial; in the 50% one-pellet test, 1 for in serotonin activation trial; in the 25% three-pellet test, 4 for serotonin activation trial and 2 for serotonin no-activation trial; in the 25% one-pellet test, 1 for serotonin activation trial and 1 for serotonin no-activation trial; in the D10 test, two for serotonin no-activation trial). These data were excluded from the analysis. Statistical analyses were performed using SPSS, Matlab (MathWorks), and R. Bayesian decision model of waiting Each trial had a hidden state X = {reward, no-reward}, and for a reward trial, the timing of reward delivery was given by a Gaussian distribution N ( t ; μ , σ 2 ). Given an observation that a reward had not been delivered by time t , the likelihood for a reward trial was 1 – f ( t ; μ , σ 2 ), where f is the cumulative Gaussian density function, whereas the likelihood for a no-reward trial was one. The posterior probability for a reward trial, given observation of no reward by time t is $${P}\left({{\mathrm{reward}|t}} \right) = {P}\left( {{\mathrm{reward}}} \right) \times (1-{f}({t};\mu,\sigma^{2}))/[P\left( {{\mathrm{reward}}}\right) \times ({\mathrm{1}}-{f}({t};\mu,\sigma^{2})) \\ + {P}\left({\rm{no}}\,{\rm{reward}}\right)],$$ where P (reward) and P (no reward) are prior probabilities of reward and no-reward trials. The expected reward to keep waiting was V (wait| t ) = P (reward| t ) for a unit of reward, while the expected reward for quitting was V (quit| t ) = 0 as no reward is obtained by quitting. By assuming a softmax action selection, the choice probability to keep waiting at time t is $${{P}}\left( {{\mathrm{wait|}}{t}} \right) = {\mathrm{1/(1}} + {\mathrm{exp[}}-{\beta} \times {P}\left( {{\mathrm{reward|}}{t}} \right){\mathrm{]),}}$$ where β is the inverse temperature parameter regulating the stochasticity of choice. The distribution of the time of quitting P quit ( t ) is given by sequential decisions: $$\begin{array}{l}{P}_{{\mathrm{wait}}}\left( {\mathrm{0}} \right) = {\mathrm{1,}}\\ {P}_{{\mathrm{wait}}}\left( {{t}} \right) = {P}_{{\mathrm{wait}}}{(t}-{\tau}) \times {P}\left( {{\mathrm{wait|}t}} \right){\mathrm{,}}\\ {P}_{{\mathrm{quit}}}\left( {t} \right) = {P}_{{\mathrm{wait}}}{{(t}}-{{\tau)}} \times \left( {{{1}}-{P}\left( {{\mathrm{wait|}t}} \right)} \right){\mathrm{,}}\end{array}$$ where P wait ( t ) is the probability of continuing to wait until time t and τ is the interval of repeated decision to wait or to quit. In Fig. 7 , we used parameters τ = 0.1 s and β = 50. The code of the Bayesian waiting decision model was written in Python. Code availability The code used to generate the results that are reported in this study are available from the corresponding author to responsible request. Data availability Data from the experiments presented in this study are available from the corresponding author to responsible request.

People usually have the ability to put aside the desire for immediate gratification in anticipation of something good. But this isn't just a human trait—a new study shows that mice can be patient, too, revealing a link between the brain's chemical system and the mice's belief about how waiting will pay off. The effect of the neuromodulator serotonin on mice's ability to stay patient when waiting for a reward is at the core of a new study published in Nature Communications. The authors, Dr. Katsuhiko Miyazaki and Dr. Kayoko Miyazaki, analyzed how the rodents behaved under the influence of serotonin, as part of a study conducted in the Neural Computation Unit at the Okinawa Institute of Science and Technology Graduate University (OIST). Serotonin is a chemical messenger that influences neuron functions. It has been linked to a wide array of behaviors encompassing mood, sleep, cravings and spontaneity. The power of the chemical over human behavior has made it a key focus in the treatment of mental conditions such as depression by selective serotonin receptor inhibitors (SSRIs), which slow down the reabsorption of serotonin and keep it active in the brain. "Serotonin has had a lot of study in pharmacology, and serotonergic drugs are commonly prescribed," said Katsuhiko Miyazaki, "but the role that serotonin has over behavior isn't clear." The team investigated for a causal relationship between serotonin levels and behavior in mice. The mice were trained to perform a task to obtain a food reward, placing their noses into a small hole and waiting—a behavior called a "nose poke." After a pre-set duration, the reward was delivered. In a previous study, the team used a method called optogenetics, a method allowing scientist to use light to stimulate specific neurons with precise timing. These neurons are genetically modified to a produce a light-sensitive protein and are then stimulated by shining light along a fiber optic implanted in the brain. In the study, serotonin-producing neurons were optogenetically stimulated in a part of the brain called the dorsal raphe nucleus (DRN). These neurons then fired widely into the forebrain. The result was that increasing the activity of serotonin neurons in the DRN drastically increased the amount of time mice were willing to wait for a food reward. In a test where food rewards were always delivered after 6 seconds, serotonin's effect of extending nose-poke time was small (left). But in tests where food goes was delivered after two, six, or ten seconds later, serotonin boosted nose poke times significantly (right). Credit: OIST While the study showed that serotonin increased patience, the latest study tested whether mice respond similarly in circumstances when getting a reward was uncertain. Would mice wait for food regardless of the probability and timing of its presentation, or would they give up if they predicted a low chance of return on their time investment? The new trials showed there are limits to serotonin's ability to enhance patience. Mice were given a nose-poke trial with a 75 percent chance of a reward, with a three second waiting period before the reward was delivered. When these mice were subject to a no-reward outcome, their waiting time was prolonged, as expected from the previous paper. However, in tests where the chance of reward delivery following a nose-poke was 50 percent or 25 percent, increasing serotonin had no effect on the mice's waiting time. "The patience effect only works when the mouse thinks there is a high probability of reward," said Dr. Miyazaki. They also found that serotonin stimulation made the mice to wait longer when the timing of a reward was harder to predict. In some sessions with a 75 percent chance of getting a reward, mice were rewarded after precise periods, while in other sessions, they were rewarded after randomized timing. The extended waiting times by serotonin neuron stimulation were more prominent when the reward timing was randomized. (Fig. 2). To explain the results of their experiment, the team constructed a computational model to explain the experimental data. In the model, the mice were able to expect when a food reward would be delivered, and to judge when they were subject to a no-reward trial. The model could reproduce the experimental results by assuming that serotonin affects confidence of receiving a reward when subjective confidence is high. In a 75 percent reward probability trial, for example, serotonin made the mice behave as if there was a 95 percent chance of reward. The model also reproduced the result of timing uncertainty. When the mice were uncertain of the timing of reward delivery, it became difficult for them to judge whether they were waiting in a reward trial or no-reward trial. Serotonin stimulation increased the mice's belief that they were in a reward trial, delaying their judgment further as reward timing was less clear. The findings show that the relationship between the activation of serotonin and subsequent behavior is highly dependent on the animals' belief about the circumstances. These results may have implications for our understanding of how humans taking serotonin boosting drugs can also be affected. "This could help explain why combined treatment of depression with SSRIs and Cognitive Behavior Therapy (CBT) is more effective than just SSRIs alone," said Dr. Miyazaki. "The psychological boost of the therapy is enhanced by raised serotonin levels."

10.1038/s41467-018-04496-y

Medicine

Activation of a key protein that allows glioblastoma cells to complete apoptosis

Laura Martínez-Escardó et al, Gossypol Treatment Restores Insufficient Apoptotic Function of DFF40/CAD in Human Glioblastoma Cells, Cancers (2021). DOI: 10.3390/cancers13215579

http://dx.doi.org/10.3390/cancers13215579

https://medicalxpress.com/news/2021-11-key-protein-glioblastoma-cells-apoptosis.html

Abstract Guidelines Hypothesis Interesting Images Letter New Book Received Obituary Opinion Perspective Proceeding Paper Project Report Protocol Registered Report Reply Retraction Short Note Study Protocol Systematic Review Technical Note Tutorial Viewpoint All Article Types Advanced Search Section All Sections Cancer Biomarkers Cancer Causes, Screening and Diagnosis Cancer Drug Development Cancer Epidemiology and Prevention Cancer Immunology and Immunotherapy Cancer Informatics and Big Data Cancer Metastasis Cancer Pathophysiology Cancer Therapy Clinical Trials of Cancer Infectious Agents and Cancer Methods and Technologies Development Molecular Cancer Biology Pediatric Oncology Systematic Review or Meta-Analysis in Cancer Research Transplant Oncology and Cancer Nursing Care Tumor Microenvironment All Sections Special Issue All Special Issues "The 10th International MDM2 Workshop"—Opening Up New Avenues for MDM2 and p53 Research T Cells and Myeloid Cells in Cancer Immunotherapy Helicobacter pylori Associated Cancer TP53 in Solid Tumors and Hematological Malignancies How Does Obesity Cause Cancer? Methylation Changes in Early Stage Non-small Cell Lung Cancer: New Evidences, Methodological Approaches and Applications Cancer and Liposomes 10th Anniversary of Cancers—Targeted Therapies for Ovarian Cancer Treatment from Bench to Bedside 2nd Edition: Adolescent and Young Adult Oncology 2nd Edition: Advances in the Management of Thyroid Cancer 2nd Edition: Colorectal Cancers 2nd Edition: Combination and Innovative Therapies for Pancreatic Cancer 2nd Edition: Estrogen Receptor-Positive (ER+) Breast Cancers 2nd Edition: New Perspectives of Ocular Oncology 2nd Edition: Resistance Mechanisms in Malignant Brain Tumors 3rd Etnean Occupational Medicine Workshop—Breast Cancer and Work 3D Cell Culture Cancer Models: Development and Applications A Deeper Dive into Signaling Pathways in Cancers A Selection of Papers from the Cancer Therapy 2021 Virtual Scientific Meeting A Selection of Papers from the Meeting on Cancer Science and Targeted Therapy Conference A Special Issue on Tumor Stroma Actionable Mutations in Lung Cancer Acute Myeloid Leukemia Acute Myeloid Leukemia (AML) Acute Myeloid Leukemia: From Diagnosis to Treatment Acute Myeloid Leukemia: The Future Is Bright Acute Promyelocytic Leukemia Acute Promyelocytic Leukemia (APML) Adaptation Strategies of Circulating Tumor Cells Adhesion and Integrins Adjuvant Chemotherapy for Colorectal Cancer Adrenocortical Carcinoma Advance and New Insights in Bladder Cancer Advance in Colorectal Cancer: A Themed Issue in Honor of Prof. Bengt Glimelius Advance in Computational Methods in Cancer Research Advance in Morbidity and Survivorship of Breast Cancer Advance in Supportive and Palliative Care in Cancer Advance Research in Thrombosis and Hematologic Malignancies Advanced Cancer Nanotheranostics Advanced Neuroendocrine Tumors Advanced NSCLC with Oncogene Addiction: An Ongoing Revolution Advanced Ovarian Cancer Advanced Pancreatic Cancer Advanced Prostate Cancer: From Bench to Bedside Advanced Research in Cancer Initiation and Early Detection Advanced Research in Glycoproteins and Cancer Advanced Research in Oncology in 2022 Advanced Research in Oncology in 2023 Advanced Research in Organs-on-a-Chip and Cancer Advanced Research in Pancreatic Ductal Adenocarcinoma Advanced Therapies for Hematological Cancers Advances and Research Progress in Hepatocellular Carcinoma Advances in Adrenocortical Carcinoma Research: Diagnosis, Treatment and Prognosis Advances in Antitumor Molecular-Targeted Agents of Urological Cancers Advances in Bone Metastatic Cancer Research Advances in Breast Cancer Brain Metastases Advances in Breast Cancer: A Themed Issue in Honor of Prof. Fernando Schmitt Advances in Breast Cancer: From Pathogenesis to Therapy Advances in Cancer Chemoprevention Advances in Cancer Cachexia Advances in Cancer Data and Statistics Advances in Cancer Disparities Advances in Cancer Epigenetics Advances in Cancer Radiotherapy Advances in Cancer Stem Cell Research Advances in Chordoma Advances in Chronic Lymphocytic Leukaemia (CLL) Research Advances in Diagnosis and Treatment for Bone and Soft Tissue Sarcoma Advances in Diagnosis, Treatment and Management of Endocrine Neoplasms Advances in Diagnostics and Treatment of Head and Neck Cancer Advances in Endometrial Cancer: From Pathogenesis, Pathology Diagnosis and Molecular Classification to Targeted Therapy Advances in Experimental Radiotherapy Advances in Follicular Lymphoma Advances in Genetic and Molecular Approaches to Skin Cancer Advances in Genetics and Epigenetics of Bladder Cancer Advances in Gynecological Oncology: From Pathogenesis to Therapy Advances in Head and Neck Cancer Research Advances in Head and Neck Squamous Cell Carcinoma Advances in Hematological Neoplasms: A Wide Perspective on the 2022 WHO and International Consensus Classifications Advances in HPB/GI Imaging Advances in Human-Papillomavirus-Related Squamous Cell Carcinoma: From Pathogenesis to Treatment Advances in Integrins in Cancer Advances in Locally Advanced and Metastatic Kidney Cancer Advances in Lymphoma, Plasma Cell Myeloma, and Leukemia Diagnostics Advances in Modern Radiation Oncology Advances in Neuroendocrine Neoplasms Advances in NK/T-cell Lymphoma, Epidemiology, Biology and Therapy Advances in Oral Cancers and Precancers Advances in Our Understanding of ALK-Related Cancers: A Selection of Papers from the Joint Annual Meeting of the European Research Initiative for ALK-Related Malignancies (ERIA) and the European Union Marie Curie European Training Network ALKATRAS Advances in Pancreatic Cancer Research Advances in Pancreatic Ductal Adenocarcinoma Diagnosis and Treatment Advances in Papillary Thyroid Cancer Research Advances in Parathyroid Carcinoma: From Bench to Bedside Advances in Plasma Cell Dyscrasias Advances in Precision Medicine: Targeting Known and Emerging Oncogenic Targets in Lung Cancer Advances in Research, Diagnosis and Treatment of Brain Metastases Advances in Salivary Gland Carcinoma Advances in Soft Tissue and Bone Sarcoma Advances in Stimuli-Responsive Nanostructures for Cancer Therapy and Diagnosis Advances in Surgical Management of Colorectal Liver Metastases: Toward a Better Patient Selection, Lower Surgical Stress, and Multidisciplinary Approach Advances in the biological responses to radiation-induced DNA damage: A selection of papers from the Joint 43rd European Radiation Research Society (ERRS) and 20th German Society for Biological Radiation Research (GBS) Annual Meetings Advances in the Diagnosis, Prognosis and Treatment of Diffuse Large B-cell Lymphoma Advances in the Management of Hepatocellular Carcinoma Advances in the Management of Oligometastatic Disease in Non-colorectal Non-neuroendocrine Tumors Advances in the Management of Thyroid Cancer Advances in Thoracic Carcinoma and Translational Research Advances in Thoracic Oncology Advances in Thymic Tumors Advances in Translational Ovarian Cancer Research Advances in Translational Research for Soft Tissue Sarcomas Advances in Treatment for Hepatobiliary and Pancreatic Cancers: Multi-Disciplinary Strategies and Outcome Evaluation Advances in Treatment of Rare Tumors Advances in Triple-Negative Breast Cancer Advances in Tumor Angiogenesis Advances of Brain Mapping in Cancer Research Advances in Head and Neck Cancer Biology and Clinical Management Advances in In Vivo Quantitative and Qualitative Imaging Characterization of Gliomas Advancing Cancer Research by Exploring the Tissue Engineering Toolbox AGC Kinases and Cancer Aggressive Prostate Cancer Aging and Cancers Alcohol and Cancer Algorithms and Data Analysis of High Throughput Sequencing in Cancers Alternative Lengthening of Telomeres in Neoplasia An Update on Surgical Treatment for Hepato-Pancreato-Biliary Cancers Androgen Receptor in Cancers: Not Only Prostate Anesthesia and Cancer Recurrence: A New Sight Anesthesia and Cancers Animal Models for Radiotherapy Research Annexin Proteins Family in Cancer Antibody-Drug Conjugates—a Coming of Age Anticancer Immunity with Physical Treatment Modalities Antigens and Cancer Therapy Antioxidants in Cancer Apoptosis in Cancer Application of Bioinformatics in Cancers Application of Emerging Technologies in Zebrafish and Mammalian Cancer Models for Disease Characterization and Therapeutic Target Discovery Application of Multi-Omics Analysis in Cancer Diagnosis, Treatment and Prognosis Application of New Molecular Probes in the Diagnosis and Treatment of Malignant Tumors Application of Next-Generation Sequencing in Cancers Application of Proteomics in Cancers Application of Ultrasound in Breast Cancer Applications of Different Knowledge Graphs and Large Language Models in Diagnosis Cancers Applications of Machine Learning and Statistical Modeling in Precision Oncology Approaches to Improve the Prognosis of Head-and-Neck Cancer Aptamers: Promising Tools for Cancer Diagnosis and Therapy AR Signaling in Human Malignancies: Prostate Cancer and Beyond Artificial Cells for Use in Cancers Artificial Intelligence and Deep Learning in Radiology Oncology Artificial Intelligence and Machine Learning in Cancer Research Artificial Intelligence and MRI Characterization of Tumors Artificial Intelligence in Cancer Research: Knowledge Representation and Data Perspectives Asbestos and Cancer Autophagy and Cancer B Cell Malignancies (including B Cell Lymphoma, Multiple Myeloma and B-CLL) Basal Cell Carcinoma of the Head and Neck Basic and Translational Research on Cancer Immunology and Immunotherapy – Selection Papers from The 8th Symposium on Advances in Cancer Immunology and Immunotherapy (SACII) Basic Research of Hepatopancreatobillary Tumor Beyond JAK Inhibition: Molecular Pathogenesis and Novel Therapeutic Strategies for the Treatment of Myeloproliferative Neoplasms (MPNs) Biliary Tract Tumors: Update in Diagnosis and Treatment Biogenesis and Function of Extracellular Vesicles in Cancers Biological Function for Laryngeal Cancer in Immunotherapy Biomarker of Lung Cancer: Early Detection, Chemoprevention and Treatment Biomarkers, Diagnostic Tools, Treatment Outcomes, and Late Complications in Neuro-Oncology Biomarkers: Oncology Studies Biomaterial-Assisted 3D In Vitro Tumor Models: From Organoid towards Cancer Tissue Engineering Approaches Biomedical Informatics and Cancer Blood Immune Cell and Cancer Therapeutics Body Composition in Oncology and Beyond Bone and Soft Tissue Tumors BRAF Mutation in Colorectal Cancer BRAF Mutations and Rare Aggressive Variants in Thyroid Cancer Brain Cancer Radiotherapy Brain Cancer: Use of Natural Derivatives as Anticancer Agents Brain Metastases Research Updates Brain Metastasis in Breast Cancer Brain Tumor: Recent Advances and Challenges BRCA Mutations and Cancer Breast Cancer Breast Cancer – Therapeutic Challenges, Research Strategies and Novel Diagnostics Breast Cancer and Hormone-Related Therapy Breast Cancer Biology and Treatment Breast Cancer Biomarkers and Clinical Translation Breast Cancer in Young Women Breast Cancer Metastasis: Novel Insights into Molecular Mechanisms and Treatments Breast Cancer Radiation Therapy Breast Cancer Recurrence: Symptoms & Treatment Breast Cancer Stem Cells: Therapy Resistance and Novel Therapeutic Targets Breast Cancer Survivors and Supportive Therapies Breast Cancers: Pathology and Biomarkers Breast Development and Cancer (Volume II) Bridging the Gap between Translational Research and Treatment in Diffuse Large B-cell Lymphoma Cancer and Chronic Illness Cancer and Diabetes: What Connections Lie between Them? Cancer and Non-cancer Effects following Ionizing Irradiation Cancer and Pregnancy Cancer Biomarkers Cancer Cachexia Cancer Cachexia: Molecular Insights and Clinical Implications Cancer Cell Imaging Cancer Cell Invasion Cancer Cell Metabolism, Glycolysis, Lactate Production and Transport and Potential Therapeutic Options for Inhibitors Cancer Cell Motility Cancer Cell Proliferation Cancer Chemoresistance Cancer Detection in Primary Care Cancer Diagnosis and Targeted Therapy Cancer Disease: Beyond the Border of Therapy Cancer Dormancy: Linking Laboratory and Clinical Findings Cancer Drug Resistance Cancer Epigenetics Cancer Evolution Cancer Genetics and Epigenetics: Their Roles and Clinical Implications Cancer Immunotherapies Cancer Invasion and Metastasis Cancer Metabolic Landscapes and Interactions Cancer Metabolism Cancer Minimally Invasive Surgery Cancer Molecular Imaging Cancer Nanomedicine Cancer Nanotherapy and Nanodiagnostic Cancer Neuroscience Cancer Organoids in Basic Science and Translational Medicine Cancer Pain: From Basic Research to Drug Discovery and Clinical Studies Cancer Pains Cancer Predisposition Syndromes: Genomics, Surveillance and Treatment Paradigms Cancer Signaling Pathways and Crosstalk Cancer Stem Cells Cancer Stem Cells and Personalized Medicine for Gynecologic Cancers Cancer Stem Cells and Tumor Microenvironment Cancer Therapy Targeting the Fibrinolytic System Cancer Vaccines and Immunotherapy Cancer Vaccines: Research and Applications Cancer-Associated Cachexia Cancer-Associated Fibroblasts Cancer-on-a-Chip: Applications and Challenges Cancers and Aging Cancers Gene Therapy Cancers Precision Immunotherapy Cancers: Molecular Imaging and Therapy CAR-T Cell in Human Cancers: Combinations, Gene-Editing, Payload Delivery, Autonomous Control and Synthetic Biology CAR-T Cell Therapy against Different Types of Cancer CAR-T Cell Therapy-Novel Approaches and Challenges CAR-T Cells: Past, Present, and Future Carbon-Ion Radiotherapy for Cancer Treatment Cardio-Oncology: An Emerging Paradigm in Modern Medicine CDK Targeting in Cancer Therapy CDK4/6 Inhibitors in Breast Cancer CelebratING 25 Years of the ING Family Proteins as Epigenetic Regulators in Cancer Cell Cycle Deregulation in Cancers Cell Death and Cancer Cell Death in Cancer Cell Invasion in Cancer Metastasis Cell Signaling in Cancer and Cancer Therapy Cellular Differentiation in Melanoma Development Cellular Plasticity and the Untapped Therapeutic Potential in Cancer Cellular Stress in Cancer Progression, Drug Resistance and Treatment Central Nervous System Tumors—the 2021 WHO Classification and Beyond Cervical Carcinoma Challenges and Opportunities in Implications of Omics for Breast Cancer in the Era of Precision Medicine and Beyond—Encompassing a Global View Challenges in Cancer-Associated Thrombosis Changing Landscape of Hereditary Breast and Ovarian Cancer 2.0 Chemoprevention Advances in Cancer Chemoradiotherapy for Head-and-Neck Cancer in the Elderly Chemoresistance in Solid Tumours Childhood Cancer in the Genomic Era: Experimental and Theoretical Approaches Childhood Leukemia Chromatin as a Target for Cancer Therapy Chromosomal Instability and Cancers Chromosomal Rearrangements in Haematological Malignancies Chromosome Instability and Aneuploidy in Cancer: State of the Art and Future Perspectives Chronic Lymphocytic Leukemia Chronic Lymphocytic Leukemia: Identification of Novel Prognostic Markers and Their Clinical Application Circadian Rhythms, Cancers and Chronotherapy Circulating Tumor Cells (CTCs) Circulating Tumor Cells in Cancers Circulating Tumor Cells' Heterogeneity and Precision Oncology Clear Cell Renal Cell Carcinoma 2022–2023 Clear Cell Renal Cell Carcinoma: From Biology to Treatment Clinical and Genetic Findings in Patients with Neurofibromatosis Type 1 Clinical and Translational Research in Gastrointestinal Cancers Clinical and Translational Research in Pediatric Surgical Oncology Clinical Application of Head and Neck Cancer Research Clinical Outcomes and Follow-Up Care in Gynecological Cancers Clinical Pharmacology in Cancer Clinical, Pathological and Molecular Peculiarities of Lobular Breast Cancer Clinical, Pathological, and Molecular Characteristics in Colorectal Cancer Cognitive Outcomes in Cancer: Recent Advances and Challenges Colorectal Cancer Heterogeneity and the Impact on Metastasis Formation and Therapy Efficacy Colorectal Cancers Colorectal Cancers: From Present Problems to Future Solutions Colorectal Cancers: From Present Problems to Future Solutions 2.0 Colorectal Liver Metastasis (Volume II) Combination Therapies in Cancers Commemorates Reaserches From Les Compagnons Hepat Biliaires 2023 Communication and Accessibility in the Tumor Microenvironment as a Therapeutic Target Comprehensive Review on Upper Tract Urothelial Carcinoma: An Update in 2023 Connexins in Cancer Conservative Axillary Surgery for Breast Cancer Contemporary Management for Gallbladder Cancer: From Diagnosis to Treatment Contemporary Perspectives and Emerging Trends in the Management of Gastric Cancer Coordination of p53 Functions in Normal and Cancer Cells Correlating Patient-Reported Quality of Life before, during, and after Therapy with Objective Measures and Outcomes Crosstalk between Cancer-Associated Fibroblasts and Cancer Cells Crosstalk between Inflammation and Carcinogenesis Curative Strategies for the Management of Hepatocellular Cancer Current Advances in Systemic Therapy for Unresectable Hepatocellular Carcinoma Current and New Insights in Theranostics of Endocrine Tumors Current Challenge and Future Advances for Lung Cancer: Genetics, Instrumental Diagnosis and Treatment 2.0 Current Challenges and Opportunities in Treating Glioma Current Challenges in Geriatric Oncology Current Concepts in the Diagnosis and Treatment of Cutaneous Melanoma Current Management of Castration-Resistant Prostate Cancer (CRPC) Current Progress and Research Trends in Ocular Oncology Current Status and Future Prospects for Oesophageal Cancer Current Status of Neuroendocrine Tumors with a Special Focus on Diagnosis and Novel Treatments-Volume II Current Topics in Cutaneous Melanoma Current Trends for Sinonasal Carcinoma Current Trends in Epigenetics of Brain Tumors: Basic and Translational Research Current Understanding of RAD52 Functions: Fundamental and Therapeutic Insights Current Updates and Future Directions in Neuroendocrine Neoplams (NENs): Understanding Biology, Diagnosis, Management and Research Efforts in NENs Current Use of PSMA in Prostate Cancer Treatment Cutaneous Lymphomas Cyclooxygenase (COX) and Lipoxygenase (LOX) in the Inflammogenesis of Cancer Cytokines and Cytokine/Chemokine Receptors in Lymphoma, Leukemia and Multiple Myeloma Cytokines in Cancer Cytokines in Cancer Immunotherapy 2.0 Cytologic Features of Tumor Decision-Support Systems for Cancer Diagnosis and Prognosis Deep Neural Networks for Cancer Screening and Classification Deregulation of Cell Death in Cancer (Volume II) Designification & Intelligentsia of Humanized Rodent Models in Cancer Research Desmoid Tumors Detecting and Targeting Mechanisms of Genomic Instability in Breast Cancer Development of Immunotherapeutic Agents against Intraepithelial Neoplasia Developments in Artificial Intelligence and Advanced Medical Imaging in Cancers Diabetes and Breast Cancer Diacylglycerol Kinases in Cancer Diagnosis and Staging of Gastroesophageal Cancer Diagnosis and Therapeutic Management of Gastrointestinal Cancers Diagnosis and Treatment for Bone Tumor and Sarcoma Diagnosis and Treatment for Hepatocellular Tumors Diagnosis and Treatment of Gastroenteropancreatic Neuroendocrine Neoplasms Diagnosis of Melanoma and Non-melanoma Skin Cancer Diagnosis, Treatment and Prevention of Gastrointestinal Cancer Diagnosis, Treatment and Prognosis of Osteosarcoma Diagnostic and Therapeutic Progress in Aggressive Lymphoma Differential Scanning Calorimetry and Related Thermal Analysis Techniques as Complementary Approaches for Cancer Diagnosis, Prognosis, Monitoring and Assessment of Treatment Response Digital Pathology: Basics, Clinical Applications and Future Trends Diversity and Biology of Cancer-Associated Fibroblasts and the Novel Targeting Therapy DNA Damage and Repair in Cancer Risk Prediction DNA Damage Response Targeting: Challenges and Opportunities DNA Methylation Markers in Liquid Biopsies DNA Methylation Markers in Liquid Biopsies (Volume II) DNA Repair Pathways in Cancer DNA Viruses in Human Cancer Does Breast Cancer Surgery Initiate Relapse, What Is the Evidence and How May This Be Arrested? Drug Resistance in Cancers Drug Targeting Therapy in Multiple Myeloma Drug/Radiation Resistance in Cancer Therapy Dynamics of Cancer: Complexity and Hierarchy on Cancer Cells and Tissues E-cadherin Mutations in Cancer Early Detection and Surgery for Pancreatic Cancer Early Gastric Cancer Efficacy and Complications of Liver Resection for Liver Cancer Efforts to Mitigate the Toxicity of Cancer Therapeutics EGFR Family Signaling in Cancer EGFR in Cancer: Innovative Insights into Signalling, Mutation and Therapeutic Targeting EGFR-Mutated Non-small Cell Lung Cancer Electric Field Based Therapies for Cancer: A Selection of Papers from the 2nd World Congress on Electroporation Electric Field Based Therapies in Cancer Treatment: a Selection of Studies Presented at the 3rd World Congress on Electroporation Electrochemotherapy as Treatment for Head and Neck Tumors Electroporation-Based Cancer Treatment. Selected Papers from the 4th World Congress on Electroporation Emerging Biomarkers and Molecular Characterization of Renal Cancers Emerging Concepts in Treatment of Laryngeal Cancer Emerging Roles of Non-coding RNAs in Gynecological Cancer Metastasis and Drug Resistance Emerging Trends in Immunotherapy for Triple Negative Breast Cancer End-of-Life Cancer Care Endometrial Cancer: Old Questions and New Perspectives (Volume II) Endoscopic Diagnosis and Treatment of Early Gastric Cancer: Current Evidence and What the Future May Hold Endoscopic Management of Gastrointestinal, Hepatobiliary and Pancreatic Malignancies Endoscopic Management of Liver and Pancreatic Cancer Endoscopic Ultrasound Fine-Needle Biopsy of Gastrointestinal Tumors Endoscopic Ultrasound in Gastrointestinal Cancers Endothelial Cells in Inflammation, Tissue Repair, Ageing and Cancer Engaging Nanotechnology and Artificial Intelligence Tools for Early Cancer Detection and to Personalize Treatment Environmental Factors in Endocrine-Related Cancers Epidemiologic Research and Cancer Epidermal Growth Factor Receptor (EGFR) Signaling in Cancer Epidermal Growth Factor Receptor Signaling in Cancer Epigenetic Dysregulation in Cancer: From Mechanism to Therapy Epigenetic Influence on Cancer Metastasis and/or Treatment Resistance Epigenetic Regulation in Human Cancers Epigenetics of Cancer Progression Epstein-Barr Virus Infection in Cancer Epstein–Barr Virus Associated Cancers Esophageal Squamous Cell Carcinoma Estrogen Receptor (ER) Signalling Pathway in Cancers Ethical Implications in Cancer Research Ewing Sarcoma Ewing Sarcoma: Basic Biology, Clinical Challenges and Future Perspectives Exosomes in Cancer Development Experimental and Clinical Advances in Counteracting Progression of Solid Cancers Volume II Experimental and Modeling Efforts to Target Metabolism in Cancer Exploring Inflammation in Cancers Exploring Microenvironment Intricacies as Putative Common Targets in Tumors Extracellular Vesicles in Cancer Progression and Drug Resistance FAK Signaling Pathway in Cancers FAP-Ligands and Its Clinical Translation in Cancers Feature Paper from Journal Reviewers Feature Review Papers on Advanced Gastric Cancer Feature Review Papers on Gastroesophageal Junction and Gastric Cancers Fertility and Pregnancy in Cancer Patients: Illusion or Reality Fertility Issues in Cancer Survivors Fertility Preservation in Oncology Fibroblasts and Growth Factors in Cancer Fibroblasts as Playmakers of Cancer Progression: Current Knowledge and Future Perspectives Flow Cytometric Analysis in Cancer Fox Proteins and Cancers: Old Proteins with Emerging New Tales From Bench to Bedside in the Management of Cholangiocarcinoma Frontiers in Hodgkin Lymphoma Frontiers in Neurofibromatosis Frontiers in Radiotherapy Function Sparing Approaches in Pelvic Malignancies Functional and Structural Insights of Non-coding RNA in Cancer Functional Neuro-Oncology Functional Neuro-Oncology—Volume II Fusion Protein-Driven Human Sarcomas: New Molecular Insights and Clinical Opportunities Future Trends and Therapies of Pancreatic Cancer—Where Are We Going from Here On? Gammopathies of Certain Significance: Managing MGUS and Smoldering Myeloma Gap Junctions and Connexins in Cancer Formation, Progression, and Therapy Gastric Cancer Gastric Cancer Metastasis Gastric Cancer Screening in the West Gastric Cancers: Molecular Mechanisms, Novel Targets and Immunotherapies: From Bench to Clinical Therapeutics Gastric-Type Mucinous Carcinoma (GAS) of the Uterine Cervix 2.0 Gastrointestinal Cancers: Advances in Diagnostic, Prognostic, and Therapeutics Gastrointestinal Oncology: Clinical Management Gastrointestinal Stromal Tumors - Recent Progress and Upcoming Challenges in a Diverse Disease Gene Expression Studies in Cancer Research Gene Regulatory Networks in Cancers Genes in Cancer Genetic and Epigenetic Regulation of Tissue Homeostasis in Cancer Genetic and Molecular Epidemiology of Breast Cancer Genetic Findings in Acute Myeloid Leukemia Genetic/Non-genetic Tumor Heterogeneity Genetics and Cancer: Recent Advances and Challenges Genetics and Epigenetics of Leukemia and Lymphoma Genetics and Heterogeneity of Colorectal Cancer Genetics of Ovarian Cancer Genome Maintenance in Cancer Biology and Therapy Genome Maintenance Systems in Cancer Genomic Instability and Cancers Genomic Instability in Multiple Myeloma and Solid Malignancies: Role of DNA Repair from Prognostic Marker to Therapeutic Target Genomics of Chronic Lymphocytic Leukemia (CLL) Genomics of Hematologic Cancers Genomics of Rare Hematologic Cancers Germ Cell Tumors Germline Mutations in Cancer—Implications for Practice Giant-Cell-Containing Tumors of Bone—New Insights into Pathobiology, the Clinical Setting and Targeted Therapies Glioblastoma Glioblastoma: Advances in Molecular Insights and Therapeutic Strategies Glioblastoma: State of the Art and Future Perspectives Global Management of Sarcoma Data: Is Real-Time Predictive Analytics on the Horizon? Gynaecological Cancer and Surgery: Current Practice, Novel Technologies and Future Developments Gynecologic Cancers: Clinical and Translational Research Gynecologic Oncology: Prevention, Screening and Treatment Innovations Hallmark Properties and Behind-the-Scenes Role of Non-coding RNAs in Gastrointestinal Cancers’ Onset, Progression and Metastasis Harnessing the Therapeutic Potential of Targeting Matrix Metalloproteinases for Gastrointestinal Cancer Head and Neck Cancer Head and Neck Cancer Genomics and Translational Applications Heat Shock Proteins in Cancer: Chaperones of Tumorigenesis Heat Shock Proteins in Cancers Hedgehog Signaling in Cancer Hedgehog Signaling Pathway in Cancer: Smoothened and GLI Take Center Stage Hematologic Malignancies: Challenges from Diagnosis to Treatment Hematologic Malignancy Hematologic Malignancy (Volume II) Hepatobiliary Cancers Hepatoblastoma and Pediatric Liver Tumors Hepatocellular Cancer Treatment Hepatocellular Cancer Treatment 2.0 Hepatocellular Cancer: Molecular Mechanisms, Diagnosis and Therapy Hepatocellular Carcinoma: Advances and Challenges in Research and Treatment Hepatocyte Growth Factor Pathway in Cancer Hereditary Syndromes and Radiation High-Risk Localized and Locally Advanced Prostate Cancer Hippo Pathway in Cancer, towards Realization of the Hippo-Targeted Therapy Hippo Signaling in Cancer Hodgkin Lymphoma (Volume II) Hodgkin Lymphoma: Present Status and Future Strategies Hodgkin Lymphoma: Recent Advances and Challenges Hodgkin's Lymphoma Hormone Involvement in Tissue Development, Physiology, and Oncogenesis Hormone Receptors in Genitourinary Tumors Hormone Signaling in Cancer Hormones and Carcinogenesis Hot Topics in Neuro-Oncology How to Improve Chondrosarcoma Treatment? From Fundamental Research to Biomarker Discovery and Clinical Applications HOX Genes in Cancer HPV Associated Cancers Human Hepatocellular Carcinoma (HCC) Human Hepatocellular Carcinoma (HCC) 2.0 Human Papillomavirus (HPV) Associated Head and Neck Cancers: From Basic Biology to Clinical Challenges Human Papillomavirus and Cancers Hyaluronan Family Members in Carcinogenesis Hyperthermia in Cancer Therapy Hyperthermia-based Anticancer Treatments Hyperthermic Intraperitoneal Chemotherapy in Ovarian Cancer Identification of Candidate Genes in Breast and Ovarian Cancer IECC2021: Exploiting Cancer Vulnerability by Targeting the DNA Damage Response IECC2022: Tumor Microenvironment Heterogeneity in Cancer Progression: Challenge or Opportunity IECC2023: New Targets for Cancer Therapies IL-6 and IL-6-Type Cytokines in Cancer Immunotherapy: Signaling, Receptor Blockade and Designer Proteins Imaging and Liquid Biopsy Biomarkers for Cancer Diagnosis and Treatment Imaging in CAR-T Cell Therapy on Cancers Imaging of Gynecologic and Genitourinary Malignancies Immune Checkpoint Inhibitors in Cutaneous Oncology Immune Responses to Human Prostate Cancer Immunity in Melanoma Immuno-Competent 3D Tumour Models to Predict Patient Response Immunohistochemistry and Cancer Diagnosis Immunohistochemistry in Translational Research and Diagnostics of Breast Cancer Immunometabolism and Cancer: Localized and Systemic Metabolic Interactions That Shape the Evolution of Malignancy Immunotherapy for Pancreatic Cancers—Challenges and Perspectives Immunotherapy in Ovarian Cancer Incidence, Mortality, Trend, and Survival of Cancer Infiltrative Gliomas: Emerging Insights in Pathophysiology, Diagnosis, and Management Inflammation and Cancer Inflammation and Cancer Metastasis Inflammatory and Immunological Markers in Liver Cancers Innate Immunity in Cancer Therapy Innovation in Esophageal Cancer Innovations in Diagnosis and Treatment of Colon and Rectal Cancer: Preoperative Optimisation, Multidisciplinary Management, and Surgical Technology Advancement Innovations in Diagnosis, Prognostic Evaluation, and Therapeutic Management of Gynecologic Tumors Innovations in Early Cancer Diagnostics and Therapeutics Innovations in Endocrine Cancer—Technology, Techniques and Therapy Innovative Approaches in the Management of Sinonasal Cancers Innovative Immunotherapies: CAR-T Cell Therapy for Cancers Insights and Advances in the Surgical Management of Hepatocellular Carcinoma Insights in Cancer Endocrinology Insights into Cancer Metabolism from Metabolomics Insights into Urologic Cancer Integrated Management of Cancer Integrating Loco-Regional Hyperthermia in Clinical Oncology Integrating Tumor Evolution Dynamics into the Treatment of Cancer Integrins and Tumor Microenvironment, New Perspectives in Targeted Treatments Integrins in Cancer Intensity Modulated Radiation Therapy Intensive Care and Cancers Intercellular Communication between Tumor and Stromal Cells in Endocrine-Related Cancer Interconnectivity of Cell Death Pathways in Cancer Interdisciplinary Management of Colorectal Liver Metastases in the Era of Precision Medicine Interplay Networks of Driver Oncogenes—Mechanisms and Therapeutic Targeting in Cancer Interventional Oncology: A Theme Issue in Honor of Professor Luigi A. Solbiati Interventional Radiotherapy in Gynecological Cancer Intraductal Cancer of the Prostate (IDC-P): Diagnosis and Characterization Intraoperative Visualization Techniques and Advanced Imaging in Brain Tumors Invasive Skin Cancers and Underlying Compartments Interactions Ion Channels in Cancer Ionizing Radiation in Therapy and Biology of Cancer: Role of Monte Carlo simulations, Biophysical Modeling, and Radiobiological Techniques Iron and Cancer JAK-STAT Signalling Pathway in Cancer Kinase Signaling in Cancer Kinases and Cancer Larynx Cancer: From Diagnosis to Treatment and Rehabilitation Late Recurrence in Breast Cancer Latest Advances in Research on Chronic Lymphocytic Leukemia Latest Development in Melanoma Research Latest Development in Multiple Myeloma Latest Development in Pancreatic Cancer Latest Research in Cartilaginous Neoplasms Leukemia Lifestyle Modifications and Survival of Cancer Patients Linking Obesity to Colorectal Cancer Lipids and Small Metabolites in Cancer Liquid Biopsy for Cancer Liquid Biopsy in Breast Cancer Liquid Biopsy in Hematologic Malignancies Liquid Biopsy in Lung Cancer Liver Cancer and Liver Cirrhosis with Portal Hypertension Liver Cancer and Potential Therapeutic Targets Liver Cancer: Current Surgical Management Liver Interventional Oncology Liver Metastasis of Cancer Locally Advanced and Recurrent Rectal Cancer Locally Advanced and Recurrent Rectal Cancer (Volume II) Locally Advanced Non-small Cell Lung Cancer—Challenges and Current Treatment Options Loco-Regional Arterial Chemotherapies Alone or in Combination with Sistemic Treatments for Primary and Secondary Hepato-Pancreatic Tumors Locoregional Treatment and Gene Targeted Therapies for Cancer Metastasis Long-Read Sequencing in Cancer Low Grade Gliomas Lung Cancer Lung Cancer - Molecular Insights and Targeted Therapies Lung Cancer Biomarkers Lung Cancer: Molecular Pathways, Current Therapies and New Targeted Treatments Lung Cancer—Molecular Insights and Targeted Therapies (Volume II) Lymph Node Dissection for Gynecologic Cancers Lymph Node Dissection in Colorectal Cancer Lymphoma Lynch Syndrome: State of the Art Lysine-Specific Demethylase 1 (LSD1): A Multifaced Epigenetic Enzyme with Multifunctional Roles in Cancer Lysophosphatidic Acid Signalling in Cancer Macrophage Polarization States in Cancer Tumor Microenvironment Macrophages in Cancer Progression, Diagnosis and Treatment Magnetic Resonance Imaging of Brain Tumor Malignant Adrenal Tumors – from Bench to Bedside Malignant Mesothelioma Malignant Pleural Mesothelioma (MPM) Management of Gastric Cancer Management of Glioblastomas Management of Hepatocellular Carcinoma in Liver Disease Management of Locally Advanced Cervical Cancer Management of Neuroendocrine Neoplasms Management of Pancreatic Cancer Management of Pancreatic Cancer: Prediction and Prognostic Factors Management of Relapsed and Refractory Lymphomas Management of Side Effects of Cancer Treatments: New Approaches Mantle Cell Lymphoma: From Biology to Therapy MAPK in Cancers: From Signalling Pathways to Therapeutic Targets Mathematical Oncology: Using Mathematics to Enable Cancer Discoveries Matrix Metalloproteinases in Cancer Progress Measurable Residual Disease in Cancer Mechanisms of Cancer Stem Cells in Melanoma Progression and Treatment Resistance Mechanisms of Genetics in Acute Myeloid Leukemia and Their Influence of Metabolism Mechanisms of mRNA Translation in Pancreatic Cancer Medical Complications and Supportive Care in Patients with Cancer Melatonin and Cancer: Current Challenges and Future Perspectives Meningioma: From Bench to Bedside Meningiomas and Low Grade Gliomas Meningiomas: Update on the Diagnosis and Management Merkel Cell Biology through Clinical Research Merkel Cell Carcinoma Merkel Cell Carcinoma: An Update and Review MET in Cancer Metabolic Reprogramming and Vulnerabilities in Cancer Metastasis and Tumor Cell Migration of Solid Tumors Metastatic Colorectal Cancer Metastatic Lung Cancer Metastatic Progression and Tumour Heterogeneity Microbiome in Cancer: Role in Carcinogenesis and Impact in Therapeutic Strategies-Volume II Microbiota in Colorectal Cancer MicroRNA-Associated Cancer Metastasis Microtubule-Associated Proteins (MAPs) and Cancers Minimally Invasive Surgery in Ovarian Cancer miRNA in Colorectal Cancer Mitochondria and Cancer Mitochondrial function and dysfunction in cancer and their potential as anti-cancer targets Mitochondrial Function and Dysfunction in Cancer and Their Potential as Anti-cancer Targets (Volume II) Models of Experimental Liver Cancer Molecular Advances and Targeted Therapy in Asian Thyroid Practice Molecular and Image Diagnosis in Endometrial Cancer 2024 Molecular Biology and Therapeutic Perspectives for K-ras Mutant Cancers Molecular Diagnostics and Targeted Therapy in Patients with Metastatic Breast Cancer Molecular Genetics and Signaling Pathways in Liver Cancer Molecular Genetics of Pancreatic Cancer and Translational Challenges Molecular Imaging in Tumor Evolution and Therapy Response Molecular Mechanisms and Clinical Implications in Thoracic Cancers Molecular Mechanisms and Clinical Implications of Pancreatic Cancer Molecular Mechanisms in Head and Neck Cancer Molecular Mechanisms of Gastric Cancer Development Molecular Mechanisms of Lung Cancer and Mesothelioma Molecular Pathogenesis of Cervical Cancer Molecular Pathogenesis of Cervical Cancer (Second Edition) Molecular Pathogenesis of Liver Cancers Molecular Pathways in Cancers Molecular Pathways in Metastasis of Lung Cancer Molecular Profiling of Lung Cancer Molecular Signaling Pathways and Networks on Cancer Molecular Stress Response Dysregulation in Cancer: Therapeutic Targets and Opportunities Molecular Targeted Therapy in Cancer Molecular Testing for Thyroid Nodules and Cancer Molecular Tumor Boards: Promise and Limitations for Personalized Cancer Therapy MRI in Breast Cancer mTOR Pathway in Cancer mTOR Signaling in Cancer Development and Growth Multi-Faceted Epigenetic Dysregulation in Acute Myeloid Leukemia Multimodality Management of Sarcomas Multiple Myeloma — Biology, Diagnosis, Treatment and Prognosis Multiple Myeloma: Recent Advances in Diagnosis, Analysis and Therapeutic Management Multiple Signaling Pathways in Ovarian Cancer Multiplexing Immunohistochemistry as an Approach for Diagnosis, Research and Bases for Targeted Therapy Mutation Profiles, NGS and Heterogeneity of Ovarian Cancer Nanobiomaterials for Cancer Early Detection and Therapy Nanotechnology and Cancer Nanotechnology and Cancer Therapeutics Nanotherapeutics in Cancer Management Natural Killer Cells and Cancer Therapy Neoadjuvant and Adjuvant Therapy for Gynecologic Malignancies Neoadjuvant Treatments in Breast Cancer Patients Neuroendocrine Neoplasms: Current Challenges and Advances in the Biological Aspects, Diagnostic and Therapeutic Management Neuroendocrine Neoplasms: Current Challenges and Advances in the Biological Aspects, Diagnostic and Therapeutic Management (Volume II) Neuroendocrine Prostate Cancer Neuroendocrine Tumors Neuroendocrine Tumors: Treatment and Management 2.0 Neurofibromatosis Type 1 (NF1) Related Tumors New Advances and Perspectives for Relapsing Brain Tumors New Advances in Breast Cancer Surgery New Advances in High-Grade Glioma Research New Advances in Metastatic Prostate Cancer New Advances in Urothelial Cancer: Diagnosis, Therapy and Prognosis New and Special Subtypes of Breast Cancer New Approaches in the Management of Head and Neck Cancer New Biomarkers in Cancers New Biomarkers in Cancers (Volume II) New Challenges in Breast Cancer Diagnosis and Management New Challenges in Cancer Imaging New Concepts and Recent Advances in the Management of Skin Cancer New Developments in Radiotherapy New Diagnostic and Therapeutic Tools Against Multidrug Resistant Tumors (STRATAGEM Special Issue, EU-COST CA17104) New Discoveries in Radiation Science: Selection of Papers from the 44th Congress of the European Radiation Research Society (Pécs, Hungary) New Era of Cancer Research: From Large-Scale Cohorts to Big-Data New Era of Neuroblastoma Treatment: DNA Project New Horizons and Surgical Decision Making in HPB Cancer New Insight of Non-small Cell Lung Cancer New Insights and Future Directions in Palliative Care across the Cancer Continuum New Insights in Lymphedema after Cancer to Enhance Clinical Practice New Insights in Neuroendocrine Neoplasms (NENs) and Thyroid Cancer: Focus on Basic, Translational, and Clinical Research New Insights in the Genetics and Genomics of Adrenocortical Tumors and Pheochromocytomas New Insights in the Genetics and Genomics of Adrenocortical Tumors and Pheochromocytomas New Insights in Thoracic Sarcoma New Insights in Tumor-Infiltrating Lymphocytes New Insights into Breast and Endometrial Cancer New Insights into Myeloproliferative Neoplasms New Insights into Neurofibromatosis New Insights into Oligo-Recurrence of Various Cancers New Insights into the Management of Intrahepatic Cholangiocarcinoma New Insights into the Use of Cytotoxic Agents for Cancer Treatment New Insights into Thymic Epithelial Tumors New Insights into Tumour pH Regulation New Insights of Hematology in Cancer New Insights of Malignant Pleural Mesothelioma New Insights on the Hippo-YAP/TAZ-TEAD Pathway and Its Roles in Cancer New Perspectives of Ocular Oncology New Sights in Cancer Survival Analysis: Non-clinical Determinants of Survival for Patients with Cancer 2.0 New Technologies and Advancements in Gastro-Esophageal Cancer Surgery New Therapies for Prostate Cancer New Trends in Esophageal Cancer Management New Trends in Esophageal Cancer Management (Volume II) New Trends in Surgery for Non–Small-Cell Lung Cancer Next Generation Sequencing Approaches in Cancer NF-kappaB signalling pathway Non-coding RNA and Cancer Non-Coding RNA in Pancreatic Ductal Adenocarcinoma: New Perspectives for the Clinical Practice Non-coding RNAs and Extracellular Vesicles in Cancer Crosstalk: Diagnostic and Therapeutic Implications Non-Coding RNAs as Emerging Regulators of Signalling Pathways and Novel Therapeutic Targets in Human Cancers Non-Coding RNAs in Cancers Non-Melanoma Skin Cancer: Advances Towards Prevention and Personalized Medicine in Clinical Practice Non-Small Cell Lung Cancer Therapies Nonalcoholic Fatty Liver Disease-Related Hepatocellular Cancer (NAFLD-HCC) Noncoding Landscapes of Uveal Melanoma Noninvasive Diagnostic Imaging and Management of Skin Cancer Novel Approaches in Thymic Epithelial Tumors Diagnosis and Treatment Novel Biomarkers in Acute Myeloid Leukemia (AML) Novel Biomarkers in Pancreatic Cancer Novel Biomarkers of Hepatobiliary Tumors Novel Computational and Artificial Intelligence (AI) Models in Cancer Research Novel Concepts of Metastatic Cancer Progression Novel Developments on Skin Cancer Diagnostics and Treatment Novel Diagnostic and Therapeutic Approaches in Diffuse Gliomas Novel Insight in the Etiology of CRC: Genetics, Diagnosis, Management and Risk Assessment Novel Insight of MRI for Lung Cancer and Thoracic Neoplasm Novel Insights in Acute Lymphoblastic and Myeloblastic Leukemia Novel Insights in Myeloma Novel Insights in Ocular and Orbital Oncology: From Molecular Biology to Treatment Strategies Novel Insights in the Biology and Clinical Management of Breast Cancer during Pregnancy Novel Insights into Biology and Cancers Novel Insights into the Hallmarks of Breast Cancer Progression Novel Personalized Therapeutic Strategies for Breast Cancer Novel Predictive and Prognostic Biomarkers for Locoregionally Advanced Nasopharyngeal Carcinoma Novel Strategies in the Prevention/Treatment of Colorectal Cancer Novel Targeted Therapies in Brain Tumors Novel Targets in Triple-Negative Breast Cancer Novel Therapeutic Considerations in Bone and Soft Tissue Sarcoma Novel Therapeutic Strategies for Neuro-Oncology Novel Therapies for Pediatric Acute Myeloid Leukemia Novel Treatment for Glioblastoma Targeting Heterogeneity and Cellular Plasticity NSCLC—Tumor Microenvironment and Metastasis NUTM1-Rearranged Neoplasia: Understanding an Expanding Family of Aggressive Cancers Obesity as a Risk Factor for Cancer Oesogastric Cancer: Treatment and Management Oesophago-Gastric Cancer Surgery Old Drugs in a New Package: Future of Cancer Nanomedicine Oligometastatic Disease Oligoprogression in the Non-small Cell Lung Cancer (NSCLC) Oncogenes and Tumor Suppressor Genes in Brain Tumor Oncogenic Forms of BRAF as Cancer Driver Genes Oncogenic Metabolic Reprogramming in Cancer and Metastasis Oncogenic Virus-Associated Nasopharyngeal and Oropharyngeal Carcinoma. How Should We Treat It? Oncologic Emergencies: The Emergency Care of Cancer Patients Oncological Safety of Endoscopic and Robotic Thyroidectomy Oncology: State-of-the-Art Research and Initiatives in Japan Oncology: State-of-the-Art Research in France Oncology: State-of-the-Art Research in Germany Oncology: State-of-the-Art Research in Spain Oncology: State-of-the-Art Research in the USA Oncology: State-of-the-Art Research in UK Oncology: State-of-the-Art Researches in Poland Oncolytic Virotherapy Oncolytic Virus Therapy Against Cancer Oncolytic Viruses: A key Step Toward Cancer Immunotherapy Oral Cancer Risk and Its Management: What Is New? Oral Squamous Cell Carcinoma – from Diagnosis to Treatment Organ-Specific Metastasis Formation Organotypic 3D In Vitro Tumor Models: Bioengineering and Applications Oropharyngeal Squamous Cell Carcinoma, Challenges and Opportunities Ovarian Cancer Metastasis Ovarian Cancer Progression: From Experimental Models to Clinical Applications Ovarian Cancer Proliferation and Progression Oxidative Stress and Cancer p21 – An Underestimated Driver for Cancers p53 Family in Cancer: How Close Are We to the Clinic? p53 Signaling in Cancers P53, EMT and DNA Repair: Novel Links Impacting Cancer Progression and Drug Response Palliation of Gastrointestinal Tumors with Lumen Apposing Metal Stents Palliative Care for Patients with Cancer Palliative Radiotherapy in Cancer Palliative/Supportive Care Pancreatic and Duodenal Neuroendocrine Tumors Pancreatic Cancer Pancreatic Ductal Adenocarcinoma Pancreatic Neuroendocrine Tumors PARP Enzymes, ADP-Ribose and NAD+ Metabolism in Cancer PARP Inhibitors in Cancers PARP Inhibitors: Targeting DNA Damage Repair in Cancer Treatment PARPs, PAR and NAD Metabolism and Their Inhibitors in Cancer Past, Present, and Future Strategies in the Treatment and Management of Gliomas Pathogenesis and Diagnosis of Genitourinary Cancer Pathogenesis, Prognosis and Prediction in Breast Cancer Pathology and Genetics of Glioblastoma Pathology of Acute Myeloid Leukemia (AML) Pathology of Hematologic Malignancies Pathophysiology and Treatment of Peritoneal Metastasis Patient-Derived Xenograft-Models in Cancer Research Patient-Derived Xenografts in Cancer PD-L1/PD1 Modulation Mechanisms in Lung Cancer: From Basic to Translational Evidences Pediatric Brain Tumor Pediatric Cancer Predisposition Pediatric/Adolescent Cancer and Exercise Penile Carcinoma Perihilar Cholangiocarcinoma Perioperative Care and Pain Management in Cancer Patients: From Basic Science to Clinical Practice Perioperative Care in Gynecologic Oncology Perioperative Chemotherapy for Liver Metastasis of Colorectal Cancer Perioperative Imaging and Mapping Methods in Glioma Patients Perioperative Interventions and Oncological Outcome in Surgical Cancer Patients Personalized Medicine: Recent Progress in Cancer Therapy Personalized Medicine—Guided Synthetic Lethality Targeting DNA Repair Pathways Personalized Therapy of Sarcomas Perspectives on Early-Stage Medullary Thyroid Cancer Treatment PET and MRI Radiomics in Cancer Predictive Modeling PET Imaging in Prostate Cancer PET/CT and Conventional Imaging in Cancers PET/CT in Head and Neck Cancer PET/CT in Multiple Myeloma Patients Pheochromocytoma (PHEO) and Paraganglioma (PGL) Photodynamic Cancer Therapy Photodynamic Therapy (PDT) in Oncology PI3K Pathway in Cancer PI3K/PDK1/Akt Pathways in Cancer Pituitary Tumors: Molecular Insights, Diagnosis, and Targeted Therapy (Volume II) Pituitary Tumors: New Insights into Molecular Features, Diagnosis and Therapeutic Targeting Plasma Cell Heterogeneity in Humoral Responses and Malignancies Plasma in Cancer Treatment Platelets and Cancer Platinum-Based Therapeutics for Cancers Polyphenols in Cancer Treatment Possible Biomarkers in Oral Tumors and Their Clinical Significance Precision Medicine in Gastrointestinal Oncology Precision Medicine in Myeloma: Current, Past and Future Precision Oncology: Bioinformatics and Experimental Validation Precision Urologic Oncology—A Blueprint for Clinical Practice and Research for the Decade (2023 to 2033) to Come Preclinical and Clinical Advances in Ovarian Cancer Preclinical and Clinical Studies of Novel Therapies in Leukemia and Lymphoma Predictive and Prognostic Factors in Mesothelioma Predictive Biomarkers for Colorectal Cancer Predictive Biomarkers for Treatment Response for Head and Neck Cancers Prevention, Diagnosis and Treatment of Oropharyngeal Cancers Primary and Secondary Liver Tumors Primary Hepatobiliary Tumor Primary Liver Cancer Primary Liver Cancers in Autoimmune Liver Diseases Prognostic and Predictive Biomarkers in Malignant Mesothelioma: From Bench to Bedside and Return Prognostic and Predictive Biomarkers of Prostate Cancer Prognostic and Predictive Factors in Colorectal Cancer Prognostic and Therapeutic Implications of Tumor Biology in Colorectal Liver Metastases Prognostic Factors after Surgery for Salivary Gland Cancer; What's New, and What's Next? Prognostic Factors in Prostate Cancer Prognostic Factors in Urologic Cancers — Assessment and Integration into Clinical Care Prognostic Factors Research in Breast Cancer Patients Prostate Cancer Prostate Cancer and Radical Prostatectomy; Controversies in Anatomy, Surgical Techniques and Outcome Prostate Cancer Progression Prostate Cancer Radiotherapy Prostate Cancer: Past, Present, and Future Protein Kinases and Cancers Protein Regulatory Mechanisms in Tumorigenesis Protein Structure and Cancer Protein Synthesis in Cancer Cells: Mechanisms and Novel Targeted Therapies (Volume II) Protein Ubiquitination and Degradation in Tumor Cells Proton and Carbon Ion Therapy Proton Therapy for Cancer Proton Therapy Promises and Perils: What Progress Has Been Made? PTEN: A Multifaceted Tumor Suppressor PTEN: Regulation, Signalling and Targeting in Cancer Pulmonary Oncology Research Quality of Life and Side Effects Management in Cancer Treatment Quality of Life and Side Effects Management in Cancer Treatment (Volume II) Quality of Life for Cancer Patients Quality of Life in Patients with Gynecologic Cancer Quantitative Image Tissue Analysis Based on Multiplexed Platforms for Immuno-Oncology Applications Quantitative Technologies to Decipher Functional Phenotypes of Tumor-Associated Macrophages Radiation and Cancers Radiation Dose in Cancer Radiotherapy Radiation Therapy in Primary Liver Cancers Radiation Therapy in Thoracic Tumors: Recent Trends and Current Issues 2.0 Radiation-Induced Carcinogenesis Radioimmunetheranostics – An Emerging Approach in Personalized Oncology Radiomics in Brain Tumor Imaging Radiomics in Head and Neck Cancer Care Radionuclides in Diagnostics and Therapy of Malignant Tumors: New Development (Volume II) Radiopharmaceuticals for Oncological Diseases Radioprobes and Other Bioconjugates for Cancer Theranostics Radiotherapy for Gastrointestinal Cancer Radiotherapy in Endometrial Cancer Rapid Diagnostics for Antimicrobial Resistance in Cancer Patients Rare Tumors Involving Bone – Insight into Their Biology, Novel Diagnostic and Therapeutic Tools, including Targeted Therapies RASSF Signalling in Cancer Re-Irradiation, Chemotherapy, New Drugs for the (Re)-Treatment of Recurrent Gliomas Re-Irradiation: New Challenges and Perspectives Recent Advance in Antibody–Drug Conjugates for Cancer Therapy Recent Advance in Thoracic Cancers Progressing after Chemo-/Immunotherapy Recent Advances and Challenges in Breast Cancer Surgery Recent Advances in Basic and Clinical Colorectal Cancer Research Recent Advances in Cutaneous Squamous Cell Carcinoma Recent Advances in Deep Learning and Medical Imaging for Cancer Treatment Recent Advances in Gastric Cancer Recent Advances in Liver Transplantation for Cancer Recent Advances in Nanotechnologies for Cancer Detection and Treatment Recent Advances in Non-Small Cell Lung Cancer Recent Advances in Oncology Imaging Recent Advances in Orthopaedic Oncology Recent Advances in Ovarian Cancer Surgery Recent Advances in Pancreatic Ductal Adenocarcinoma Recent Advances in Particle Therapy for Cancers Recent Advances in Rare Cancers: From Bench to Bedside and Back Recent Advances in the Management of Neutropenia in Cancer Patients Recent Advances in the Pathogenesis of B Cell Malignancies Recent Advances in the Treatment of Peritoneal Metastasis, with Special Reference to the Role of Hyperthermic Intraperitoneal Chemotherapy (HIPEC) Recent Advances in the Understanding of Myelodysplastic Syndrome and Acute Myeloid Leukemia Recent Advances in Trachea, Bronchus and Lung Cancer Management Recent Advances on the Pathobiology and Treatment of Multiple Myeloma Recent Developments of Hematologic Diagnostics in the Interplay with Evolving Treatment Developments—Volume II Recent Highly Advanced Surgery for Pancreatic Cancer Recent Perspectives on Mechanisms of Radiation-Mediated DNA Damage Induction and Response in Cancers Recent Progress in the Diagnosis and Treatment of Melanoma and Other Skin Cancers Recent Research of Geriatric Hematology Recent Research on Gastrointestinal Carcinoma Recent Scientific Developments in Metastatic Prostate Cancer Recent Updates on Salivary Gland Tumors Recurrent Glioblastoma Redox Dysregulation and Oxidative Stress in Cancer: Therapeutic Targets and Opportunities Redox Mechanisms in Infection-Associated Cancers Regulatory and Non-Coding RNAs in Cancer Epigenetic Mechanisms Renal Cell Carcinoma Research Advances in Genetic Variants Associated with Cancer Research Advances in Paediatric Tumour of the Nervous System Research on Ocular and Intraocular Tumors Research Progress of Prostate Cancer Stem Cell Inhibitors Resistance in Breast Cancer Rhabdomyosarcoma: Still Unresolved Questions but New Perspectives Rho Family of GTPases in Cancer Risk Factors for Bladder Cancer – Clinical and Molecular Insights Risk Prediction and New Prophylaxis Strategies for Thromboembolism in Cancer Risk Stratification of Thyroid Nodule: From Ultrasound Features to TIRADS Robot-Assisted Urologic Cancer Surgery: Current Standards and Future Trends Robotic Assisted Surgical Approach for Genitourinary Tract Malignancies: State of The Art Robust Methodology for the Network-Centered “Downstream” scRNA-seq Cancer Data Analysis Role of Epigenetic Modifications in Cancers Role of Immunotherapy in Gastroesophageal Cancers: Advances, Challenges and Future Strategies Role of Medical Imaging in Cancers Role of Mesenchymal Stem Cells and Exosomes on Cancer Role of Methylglyoxal and other Oncometabolites in Linking Metabolism and Cancer Role of miRNAs in Cancer—Analysis of Their Targetome Role of Mitochondria in Cancer: Past, Present and Future Role of Natural Bioactive Compounds in the Rise and Fall of Cancers Role of New Clinical, Biologic Factors and Prognostic Systems in the Clinical Management of Chronic Lymphocytic Leukemia Role of Oxidatively-Induced DNA Damage in Carcinogenesis Role of Platelet-Derived Extracellular Vesicles in Cancer and Metastasis Roles of Exosomes/Microvesicles in Stromal-Epithelial Interaction and Cancer Progression Salivary Glands Tumors, Head and Neck Tumors and Thymoma: Current Understanding and Future Personalized Therapeutic Approach Sarcoma and Bone Cancer Awareness Month Sarcomas of Extra-Mesenchymal Sites Sarcopenia and Frailty as a Prognostic/Outcome Biomarker of Urological Cancer Patients Sarcopenia in Cancer Patients and Tumor Bearing Animal Models Selecting the Best Approach for Single and Multiple Liver Tumors Seminal Discoveries Relating to Chromatin Remodeling and Its Regulation of Cancer Biology Senescence and Cancer Sensitization Strategies in Cancer Treatment Setting the Standards for Malignancies of the Nose Vestibule Sex Differences in Cancer Signaling Pathways and Immune Checkpoint Regulation in Cancer Signaling Pathways in Gliomas Signaling Pathways Involved in Liver Cancer Development and Progression Signaling Pathways of Breast Cancer Signalling Pathways in Glioblastoma Significance of Altered (Glucose) Metabolism in Cancers Single Cell and Spatial Analysis of Solid Cancers Skin Cancer Skin Cancer and Enviornment Skull Base Reconstruction Following Surgical Treatment of Sinonasal and/or Intracranial Tumors Skull Base Tumors Small GTPases in Cancer Soft Tissue and Bone Sarcoma Soft Tissue Sarcoma: Imaging, Mechanisms and Therapy Solitary Fibrous Tumor Splicing in Cancer Research STAT3 Signalling in Cancer: Friend or Foe State of the Art and New Approaches to Spinal Cord Tumors State-of-the Art Updates in the Molecular Characteristics of Thyroid Cancer State-of-the-Art and Perspectives in the Treatment of Hormone-Receptor-Positive Breast Cancer State-of-the-Art Cancer Immunology and Immunotherapy in the USA State-of-the-Art in Cancer Cachexia Diagnostic, Prognostic and Therapy State-of-the-Art Research on Multiple Myeloma Progression State-of-the-Art Strategies for Non-coding RNA Function Detection and Regulation in Cancer State-of-the-Art Treatment on Chemotherapy and Immunotherapy for Urological Cancer State-of-the-Art in Eye Cancer Stem Cell Origin of Cancers: Biological and Clinical Implications of a Unified Theory of Cancer Stem Cells in Oral Cancer Stereotactic Body Radiotherapy for Various Cancers: Recent Advances and Future Stress Responses in Tumors and The Tumor Microenvironment Study on the Complex Melanoma Study on the Complex Melanoma 2.0 Subcutaneous Melanoma Supportive Care for Patients with Cancer Surgery for Osteosarcoma Surgery in Metastatic Cancer Surgery Induced Tumorigenesis in Breast and Other Cancers: An Inconvenient Truth? Surgical Management of Gastric Cancer: New Insights and Future Prospectives Surgical Pathology in the Digital Era—Volume II Surgical Treatment for Urogenital Cancers Surgical Treatments and Modern Techniques in Colorectal Cancer System Biology in Cancer Research Systems Biology of Tumor Immune Microenvironment and Immuno-Oncology T-Cell Lymphomas TAM family receptors in cancer biology and therapeutic resistance Targeted Intervention for Pancreatic Cancer Associated with Smoking, Alcohol Abuse and Psychological Stress Targeted Modalities for Individualized Cancer Treatment: Radiotherapy, Chemotherapy, Immunotherapy and Rationales for Their Combination Targeted Radiation Therapy and Molecular Imaging in Neuroendocrine Cancer Targeted Radiation Therapy in Spinal Metastases Targeted Therapies for Melanoma Targeted Therapies/Targetable Molecules for Treatment of Cancer and Diseases That Could Predispose to Cancer Targeted Therapy for Small Cell Lung Cancer Targeted Treatment for Soft Tissue Sarcoma and Bone Sarcoma Targeting ALK in Cancer Targeting Bone Metastasis in Cancer Targeting Bone Metastasis in Cancers Targeting Channel Proteins in Cancer Targeting FLT3 Mutations in AML (Acute Myeloid Leukemia) Targeting Head and Neck Cancer Targeting Immune Checkpoints for Cancer Therapy: Potential and Challenges Targeting Innate Immunity Cells in Cancer Targeting Metabolic Vulnerabilities in Cancer Targeting Novel Immunotherapy in Cancers: Selection of Papers from the 4th Congress of the International Symposium on Immunotherapy (London, United Kingdom) Targeting Raf Kinase Inhibitor Protein (RKIP) in Cancer Targeting Ras/RASSF in Cancer Targeting Signal Transduction Pathways in Cancer Targeting Solid Tumors Targeting STAT3 and STAT5 in Cancer Targeting STATs for Anti-cancer Therapy Targeting the JAK–STAT Signaling Axis in Cancer Targeting the Sphingolipid Metabolic Pathway: Promotion from Benchwarmer to the Starting Lineup—A Themed Issue in Memory of Dr’s. Lina Obeid and Mark Kester Targeting the Ubiquitin Pathway in Cancer Targeting Tumor Niches for Cancer Chemoprevention and Treatment Targeting Wnt Signaling in Cancer TGF-β Signaling and Its Roles in Cancers TGF-Beta Signaling in Cancer The 5th ACTC: “Liquid Biopsy in Its Best” The Asymptomatic Version of Myeloma: MGUS and Smoldering Myeloma The Chorioallantoic Membrane (CAM) Model – Traditional and State-of–the Art Applications: The 1st International CAM Conference The Complex and Evolving World of Thyroid Cancer: From Basic to Clinical Studies The Computational Methods for Anticancer Drug Development The Current Staging Systems of Tumor and Their Pitfalls The Current Status of CT, MRI and Molecular Imaging in Brain Tumors The Development of Effective Therapy Targeting the Microenvironment of Cancer The Dual Roles of Telomeres and Telomerase in Aging and Cancer The Effect of Radiation Therapy on the Tumor Ecosystem The Epithelial-to-Mesenchymal Transition (EMT) in Cancer The Evolving Landscape of Treatment against Unresectable Malignant Pleural Mesothelioma (MPM): Novel Options and Future Outlook The Future of Radiation Research in Cancers The Impact of Iron Metabolism in Cancer The Influence of Advances in Head and Neck Imaging on Diagnosis and Treatment of Head and Neck Squamous Cell Carcinoma The Landscape of Transcriptomic Diversity in Oncology The Latest Findings of the Comprehensive Management of Intrahepatic Hepatocellular Carcinoma through the Combined Use of Lenvatinib and Locoregional Therapies The Mouse Xenograft Model in Cancer Research The Multidimensional Landscape of Pancreatic Cancer Research The p53 Family in Lung Cancer The p53 Pathway in Cancer Research The p53 Pathway in Cancers The Parallel Universe of RNA beyond the Codifying Genome: ncRNAs, RNA Editing, RBP, and Epitranscriptome in Cancer The Portrait of Cancer Immunotherapy: Tumor Microenvironment, Biomarkers and Immune Resistance—Volume II The Progressive Skeletal Muscle and Body Weight Loss in Cancer Patients The Recent Updates in Primary CNS Tumors The Role of Adenovirus in Cancer Therapy The Role of Alternative Splicing in Cancer The Role of Aptamers in Cancer Diagnostics and Therapy The Role of Autophagy in Brain Tumors The Role of Autophagy in Brain Tumors (Volume II) The Role of Autophagy in Cancer Progression and Drug Resistance The Role of Bcl-2 Family Proteins in Cancer The Role of Cancer Stem Cells in Cancer Targeted Therapy The Role of Cell Death in Cancer Research The Role of Chromosomal Instability in Cancer The Role of CXCR4 in Cancer The Role of Epithelial Signalling Pathways on Tumour Progression The Role of Epithelial-Mesenchymal Transition in Therapies Resistance and Cancer Metastasis The Role of Extracellular Vesicles (EVs) in Cancer Immunotherapy The Role of Hypoxia Inducible Factor (HIF) in Cancers The Role of Immunotherapy in Hematological Malignancies Volume II The Role of Integrins in Cancer The Role of Lactate Isomers in Cancer The Role of Long Non-coding RNA in Solid Tumors The Role of Molecular Medicine in the Targeted Treatment of Gastric Cancer The Role of PET Imaging in Oncology: New Advancements in Clinical and Research Setting The Role of Rho GTPases in Cancer The Role of SBRT/SABR in Prostate Cancer Radiotherapy The Role of Src Kinase Family in Cancer The Role of T/NK Cells in Anti-tumor Immunity The Role of Telomeres and Telomerase in Cancer The Role of the Advanced Patient Derived Models in Tumors and Translational Medicine The Role of Thrombosis and Haemostasis in Cancer The Role of Vitamin D in Cancer The Roles of microRNA in Tumor Initiation and Development: Diagnostic and Therapeutic Potential The Sphingolipid Pathway in Cancer The Study of Cancer Susceptibility Genes The Study of Cancer Susceptibility Genes (Volume II) The Study of Molecular Pathogenesis and Therapeutic Strategies of Pancreatic Cancer The Survival of Colon and Rectal Cancer The Theragnostics Era: New Radiopharmaceuticals for Diagnostics and Therapy The Tumor Microenvironment of High Grade Serous Ovarian Cancer The Tumor Neuroenvironment The Tumor–Immune Interface for Next-Generation Immunotherapy The Use of Real World (RW) Data in Oncology The Warburg Effect in Cancers Theranostic Imaging and Dosimetry for Cancer Therapeutic Approaches in Chronic Lymphocytic Leukemia Therapeutic Monitoring of Anti-cancer Agents Therapeutic Monoclonal Antibodies and Antibody Products, Their Optimization and Drug Design in Cancers Therapeutic Strategies for Metastatic Melanomas Therapeutic Targeting of the Unfolded Protein Response in Cancer Therapeutic Targets in Chronic Lymphocytic Leukemia Therapeutics of Ovarian Cancers: State of the Art and Science Therapy for Human Endometrioid - Endometrial Carcinoma and Endometriosis Thermal Ablation in the Management for Colorectal Liver Metastases Third Edition of Gynecological Cancer Thoracic Malignancies Surgery Thoracic Neuroendocrine Tumors and the Role of Emerging Therapies Thromboembolism in Breast Cancer: Evidence in Context Thymic Tumors: New Developments and Future Directions in Molecularly Informed Therapies Thyroid Cancer Thyroid Cancer in the Elderly Thyroid Cancer: Diagnosis, Prognosis and Treatment Thyroid Cancer: New Advances from Diagnosis to Therapy Thyroid Cancer: Translational and Clinical Studies Thyroid Cancers Time for a Consolidated Approach for the Integration of Precision Surgery in the Treatment of Colon and Rectal Cancer. Focus on Laparoscopic and Robotic Colorectal Surgery Tissue Agnostic Drug Development in Cancer Tobacco-related Cancers Topical and Intralesional Immunotherapy for Skin Cancer Towards New Promising Discoveries for Lung Cancer Patients: A Selection of Papers from the First Joint Meeting on Lung Cancer of the FHU OncoAge (Nice, France) and the MD Anderson Cancer Center (Houston, TX, USA) TRAIL Signaling in Cancer Cells Transcription Factor Regulation and Activities in Cancer Translational Research in Gynecologic Cancer Treatment Advancement in Localized and Metastatic Renal Cell Carcinoma Treatment Intensification in Localized Prostate Cancer: What Route with What Car? Treatment of Cancer-Associated Thrombosis Treatment of Hepatocellular Carcinoma and Cholangiocarcinoma Treatment of Lung Cancer Treatment of Older Adults with Acute Myeloid Leukemia Treatment Strategies and Emerging Biomarkers in High Risk Early-Stage Melanoma Treatment Strategies and Survival Outcomes in Breast Cancer Treatment Strategies for Recurrent Cancers in Head and Neck Oncology Triple Negative Breast Cancer: From Biology to Treatment Tumor and Metabolism Tumor Angiogenesis: An Update Tumor Associated Fibroblasts on Tumor Immune Response Tumor Associated Macrophages Tumor Cell Genesis and Its Microenvironment: Chicken or the Egg Tumor Evolution: Progression, Metastasis and Therapeutic Response Tumor Heterogeneity Tumor Heterogeneity in Pancreatic Cancer Tumor Markers in the Diagnosis of Urological Malignancies Tumor Metabolome: Therapeutic Opportunities Targeting Cancer Metabolic Reprogramming Tumor Microenvironment and Treatment in Uveal Melanoma Tumor Models and Drug Targeting In Vitro (Volume II) Tumor Radioresistance Tumor Stroma Tumor Xenografts Tumor-Associated Myeloid Cells Tumor-Initiating Cells in Breast Cancer: From Bench to Bedside Tumor, Tumor-Associated Macrophages, and Therapy Tumorigenesis Mechanism of Colorectal Cancer Tumors of the Central Nervous System: An Update Tumour Angiogenesis Tumour Associated Dendritic Cells Two Years into the COVID-19 Pandemic: What It Means for Our Cancer Patients Tyrosine Kinase Inhibitors for Lung Cancer Tyrosine Kinase Signaling Pathways in Cancer Ubiquitin-Related Cancer Unmet Need for Evidence on the Possible Role of Neoadjuvant Chemotherapy in Gynaecologic Malignant Disease Unraveling an Aggressive Cancer: The Role of Epigenetics in Pancreas Cancer Unravelling Gastric Cancer Pathobiology: From Correa Cascade to Patchwork Vision Update in Ocular Oncology Update on Pathogenesis and Treatment of Kaposi’s Sarcoma Update on the Management of Head & Neck Paragangliomas: Papers from the First International Congress, Piacenza, Italy September 20-22, 2023 Updates in Acute Myeloid Leukemia Updates in Diagnosis and Management of Bladder Cancer Patients: From Prevention to Surgery and Beyond Updates in Thyroid Cancer Surgery Updates on Chronic Lymphocytic Leukemia Updates on Epigenetics of Brain Tumor Updates on the Genetics of Myeloid Malignancies Updates on the Molecular Profile of Gastrointestinal Stromal Tumors (Volume II) Updates on Urologic Oncology: From Diagnosis to Localized and Systemic Therapy Options Urologic Cancer: Endoscopic, Laparoscopic, and Robot-Assisted Surgery Managment Urology Cancers: Drug Resistance and Signaling Mechanism Urothelial Carcinoma of the Upper Urinary Tract: What Changed? Uveal Melanoma Vaginal Cancer: From Pathology to Treatment Venous Thromboembolism and Cancer Views and Perspectives of Cutaneous Squamous Cell Carcinoma Views and Perspectives of Robot-Assisted Liver Surgery Vitamin D: Role in Cancer Causation, Progression and Therapy What Is New in the Treatment of Intraocular (Uveal) Melanoma Whole Breast Radiotherapy versus Endocrine Therapy in Early Breast Cancer Wnt Signaling in Cancer Women’s Special Issue Series: Oncology World Lung Cancer Awareness Month All Special Issues Volume Issue Number Page Logical Operator Operator AND OR Search Text Search Type All fields Title Abstract Keywords Authors Affiliations Doi Full Text

Apoptosis is the mechanism by which defective cells promote their own death to protect the body. It is a complex process, with many stages, in which the different parts of the cell gradually degrade. In glioblastoma cells, even when apoptosis begins, the process stops at one of the stages and allows the cell to survive. In previous studies, researchers had already demonstrated that glioblastoma cells have insufficient levels of DFF40/CAD, a protein that, during apoptosis, orchestrates the breakdown of the cell nucleus. This shortage means that the fragmentation step is not completed and the cell can recover. In the article, published in Cancers and coordinated by Dr. Víctor J. Yuste, professor at the Department of Biochemistry and Molecular Biology at the UAB Faculty of Medicine and researcher at the Institut de Neurociències of the UAB (INc-UAB), the researchers administered to the tumor cells a substance derived from the cotton plant, gossypol. This molecule enhances the activity of DFF40/CAD. The result is that, in the treated cells, the nuclear fragmentation process is completed and the cell dies. Dr. Laura Martinez-Escardó, researcher at the Department of Biochemistry and Molecular Biology of the UAB and first author of the work explains the finding, says, "We have seen that, contrary to other drugs, gossypol allows DFF40/CAD to stay in the nucleus for longer to promote its fragmentation. With our study, we demonstrated that compounds such as gossypol can push these cells to a point of no return after starting the cell death process without modifying them genetically." Dr. Yuste says, "Promoting apoptosis to end properly in tumor cells could be a good therapeutic strategy to treat glioblastoma. The findings presented are promising and encourage us to carry out further research. The new results help us better understand the biology of this aggressive tumor and may provide us with new tools for the development of more effective strategies. This is especially interesting because there is currently no cure for this disease."

10.3390/cancers13215579

Space

Data from Mars rover Zhurong shows evidence of wind, and possibly water, erosion

L. Ding et al, Surface characteristics of the Zhurong Mars rover traverse at Utopia Planitia, Nature Geoscience (2022). DOI: 10.1038/s41561-022-00905-6 Journal information: Nature Geoscience

http://dx.doi.org/10.1038/s41561-022-00905-6

https://phys.org/news/2022-03-mars-rover-zhurong-evidence-possibly.html

Abstract China’s Mars rover, Zhurong, touched down on Utopia Planitia in the northern lowlands of Mars (109.925° E, 25.066° N) in May 2021, and has been conducting in situ investigations of the landing area in conjunction with the Tianwen-1 orbiter. Here we present surface properties derived from the Zhurong rover’s traverse during the first 60 sols of rover operations. Our analysis of the rover’s position from locomotion data and camera imagery over that time shows that the rover traversed 450.9 m southwards over a flat surface with mild wheel slippage. Soil parameters determined by terramechanics, which observes wheel–terrain interactions, indicate that the topsoil has high bearing strength and cohesion. The soil’s equivalent stiffness is estimated to range from 1,390 to 5,872 kPa per m N , and the internal friction angle ranges from 21° to 34° under a cohesion of 1.5 to 6 kPa. Aeolian bedforms in the area are primarily transverse aeolian ridges, indicating northeastern local wind directions. Surface rocks imaged by the rover cameras show evidence of physical weathering processes, such as wind erosion, and potential chemical weathering processes. Joint investigations utilizing the scientific payloads of the rover and the orbiter can provide insights into local aeolian and aqueous history, and the habitability evolution of the northern lowlands on Mars. Main The Mars Exploration Probe Tianwen-1 lander, including the Zhurong rover, landed on Utopia Planitia of Mars on 15 May 2021 1 and successfully fulfilled the goals of orbiting, landing and roving on Mars. The Zhurong rover (Extended Data Fig. 1 ), a six-wheeled solar-powered robot with rocker-bogie suspension, is the first rover to land on the Northern Plain of Mars. Zhurong, working in high-mobility performance (Extended Data Table 1 ) and carrying six scientific payloads 2 , 3 , 4 , 5 , 6 , 7 , has been roving and conducting in situ investigations for more than the designed 90-sol primary mission at the time of this writing, and collected complementary scientific data at different scales and precisions for five primary scientific objectives 7 . Seven gigabits of scientific data have been relayed via the Tianwen-1 orbiter by the end of the first 60 sols, and some of them are currently under study. Here we present preliminary surface characterization results of the landing site based on the first 60 sols of cooperative observations from the Zhurong rover and the Tianwen-1 orbiter. Landing site description The landing site is localized to 109.925° E, 25.066° N in the planetocentric frame tied to the IAU/IAG 2000 Mars coordinate system using orbital, descent and rover images 8 , and mapped to the Tianwen-1’s high-resolution imaging camera (HiRIC)-derived digital elevation map at an elevation of −4,099.8 m (Fig. 1a ). The landing site has been further confirmed by the high-resolution imaging science experiment (HiRISE) onboard the Mars reconnaissance orbiter (MRO; Fig. 1b–e ), in which the lander and the rover can be identified. The landing site lies in the Late Hesperian lowland unit, dated to ~3.32–3.36 billion years ago 9 . The region contains hundreds of superposed pedestal-crater forms, thumbprint terrain, topographically subdued wrinkle ridges and narrow grabens northeast of Alba Mons 10 . We highlight the local area in a circular shape that might be within reach of the rover (centred on 109.925° E, 24.980° N; diameter, 20 km; Extended Data Fig. 2a ). The elevation of the landing area gradually increases heading south from −4,219 to −3,989 m, with an average elevation of −4,095 m and a standard deviation of 42.11 m (based on the Mars orbiter laser altimeter (MOLA) 11 digital elevation map 128PPD; 463 m pixel −1 ). Approximately 99% of the area consists of slopes less than 5°, and relatively high slopes are primarily distributed at impact crater rims (Extended Data Fig. 2b ). Therefore, the area is flat and may facilitate long-distance traversal and exploration of the Zhurong rover. Fig. 1: Regional geomorphological features of the Tianwen-1 landing site. a , Major physiographic features of the Tianwen-1 touchdown spot (in southern Utopia Planitia) on a base map of MOLA. Also shown are previous landings or roving missions (Viking 2, InSight, Mars Science Laboratory (MSL) Curiosity 39 , Mars Exploration Rover (MER) Spirit). The base map is a portion of the MOLA shaded-relief topographic map of Mars with geoid elevations (available at ). b – e , Orbital view ( b ) and enlarged images of the Tianwen-1 lander and Zhurong rover ( c ), parachute and backshell ( d ) and heatshield ( e ). The HiRISE image is ESP_069665_2055 at 29.2 cm pixel −1 (with 1 × 1 binning) acquired by NASA’s Mars reconnaissance orbiter on 6 June 2021. c , The Tianwen-1 lander (large bright green spot) and Zhurong rover (small bright green spot). d , The parachute (long bright white spot) and backshell (round bright green spot) located 109.923° N, 25.059° E at ~350 m to the southwest of the lander. e , The heatshield (109.901° N, 25.048° E) is ~1,617 m southwest of the lander in the panchromatic image. f , A 360° panorama of the Tianwen-1 landing site taken by the NaTeCam on sol 6 while the Zhurong rover was sitting on the lander platform. The panorama is composed of 12 images taken by the NaTeCam at a beginning azimuth of 164° at 30° intervals. The surface is smooth and flat without large boulders. Two plume-impinged traces reveal dark materials when removing the bright-toned dust and topsoil. On the southeast of the lander, there are several bright transverse aeolian ridges and small craters surrounded by several dark-toned rocks, probably ejected from the crater. Image credit: CNSA/BACC. Full size image The 360° panorama composed of 12 images taken by the navigation and terrain camera (NaTeCam) (Fig. 1f ) shows an overview of gentle topography at the landing site, with major surface geological features including aeolian bedforms, small craters and rocks. The surface is smooth and flat, without boulders. A bedrock is semi-buried in the reddish soil close to the plume-impinged traces. Lander jets removing bright-toned dust and topsoil reveal the originally covered dark rock clasts. On the southeast of the lander, there are several bright sand deposits and a small crater surrounded by several dark-toned rocks, probably ejected from the crater. Soil strength and friction analysis Zhurong rover is travelling south, aiming for a variety of geological targets including craters, cones, ridges, troughs, transverse aeolian ridges (TARs) and the potential ancient shoreline 12 . It has travelled 450.9 m heading south during the first 60 sols (Fig. 2a ), with waypoints calculated by cross-site visual localization 13 , and performed several in situ investigations (Extended Data Table 2 ). Zhurong rover’s locomotion is planned by the operations team on Earth on the basis of terrain analysis, considering slopes, roughness, steps and terrain mechanical properties. In the planning, textureless surfaces, sand traps, large or sharp rocks, densely rock-packed regions and large slopes are avoided, while flat surfaces or bedrock embedded regions, even those with small rocks, are considered safe to traverse. Sandy surfaces or small sand slopes are risky for rover traversal from a terramechanical perspective and are usually bypassed, because they are prone to cause rover wheel slippage and sinkage, or even wheel embedding. The planned path for the rover to traverse is evaluated by an off-line dynamic mobility simulation 14 (Extended Data Fig. 3 ) implemented based on Terramechanics Dynamics (RoSTDyn) 15 in terms of safety and efficiency before uploading locomotion commands. Compared with the postures derived from the visual localization method, the simulations of planned traverses deviate by ≤2%, and the average azimuth error is ≤1°. Due to inevitable wheel slippage, the planned destination is not always accurately reached, particularly when traversing through sandy soils. The wheel slip ratio (denoted by s ) over each continuous steady traverse is calculated using the guidance, navigation and control system-derived rover velocity, and the wheel angular velocity is derived from the encoder ( Methods ). The distribution of the wheel slip ratio along the route (Fig. 2b ) shows that the Zhurong rover primarily operates under mild wheel slippage (slip ratio s < 0.2, that is, less than 20% wheel translational velocity loss). The average slip ratio of a steady climb of ~0.55° is 0.056, consistent with the elevation increase of 4.34 m during the 60-sol journey (Fig. 2c ). Occasionally, the rover suffers wheel skids ( s < 0) on localized down slopes, for example, when traversing through a small down-slope area of ~−3° on sol 34 before approaching a crescent-shaped TAR. Patterns of wheel tracks (Extended Data Fig. 4a ) left by the Zhurong rover also provide clues on the severity of wheel slippage in a more detailed view. By extracting the wheel track unit 16 , the average wheel slip ratio on discernible parts of the wheel track (created after Zhurong traversed from the end waypoint of sol 8 to that of sol 11; Extended Data Fig. 4b ) is calculated to be 0.05 (that is, 5% wheel translational velocity loss), consistent with telemetry-derived results. Fig. 2: The routing path of the Zhurong rover and the associated wheel slippage for the first 60 sols. a , The routing path of the Zhurong rover. The green dots represent the end waypoint on each sol. The base image is a high-resolution (0.7 m pixel −1 ) digital orthophoto map generated by the HiRIC 9 . Image credit: CNSA/BACC. b , Boxplot of Zhurong rover’s wheel slip ratios when moving in high-efficiency mode (from sol 23 onwards). The lower and upper bounds of each box represent the first and the third quartile of slip ratios on that sol, respectively. The red line inside each box represents the statistical median of slip ratios. The lower and upper whiskers of each box represent the minimum and maximum of slip ratios experienced on that sol. The red crosses represent outliers of slip ratios. Travelling distances on sols 42–48 (grey shading) are short, causing relatively large locomotion measurement errors and resulting in the slip ratio being in disagreement with the elevation trend. c , The rover elevation along the traversed path. The blue dots represent the rover elevation at end waypoints on each sol. The grey dashed line divides the traverse according to the work mode, and the right side of the grey dashed line is the traverse performed under the high-efficiency mode. The grey-shaded area is the traverse on sols 42–48. The rover elevation at the end waypoint of sol 8 (the initial waypoint on the surface) is taken as the baseline (elevation of 0), and the elevation varies from 0 m (on sol 8) to 4.34 m (on sol 60). Source data Full size image The mechanical properties of the surface soils are estimated using the grouser wheels of Zhurong as test devices. Using stereo images taken by the rear hazard-avoidance camera (HazCam), the soil surface disturbed by the grouser wheels can be reconstructed and the wheel sinkage can be estimated by extracting the wheel track unit, as was done for the Chang’E-3 mission 17 . In discernible parts of the wheel tracks (Extended Data Fig. 4c ), the equivalent wheel sinkage (the sum of the contribution of the grousers and the drum-shaped wheel) is approximately 10 mm. However, most wheel tracks are incomplete, generally lacking a clear well-trimmed imprint of the wheel edge (Extended Data Fig. 4d ). The wheel tracks are formed such that the 5-mm-high wheel grousers almost completely immerse into the soil, with the wheel rim being several millimetres above the surface (as observed in the images on sol 12 taken by the Wi-Fi camera deployed on the ground; Extended Data Fig. 4e ). Therefore, the equivalent wheel sinkage is estimated to be about 5 mm. Wheel sinkage is sometimes interrupted by protruding gravels on the surface, leading to smaller values (about 2 mm), or increases to approximately 15 mm when the wheel rim sinks below the surface (Extended Data Fig. 4f ). Compared with the Yutu-2 rover of the Chang’E-4 lunar exploration mission, the load on each wheel of Zhurong (~148.8 N under Martian gravity) is much larger than that on the wheels of the lunar rover (~36.5 N under lunar gravity). According to terramechanics, the wheel sinkage approximately increases linearly with the wheel load for wheels of the same size 18 . However, the wheel sinkage during the Zhurong traverses (5–10 mm) is not deeper than that of the Yutu-2 rover (~10 mm). Considering the rover size (Extended Data Table 1 and Supplementary Table 1 ), the wheel area on the surface of the Yutu-2 rover is estimated to be ~160 cm 2 , which is within that of the Zhurong rover (150–210 cm 2 ). Therefore, it could be inferred that the Martian soil at the Zhurong landing site has a higher bearing strength than the lunar regolith at the Yutu-2 landing site. On the basis of the terramechanics model 19 for the rigid grouser wheels, characteristic curves of the sinkage exponent (denoted by N ) and the equivalent stiffness (denoted by K s ) were predicted under different sinkage conditions (Fig. 3a ). We set the upper bound of N to 1.0 (that is, the typical value of lunar regolith 20 ). As N increases, the corresponding soil softens and produces a sharp increase in wheel sinkage. We set the lower bound of N to 0.7 to maintain K s within a reasonable range because the value of K s decreases with a decrease in N . Considering soil characteristics within curves corresponding to a wheel sinkage of 2–5 mm and representative terrestrial soil types, the equivalent stiffness K s is estimated to be 1,390–5,872 kPa per m N , shown as the rectangular region in Fig. 3a . Fig. 3: Analysis of soil mechanical parameters at the Zhurong landing site. a , Soil bearing characteristic curves under different wheel sinkages for Zhurong’s wheels and the bearing parameters of typical soil samples on Earth. The bearing parameters of soil samples at the Tianwen-1 site are estimated to be within the area (in light orange) containing five soil samples. N is the variable sinkage exponent of wheel–terrain interaction; for scatter points representing parameters of terrains on Earth, N = n (the intrinsic sinkage exponent of the terrain). b , Soil shearing characteristic curves derived from rover driving torque for Zhurong’s wheels and the shearing parameters of typical soil samples on Earth. The bearing and shearing parameters of typical soil samples on Earth in a and b are listed in Supplementary Table 2 . LLL, Land Locomotion Lab. c , Soil shearing parameters of the Tianwen-1 landing site compared with those of other Mars landing sites. The shearing parameters of soil samples on other Mars landing sites are listed in Supplementary Table 3 . Source data Full size image Using the wheel–terrain interaction model, the characteristic shearing curves under different driving torques (4.0–8.0 N m) were predicted (Fig. 3b ). The larger driving torque used for traverse represents stronger shearing resistance of the soil. Using the wheel driving motor current of the moving rover, the average wheel driving torque of each continuous traverse during sols 23–34 range from 3.7–7.0 N m (Extended Data Table 3 ). The maximum wheel driving torque reaches 8–10 N m during some sampling periods, although the corresponding slip ratio increases accordingly. Therefore, 8 N m was taken as the upper bound of the driving torque. Taking the cohesion (denoted by c ) of 6 kPa (the upper bound; close to the maximum soil cohesion of North Gower clayey loam 21 in a reasonable range) and 1.5 kPa (the lower bound; the minimum cohesion of representative soil types on Earth), the internal friction angle (denoted by φ ) is limited to ~21°–34° in the area enclosed by these curves. The friction characteristics of the soil at the Tianwen-1 site are less than those of the Viking 2, InSight, Phoenix and Mars Pathfinder landing sites. At the same time, the soil cohesion of the Tianwen-1 site is relatively high, resulting in soil adhering to the wheel surface during the traverse (Extended Data Fig. 4e ). Compared with the soil shearing properties in other Mars missions 22 , 23 , 24 , 25 , 26 , 27 , 28 , the in situ results of Tianwen-1 are within the envelop region of the others and closest to that of the Viking 1 and Curiosity rover sites (Fig. 3c ). Aeolian bedforms, craters and diverse rocks One distinct feature of the landing site is the distribution of aeolian bedforms in the form of TARs. The TARs in the landing area are of three major shapes, including solo crescent-shaped, seagull-shaped formed by two merging ripples, and solo straight-crested (enlarged views in Fig. 4a ). The TARs are bright (high albedo) from the orbit and mostly have stoss oriented approximately north-eastwards, indicating a NE–SW local wind direction (Fig. 4a ). The first TAR encountered by the Zhurong rover on sol 50 is 0.6 m in height, 40 m in length and 8 m in width (Fig. 4b ). The height/width ratio of 0.075 and a width of 8 m make it a megaripple or a small TAR, consistent with the previous classification 29 . Detailed investigations on these TARs by Zhurong, along with wind data, provide an excellent opportunity to look into the formation and evolution of these unique aeolian bedforms that are rare on Earth. Fig. 4: The geological features at the landing site. a , Distribution and characteristics of aeolian bedforms around the Tianwen-1 landing site on Mars. The three insets show the morphology of the transverse aeolian ridges (from top to bottom): the seagull shape, the crescent shape and the straight-crested shape. Most TARs in this area are crescent-shaped and their windward slopes (yellow arrows) are almost oriented in the NE–SW direction. The HiRISE image is ESP_069665_2055 at 29.2 cm pixel −1 (with 1 × 1 binning) acquired by NASA’s Mars reconnaissance orbiter on 6 June 2021. b , c , The first TAR that Zhurong rover encountered. The stoss slope is relatively gentler than the leeward one. Images were taken on sol 50. d , A small crater surrounded by dark-coloured rocks, with sand deposited at its bottom. The image was taken by NaTeCam on sol 34 (17 June 2021). e , A crater showing severe erosion, with severely damaged rims and lost clear-impact structure that shows as a depression. The image was taken by NaTeCam on sol 57 (11 July 2021). f , A mini-crater (~0.95 m diameter) right underneath the lander, formed by the thrust engine plume of the Tianwen-1 probe during landing. g – i , Rocks riddled with a large number of closely spaced pits on the surface. The rocks appear relatively light in tone through the exposed surface, with less dust and soil covering. j , k , Rocks showing layered structures, with about three layers (divided by the red dashed lines) visible on the sides. The rock flakes of the top and the middle layers seem different in direction (yellow arrows). l , Rocks with one windward face full of grooves, probably worn and shaped by wind-blown particles. Image credit: CNSA/BACC. Full size image The flat nature of the landing area is occasionally interrupted by small craters. More than 2,000 craters were identified on the HiRISE image (~0.3 m pixel −1 ) within the circular area surrounding the Tianwen-1 site (centred at 109.925° E, 25.048° N; 4 km in diameter; Extended Data Fig. 5a ), and size-frequency distribution analysis was conducted (Extended Data Fig. 5b ). The diameters of craters in this area range from 1.1 to 287.6 m, with a median diameter of 5.2 m. The arithmetic mean of these craters is 10.6 m, with a standard deviation of 20.2 m. About 79% of these craters are less than 10 m in diameter, and most are probably secondary craters with smaller radii and distribution in chains or clusters 30 , and distinguished from primary craters. Relatively large craters (≥200 m in diameter) (such as C1, C2, C3 in Extended Data Fig. 5a ) show clear bowl-shaped structures, but have broken or degraded rims. The surface age of the geological units in the landing region estimated on the basis of the crater chronology function is Late Hesperian 10 , which is consistent with the geological age division given by Tanaka et al. 11 . Craters encountered by Zhurong are relatively small depressions (<10 m in diameter; Fig. 4d–e ) surrounded by dark-toned rocks (probably ejecta). Many of the craters appear hollow or sediment filled (Fig. 1e and Extended Data Fig. 5c,d ), suggesting that they have been subjected to long-term weathering, and the surface shows signs of erosion and degradation. A fresh mini-crater (~0.95 m in diameter) beneath the lander created by the landing plume was observed in the NaTeCam image, providing clues on a shallow subsurface layer (Fig. 4f ). Gravels (1–4 cm in diameter), splashed out around the lander struts or settled on the crater rim, exhibit a dark-brown tone in sharp contrast to those semi-buried by the dust. The surface traversed by the Zhurong rover is primarily free of large boulders, but is littered with small rocks and clasts bearing distinct features (Fig. 4g–l and Extended Data Fig. 6a–c ). Most rocks are fine-grained and angular to subangular, with low roundness. Some rocks with pitted surfaces show similar morphology to igneous rocks observed in previous missions (for example, by the Viking 1 lander and the Spirit rover) 31 , 32 , which are hypothesized to have been formed via brine-related dissolution processes under cold environments 32 , 33 . Similar rock features may provide insights into the climate and geological processes of the Tianwen-1 site, but require detailed investigations by the scientific payloads of the rover. Some rocks also show flaky textures, similar to the rock targets observed on the Gusev Plain, such as ‘Mimi’ 34 . The flake textures might be related to aqueous alterations in which deliquescence water goes through insolation cracking to flake the rocks, and brine and salt may work to cement these flakes 35 . In addition, rocks with grooves and etchings on the windward sides are ubiquitous on the landing site, usually interpreted as ventifacts resulting from intense wind erosion with sand 36 , 37 , 38 . Therefore, the rock textures observed at the site thus far may indicate both the presence of physical weathering (for example, impact sputtering, wind erosion and potential freeze–thaw weathering) and aqueous interactions involving salt and brine. These rock and soil targets provide excellent opportunities to peek into the aqueous history and climate evolution of the northern lowlands, and shed light on the habitability evolution of Mars. During the first 60-sol traverse (450.9 m) after landing on the southern Utopia Planitia, the Zhurong rover has obtained data on major surface characteristics including soils, TARs, craters and rocks in the local area. The local topography is flat and gentle, facilitating long-distance exploration of the Zhurong rover to more geological targets of interest during the extended mission period. Local soil properties inferred from the wheel–terrain interaction indicate that the Martian soil on the landing site has a higher bearing strength than the lunar regolith on the Yutu-2 landing site. Compared with previous landing sites, Tianwen-1 soil has the lowest friction characteristics of all sites in the northern lowlands, the soil being further characterized to have about 21–34° internal friction angle under a cohesion of 1.5–6 kPa. The primary form of aeolian bedforms in the area is megaripples or small TARs, which have high albedo in the orbital images and are brighter than the local soils. These TARs are generally oriented in the NE–SW direction, and some show dust-covered surfaces suggesting that these ripples may be inactive. About 79% of craters in the area are less than 10 m in diameter and highly degraded, being subjected to long-term weathering and erosion. The variety of rock textures in the area, including pitted surface, layered and flaky structures, or ventifact morphology, indicates surface alteration processes, such as wind erosion and aqueous-related chemical weathering processes. All these geological features warrant further investigation. In addition, from an engineering operation perspective, the obtained surface characteristics at the Zhurong site can improve the fidelity and reliability of rover locomotion prediction and performance analysis, and more terrain-adaptive control strategies can be designed to operate the rover with improved efficiency. Advanced locomotion modes, such as creeping mode, can be employed with improved locomotion parameters, and support future exploration of hazardous terrains, such as sand deposits or cones. Methods Instruments and data description The Zhurong rover is a six-wheeled solar-powered robot with an improved active rocker-bogie suspension. Benefitting from its active suspension 40 , 41 , which is a rocker-bogie design, the rover is not only able to work in basic wheeled movement modes, but is also capable of moving in gaits, such as a crab gait for cross-walking, a creeping gait for better slope climbing capability, a wheel uplift gait to escape from wheels being stuck 42 and a body uplift/settlement gait. The strong terrain adaptability and resilient fault recovery capability allow the rover to access dangerous but scientifically beneficial regions, such as dunes and crater rims, leading to notable scientific discoveries. The grouser wheels on the Zhurong rover are used as devices for soil mechanical parameter analysis based on terramechanics. Considering the gravity on Mars, the vertical load on each wheel is estimated to be 148.8 N under a quasi-static state based on the rover’s configuration and mass distribution. The parameters of the drum-shaped wheel are shown in Extended Data Table 1 . The data used in this study include images from a NaTeCam, two HazCams and a Wi-Fi camera, alongside locomotion data from the onboard inertial measurement unit and wheel encoders. The NaTeCam, one of the scientific payloads, is mounted on the mast of the Zhurong rover. It consists of two optical systems with identical functions, performances and interfaces. The parameters of the NaTeCam are listed in Supplementary Table 4 . The NaTeCam is used for three-dimensional panoramic imaging of the Mars surface and for studying the topography and geological structure of the roving area. It takes 12 pairs of images in sequence to compose a complete 360° mosaic. The HazCam and Wi-Fi camera are engineering payloads onboard the Zhurong rover. There are two pairs of HazCams installed separately on the rover’s carriage. One pair is forward as shown in Extended Data Fig. 1b , while the other pair is installed backward. Both pairs are configured for stereo terrain observation at a close range. The parameters of the HazCam are also in Supplementary Table 4 . The Wi-Fi camera is hidden at the bottom of the rover’s carriage. It is equipped with Wi-Fi components and has Wi-Fi communication capabilities. The camera was released at sol 17 to take a selfie of both the rover and the lander. There is also a Wi-Fi receiving device on the rover, responsible for receiving Wi-Fi image data. The received data are then transmitted back to Earth. The locomotion data used in this study included the rover’s position ( x , y ) at the landing site cartographic coordinate frame, the rover’s elevation at the landing site local coordinate frame, the rover’s mileage, the rover’s forward velocities at the rover coordinate frame and the angular velocities of each wheel. The rover’s positions are derived from visual localization, mainly based on NaTeCam images. The rover’s elevation and mileage are derived from the onboard inertial measurement unit, and the wheels’ angular velocities are derived from the angle recorded by wheel encoders. The landing site cartographic coordinate frame is a fixed east-north-up right-handed coordinate system, with its origin at the lander. Its x axis points to the east, its y axis points to the north pole, and its z axis points upward. The landing site local coordinate frame is a fixed north-east-down right-handed coordinate system, with its origin also at the lander. Its z axis points down, its x axis points to the north pole, and its y axis is orthogonal to the x and z axes. The rover coordinate frame is an east-north-up right-handed local system, whose origin is at the rover’s centre, and its x axis pointing to the forward direction of the rover. Its z axis points up, which is opposite of the landing site local coordinate frame, and its y axis is orthogonal to the x and z axes. The rover positions were measured on each waypoint after traverse by the operations team on Earth. Other data were recorded onboard at a higher frequency, but the data transmitted to Earth are only at a frequency of 1/15 Hz due to communication channel restriction, with both ‘recorded’ and ‘received’ timestamps. Wheel slip ratio estimation based on telemetry data The slip ratio s of a grouser wheel at each moment t is defined as follows: $$s\left( t \right) = \left\{ {\begin{array}{*{20}{l}} {1 - v\left( t \right)/\left( {r_{{{\mathrm{s}}}}\omega \left( t \right)} \right)\quad \quad \left( {r_{{{\mathrm{s}}}}\omega \left( t \right) \ge v\left( t \right),0 \le s\left( t \right) \le 1} \right)} \hfill \\ {r_{{{\mathrm{s}}}}\omega \left( t \right)/v\left( t \right) - 1\quad \quad \left( {r_{{{\mathrm{s}}}}\omega \left( t \right) < v\left( t \right), - 1 \le s\left( t \right) < 0} \right)} \hfill \end{array}} \right.,$$ (1) where ω ( t ) is the angular velocity function, r s is the equivalent shearing radius and v ( t ) is the linear velocity function. The wheel angular velocity is recorded by wheel encoders, and its linear velocity is derived on the basis of the rover’s linear velocity (recorded by the onboard inertia measurement unit) and the curvature of the trajectory. To characterize the average wheel slippage of the Zhurong rover, the angular velocity of six wheels is averaged for the wheel slip ratio calculation at each moment. The equivalent shearing radius \(r_{{{\mathrm{s}}}}\) can be computed as 43 : $$r_{{{\mathrm{s}}}} = r + \lambda _{{{\mathrm{s}}}}h,$$ (2) where r is the wheel radius, h is the grouser height and λ s (0 ≤ λ s ≤ 1) is the grouser coefficient determined by parameters of the grousers and the internal friction angle of the soil. As Zhurong’s wheels are drum-shaped, its radius r in equation ( 2 ) was set as a value between the largest and the smallest radius. Here it was set as the mean of the wheels’ largest and smallest radius. Considering that Zhurong’s wheels have evenly arranged 5-mm-high grousers, the value of λ s was set at approximately 0.5, according to the experimental results in ref. 43 . A slip ratio s > 0 indicates that the wheel slips, which generally occurred when the wheel was moving on flat terrain or climbing up a slope; s = 0 indicates that the wheel rotates without slipping or skidding; and s < 0 indicates that the wheel skids, which usually occurred when moving down a slope, with | s | being the value of the skid ratio 42 . Soil parameter analysis Zhurong’s wheels were used to analyse soil parameters on the basis of wheel–soil interactions. The soil’s bearing and shearing parameters can be analysed through the wheel sinkage and wheel–soil interaction force, similar to ref. 16 . For a grouser rover wheel moving on the soil with an angular velocity ω , the wheel is applied with a vertical load W and a resistance force f DP from the vehicle suspension, as well as a driving torque T at the wheel rotational axis by an actuator. The soil interacts with the wheel circumference in the contact region, which corresponds to the angle divided into two parts: the entrance angle θ 1 from the vertical at which the wheel first makes contact with the soil, and the exit angle θ 2 from the vertical at which the wheel loses contact with the soil. In the wheel–terrain interaction region ( θ 1 + θ 2 ), the continuous normal stress σ to support the wheel and the shearing stress τ due to the relative movement are exerted on the wheel surface. The point of maximum stress is denoted as θ m , according to which the stress region is divided into a forward part ( σ 1 , τ 1 ), corresponding to the angle from θ 1 to θ m , and a rear part ( σ 2 , τ 2 ), corresponding to the angle from θ m to θ 2 . When θ approaches θ m , the corresponding normal stress and shearing stress approach their maximum as 44 $$\sigma _{{{\mathrm{m}}}} = (\frac{{k_{\it{c}}}}{b} + k_{\it{\upvarphi }})r^N(\cos \theta _{{{\mathrm{m}}}} - \cos \theta _1)^N,$$ (3) $$\begin{array}{l}\tau _{{{\mathrm{m}}}} = (c + \sigma _{{{\mathrm{m}}}}\tan \varphi ) \times \\ \left\{ {1 - \exp ( - r_{{{\mathrm{s}}}}\left[ {(\theta^{\prime}_1 - \theta _{{{\mathrm{m}}}}) - (1 - s)(\sin \theta^{\prime}_1 - \sin \theta _{{{\mathrm{m}}}})} \right]/K)} \right\},\end{array}$$ (4) where k c is the cohesive modulus of the soil, b is the width of the wheel, k φ is the frictional modulus of the soil and N is a variable soil sinkage exponent ( N = n 0 + n 1 s ; n 0 is the static sinkage exponent and n 1 is the dynamic sinkage resulting from wheel slippage) 19 , c is the cohesion of the soil, φ is the internal friction angle, K is the shearing deformation modulus and \(\theta^{\prime}_1\) is the equivalent entrance angle, which can be calculated as $$\theta^{\prime}_1 = \arccos [(r - z)/(r + h)].$$ (5) When the wheel is in a quasi-static state, the effect of the distributed stress (normal stress σ and shearing stress τ ) can be simplified to the normal force F N , drawbar pull F DP and driving resistance torque M R by integrating along with the wheel–terrain interaction area, and the simplified F N , F DP and M R are balanced with the wheel load W , resistance force f DP and driving torque T , respectively. For the wheels of Zhurong moving almost on flat terrain with a maximum speed of 5.6 cm s −1 , the dynamic effects are negligible at low speeds, thus the rover wheel can be viewed in quasi-static condition. Therefore, the force/torque balance equations for Zhurong’s grouser wheels are expressed as: $$F_{{{\mathrm{N}}}} = b{\int}_{\theta _2}^{\theta _{{{\mathrm{1}}}}} {\left[ {r\sigma \left( \theta \right)\cos \theta + r_{{{\mathrm{s}}}}\tau \left( \theta \right)\sin \theta } \right]{{{\mathrm{d}}}}\theta } {{{\mathrm{ = }}}}W,$$ (6a) $$F_{{{{\mathrm{DP}}}}} = b{\int}_{\theta _2}^{\theta _{{{\mathrm{1}}}}} {\left[ {r_{{{\mathrm{s}}}}\tau \left( \theta \right)\cos \theta - r\sigma \left( \theta \right)\sin \theta } \right]{{{\mathrm{d}}}}\theta } = f_{{{{\mathrm{DP}}}}},$$ (6b) $$M_{{{\mathrm{R}}}} = r_{{{\mathrm{s}}}}^2b{\int}_{\theta _2}^{\theta _{{{\mathrm{1}}}}} {\tau \left( \theta \right){{{\mathrm{d}}}}\theta } = T.$$ (6c) By linearizing the wheel–terrain interaction stress 45 , and the product of cos θ and stress 44 , a simplified expression of the driving torque M R can be calculated as $$M_{{{\mathrm{R}}}}{{{\mathrm{ = }}}}r_{{{\mathrm{s}}}}^2b\left( {\theta _1 - \theta _2} \right)\tau _{{{\mathrm{m}}}}/2,$$ (7) And then, ignoring the vertical component of the shearing stress, a simplified expression 43 of the normal force F N can be calculated as $$F_{{{\mathrm{N}}}} \approx rb\left( {\theta _1 - \theta _2} \right)\sigma _{{{\mathrm{m}}}}\cos \theta _{{{\mathrm{m}}}}/2,$$ (8) while equation ( 7 ) and equation ( 8 ) can be rearranged as $$\sigma _{{{\mathrm{m}}}} = \frac{{2F_{{{\mathrm{N}}}}}}{{rb(\theta _1 - \theta _2)\cos \theta _{{{\mathrm{m}}}}}},$$ (9) $$\tau _{{{\mathrm{m}}}} = \frac{{2M_{{{\mathrm{R}}}}}}{{r_{{{\mathrm{s}}}}^2b(\theta _1 - \theta _2)}}.$$ (10) Equation ( 9 ) is combined with equation ( 3 ) to obtain the relationship among bearing characteristic parameters (cohesive modulus k c , frictional modulus k φ and sinkage exponent N of the soil). When the normal force F N and the wheel sinkage z are estimated while the wheel parameters are known, and we let K s = k c / b + k φ , then the bearing characteristics curves of the soil for the sinkage exponent N and the equivalent stiffness modulus K s under different wheel sinkages can be plotted as in Fig. 3a . Regarding the analysis of the soil shearing parameters, we find that equations ( 4 ) and ( 10 ) are two expressions of the maximum shearing stress τ m , and τ m can be directly calculated according to equation ( 10 ), with the measured driving resistance torque M R . Then, all quantities in equation ( 4 ) can be measured, except for the three unknown shearing parameters (the cohesion c , the internal friction angle φ and shearing deformable modulus K ). The shearing deformable modulus K is determined by the slope of the shearing curve at the origin point and the maximum shearing stress τ m . Its value is 1/3 of the corresponding shearing deformation when the shearing stress τ is equal to 95% of the maximum shearing stress τ m . The denser the soil, the smaller the value of K . We concluded in the main text that Martian soil at the Tianwen-1 landing site has a higher bearing strength than the typical lunar regolith. The typical value of the shearing deformable modulus K of the lunar regolith is 17.8 mm. Therefore it was assumed that the soil shearing deformable modulus K at the Zhurong landing site is 5 mm. Since the soil parameters at the Zhurong landing site have not been rigorously identified, a K value of 5 mm is not accurate, but it can be used for inference. Therefore, the relationship between the cohesion c and the internal friction angle φ for soil under different driving torques can be determined. In calculations, the parameters of Zhurong rover, including the wheel radius r , wheel width b , wheel grouser height h and equivalent shearing radius r s , are constant as shown in Extended Data Table 1 . The normal force F N was estimated from the vertical load W , which was computed from a quasi-static force analysis of the rover, with knowledge of the rover configuration and mass distribution. The two key wheel motion state indicators, the slip ratio s and sinkage z , were computed using vision-based techniques or kinematic analysis of the rover suspension. On the basis of the given parameters, for wheel sinkages of 2, 5, 10 and 15 mm, the curves of the equivalent stiffness modulus K s and the sinkage exponent N were plotted (Fig. 3a ). The motor current of the driving wheel varied from 0.17 A to 0.28 A, with an average of 0.23 A when the wheel was rotating. Using the relationship between the motor current and the driving torque for the driving wheel, as shown in Extended Data Fig. 7b and the reduction ratio of the reduction drive between the wheel driving motor and the driving wheel (3.92 × 160, with an efficiency of 60%), the average driving torque for each continuous steady traverse was calculated for analysis. The minimum and maximum average driving torques are 3.7 N m and 7.0 N m, respectively. The average driving torque is mostly around 5.5 N m. Therefore, for driving torques of 4, 5, 6, 7 and 8 N m, the curves of the cohesion c and the internal friction angle φ were plotted (Fig. 3b ). Data availability The Tianwen-1 data used in this work are produced by the Beijing Aerospace Control Center (BACC). The data used in this manuscript are available at . The MOLA base map with geoid elevations is available at . Source data are provided with this paper.

A team of researchers affiliated with multiple institutions in China and one each from Canada and Germany, has found data from the Chinese Mars rover Zhurong over its first 60 sols, showing evidence of wind erosion and possibly impacts from water erosion, as well. In their paper published in the journal Nature Geoscience, they discuss what they have found thus far. China's Mars rover Zhurong has been on the surface of Mars since May of last year. During that time, it has rolled approximately 450 meters over the course of 60 Martian days (sols). Recently, the team working with Zhurong made the data from the rover public. In this new effort, the researchers have been studying the data sent back to learn more about what it has found. Zhurong was deployed on the planet's Utopia Planitia—a volcanic plain situated in the northern hemisphere. It is a site that some have suggested was likely once covered with water. Data from the rover's cameras showed that the part of the plain where Zhurong has been rolling along is generally quite flat, with very few boulders. And data from the wheels showed that the surface beneath the rover is covered with small, non-round rocks. Zhurong has also been collecting soil samples as it wanders—thus far, the composition of the soil in the area is similar to that collected by rovers on other parts of the planet. Image data has also shown that the small rocks have etched grooves on them that appear to be due to wind erosion. They also found some evidence of flakiness in some of the rocks, possible evidence of water erosion. The researchers also found evidence of mega-ripples on the surface—features formed by wind—similar to sand dunes on Earth. They found the ripples appeared as bright streaks when viewed from an orbiting craft. They theorize the reason the ripples appear so bright is because they have been covered by a very thin layer of dust. If that turns out to the be case, they note, it would suggest that the wind that had formed the ripples was no longer present.

10.1038/s41561-022-00905-6

Medicine

Immunotherapy may benefit patients with cancer that has spread to tissues around the brain

Sanjay M. Prakadan et al, Genomic and transcriptomic correlates of immunotherapy response within the tumor microenvironment of leptomeningeal metastases, Nature Communications (2021). DOI: 10.1038/s41467-021-25860-5 Priscilla K. Brastianos et al, Phase II study of ipilimumab and nivolumab in leptomeningeal carcinomatosis, Nature Communications (2021). DOI: 10.1038/s41467-021-25859-y Journal information: Nature Communications

http://dx.doi.org/10.1038/s41467-021-25860-5

https://medicalxpress.com/news/2021-10-immunotherapy-benefit-patients-cancer-tissues.html

Abstract Leptomeningeal disease (LMD) is a devastating complication of solid tumor malignancies, with dire prognosis and no effective systemic treatment options. Over the past decade, the incidence of LMD has steadily increased due to therapeutics that have extended the survival of cancer patients, highlighting the need for new interventions. To examine the efficacy of immune checkpoint inhibitors (ICI) in patients with LMD, we completed two phase II clinical trials. Here, we investigate the cellular and molecular features underpinning observed patient trajectories in these trials by applying single-cell RNA and cell-free DNA profiling to longitudinal cerebrospinal fluid (CSF) draws from enrolled patients. We recover immune and malignant cell types in the CSF, characterize cell behavior changes following ICI, and identify genomic features associated with relevant clinical phenomena. Overall, our study describes the liquid LMD tumor microenvironment prior to and following ICI treatment and demonstrates clinical utility of cell-free and single-cell genomic measurements for LMD research. Introduction LMD—the infiltration of tumor cells into the leptomeninges and CSF—is an especially devastating complication of solid tumor malignancies, as it is usually rapidly fatal, with a median survival of about 4–6 weeks 1 . Approximately 5–8% of all cancer patients develop LMD 1 , 2 , 3 , with common histologies including breast cancer, lung cancer, and melanoma 2 , 3 . Furthermore, over the past decade, the incidence of LMD has risen due to increased patient survival through better tolerated and more effective treatment strategies. An effective systemic therapy for LMD is thus urgently needed, as current measures (e.g., craniospinal radiation and intrathecal therapies) have uncertain benefit and significant adverse effects 4 . Immune checkpoint inhibitors (ICI) have revolutionized the field of oncology, and demonstrated remarkable response rates in a variety of metastatic, chemotherapy-refractory solid tumors 5 , 6 , 7 . More recently, ICI has emerged as a promising option for central nervous system (CNS) metastases. Preclinical data have demonstrated infiltration of T cells and programmed death-ligand 1 (PD-L1) expression in brain metastases (BM) of various histologies 8 , 9 , suggesting the potential for ICI to be efficacious in the CNS. Relatedly, ICI for metastatic melanoma and non-small cell lung cancer parenchymal brain metastases has demonstrated objective intracranial responses at a rate similar to systemic disease 10 , 11 . To our knowledge, however, ICI as treatment for LMD has not been evaluated in prospective clinical trials. To address this unmet need, we initiated two phase II clinical trials of ICI in patients with LMD of any histology (NCT02886585, NCT02939300; see Methods). One trial (NCT02886585) evaluated the efficacy of an antibody targeting programmed cell death protein 1 (PD-1, drug name pembrolizumab) and the other (NCT02939300) evaluated targeting of PD-1 (drug name nivolumab) and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4, drug name ipilimumab). Both of these treatments were administered intravenously (IV), with blood and CSF drawn prior to each dose when clinically indicated (see Methods) and sent for genomic analysis. Notably, both trials achieved primary endpoint (60% and 44% of patients were alive at 3 months after enrollment in the pembrolizumab and ipilimumab/nivolumab trials, respectively) and showed improved overall median survival (3.6 and 2.9 months in the pembrolizumab and ipilimumab/nivolumab trials, respectively) compared to historical controls 12 , 13 . Despite the promise shown in NCT02886585 and NCT02939300, questions remain regarding the utility and long-term efficacy of ICI to treat LMD. For example, it is not known whether the clinical benefit observed in these patients is strictly a result of the systemic effects of ICI administration, or whether these extend to the CNS. Additionally, the cellular and molecular features that underlie patient response have yet to be elucidated. Here, we applied single-cell RNA-sequencing (scRNA-Seq) and cell-free DNA-sequencing (cfDNA-Seq) in conjunction with conventional clinical assays to serial CSF and peripheral blood leukocyte (PBL) samples from patients on these two ICI trials to: (1) describe the cellular composition of the LMD tumor microenvironment (TME); (2) assess inflammatory immune responses within the CSF; and, (3) identify potential factors informing the clinical courses observed in individual patients. Results scRNA-Seq of the LMD TME We performed longitudinal high-throughput scRNA-Seq (Fig. 1a ) on 12 pre-treatment and 25 post-treatment low-input CSF samples from 19 total patients enrolled in NCT02886585 and NCT02939300, including 9 patients sampled at multiple time points. After filtering for low quality cells, we retained 34,742 single cells from available clinical trial samples (Supplementary Fig. 1 ), which we further classified and visualized using dimensionality reduction by principal component analysis (PCA) 14 and uniform manifold approximation and projection (UMAP; Methods) 14 , 15 . Our analyses reveal 17 distinct clusters, which we identified through differential gene expression (Supplementary Data 2 ) as adaptive immune cells (including T cells, immunoglobulin-expressing B cells), innate immune cells (including dendritic cells, monocytes, and macrophages), and non-immune cells (Fig. 1b, c ). The non-immune cell clusters ( n = 11) exhibited strong patient-specific representation while the immune clusters ( n = 5) grouped by phenotype rather than patient (Supplementary Fig. 2 ), consistent with previous observations derived through scRNA-Seq of human tumors 16 , 17 . Fig. 1: Development of a pipeline for scRNA/cfDNA from longitudinally sampled CSF samples before and after ICI therapy. a Schematic representation of the longitudinal sampling performed on patients in this study. b Longitudinal sampling overview from patients in each study, including trial primary endpoint (dashed line), and date of patient mortality, when known. c UMAP of single-cell transcriptomes from all captured CSF cells in both trials, colored by patient, with cell type of origin indicated. Full size image After initially identifying non-immune clusters via unsupervised clustering, we confirmed their malignancy status for samples from NCT02886585 by inferring copy number variation (CNV) profiles for each cell and matching to DNA-based profiling results 17 , 18 , 19 . DNA-derived CNV profiles were obtained via whole exome sequencing (WES) of cell-free DNA (cfDNA) extracted from the CSF. Using these data, we confirmed that patient-specific non-immune clusters shared the inferred CNV profiles (as previously described) of their time-point matched cfDNA counterparts (Methods, Supplementary Fig. 3b ). The proportions of tumor cells captured by Seq-Well for available time points (see Supplementary Data 1 for all available cytology reports) correlated significantly (Kendall’s т correlation = 0.62, p = 0.0027; Pearson correlation coefficient = 0.89, p = 1.8 × 10 −5 ) with the reported tumor cell fraction detected by CytoSpin from CSF (Supplementary Fig. 3c ). We additionally obtained PBL-derived scRNA data on 810 cells from patients P010, P043, P046, and P073 of trial NCT02886585 (Supplementary Fig. 4 ). CD8 + T Cells in CSF following intravenous ICI administration We found that CD8 + T cells in the CSF are more abundant (NCT02886585) and proliferative (both trials) in samples treated with immune checkpoint inhibitors relative to untreated samples. We first performed unsupervised analysis of the T/NK cluster (Fig. 2a, b ), and calculated the proportion of CD4 + T cells, CD8 + T cells, and NK cells (which can co-segregate with T cells during high-level analyses of scRNA-Seq data based on gene-expression similarity 20 ) in each sample in our dataset (Supplementary Data 3 - 4 ). The proportion of CD8 + T cells in post-treatment CSF samples from NCT02886585 was significantly higher than in pre-treatment samples in evaluable samples (Cohen’s d = 0.87, Two-sided Wilcoxon’s rank-sum p = 0.03, N = 24 samples; 11 pre-treatment, 13 post-NCT02886585, see Methods), while there was no statistically significant difference in the proportion of CD8 + T cells in the post-treatment CSF samples of NCT02939300 relative to pre-treatment (Fig. 2c ). Unsupervised analysis of the T cells also revealed a cluster of cells with increased expression of genes associated with proliferation, including MKI67 , BIRC5 , and TOP2A , among others 16 , 18 , 21 . Increased proliferation following ICI administration has previously been reported in the peripheral blood of patients undergoing systemic treatment 19 , 22 , 23 , 24 . We calculated the fraction of proliferating CD8 + T cells for each sample. Samples treated with ICIs had a significantly greater fraction of proliferating CD8 + T cells compared to untreated samples in evaluable samples (Cohen’s d = 0.62, 0.60; Two-sided Wilcoxon rank-sum p = 0.02, N = 21, 19, 9 pre-treatment, 12 post-NCT02886585, 10 post-NCT02939300; Fig. 2d , see Methods. Accompanying analysis of longitudinally matched samples across patients in Supplementary Fig. 5 ). These data suggest that the abundance and rates of proliferation of T Cells in the CSF increased post-treatment. Fig. 2: T cells in CSF exhibit strong differences in the expression of interferon-induced, cytotoxic, and exhaustion genes following ICI. a , b UMAP calculated over all T/NK cells ( n = 16,954), colored by cohort ( a ) and canonical cell type ( b ) as identified via iterative subclustering (see Methods). c Percentage of CD8 + T Cells in pre-treatment, post-treatment cohort 1, and post-treatment cohort 2 samples; only samples with more than 20 T cells considered. d Proportion of CD8 + T cell cycling in pre-treatment, post-treatment cohort 1, and post-treatment cohort 2; only samples with more than 10 CD8 + T cells considered. e Effector vs naïve gene expression in pre-treatment, early post-treatment (<30 days post-treatment), and late post-treatment (≥30 days post-treatment) CD8 + T cells, N = 6,133 CD8 + T cells. f IFN-γ response in pre-treatment, early post-treatment (<30 days post-treatment), and late post-treatment (≥30 days post-treatment) CD8 + T cells. g Median IFN-γ response across samples in tumor cells vs CD8 + T Cells. In c–d data are represented as boxplots where the middle line is the median, the lower and upper hinges correspond to the first and third quartiles, the upper whisker extends from the hinge to the largest value no further than 1.5× IQR from the hinge (where IQR is the interquartile range) and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge. In c – f , indicated p-values are two-sided, calculated from a Wilcoxon rank-sum test, and in figure g, p -value of the Kendall-Tau correlation is two-sided (see Methods). Full size image CD8 + T cells from samples treated with ICI transiently exhibited higher levels of genes associated with effector function and IFN-γ signaling relative to untreated samples, which were more naïve-like. Principal component analysis revealed that the first two principal components distinguished early post-treatment (<30 days since initial treatment) and pre-treatment samples (Supplementary Fig. 6 ). Loadings for these principal components were driven by genes associated with IFN-γ signaling as well as effector/naïve phenotypes of CD8 + T cells 25 , 26 , 27 . Indeed, we detected a significant increase in genes related to effector-like function and IFN-γ signaling (Fig. 2e, f ; gene lists in Supplementary Data 5 ) in CD8 + T cells in recently treated (<30 days since initial treatment) samples vs. pre-treatment in both NCT02886585 (Cohen’s d = 0.91 and 0.65, p < 0.001 for both IFN-γ and effector/naïve signatures respectively) and NCT02939300 (Cohen’s d = 0.75 and 1.07, p < 0.001 for both IFN-γ and effector/naïve signatures, N = 6,133 CD8 + T cells). Furthermore, we found that the mean level of IFN-γ signaling in the T cells was strongly correlated with the mean IFN-γ response in tumor cells at the same time points (Kendall’s τ correlation = 0.67, p = 0.003, Fig. 2g ), suggesting that inflammatory response is consistent across cell types in the same CSF sample. Longitudinal scRNA-seq reveals transient IFN-γ response and antigen presentation signatures following ICI administration IFN-γ response (Fig. 3a–c ) and antigen presentation (Supplementary Fig. 7a–c ) signatures exhibited transient upregulation immediately following ICI administration, which was observed across patients in multiple cell types. We found temporary elevation in the module scores for both signatures, with the maximum occurring at early time points (<30 days after initial administration; p = 0.00405, 0.00262, 0.0774 for IFN-γ response in lymphoid, innate, and tumor compartments respectively; p = 0.00406, 0.00871, 0.0938 for antigen presentation in lymphoid, innate, and tumor compartments respectively; Two-sided Wilcoxon rank-sum test throughout), and a significant reduction in the majority of these signatures at later time points (≥30 days after initial administration; p = 0.03212, 0.16491, 0.08648 for IFN-γ response in lymphoid, innate, and tumor compartments respectively; p = 0.022271, 0.049141, 0.032125 for antigen presentation in lymphoid, innate, and tumor compartments, respectively; Two-sided Wilcoxon rank-sum test throughout; N = 12 pre-treatment, N = 7 0–30 days post-treatment, N = 6 36+ days post-treatment for lymphoid, N = 12 pre-treatment, N = 8 0–30 days post-treatment, N = 7 36+ days post-treatment for innate, N = 11 pre-treatment and N = 7 0–30 days post-treatment, N = 6 36+ days post-treatment for tumor, see Methods). Taken together, these results suggest the onset of either an acute inflammatory response within the CSF to intravenously administered ICI or the infiltration of ICI-activated immune cells to this compartment, which may potentially explain the clinical efficacy of intravenous ICI against LMD in NCT02886585 and NCT02939300 12 , 13 , 28 . Fig. 3: Acute immune response in CSF subsequent to intravenous ICI and relationship to survival. a – c Mean module score for IFN-γ response within samples over time points for lymphoid ( a ), myeloid ( b ), tumor cells ( c ). Samples from a single patient are connected with a dashed line. The size of markers is proportional to the number of relevant cells in a sample; only samples with more than 5 cells of the corresponding type are considered. Points at 0 days relative to ICI administration are pre-treatment. d Violin plots of IFN-γ response for tumor cells plotted against survival (time-on-trial), for samples taken <30 days after initial ICI administration. Medians and upper and lower quartiles are indicated in each violin plot by dashed lines. Full size image IFN-γ response and antigen presentation within malignant cells shortly (<30 days) after ICI administration correlates with time on trial To see whether the previously described ICI response had prognostic value in this patient population, we compared IFN-γ response and antigen presentation within cell types with time-on trial. In total, 6 samples were obtained from individuals having received their initial ICI dose <30 days prior, for which we had known dates of death. We observed that in malignant cells there was a relationship between survival beyond primary endpoint and the mean module score of IFN-γ response ( p = 0.0526, single-sided Wilcoxon rank-sum test 29 Fig. 3d ). This relation did not hold for non-malignant cells (Supplementary Fig. 8 ). Inflammatory signatures appear greater in the CSF than in the blood We observed more pronounced inflammatory signatures (antigen presentation and IFN-γ response) in CSF than peripheral blood lymphoid and myeloid cells in post-treatment timepoint matched samples ( p -values in Supplementary Data 6 ) while using a down-sampling procedure (Methods) to adjust for differences in cell quality. Furthermore, whereas we observed a significant increase in antigen presentation and IFN-γ response in CSF-derived innate and lymphoid immune cells in P043 immediately after treatment, this was not observed in PBL-derived innate and lymphoid immune cells ( p -values provided in Supplementary Data 6 ). Moreover, while we observed a significant increase in M1-like phenotype 30 , 31 in CSF-derived myeloid cells over time in P043 ( p -values in Supplementary Data 6 ), we observed a decrease in the same signature in PBL-derived myeloid cells in that patient ( p -values in Supplementary Data 6 ). This suggests that post-ICI inflammation in patients with LMD may be particularly elevated in the CNS, and that different compartments within the body may express divergent levels or stages of overall immune response, warranting further studies across multiple sites and timepoints to fully characterize patients’ overall response to therapy. Differential expression rankings (Methods) between PBL and CSF-derived lymphoid and myeloid cells are provided in Supplementary Data 7 . An adaptive selection of a less immunogenic subclone coincides with transient response in one patient We investigated cellular behavior underlying treatment response in a particular patient, P043, who showed unique clinical, phenotypic, and inferred CNV-based dynamics over the course of treatment. Three weeks following initial pembrolizumab administration, P043 showed a reduction in tumor burden according to both cytology and Seq-Well (Fig. 4a ). At 6 weeks following initial ICI administration, and consistently from that point onward, the reported tumor cell fraction by cytology increased progressively until the patient came off study, with both Seq-Well and cytology showing an increase in tumor burden (Fig. 4a ). This was accompanied by an interval increase in LMD-associated enhancement on the patient’s brain MRI at 6 weeks following treatment. At 12 weeks, malignant cytological fraction was above pre-treatment levels according to cytology, and MRI scans showed further LMD progression at 12 weeks, again indicative of LMD progression (Fig. 4a, b ). Fig. 4: Longitudinal scRNA and cfDNA from P043 suggest the adaptive selection of a less-immunogenic over a more-immunogenic subclone. a Tumor fraction within CSF, as measured by Seq-Well and cytology, overlaid with tumor purity inferred by ABSOLUTE run on CSF-derived cfDNA. b MRI imagery at 0, 6, and 12 weeks relative to treatment; LMD-indicative enhancement indicated by red arrow. c Unsupervised clustering of inferred copy number profiles (left, see Methods) and expression (right) reveals intercellular heterogeneity, possibly explainable by the presence of subclones. d Relative proportions of subclones as a function of time. Darker purple and lighter purple denote the descendant and ascendant subclone, respectively. e IFN-γ response expression in subclones over time (*** p = 0.001, ** p = 0.01, Wilcoxon ranked-sum test, Cohen’s d = 1.4, N = 52 for descendent, 19 for ascendant at P043-3; Cohen’s d = 1.4, N = 14 for descendent, 31 for ascendant at P043-4). Darker purple and lighter purple denote the descendant and ascendant subclone, respectively. Full size image Unsupervised analysis of tumor cells from P043 revealed heterogeneity in both the gene expression and inferred copy number (rWME, see Methods) profiles that was suggestive of adaptive selection leading to acquired ICI resistance (see Fig. 4c ). For all patients, we assessed the possibility of subclonal tumor heterogeneity 32 , 33 , 34 , 35 , 36 by inferring single-cell CNV profiles 16 , 17 , 37 via clustering in windowed mean expression (WME, see Methods) space. Visualizing these clusters with reduced dimensions, we found little structured CNV heterogeneity within 14 of the 15 evaluable patients (Supplementary Fig. 8). For example, in P029, we observed clusters in gene expression space (primarily attributable to cycling status) but not in inferred CNV space; in other patients (P014, P050, P061), the inferred CNV variation was limited, restricted to only a few loci, or sampled at only one time point. In P043, however, we detected the presence of two distinct copy number profile clusters, which suggested the presence of tumor subclones (Fig. 4c , Methods). Moreover, the fractional abundance of these two subclones shifted over time, with one clone lowly abundant pre-treatment and monotonically increasing at the expense of the other (Fig. 4d ). We therefore hypothesized that a minority subclone was adaptively selected in P043 as a result of ICI administration. We noted that regions of high divergence in inferred CNV between clusters corresponded to focal genomic amplifications distinguishing pre- and post-treatment cfDNA-derived copy-number profiles (Supplementary Fig. 9 ). Notably, the CNV-associated clusters were highly concordant with gene expression-derived clusters, suggesting correspondence between genetic and phenotypic heterogeneity in the tumor cells of P043 (Fig. 4c ). We plotted IFN-γ response scores for each cluster at the three time points measured for P043 (Fig. 4e ). These data show that the descendent clone exhibited higher IFN-γ response genes at P043-3 ( p < 2 × 10 −3 for post-treatment time points, Two-sided Wilcoxon rank-sum test)–the time point immediately following treatment (Fig. 4e ). In contrast, the ascendant clone exhibited consistently lower IFN-γ response across this trajectory, eventually predominating at the last time point (P043-4). Differences in antigen presentation between subclones ( p = 0.01 at P043-3, p = 0.0015 at P043-4; N = 52 for descendent, 19 for ascendant at P043-3; N = 14 for descendent, 31 for ascendant at P043-4) are given in Supplementary Fig. 11d . To support the subclonal hypothesis without relying on either inferred CNV profiles or unsupervised clustering thereof, we performed a supervised comparison of single-cell expression profiles at each time point to both the early and late cfDNA-derived WES copy ratios (Supplementary Fig. 11c ). This analysis revealed that cells collected at P043-4 had gene expression profiles more concordant with the copy number profile calculated from cfDNA obtained at P043-4, whereas cells collected at P043-2 had gene expression profiles more concordant with the copy number profile from cfDNA obtained at P043-2. At post-treatment time points, cells with gene expression profiles correlating more strongly with copy number profiles from the later cfDNA (from P043-4) tended to have lower expression of IFN-γ response related genes while cells with gene expression profiles correlating more strongly with copy number profiles from the earlier cfDNA (from P043-2) tended to have higher expression of IFN-γ response related genes (Theil-Sen slope = −15.9 IFN-γ response/correlation difference for P043-3 and −8.79 IFN-γ response/correlation difference for P043-4). Genes with large cfDNA-derived CNV difference between the time points P043-2 (enriched for the descendant subclone) and P043-4 (enriched for the ascendant subclone) represent hypothetical drivers for observed difference in immunogenicity between subclones. Supplementary Data 8 contains the top genes with the largest (respectively smallest) fold change in cfDNA-derived copy ratio at P043-2 and P043-4, and the corresponding mean expressions of those genes in the ascendant and descendant subclones; RAD21 , for example, which has been reported to predict poor prognosis 38 , 39 , is copied roughly 5 times more in the cfDNA at P043-4, and is roughly 3 times more highly expressed in the ascendant subclone when controlling for time point (i.e., at time point P043-3). Discussion Here, we have used low-input profiling techniques to perform scRNA-Seq characterization of cell types in the liquid component of the human LMD TME. We catalog the presence of tumor cells, lymphoid cells, and myeloid cells, examining shifts in their abundance and phenotype in conjunction with two clinical trials of ICI efficacy. These highly-resolved data enable further study of LMD- and treatment-associated phenomena that are unique to the CSF both within and across patients. We calculated statistically significant increases in the abundance and outgrowth of CD8 + T cells in the CSF following ICI administration relative to pre-treatment in NCT02886585. Additionally, we detected higher overall levels of IFN-γ signaling and cytotoxicity in CD8 + T cells post-treatment in both trials. These results suggest that intravenous ICI administration modulates the immune microenvironment in the CSF of a subset of patients in these clinical trials, and that this may have been associated with observed clinical benefit. We detect suggestive evidence that the magnitude of the initial inflammatory response among malignant cells may have prognostic value, warranting further investigation. Investigation of IFN-γ response and antigen processing signatures in the tumor compartments of patients in these trials reveals a distinct increase across patients immediately following the first ICI dose followed by a steady decrease over time. While this observation is consistent with reports of anti-PD-1 administration in the peripheral blood 40 , 41 , it also illustrates a potential limitation in ICI efficacy for LMD. Additionally, these results underscore the importance of sampling time, as well as the value of longitudinal profiling, in scRNA-seq studies of response to therapies, as these responses are liable to exhibit transient transcriptional effects; more time-resolved sampling may be necessary to properly characterize the dynamic phenotypic processes presented here. Finally, in a particular patient, we note evidence suggesting the existence of subclones that might underlie the transient response to ICI observed in that individual. We use that evidence to propose hypothetical drivers of increased or decreased immunogenicity 42 . These results collectively support the findings of Brastianos et al. 12 and Brastianos et al., 13 —namely, that ICI shows clinical efficacy in patients with LMD. Additionally, we observe strong compartmental and temporal variation in inflammatory signatures in the post-therapy TME, with significant implications for future study design in LMD and in other cancer types. Due to the limited input volumes obtainable from clinical CSF samples, we were not able to obtain longitudinal data from all patients at all time points. Moreover, the diversity of the cohort—containing diversity in both primary histology and histological subtype—recommends that follow-up studies control for these factors, so as to confirm whether the biomarkers of response suggested above have clinical utility. Future work with larger cohort sizes, and more frequent longitudinal—as well as multi-site—sampling will enable the characterization of genotypic factors in ICI response in LMD, support comparisons of the effects of multiple vs. single-drug treatments, and further test and refine the prognostic biomarkers suggested in this work. Methods Study design: patients These clinical trials (Clinicaltrials.gov identifier NCT02886585 and NCT02939300) were designed by the principal investigators and the Dana-Farber Harvard Cancer Center (DF/HCC) Institutional review board approved the protocol. The study was designed by the principal investigators and conducted in accordance with the provision of the Declaration of Helsinki and Good Clinical Practice guidelines. The Dana-Farber/Harvard Cancer Center institutional review board approved the protocol. All patients provided informed consent. Eligible patients had histologically confirmed disease from any solid tumor and LMD defined by positive CSF cytology for malignant cells. Other key inclusion criteria included the following: an ECOG performance status ≤2, normal organ and marrow function, and a stable dose of dexamethasone of 2 mg or less for 7 days prior to initiation of treatment. Given the frequent occurrence of neurologic symptoms (e.g. headaches) associated with LMD, 11/20 patients included in this study were on a steroid regimen at the time of enrollment and 10/20 patients were treated with steroids while receiving ICI 13 . Written informed consent was obtained for all participants. Further details of the subjects’ clinical courses, including cytology, steroid dosage, and Ommaya shunt status, are provided in Supplementary Data 1 . Study design: treatment and endpoints These studies were designed as open-label, single arm Phase-II clinical trials to evaluate ICI in patients with LMD of any histology. For the first trial (NCT02886585), patients with CNS metastases were enrolled across multiple cohorts. Cohorts A, B, and D include patients with parenchymal brain metastases. The LMD cohort was Cohort C of this study. Pembrolizumab was administered intravenously at 200 mg every 3 weeks until disease progression, death, or grade 3–4 toxicity. A brain MRI and CT chest/abdomen/pelvis were obtained every 6 weeks for restaging purposes. The primary endpoint of the LMD cohort was the rate of overall survival at 3 months (OS3). 11 patients with LMD enrolled to NCT02886585 were included for single-cell analysis; 4 of whom had serum and CSF sampling at multiple time points during treatment. All 11 patients had metastatic breast cancer (Supplementary Data 1 ). Treatment in the second trial (NCT02939300) consisted of ipilimumab and nivolumab. Ipilimumab and nivolumab were administered intravenously every 3 weeks for 4 doses. Afterwards, ipilimumab was given every 6 weeks and nivolumab was given every 2 weeks (for non-small cell lung cancer and head and neck cancer) or 4 weeks (for all other malignancies). Treatment was continued until disease progression, death, or grade 3–4 toxicity. A brain MRI and CT chest/abdomen/pelvis were obtained every 6 weeks for restaging purposes. The primary endpoint was OS3. 9 patients on trial were included for single-cell analysis; 5 of whom had serum and CSF sampling at multiple time points during treatment. 5 patients had metastatic breast cancer, 2 patients had high-grade glioma, 1 patient had ovarian cancer, and 1 patient had esophageal cancer. Further details of the subjects’ clinical courses, including cytology, steroid status, and Ommaya/VP shunt status are provided in Supplementary Data 1 . CSF cell extraction Cerebrospinal fluid (CSF) from patients was extracted via an Ommaya reservoir or ventriculoperitoneal shunt (VPS) as part of clinical care. CSF not required for clinical testing was spun at 800 G for 5 min to pellet cells, and resuspended in PBS (ThermoFisher 10010023, Ca/Mg-free). Red blood cells (RBCs) were lysed using ACK lysis buffer (ThermoFisher A1049201) for 4 min on ice to remove RBCs. Cells were then washed with sterile PBS and spun down at 800 G for 5 min, resuspended as a single-cell suspension in RPMI (Corning) with 10% FBS (ThermoFisher 10082-147) for subsequent scRNA-Seq. Cytology was performed whenever possible from available CSF; cytology results for all available samples are given in Supplementary Data 1 . Peripheral blood lymphocyte (PBL) extraction A 10 cc tube of blood was collected and processed within 3 hours of blood draw. 15 mL of Lymphoprep (STEMCELL Technologies, Catalog #07801) and 10 mL of phosphate-buffered saline was added to the blood. This mixture was then centrifuged at 1200 G for 12 minutes. The supernatant was then poured out and 10 mL of phosphate-buffered saline was added. This mixture was centrifuged a second time at 500 G for 5 min. This supernatant was poured out and 1 mL of CryoStor CS10, Cryopreservation Freeze Media (STEMCELL Technologies, Catalog MSPP-07930) was added to the pellet. This mixture was frozen at −80 °C, then later thawed on ice, then to room temperature, then processed using Seq-Well as described below for CSF-derived cells. Extraction and sequencing of cell-free DNA For blood, a 10 cc tube was first centrifuged at 500 g for 10 min. Afterwards, the supernatant was extracted and then centrifuged again at 1000 g for 10 min. The second supernatant was then used for serum cell-free DNA extraction and sequencing. For CSF samples, a 3 cc tube of sample was centrifuged at 400 G for 5 min. Extraction of cell-free DNA from banked plasma and centrifuged CSF was done using an automated liquid handler at the Broad Institute’s Blood Biopsy Lab. Sequencing was then conducted by the Broad Institute’s core facility. scRNA-Seq with Seq-Well Resuspended CSF cells were profiled using the Seq-Well platform for massively parallel, high-throughput scRNA-seq for low-input clinical samples. A complete description of methods is available online 43 . A complete list of primers described in Gierahn et al. 43 is additionally provided in Supplementary Data 10 . Briefly, cells from each CSF sample were manually counted (Bal Supply 808CI) and loaded onto one array preloaded with barcoded mRNA capture beads (ChemGenes). All samples retained fewer than 10,000 cells with the exception of two (CSF029-4 & DFCI010-4; ~100,000 cells). Thus, all available cells were loaded onto a single array, except CSF029-4 and DFCI010-4 where ~10,000 cells were loaded. The loaded arrays containing cells and uniquely barcoded oligo-dT beads were then sealed using a polycarbonate membrane with a pore size of 0.01 μm, which allows for the exchange of buffers but retains biological molecules confined within each nanowell. Subsequent buffer exchanges facilitated cell lysis, transcript hybridization, and bead recovery before performing reverse transcription en masse . Following reverse transcription using Maxima H Minus Reverse Transcriptase (ThermoFisher EP0753) and an Exonuclease I treatment (NewEngland Biolabs M0293L) to remove excess primers, PCR amplification was carried out using KAPA HiFi PCR Mastermix (Kapa Biosystems KK2602) with approximately 2,000 beads per 50 μl reaction volume. Libraries were then pooled into one tube (with the exception of CSF014-4, CSF029-4, CSF046-2, CSF104-1, CSF104-3, and CSF123-4, which were pooled to two tubes) and purified using Agencourt AMPure XP beads (Beckman Coulter, A63881) by a 0.6X SPRI followed by a 0.8X SPRI and quantified using Qubit hsDNA Assay (Thermo Fisher Q32854). The quality of each WTA product was assessed using the Agilent hsD5000 Screen Tape System (Agilent Genomics) with an expected peak ranging between 800 and 1500 bp tailing off to beyond 5000 bp, and a small or non-existent primer peak (~100–200 bp). 3′ digital gene expression (DGE) libraries were constructed using the Nextera XT DNA tagmentation method (Illumina FC-131-1096) using index primers as described in Gierahn et al. 28 . Loaded samples ranged from 600 to 2,000 pg of pooled cDNA, depending on the peak distribution of the WTA product for the sample. Tagmented and amplified sequences were purified at a 0.6× SPRI ratio followed by a 0.9X SPRI yielding library sizes with an average distribution of 400–750 base pairs in length as determined using the Agilent hsD1000 Screen Tape System (Agilent Genomics). Samples DFCI010-4, CSF011-1, CSF011-7, CSF014-1, CSF014-2, and CSF014-4, CSF029-2, CSF029-5, DFCI037-1, CSF046-2, CSF050-4, CSF050-7, and CSF073-4 were sequenced using an Illumina 75 Cycle NextSeq500/550v2 kit (Illumina 20024906) at a final concentration of 2.2–2.8 pM. Samples CSF029-4, CSF043-2, CSF043-3, CSF043-4, CSF050-3, CSF050-12, CSF050-19, DFCI056-3, DFCI058-1, DFCI058-5, DFCI058-7, DFCI061-1, DFCI061-2, DFCI062-2, DFCI062-3, DFCI062-4, DFCI062-5, DFCI062-6, CSF091-3, CSF104-1, CSF104-3, CSF119-1, CSF123-4, CSF127-3, and CSF129-1 were sequenced using an Illumina 100 Cycle NovaSeq6000S kit (Illumina 20027464). The read structure in both cases was paired end with read 1 starting from a custom read 1 primer containing 20 bases with a 12-bp cell barcode and 8-bp unique molecular identifier (UMI) and read 2 containing 50 bases of transcript information. Alignment & Pre-processing of scRNA-Seq data Read alignment was performed as in Macosko et al. 21 . In brief, for each Illumina sequencing run, raw sequencing data were converted to demultiplexed FASTQ files using bcl2fastq2 based on Nextera N700 & N500 indices corresponding to individual samples/arrays. Reads were then aligned to hg19 genome using the dropseq_tools v2.1.0 pipeline maintained by the Broad Institute using standard settings. Individual reads were tagged according to the 12-bp barcode sequenced and the 8-bp UMI contained in Read 1 of each fragment. Following alignment, reads were binned onto 12-bp cell barcodes and collapsed by their 8-bp UMI with a hamming distance correction of 1. DGE matrices (genes-by-barcode) for each sample were obtained from quality filtered and mapped reads, with an automatically determined threshold for barcode count. DGEs from each sample were individually culled and inspected by unsupervised analysis before inclusion into the full analysis by a combination previously described methods 29 , 30 . Each barcode was initially scored on: (1) average expression of a list of curated housekeeping genes (Supplementary Data 5 ) and (2) complexity, estimated by the total number of genes detected. All sequenced samples were cut to exclude barcodes with low complexity/housekeeping gene expression (400 gene complexity cutoff, housekeeping gene expression cutoff of 1.6 log 2 (tp10k)). Each sample was then inspected using unsupervised analysis to further identify and curate potential analyzable single cells. Individual arrays were analyzed to determine the extent of cross-cell type gene expression contamination. Minimal cross-cell type gene expression contamination existed between immune subsets, and select barcodes were filtered out according to cross expression of marker genes from other immune subsets. Further restrictive analyses incorporating lowered complexity thresholds and count-based downsampling was performed to control for technical confounders wherever relevant. Following curation, all samples were combined and genes expressed in at least 1% of the remaining barcodes were retained (in the case of WME calculations used in Fig. 4c we retained genes expressed in at least 0.0875% of remaining barcodes; in the case of WME calculations used in Supplementary Fig. 9 , we retained all genes). Consecutive samples from the same patient were combined by assigning zeros to all undetected genes per sample and concatenating columns. miRNA and T cell receptor chain genes were subset and saved before cutting genes to ensure information was not lost. This curated, UMI-collapsed data was then normalized to 10,000 UMIs per barcode (tp10k) and log 2 -normalized before being input into Seurat 14 v2.3.4 ( ) for further analysis. This yielded a Seurat object of 34,742 single cells and 8,156 genes, with different genes being used for more specific analysis (such as T cell analysis). The 37 individually sampled time points averaged 890.8 cells per sample with a range between 103 cells and 1,946 cells (Supplementary Data 1 ). Alignment & Pre-processing of PBL-derived scRNA-Seq data Read alignment was performed identically as with CSF-derived scRNA-Seq data. Barcodes were selected from DGEs with a 200 gene complexity cutoff. Unsupervised analysis was performed jointly with CSF-derived cells, with the same complexity cutoff, from patients with PBL-derived data (P010, P043, P046, P073). All genes were retained. In total, 810 PBL-derived immune cells were detected. For PBL vs CSF comparison, to account for differences in cell qualities between these samples, all cells had their total UMI count adjusted to 500 (the approximate mean UMI count of PBL-derived cells with complexity greater than 200). This was done by randomly selecting 500 UMIs for each cell, with sampling probabilities given by the pre-adjusted UMI count for that gene, in that cell. This data was then normalized to 10,000 UMIs per barcode (tp10k) and log 2 -normalized. Differential expression and module score calculations were performed as with CSF-derived scRNA-Seq data. Unsupervised transcriptomic analysis Before performing dimensionality reduction, a list of the 2,359 most variable and highly expressed genes was generated by including genes with an average normalized and scaled expression value greater than 0.1 and with a dispersion (variance/mean) greater than 0.1. We then performed principal component analysis (PCA) over the list of variable genes. For both uniform manifold approximation and projection (UMAP) and SNN (shared nearest neighbor) clustering, we used the first 30 principal components. We used FindClusters within Seurat (which utilizes a SNN modularity optimization-based clustering algorithm) with a resolution of 0.7 and UMAP with minimum distance of 0.2 and number of neighbors of 50 to identify 27 clusters across the 37 input samples. 11 of these clusters were collapsed due to gene expression similarity evaluated by prior biological knowledge (7 extraneous divisions in cluster 0, 1 extraneous division in cluster 1, 2 extraneous divisions in cluster 3, and 1 extraneous division in cluster 4) to arrive at 17 total biological clusters across all samples. Dimensional reduction on data from the CD8 + T cells and myeloid cells alone was similarly performed using PCA followed by UMAP and SNN clustering, all implemented in Seurat. For CD8 + T cells, principal components 1-6 were used with UMAP parameters of minimum distance 0.3 and number of neighbors 20; a resolution of 0.4 was used to identify clusters. Cell type identification and within cell type analysis To identify genes that defined each cluster, we performed differential expression using the “bimod” test implemented with the FindMarkers function in Seurat based on a likelihood ratio test designed for single-cell differential expression incorporating both a discrete and continuous component. Thresholds were set at an average log-fold difference 0.2 and adjusted p -value (Bonferroni) less than 0.05. Top marker genes with high specificity were used to classify cell clusters into cell types (Supplementary Data 2 , 4 ) based on literature precedent. Two closely related clusters (T/NK clusters) were merged based on biological curation and analysis of hierarchical cluster trees yielding the twelve unique clusters. For T cells, we subclustered first on treatment condition, as we found that the original clusters were dependent on this metadata. The process used for clustering and subset identification was adapted for each iteration of clustering to optimize the parameters of variable genes, principal components, and resolution of clusters desired. Following identification of canonical subsets – CD4 + T cells, CD8 + T cells, and NK cells – these identities were assigned to the main T/NK cluster cells. One cluster, cluster 15, containing 50 cells was not classified as immune or malignant. All cells in this cluster came from sample CSF011-1 and upregulated genes associated with neuronal expression, and this cluster was classified as “other.” NK cell clusters within the pre-treatment and post-treatment T/NK groups were annotated based on expression of CD2 and FCRG3A ( CD16 ), lack of expression of CD3 genes ( CD3D , CD3E , CD3G ). The plasmacytoid DC (pDC) and conventional DC (cDC) clusters were distinguished from the other innate cells as dendritic cells, and then the differentially expressed genes between the two clusters were enriched using GSEA. The top GSEA hits on both gene lists distinguished cDCs and pDCs (Supplementary Data 9 ). Comparisons of abundance of T cells were made across time points with at least 20 T cells detected (34 of 37 time points). Comparisons of proliferation of CD8 + T cells were made across time points with at least 15 CD8 + T cells detected (31 of 37 time points). Differential expression and scoring over gene sets To identify differentially expressed genes within cell types and subtypes across treated and untreated conditions, we again used the ‘bimod’ setting in FindMarkers implemented in Seurat. To determine the scores of gene sets and pathways, such as IFN-γ response and antigen processing, we used the ‘AddModuleScore’ function in Seurat to construct a mean score of supplied genes subtracting a background score constructed from a random selection of genes in bins of average expression across all cells. When comparing scores within a specific subset of cells, AddModuleScore was constructed only over that subset, and recalculated if the subset was further partitioned. Tumor cell scores were calculated both across all patients (to compare pre-treatment and post-treatment time points across patients) and within individual patients (to compare across time points within patients). For specific comparisons of AddModuleScore-derived signatures with large differences in complexity between groups of cells, an upper complexity threshold and count-based downsampling were used to examine the possibility of complexity-confounded effects. No such effects were observed in comparing between tumor cells across patient and within patient. IFN-γ Response, Antigen Presentation, and Exhaustion Signatures IFN-γ response signature and exhaustion signatures were obtained from GSEA (HALLMARK_INTERFERON_GAMMA_RESPONSE, various signatures from Wherry et al. 2007), provided in Supplementary Data 5 . Antigen presentation signature was compiled following a search of the literature and is provided in Supplementary Data 5 . Inferred CNV analysis of Patient 043 To more accurately infer CNV patterns in high-complexity (complexity > 1000) tumor cells with sub-chromosomal resolution, we group genes into fixed length windows of 200 genes consecutive along the genome, removing from consideration those genes in the uppermost decile of dropout rate, as well as all immunoglobulin genes. All possible windows were considered where all included genes reside on the same chromosome. We converted the log (TP10k + 1) gene expression profiles to TP10k ones. We then took the mean TP10k expression over genes in a window, neglecting the highest 5 gene expressions in that window. This vector of values is hereafter referred to as the unnormalized Windowed Mean Expression (uWME). Having identified the malignant cells for each patient, we additionally computed a normalized version of the uWME as follows: the uWME from all patients’ non-malignant cells were averaged for each window across patients. HLA-* and associated genes on the 6p arm exhibited particularly strong hematopoietic expression; therefore, the means of these windows were imputed with the mean (windows) of the mean (patients) WME for all other windows. These values we refer to as the mean non-malignant uWME. We normalize uWME for malignant cells by dividing the window uWME by the mean non-malignant uWME for each window, hereafter referred to as Windowed Mean Expression (WME). To reduce possible confounding factors due to experimental or batch effects during subsequent clustering analysis, we converted the WME values in each single cell to ranks, hereafter referred to as the ranked, normalized WME (rWME). PCA and UMAP were performed on the rWME using the first 50 principal components of all tumor cells. In this P043, the UMAP-obtained clustering was concordant with that achieved via agglomerative clustering. To perform this clustering, we used as a distance metric 1-τ K , where τ K is the Kendall’s т coefficient between the WME of all pairs of cells. Agglomerative clustering was performed with a weighted linkage to obtain four clusters; two clusters contained single cells, and two other clusters contained 128 and 62 cells and were denoted descendant and ascendant respectively. To support the subclonal hypothesis without relying on either inferred CNV profiles or unsupervised clustering thereof, we performed a supervised comparison of single-cell expression profiles at each time point to both the early and late cfDNA-derived WES copy ratios (Fig. 4d ). We calculate the Kendall’s т correlation for all genes’ total copy ratio and single-cell expression, for all single cells. Then we calculate the difference in correlation for all single cells when using total copy ratio from time point 4 (late) vs. time point 2 (early). We observe that CSF043-2 single cells exhibit correlations more similar to WES from time point 2, and that CSF043-4 single cells exhibit correlations more similar to WES from time point 4. At CSF043-3, we observe bimodality in the distribution of the difference of Kendall’s т correlations. Additionally, we observe that single cells derived from post-treatment time points (CSF043-3 and CSF043-4) exhibit anti-correlation between their IFN-γ response score (Fig. 4d ) and the difference in Kendall’s т correlations between total copy ratios derived from WES at time point 4 vs. time point 2. We note that the relative populations of the two identified clusters in P043 varied significantly across time (Fig. 4e ). We plotted, for each gene, the mean purity corrected tCR vs change in the WME between all possible pairs of time points. The purity corrected tCR has the following form: $${{{{{{\rm{tCR}}}}}}}_{{{{{{\rm{corrected}}}}}}}=\frac{{{{{{{\rm{tCR}}}}}}}_{{{{{{\rm{observed}}}}}}}-\left(1-p\right)\times {{{{{{\rm{tCR}}}}}}}_{{{{{{\rm{germline}}}}}}}}{p}=\frac{{{{{{{\rm{tCR}}}}}}}_{{{{{{\rm{observed}}}}}}}-\left(1-p\right)}{p}$$ (1) where p is purity of sample calculated by ABSOLUTE 44 and tCR germline = 1. This relationship is demonstrated in Supplementary Fig. 11 , showing that the windowed expressional change between these clusters is concordant with the change in WES-derived tCR between any two time points. This concordance is robust to considering only the cells obtained at time point 3 (i.e., the correlated changes in single-cell expression and cfDNA-derived CNV profile cannot be attributed to batch effects confounding the observed scRNA-seq profiles). Windowed-mean expression results were compared to the InferCNV R package from the Broad Institute, and broad amplifications and deletions were concordant between the two approaches (Supplementary Fig. 11a,b ). Statistical analyses Statistical analyses of differential expression were performed using Seurat v2.3.4 implemented in RStudio. All statistical tests of distributions, cluster diversity, and change in representation, etc. were performed using base R packages implemented in RStudio. All statistical tests of gene set enrichment were performed using piano v1.22.0 and implemented in RStudio for all except enrichments of cluster markers for the full dataset, which was implemented in R. All violin plots and boxplots were generated using ggplot2 without modifications to smoothing or density. Boxplot rectangles encompass the 25 th to 75 th percentile with outliers as individual points above and below the rectangle. Overlapping genes between IFN response and antigen processing signatures were removed from both before utilization. As scores followed non-normal distributions as tested for using a Lilliefors normality test, we used a Wilcoxon rank-sum test where indicated for determining statistical significance. For scores in single-cell data, we report effect sizes in addition to statistical significance as an additional metric to capture the magnitude of the effect observed. The calculation was performed as Cohen’s d where: effect size d = (Mean 1 -Mean 2 )/(s.d. pooled). In Supplementary Fig. 5 , the calculation of Cohen’s d was modified to d pair = (Mean 1 -Mean 2 )/(s.d. 2 ), where the difference in means is normalized by the standard deviation of the pre-treatment group. All p -values subject to the multiple comparisons problem (such as marker identification by differential expression) were adjusted by Bonferroni correction. Wilcoxon rank-sum tests were calculated via the R command wilcox.test. Related Student’s t-test p-values were computed in python 3.7.7 using the function scipy.stats.ttest_rel from scipy v1.5.4. Theil-Sen slopes were computed in python 3.7.7 using the function scipy.stats.theilslopes from scipy v1.5.4. Kendall-tau correlations and associated p-values were computed using the function scipy.stats.kendalltau from scipy v1.5.4. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The genes-by-cells matrix and associated metadata generated from CSF draws and analyzed during the current study is available via the single-cell portal: . Raw data have been deposited in the dbGaP database under accession code phs002416.v1.p1 [ ] and the data are available under restricted access. Raw data are also available on the Broad Data Use and Oversight System (DUOS) through the accession codes DUOS-000131 and DUOS-000132: . The remaining data are available within the Article, Supplementary Information or Source Data file. Source data are provided with this paper. Code availability Panopticon v0.1 has been made publicly available ( ) 45 . R or python notebooks details of figure creation available upon reasonable request. Change history 25 November 2021 A Correction to this paper has been published:

Two new studies indicate that immunotherapy may benefit people with leptomeningeal carcinomatosis (LMD), a rare but serious complication of cancer that has spread to the brain and/or spinal cord. The research, which was led by investigators at Massachusetts General Hospital (MGH), Dana-Farber Cancer Institute and the Broad Institute, is published in Nature Communications. Although advances in cancer treatment have extended patient survival, some cancers come back, often in a different location in the body. This may in part help explain recent increases in the incidence of LMD—when tumor cells infiltrate the leptomeninges (layers of tissue that cover the brain and spinal cord) and cerebrospinal fluid. Approximately 5–8% of all patients with cancer develop LMD after first being diagnosed with breast cancer, lung cancer, melanoma or other malignancies. Current treatment options rarely benefit patients with LMD, and there is an urgent need for new therapies. Immune checkpoint inhibitors are important medications that boost the immune system's response against various cancers, but their effects against LMD are unclear. To investigate, researchers conducted two phase II clinical trials. When they collected and analyzed immune cells and cancer cells from the cerebrospinal fluid of patients in the trials both before and after treatment with immune checkpoint inhibitors, the scientists found signs that the therapy was having an effect. For example, the number of certain cancer-killing immune cells and the expression of particular genes within cells were higher following treatment. The second article in Nature Communications presents the results of one of the phase II studies, which included 18 patients with LMD who received combined ipilimumab and nivolumab (two types of immune checkpoint inhibitors) until the cancer progressed or the patient experienced unacceptable toxicity. The primary endpoint was overall survival at 3 months, and 8 of the 18 patients were alive at that time. (Historically, patients survive for a median of 3–7 weeks after being diagnosed with LMD.) One-third of patients experienced one or more serious adverse events. Two patients discontinued treatment due to unacceptable toxicity. The most frequent adverse events include fatigue, nausea, fever, anorexia and rash. The authors noted that larger, multicenter clinical trials are needed to validate their results. "In these two published studies, we demonstrated—in patients through a clinical trial and microscopically in the laboratory—that immune checkpoint blockade has promising activity for patients with LMD. More data is needed, but this is an exciting first step towards showing that immune checkpoint blockade may have a role in treating this devastating disease," says co-author Priscilla K. Brastianos, MD, who is the director of the Central Nervous System Metastasis Center at MGH and an associate professor of medicine at Harvard Medical School.

10.1038/s41467-021-25860-5

Biology

New tool predicts where coronavirus binds to human proteins

Haiyuan Yu, A 3D structural SARS-CoV-2–human interactome to explore genetic and drug perturbations, Nature Methods (2021). DOI: 10.1038/s41592-021-01318-w. www.nature.com/articles/s41592-021-01318-w Web server: 3d-sars2.yulab.org/ Journal information: Nature Methods

http://dx.doi.org/10.1038/s41592-021-01318-w

https://phys.org/news/2021-11-tool-coronavirus-human-proteins.html

Abstract Emergence of new viral agents is driven by evolution of interactions between viral proteins and host targets. For instance, increased infectivity of SARS-CoV-2 compared to SARS-CoV-1 arose in part through rapid evolution along the interface between the spike protein and its human receptor ACE2, leading to increased binding affinity. To facilitate broader exploration of how pathogen–host interactions might impact transmission and virulence in the ongoing COVID-19 pandemic, we performed state-of-the-art interface prediction followed by molecular docking to construct a three-dimensional structural interactome between SARS-CoV-2 and human. We additionally carried out downstream meta-analyses to investigate enrichment of sequence divergence between SARS-CoV-1 and SARS-CoV-2 or human population variants along viral–human protein-interaction interfaces, predict changes in binding affinity by these mutations/variants and further prioritize drug repurposing candidates predicted to competitively bind human targets. We believe this resource ( ) will aid in development and testing of informed hypotheses for SARS-CoV-2 etiology and treatments. Main The ongoing global COVID-19 pandemic has resulted in over 210 million SARS-CoV-2 infections and over 4.4 million deaths worldwide 1 . The coronavirus family of enveloped viruses causes respiratory and enteric tract infections in avian and mammalian hosts 2 . Seven well characterized human coronaviruses 3 , 4 , 5 exhibit symptoms ranging from mild respiratory illness to severe pneumonia and acute respiratory distress syndrome. These coronaviruses are either highly transmissible yet generally not highly pathogenic (for example HCoV-229E and HCoV-OC43) or highly pathogenic but poorly transmissible (SARS-CoV-1 and MERS-CoV). Unique from these, SARS-CoV-2 is both highly transmissible and capable of causing severe disease with infectivity and pathogenesis differing between individuals 6 , 7 . While ~25–35% of infected individuals experience only mild or minimal symptoms, ~1–2% of infected patients die primarily from severe respiratory failure and acute respiratory distress syndrome 8 , 9 . Differences in morbidity, hospitalization and mortality among different ethnic groups 10 , 11 , 12 , 13 , 14 , 15 are not fully explained by cardiometabolic, socioeconomic or behavioral factors, suggesting a role for human genetic variation in SARS-CoV-2 pathogenicity. Insights into the evolution of SARS-CoV-2, its elevated transmission relative to SARS-CoV-1 and dynamic range of symptoms have been key areas of interest. These traits are likely driven by molecular mechanisms of pathology, including interactions between the virus and its host, but specific causes are yet to be fully characterized. Networks of protein–protein interactions between pathogens and their hosts provide one avenue to understand mechanisms of infection and pathology. Viral–human interactome maps have been compiled for SARS-CoV-1 (ref. 16 ), HIV 17 , Ebola virus 18 and Dengue and Zika viruses 19 among others. Recent, affinity-purification mass spectrometry experiments on 29 SARS-CoV-2 proteins identified 332 viral–human interactions 20 . Interspecies interactions contribute to disease progression by facilitating pathogen entry into host cells 21 , 22 , 23 , 24 , 25 , 26 , inhibiting host response proteins and pathways 27 , 28 , 29 and hijacking cell signaling or metabolism to accelerate cellular (and consequentially viral) replication 30 , 31 , 32 . Structures and dynamics of these interactions can provide insights into their roles. For instance, the viral–human binding interface between poxvirus chemokine inhibitor vCCI and human MIP-1β is shown to occlude domains vital to chemokine homodimerization, receptor binding and interactions with GAG, thus explaining the inhibitory effect of poxvirus on chemokine signaling 29 . Additionally, the dynamics of a herpesvirus cyclin and human CDK2 interaction induce a conformational change on CDK2 that matches its interaction with human cyclin A, leading to dysregulated cell cycle progression 31 . Because protein–protein interactions mediate the majority of protein function 33 , 34 , 35 , targeted disruption by small-molecule inhibitors that compete for the same binding site provide a precise toolkit to modulate cellular function 33 , 35 , 36 , 37 , 38 . For instance, BCL-2 inhibitors that displace bound anti-apoptotic BCL-X interactors can treat chronic lymphocytic leukemia pathogenesis 39 . This approach can be particularly effective in viral networks and several potent inhibitors of key interactions have been developed. Disruption of viral complexes involved in viral replication has been successful in vaccinia virus 40 and human papilloma virus therapies 41 , 42 . Specifically, disruption of viral–host protein–protein interactions involved in early viral infection is an important therapeutic strategy. Discovery that a population variant in the membrane protein CCR5 conferred resistance to HIV-1 by disrupting its interaction with the viral envelope glycoprotein led to the development of Maraviroc as a US Food and Drug Administration-approved treatment for HIV-1 that functions by blocking the interface for this interaction 23 , 43 . Here we apply a full-interactome modeling framework to construct a three-dimensional (3D) structural interactome between SARS-CoV-2 and human proteins. Our framework first applies our previous ECLAIR framework 44 to identify interface residues for the whole SARS-CoV-2–human interactome and leverages these predictions to guide atomic-resolution interface modeling and docking in HADDOCK 45 , 46 . We additionally carried out in silico scanning mutagenesis in PyRosetta 47 to predict the impact of mutations on interaction binding affinity and explored the overlap between protein–protein and protein–drug binding sites. All results from our 3D structural interactome are provided as a user-friendly web server allowing exploration of individual interactions or bulk download and analysis of the whole dataset. We further explore the utility of our 3D interactome modeling approach in identifying key interactions undergoing evolution along viral-protein interfaces, highlighting population variants on human interfaces that could modulate the strength of viral–host interactions to confer protection from or susceptibility to COVID-19 and prioritizing drug candidates predicted to bind competitively at viral–human interaction interfaces, some of which could potentially be used for therapeutic purposes. Cumulatively these predictions and analyses are intended as a resource to facilitate investigation and further characterization of SARS-CoV-2–human interactions. Results Enrichment of variation on the spike–ACE2-binding interface We highlight the utility of computational and structural approaches to model the SARS-CoV-2–human interactome, from the interaction between the SARS-CoV-2 spike protein (S) and human angiotensin-converting enzyme 2 (ACE2) (Fig. 1a ). This interaction mediates viral entry into human cells 3 and is among the only viral–human interactions solved in both SARS-CoV-1 (ref. 48 ) and SARS-CoV-2 (refs. 49 , 50 , 51 ). Recent sequence divergences of the S protein are highly enriched at the S–ACE2 interaction interface (Fig. 1a ; log 2 odds ratio (OR) = 2.82, P = 1.97 × 10 −5 ), indicating functional evolution around this interaction. We predicted the impact of these mutations on the binding affinity (ΔΔG) between the SARS-CoV-1 and SARS-CoV-2 versions of the S–ACE2 interaction using the Rosetta energy function 52 (Fig. 1b,c ). The negative ΔΔG value of −14.66 Rosetta Energy Units (REU) indicates an increased binding affinity using the SARS-CoV-2 S protein driven by better optimized solvation and hydrogen bonding potential fulfillment. Our result is consistent with the hypothesis that increased stability of the S–ACE2 interaction contributes to the elevated transmission of SARS-CoV-2 (ref. 53 ). Experimental kinetics assays have confirmed that compared to SARS-CoV-1, SARS-CoV-2 S protein binds ACE2 with 10–20-fold higher affinity 54 , supporting the conclusions from our computational modeling. Fig. 1: Enrichment and predicted impact of divergences between SARS-CoV-1 and SARS-CoV-2 along the S–ACE2 interface. a , Co-crystal structure of the interaction between SARS-CoV-2 S with human ACE2 (PDB 6LZG ). All 15 sequence divergences between SARS-CoV-1 and SARS-CoV-2 S interfaces are highlighted as red spheres and all 6 population variants on the ACE2 protein interface are highlighted as green (ACE2_S19P), cyan (ACE2_T27A), blue (ACE2_E35K), purple (ACE2_E37K), yellow (ACE2_M82I) and orange (ACE2_G326E) spheres. Enrichment of these variants on the interface are reported for SARS-CoV-2 (log 2 OR = 2.82, P = 1.97 × 10 −5 by two-sided z -test) and human (log 2 OR = 0.38, P = 0.30 by two-sided z -test) shown to the right. Data are presented as log 2 OR ± s.e.m. b , c , Expanded interface views for the SARS-CoV-1 S–ACE2 structure (PDB 6CS2 ) and SARS-CoV-2 S–ACE2 structure (PDB 6LZG ). Sequence divergences are highlighted as red sticks. Inter-protein polar contacts that contribute to stabilizing the interaction are shown as yellow dashed lines. The negative predicted change in binding affinity (ΔΔG = −14.66 REU) indicates the interaction is more stable (lower energy) in the SARS-CoV-2 version of the interaction. d , Predicted impact of each ACE2 population variant. Mutated structures superimposed over the wild-type structure (magenta). The mutated residue is shown as sticks. Residues contributing to the overall change in binding energy are colored from blue (decreased ΔΔG) to white (no change) to red (increased ΔΔG). The gnomAD reported allele frequency and predicted ΔΔG for each mutation are reported. Full size image A wide range in severity of and susceptibility to SARS-CoV-2 exists between individuals 6 , 7 , 55 . Genetic predisposition hypotheses explaining this range include both expression-regulating and protein-coding variants 56 , 57 . For instance, an RNA-sequencing analysis suggested higher expression of ACE2 in Asian males could facilitate viral entry and explain increased susceptibility among this population 58 . Alternatively, missense population variants in ACE2 could strengthen or weaken the S–ACE2 interaction, thereby modulating susceptibility to infection. We used a mutation scanning pipeline in PyRosetta 59 , 60 to predict the impact of six missense variants reported in gnomAD 61 that occur on the S–ACE2 interface (Fig. 1d ). The three variants with the largest predicted impact on S–ACE2 binding affinity (ACE2_E37K (ΔΔG = 1.50), ACE2_M82I (ΔΔG = 2.95) and ACE2_G326E (ΔΔG = 5.74)) were consistent with previous experimental screens identifying them as putative protective variants exhibiting decreased binding of ACE2 to S 62 , 63 . Our results highlight utility for a 3D structural interactome modeling approach in identifying interactions and mutations important for viral infection, pathogenesis and transmission. Constructing the 3D structural SARS-CoV-2–human interactome To facilitate similar investigation and hypothesis development at the full-interactome scale, we next compiled a comprehensive 3D structural interactome between SARS-CoV-2 and human proteins based on 332 viral–human interactions uncovered in an early interactome screen 20 . First, we modeled SARS-CoV-2 proteins supplementing solved structures from the Protein Data Bank (PDB) 64 (16 of 29 proteins) with homology derived from SARS-CoV-1 templates (12 of 29 proteins). Homology models added one new structure for nsp14 (Extended Data Fig. 1a ), while comparison against the available SARS-CoV-2 PDB structures from the remaining 11 validated the quality of our modeling approach (Extended Data Fig. 1b,c ). For human interactors all models were obtained from the PDB or Modbase 65 (Extended Data Fig. 2a ). We then predicted the interface residues for each interaction using our ECLAIR framework 44 . In total, our pipeline identified 679 interface residues across 21 SARS-CoV-2 proteins with an average 18.23 residues per interface and 5,790 across 189 human proteins with an average 17.4 residues per interface. To provide structural interaction models for visualization and downstream analysis we performed guided docking in HADDOCK 45 , 46 using our high-confidence ECLAIR-predicted interface residues as restraints to refine the search space. To avoid potential biases in interface identification from docking low-coverage models (Extended Data Fig. 2b ) we only performed docking for 138 out of 332 interactions for which either (1) at least 33% of the full-length proteins were covered by available structures or (2) available structures included at least one high-confidence ECLAIR prediction to use as docking restraint. In total we report 1,248 docked interface residues across 15 SARS-CoV-2 proteins with an average 33.4 residues per interface and 4,604 across 138 human proteins with an average 32.4 residues per interface. For all analyses, docked interface annotations were prioritized over initial ECLAIR predictions. The full interface annotations from our ECLAIR and docking predictions are available in Supplementary Tables 1 and 2 , respectively. Benchmarking ECLAIR and guided docking predictions Our specific applications of ECLAIR (for interspecies interactions) and HADDOCK (performing data-driven docking with computational rather than experimental priors) are unique from those these tools were previously validated for. To ensure the robustness and quality of these methods for our interface prediction task, we constructed a comprehensive human–pathogen PDB benchmark set consisting of 509 interactions between a human protein and a viral or bacterial interactor (Fig. 2a ). The full list of interactions in this benchmark set alongside the PDB sources plus true and predicted interfaces are provided in Supplementary Table 3 . Fig. 2: Validation of ECLAIR and guided docking performance. a , Steps taken to parse the PDB and construct our human–pathogen PDB benchmark set. b , Comparison of ECLAIR performance on intraspecies interactions ( n = 200 human–human interactions) against interspecies interactions ( n = 509 human–pathogen interactions). Area under the receiver operating characteristic (AUROC) evaluation indicates considerable predictive power is achieved in both tasks (intraspecies AUROC = 0.737 and interspecies AUROC = 0.690). c , Comparison of final interface predictions across all residues in 153 dockable human–pathogen interactions using either ECLAIR (precision = 0.15 and recall = 0.19), a raw docking HADDOCK protocol (precision = 0.21 and recall = 0.21) or our guided docking HADDOCK protocol implementing ECLAIR predictions as restraints (precision = 0.19 and recall = 0.29). Recall from guided docking significantly outperformed the raw docking method ( P = 5.88 × 10 −6 by two-sided two proportion z -test) without losing precision ( P = 0.15 by two-sided two proportion z -test). Data are presented as precision or recall ± s.d. as estimated by 1,000-fold bootstrapping sampling 153 interactions and interface predictions with replacement each iteration. d , Distributions of RMSD between the top-scored raw or guided docking output and the co-crystal structure ( n = 153 dockable human–pathogen interactions). Interior box plots represent the distribution quartiles with whiskers representing the most extreme non-outlier values. Average RMSD from guided docking (average RMSD = 9.45) was significantly lower than raw docking (average RMSD = 11.79) based on a two-sided t -test ( P = 0.04). An example where the guided docking accurately identifies the correct interaction orientation missed by the raw docking (human protein shown by gray surface and raw docking, guided docking and co-crystal structure viral protein shown as green, orange and yellow cartoons, respectively) (right). e , Example showing a best-case scenario where a few true interface residues predicted by ECLAIR (recall = 27.7%) (top) are successfully expanded to identify the rest of the interface by the guided docking (recall = 95.7%) (bottom). Human and viral proteins shown in green (left) and in blue (right), respectively. Residues identified as an interface in each approach are darkened. True interfaces from the co-crystal structure are outlined and shaded in yellow. Full size image To validate ECLAIR’s applicability to interspecies interactions, we compared its published performance test set of 200 human–human interactions to its performance on our human–pathogen PDB benchmark set. Both tasks achieved comparable performance (receiver operating characteristic area under the curve = 0.69 versus 0.74), although the intraspecies task slightly outperformed interspecies (Fig. 2b ). We note that feature availability between sets (for instance, coevolution features can only be calculated for intraspecies interactions) may confound direct comparisons between different interaction sets. Overall, the evaluation of our benchmark conclusively shows that ECLAIR retains predictive power for interspecies interactions. To evaluate the benefit of using ECLAIR-predicted interfaces as restraints in HADDOCK docking, we compared our ECLAIR data-driven protocol against a raw protocol with no restraints. From the original 509 interspecies interactions, 153 fit our criteria for docking. We compared interface annotations from each protocol based on precision and recall (Fig. 2c ). Overall interface quality was comparable between both raw and guided protocols (precision = 0.21 versus 0.19, P = 0.15), however, the guided docking better recovered the total interface (recall = 0.21 versus 0.29, P = 5.88 × 10 −6 ). Previous evaluation on the HADDOCK framework confirms accurate interface predictions can be achieved even if the precise binding orientation is not recovered. While our main evaluation of interest is correct identification of interface residues, by evaluating the root-mean-square deviation (RMSD) between docked and reference structures, we further demonstrate that the guided docking better recapitulated the true co-crystal structures (Fig. 2d ; average RMSD = 9.45 versus 11.79, P = 0.04). Our aim in performing guided docking based on ECLAIR-predicted interfaces was to produce atomic-resolution structures that reflected our residue-level predictions for use in downstream analyses. However, we also hypothesized that docking would be effective in expanding accurate interface annotations to nearby residues if ECLAIR only identified a few high-confidence interface residues (Fig. 2e ). Comparison of the precision and recall between ECLAIR and our guided docking (Fig. 2c ) is consistent with this hypothesis and clearly demonstrates improvement in our guided docking approach over both raw docking and ECLAIR predictions. Depletion of human disease mutation at SARS-CoV-2 interfaces We explored evidence of interface-specific variation by mapping gnomAD-reported 61 human population variants (Supplementary Table 4 ) and sequence divergences between SARS-CoV-1 and SARS-CoV-2 (Supplementary Table 5 ) onto predicted interfaces. Conserved residues generally cluster along protein–protein interfaces 66 and an analysis of SARS-CoV-2 structure and evolution similarly concluded highly conserved surface residues likely drove protein–protein interactions 67 . Consistent with these previous studies, we observed significant interactome-wide depletion for both viral and human variation along predicted interfaces comparable to that observed along solved human–human interfaces (Fig. 3a ). Fig. 3: Enrichment of sequence divergences and disease mutations across all SARS-CoV-2–human interaction interfaces. a , Enrichment across 332 human genes interacting with SARS-CoV-2 for viral sequence divergence or human population variants along viral–human (V:H, log 2 OR = −0.91, P = 2.41 × 10 −10 by two-sided z-test) human–viral (H:V, log 2 OR = −0.38, P = 7.92 × 10 −13 by two-sided z -test) or human–human (H:H, log 2 OR = −0.14, P = 9.98 × 10 −4 by two-sided z -test) interfaces. Data are presented as log 2 OR ± s.e.m. b , c , Individual enrichments (sorted from most depleted to most enriched) for human population variants and viral sequence divergences, respectively on all 332 SARS-CoV-2–human interaction interfaces. Interfaces with statistically significant log 2 OR (by two-sided z -test) are labeled and shown as bars, the remainder are plotted as a line. Data are presented as log 2 OR ± s.e.m. Clusters of SARS-CoV-2 enrichments involving the nsp4 interactions with IDE, NUP210, DNAJC11, TIMM29, TIMM9 and TIMM10 and nsp2 interactions with GIGYF2, FKBP15, WASHC4, EIF4E2, POR and SLC27A2 were labeled as a group for legibility. d , Percentage of human genes that interact with (green, n = 332) or do not interact with (orange, n = 20,018) SARS-CoV-2 that contain disease annotations in HGDM (log 2 OR = 0.57, P = 1.70 × 10 −4 by two-sided z -test), ClinVar (log 2 OR = 0.64, P = 1.05 × 10 −4 by two-sided z -test) and GWAS (log 2 OR = 0.89, P = 4.54 × 10 −5 by two-sided z -test), respectively. Genes targeted by SARS-CoV-2 proteins were significantly more likely to harbor disease mutations than non-interactors by log odds enrichment test. Data presented as percentage ± s.e.m. e , Sample of individual disease terms enriched in human genes targeted by SARS-CoV-2. Full results are reported in Supplementary Table 6 . Data are presented as log 2 OR ± s.e.m. f , Comparison of the enrichment of HGDM- or ClinVar-annotated mutations on human–viral interfaces or human–human interfaces for 332 genes interacting with SARS-CoV-2. Disease mutations were enriched on human–human interfaces (HGMD, log 2 OR = 0.84, P < 1 × 10 −20 by two-sided z -test; ClinVar, log 2 OR = 0.25, P = 2.9 × 10 −3 by two-sided z -test), whereas human–viral interfaces showed no or marginal enrichment (HGMD, log 2 OR = 0.31, P = 0.048 by two-sided z -test; ClinVar, log 2 OR = 0.15, P = 0.39 by two-sided z -test). GWAS category was excluded from this analysis because most lead GWAS single-nucleotide polymorphisms occur in noncoding regions. Data are presented as log 2 OR ± s.e.m. Full size image Nonetheless, considering each interaction individually, we identified 11 interaction interfaces enriched for human population variants (Fig. 3b ) and 4 enriched for recent viral sequence divergences (Fig. 3c ). Supplementary Table 6 provides the log odds enrichment values for each interface. Similar to the S–ACE2 interface, a high degree of variation on these viral interfaces may indicate recent functional evolution around specific viral–human interactions. Because human evolution is slower, enrichment of population variants along the human interfaces is unlikely to be a selective response to the virus. Rather, interfaces with high population variation may represent edges in the interactome most prone to modulation by existing variation between individuals or populations. Alternatively, enrichment and depletion of variation along the human–viral interfaces could help distinguish viral proteins that bind along existing (likely conserved) human–human interfaces from those that bind using new interfaces (unlikely to be under selective pressure). To further explore the functional importance of variations within human interactors of SARS-CoV-2, we considered phenotypic associations reported in HGMD 68 , ClinVar 69 or the NHGRI-EBI GWAS catalog 70 . Interactors of SARS-CoV-2 were enriched for phenotypic variants from each database (Fig. 3d ). Notably, several of the individual disease categories enriched among interactors, were consistent with SARS-CoV-2 comorbidities, including heart disease, respiratory tract disease and metabolic disease 12 , 71 (Fig. 3e and Supplementary Table 7 ). Disruption of native protein–protein interactions is one mechanism of disease pathology and disease mutations are known to be enriched along protein interfaces 72 , 73 . Variants on predicted human–viral interfaces matched allele frequency distributions of variants off the interfaces, but were considered overall to be more deleterious by SIFT 74 and PolyPhen 75 (Extended Data Fig. 3 ). However, while we showed that annotated disease mutations were significantly enriched along known human–human interfaces, enrichment was drastically reduced (HGMD) or insignificant (ClinVar) on human–viral interfaces (Fig. 3f ). This is likely because mutations that disrupt human–viral interfaces would not disrupt natural cell function and hence would be unlikely to manifest as disease phenotypes. Our finding that disease mutations and viral proteins affect human proteins at distinct sites is consistent with a two-hit hypothesis of comorbidities whereby proteins whose function is already affected by genetic background may be further compromised by viral infection. Binding affinity changes from SARS-CoV-1 to SARS-CoV-2 Using a PyRosetta pipeline 47 , 59 , 60 we predicted the impact of sequence divergences between SARS-CoV-2 and SARS-CoV-1 on the binding energy (ΔΔG) of 138 viral–human interactions amenable to docking. Although the binding energy for most interactions was unchanged, we note that the divergence from SARS-CoV-1 to SARS-CoV-2 was biased toward a decreased binding energy (that is more stable interaction) (Fig. 4a and Supplementary Table 8 ). The outliers in these ΔΔG predictions may help pinpoint key differences between the viral–human interactomes of SARS-CoV-1 and SARS-CoV-2. Fig. 4: Predicted impact of sequence divergences on the binding affinity of SARS-CoV-2–human interactions. a , Predicted impact of SARS-CoV-1 to SARS-CoV-2 sequence divergences on binding affinity from docked structure for 83 applicable SARS-CoV-2–human interactions sorted from largest decrease (most stabilized relative to SARS-CoV-2) to largest increase (most destabilized relative to SARS-CoV-1) (mean = −0.57 REU, s.d. = 5.78 REU). Interaction labels shown wherever predicted ΔΔG exceeds mean ± (1 × s.d.). b , Representative cropped western blots (among three replicates) from co-IP comparing the interaction between human PRIM2 with SARS-CoV-1 or SARS-CoV-2 nsp1. More efficient PRIM2 pulldown with SARS-CoV-2 bait validates the PRIM2-nsp1 ΔΔG prediction (ΔΔG = −17.3 REU, z score = −2.9). Docked structure for PRIM2 with SARS-CoV-2 nps1 (green and blue cartoon, respectively) (bottom). SARS-CoV-1 to SARS-CoV-2 sequence divergences are represented as spheres. Interface residues are colored relative to overall ΔΔG contribution ranging from blue (more stabilizing in SARS-CoV-2) to white (little impact on ΔΔG), to red (more stabilizing in SARS-CoV-1). Residue side chains are shown as sticks in regions with high local ΔΔG. c , Representative Y2H results (among three replicates) confirming that six interactions with no predicted ΔΔG values can be detected using either SARS-CoV-2 or SASR-CoV viral protein as bait. The docked structure (visualized as in b ) for human GFER with SARS-CoV-2 nsp10 (ΔΔG = −0.06) is shown to highlight that sequence divergences in these six interactions did not localize near the interface. d , Distribution of the predicted changes in binding affinity from scanning mutagenesis for all 2,023 human population variants on SARS-CoV-2–human interfaces. Values were z score-normalized across each residue type and on each interface. Shaded regions indicate putative interface binding energy hotspots annotated as strongly disruptive ( z score ≥2, 48 total variants), disruptive (1 ≤ z score <2, 42 total variants), stabilizing (−2 < z score ≤−1, 25 total variants) or strongly stabilizing ( z score ≤−2, 26 total variants). Interior box plot represents the distribution quartiles with whiskers representing the most extreme non-outlier values. e , Docked structure between SARS-CoV-2 N protein and human G3BP2, alongside expanded interface views comparing the wild-type interface (left) with a predicted strongly disruptive (ΔΔG = 10.3 REU, z score = 2.3) population variant, G3BP2_P121T (right). Yeast two-hybrid results confirmed that the G3BP2_P121T variant completely disrupts the G3BP2–N interaction (bottom). Source data Full size image To further explore and validate the biological relevance of these predicted changes, we performed yeast two-hybrid (Y2H) screens to test 30 human interactors against both SARS-CoV-1 and SARS-CoV-2 baits. Our Y2H experiments reconstituted six of these interactions (20%) using the SARS-CoV-2 bait. Extensive previous studies across many species and hundreds of well-validated interactions show inherent limits in assay sensitivity for all high-throughput interaction assays (detection rates span 15–25%) 76 , 77 , 78 , 79 . This is due in part to inability to match native expression, proper folding or post-translational modifications under assay conditions. Our 20% reproducibility rate (in line with expected sensitivity of the Y2H system) indicates good quality of the published interactome. In each of the six reproduced interactions we predicted no changes in binding affinity between SARS-CoV-2 and SARS-CoV-1. Consistent with this prediction, each interaction was also detected using the SARS-CoV-1 bait (Fig. 4c ). Docked models for these interactions suggest sequence divergences between SARS-CoV-1 and SARS-CoV-2 occurred away from the interface and would be unlikely to affect binding (Fig. 4c ). We additionally performed co-immunoprecipitation (co-IP) assays for the interaction between human DNA Primase Subunit 2 (PRIM2) and SARS-CoV-2 nsp1 (Fig. 4b and Source Data Fig. 4 ; predicted ΔΔG = −17.3 REU). Several deviations in nsp1 were predicted to cumulatively stabilize this interaction near the edges of its interface. Results from co-IP validated our prediction showing that SARS-CoV-2 nsp1 was more effective at pulling down human PRIM2 than was SARS-CoV-1 nsp1. Moreover, a follow-up quantitative mass spectrometry comparison of SARS-CoV-2, SARS-CoV-1 and MERS-CoV 80 included five interactions that we predicted to be more stable in SARS-CoV-2. Consistent with our predictions three of these (RNF41-nsp15, PRIM2-nsp1 and SNIP1-N) showed interaction preferences for the SARS-CoV-2 protein. Specifically, the interaction between RNF41 and nsp15 was exclusively detected in SARS-CoV-2. Overall, these independent experimental results together with our co-IP result thoroughly validate the accuracy of our 3D interactome modeling approach and demonstrate its utility in identifying functional differences between SARS-CoV-1 and SARS-CoV-2. Impact of population variants on binding affinity We hypothesized the dynamic range of patient responses and symptoms reported for SARS-CoV-2 infection can be explained in part by missense variations and their impact on viral–human interactions. This is consistent with previous reports that up to 10.5% of missense population variants can disrupt native protein–protein interactions 81 and that underlying genetic variation can explain up to 15% of variation in patient response and viral load in other viruses, including HIV 82 . To explore this hypothesis we employed a previously benchmarked scanning mutagenesis protocol provided through PyRosetta 47 , 59 to identify candidate binding energy hotspot mutations for all docked interfaces. Out of 2,023 population variants on eligible interfaces, we identify 90 (4.4%) as predicted disruptive hotspots and 51 (2.5%) as predicted stabilizing hotspots (Fig. 4d ). Fig. 5: Drug docking and prioritization of SARS-CoV-2–human interaction inhibitors. a , Validation of smina’s ability to identify the correct binding site from the full protein surface based on 4,399 drug–ligand pairs across 95 protein targets. Docking was carried out either by re-docking each ligand back into its native protein structure or cross-docking each ligand into a representative receptor structure. Baseline performance expectation derived from random selection of surface patches matching the size of the correct binding site is shown for comparison. Each line and shaded area indicates the percentage of docks that correctly identify X binding site residues ± s.d. as estimated by 1,000-fold bootstrapping sampling 95 drug–target pairs with replacement each iteration. The gray shaded area (top) indicates the maximum fraction of docks whose true binding sites contain at least X residues. b , Protein–protein and protein–drug binding sites pooled across 16 applicable drug–target pairs were significantly enriched (log 2 OR = 1.38, P = 2.1 × 10 −7 by two-sided z -test). Data are presented as log 2 OR ± s.e.m. c , Individual breakdown of the overlap between the each of the protein–protein and protein–drug binding sites as either undockable (no protein–protein docked structure available for comparison; 14 total), no overlap (7 total), partial overlap (1 total) or significant overlap (8 total). The individual log 2 OR for each of the significant drug target pairs are shown. Data are presented as log 2 OR ± s.e.m. The MARK3–ZINC95559591 pair (shown in d ) is highlighted in red. d , Docked structure for ZINC95559591 bound to human MARK3. MARK3 surface is colored either green (non-interface, n = 270), blue (orf9b interface, n = 28), red (ZINC95559591 interface, n = 3) or magenta (shared interface, n = 12). Cut-out display highlights the MARK3–ZINC95559591 binding site. Polar contacts between MARK3 and ZINC95559591 are shown as dashed lines. e , Corresponding docked structure for SARS-CoV-2 orf9b bound to human MARK3. Full size image To validate our predictions for the impact of population variants, we generated a Ras GTPase-activating protein-binding protein 2 (G3BP2) variant, G3BP2_P121T ( rs1185000405 ) using site-directed mutagenesis as described previously 83 . We annotated this variant as strongly disruptive (predicted ΔΔG = 10.3 REU) and had confirmed earlier that the interaction between N and wild-type G3BP2 could be recapitulated using Y2H. Comparing the Y2H results between wild-type and mutant G3BP2 confirmed complete disruption of the G3BP2–N protein interaction by G3BP2_P121T (Fig. 4e ). Analysis of the docked models suggests that this disruption is driven by steric clashes between the mutated residue in G3BP2 and Glu-323 and Thr-325 of the N protein. The unfavorable polar interaction and steric bulk from the hydroxyl side chain of the threonine variant was also predicted to induce a rotation in the Trp-330 of N, disrupting hydrophobic interaction with Trp-282. G3BP2 is implicated in cardiovascular diseases 84 , potentially linking this interaction to known comorbidities. Moreover, G3BP2 alongside G3BP1 is an important target in viral etiology; sequestration of both proteins by SARS-CoV-2 N protein results in an inhibition of stress granule formation and suppression of host innate immune responses 85 , 86 . Therefore, the existence of naturally occurring variation disrupting this interaction is of particular interest. Although the G3BP2_P121T variant is rare (AF = 0.00043%), it may affect SARS-CoV-2 progression in roughly 30,000 individuals who carry it worldwide. Overall, our computational and experimental work concretely shows that human population variants can modulate the SARS-CoV-2–human interactome network and that our interface and energy modeling predictions can help identify such variants. The full predicted impact of all 2,023 population variants along SARS-CoV-2 interaction interfaces is provided in Supplementary Table 9 and may inform future studies investigating genetic contribution to COVID-19. Comparing binding sites of drugs and SARS-CoV-2 proteins Drugs that directly interfere with viral–host interactions (for instance by competing for the same binding site) could provide promising clinical leads to target viral infection or replication. On this basis we consider potential for our 3D interactome modeling approach to inform drug repurposing strategies. We aimed to further prioritize a current candidate set including 76 expert-reviewed drugs targeting one or more of the 332 identified human interactors of SARS-CoV-2 (ref. 20 ) on the basis of the potential for competitive binding. We performed protein–ligand docking using smina 87 to identify drug binding sites for 30 out of 76 candidate drug–target pairs that have available human receptor structures (Supplementary Table 10 ). Smina, a fork of the widely used AutoDock Vina, competes competitively in pose prediction challenges 87 and is validated by us to robustly identify the true binding site from the full protein surface on a published benchmark set of 4,399 experimentally solved protein–ligand complexes (Fig. 5a ) 88 . We compared the overlap of predicted drug binding sites with the corresponding docked viral–human interaction interface for 16 cases with both predictions available. Overall drug binding sites were significantly enriched at the interaction interface compared to the rest of the protein surface (Fig. 5b ; log 2 OR = 1.38, P = 2.1 × 10 −7 ). Individually, we further prioritized eight drugs that exhibited significant overlap between the drug- and viral-protein-binding sites (Fig. 5c ), several of which have been explored by recent independent studies. A retroactive association study identified previous treatment with metformin as an independent factor associated with reduced mortality in diabetic patients 89 , although a precise mechanism was not explored at the time. Ongoing phase 2 and phase 4 clinical trials are being conducted or are planned for silmitasertib and valproic acid, respectively (ClinicalTrials.gov identifiers NCT04668209 and NCT04513314 ). As an example, we highlight orf9b-MARK3 interaction whose interface we predicted could be blocked by ZINC95559591 (MRT-68601 hydrochloride) (Fig. 5d,e ). MARK3 is a serine/threonine protein kinase involved in microtubule organization with implicated roles in modulating gene expression by activating histone deacetylation proteins. Our models suggest that both ZINC95559591 and orf9b bind and make several polar contacts with MARK3 (for example one with Tyr-134) near its active ATP site. Consistent with its known role as an inhibitor of MARK3 (ref. 90 ) our model shows that ZINC95559591 binds deep within the ATP active site of MARK3. By contrast the N-terminal tail of orf9b forms looser contact, only entering the periphery of the active pocket. Therefore, we suspect that ZINC95559591 may outcompete orf9b for this pocket; thus making it a prime candidate to explore targeted disruption of SARS-CoV-2–human protein–protein interactions through drug repurposing. While this example fits our criteria for prioritized drug repurposing and competitive binding, it does raise further questions to consider. Namely, the functional role of a SARS-CoV-2–human interaction (whether the viral protein co-opts versus disrupts native human protein function or whether interaction is part of an immune response against the virus) is needed to inform potential clinical utility of drug repurposing. As both orf9b and ZINC95559591 bind within the same MARK3 active site, both may induce an inhibitory effect and ZINC95559591 could be counterproductive; even if it outcompetes orf9b, it may replace a harmful viral inhibitor with a more potent chemical one. In this scenario, exploration of the predicted binding sites of SARS-CoV-2 proteins could still help to uncover an inhibitory role in viral etiology. Moreover, it may be possible to design analogs of inhibitor drugs that retain high binding affinity to their receptor but lose their inhibitor activity. Therefore, while these factors may complicate the prospects of drug repurposing, we are optimistic that our 3D interactome modeling approach can facilitate understanding of viral mechanisms and may aid development of new treatments. The SARS-CoV-2–human 3D structural interactome web server We constructed the SARS-CoV-2–human 3D interactome web server ( ) to provide our computational predictions and modeling as a comprehensive resource to the public. All results and analyses described herein are directly available for bulk download or users can quickly navigation through the reported interactome to see a summary of our analyses for specific interactions of interest (Fig. 6 ). Fig. 6: 3D-SARS2 structural interactome browser overview. Overview of the main results page for exploring a given interaction in our 3D-SARS2 structural interactome browser. The main display contains information for both the SARS-CoV-2 and human proteins, including structural displays for either the docked or single crystal structures as well as a table summarizing the interface residues for both proteins. Interface residues are colored dark blue and dark green for the viral and human proteins, respectively. By default, the page will display the docked structure if available. The display can be toggled between docked structures and single structures using the button in the bottom middle. When a single structure display is selected residues will instead be colored based on the initial ECLAIR interface definition. Four categories of expandable panels containing additional analyses are provided. The interface view shows a linear representation of the protein sequence with interface residues annotated in dark blue or dark green (top left). Interfaces for other interactors of the protein are shown underneath for easy comparison. The mutations panel summarizes either human population variants or viral sequence divergences on the protein (top right). Mutations on the interface are labeled. The ΔΔG information panel summarizes the results from in silico mutagenesis scanning along the interface (bottom left). Results for each mutation are z score-normalized relative to the rest of the interface and colored blue (negative ΔΔG, stabilizing) to yellow (minimal impact) to red (positive ΔΔG, destabilizing). The heat map can be filtered to show only values corresponding to known mutations on the interface. The candidate drugs panel shows docking information for any known drug targets of the human protein (bottom right). Full size image The interface comparison panel (Fig. 6 top left) visualizes the interface annotation along a linear sequence and provides comparison against all other known or predicted interfaces from the same protein. This comparison may reveal biologically meaningful insights about the interface overlap and possible competition between viral and human interactors. The mutations panel (Fig. 6 top right) presents information on variation within each interaction partner; divergences from the SARS-CoV-1 or gnomAD population variants. We provide a log odds enrichment or depletion of variation along the interface which can help highlight interactions undergoing functional evolution for further characterization. For interactions amenable to docking, the ΔΔG Information panel (Fig. 6 bottom left) compiles the predicted impact of all possible mutations across the docked interface on binding affinity. Individual mutations are colored by their z score normalized ΔΔG prediction and can be toggled to only show the impacts of known variants. On the viral side, a cumulative ΔΔG value compares binding affinity between the SARS-CoV-1 and SARS-CoV-2 versions of the protein. Finally, the drug panel (Fig. 6 bottom right) describes any drugs known to target human proteins and provides information for each drug alongside display options for visualizing predicted binding conformations. The overlap between the drug binding site and interface with the viral protein is reported. The SARS-CoV-2–human 3D structural interactome web server currently includes 332 viral–human interactions 20 . We will continue support for the web server with periodic updates as additional interactome screens between SARS-CoV-2 and human are published. As we update, a navigation option to select between the current or previous stable releases of the web server will be provided. Discussion Our 3D SARS-CoV-2–human interactome provides a comprehensive resource to supplement ongoing and future investigations into COVID-19. The analyses provided and discussed throughout highlight potential applications of these predictions to inform structure-based hypotheses regarding the roles of individual interactions and prioritize further functional characterization of evolutionarily relevant interactions, causal links connecting population variation with differences in response to infection and drug candidates that may interfere with interaction-mediated disease pathology. Our observation that perturbation from underlying disease mutations and viral protein binding occur at distinct sites on human proteins may warrant further investigation into whether the combined role of these two sources of perturbation is clinically relevant to mechanisms of comorbidities. Although we have experimentally validated several of our predictions, we emphasize that further experimental characterization should be conducted to corroborate any hypotheses derived from individual predictions. Moreover, these predictions are not without limitation. Interface predictions may not be applicable to some published human targets identified by mass spectrometry 20 if they represent indirect complex associations rather than direct binary interactions 76 . Further, while structural coverage from SARS-CoV-2 proteins was robust, per-residue coverage of the human proteome is less complete (Extended Data Fig. 2 ). Though we only performed molecular docking for low-coverage structures when strong prior ECLAIR interface restraints were available, coverage restrictions can nonetheless introduce bias and may prohibit identification of true interface residues. Recent advances in protein-folding predictions 91 , 92 , 93 may ameliorate this restriction in the future. In the meantime, initial ECLAIR interface annotations (not susceptible to structural coverage limitations) may provide orthogonal value to docked models. Additionally we caution that direct quantitative interpretation of Rosetta-predicted ΔΔG values is often difficult. In particular, relative importance of scoring function terms may differ between proteins and interactions of varying sizes and compositions. For these reasons, we only evaluate normalized predictions to compare the relative qualitative differences from our scanning mutagenesis results. Moreover, because mutated structure optimization focuses only on side-chain repacking, our analysis is limited to mutations at or near the interface where side-chain repacking can have a direct effect. We expect mutations that substantially impact binding affinity through refolding or other allosteric effects exist but cannot be captured by our method. Notably, users can tailor the use of our raw predictions to their own interests; thus expanding upon the concepts and applications our analyses explore. For instance, we limited investigation of druggable interactions to repurposing known drugs that overlap and might disrupt viral–host interactions which we hypothesized would elicit the most promising clinical responses. However, this approach reduces the scope of the SARS-CoV-2–human interactome to only a few interactions that already have known drug candidates. An alternative application could prioritize candidate druggable interfaces throughout the whole SARS-CoV-2–human interactome by overlapping our interface annotations with predictions of druggable protein surfaces using recent deep-learning approaches 94 with the aim of designing new protein–protein interaction inhibitors. Overall, we believe that our 3D structural SARS-CoV-2–human interactome web server ( ) will prove to be a key resource in informing hypothesis-driven exploration of the mechanisms of SARS-CoV-2 pathology and host response. The scope and potential impacts of our web server will continue to grow as we incorporate the results of ongoing and future interactome screens between SARS-CoV-2 and human data. Finally, we note our 3D structural interactome framework can be rapidly deployed to analyze future viruses. Methods Generation and validation of SARS-CoV-2 homology models Homology-based modeling of all 29 SARS-CoV-2 proteins was performed in Modeller 95 using a multiple template modeling procedure consistent with previous high-profile homology modeling resources 96 . In brief, candidate template structures for each query protein were selected by running BLAST 97 against all sequences in the PDB 64 retaining only templates with at least 30% identify. Remaining templates were ranked using a weighted combination of percent identity and coverage described previously 96 . The final set of overlapping templates to use was first seeded with the top-ranked template with additional templates being added iteratively if (1) overall coverage increase from the template was at least 10% and (2) percent identify of the new template was no less than 75% the identity of the initial seed template (that is, if the template seed showed 80% identity, additional templates with percent identity as low as 60% could be included). Query-template pairwise alignments were generated in Modeller using default settings and were manually trimmed to remove large gaps (five or more gaps in a ten-residue window). Finally, modeling was carried out using the Modeller automodel function. This approach generated homology models for 18 out of 29 proteins. Based on manual inspection of the template quality and sources, homology models were further filtered to 12 models for which a high-quality template from a SARS-CoV-1 homolog was available. Moreover, during revision of this manuscript, newly deposited PDB structures for many SARS-CoV-2 proteins ( ) allowed independent validation of homology model quality based on the RMSD following alignment and refinement in PyMol 98 . Visual representations of these alignments between modeled and solved structures are provided in Extended Data Fig. 1 . For all analyses SARS-CoV-2 PDB structures were prioritized where available and only the homology model for nsp14 was retained. Interface prediction using ECLAIR Interface predictions for all 332 interactions reported previously 20 were made in two phases. In phase one, we leveraged our previously validated ECLAIR framework 44 to perform initial residue-level predictions across all interactions. ECLAIR compiles five sets of features: biophysical, conservation, coevolution, structural and docking. In brief, biophysical features are compiled using a windowed average of several ExPASy ProtScales 99 , conservation features are derived from the Jensen–Shannon divergence 100 , 101 from known homologs for each protein, coevolution features between interacting proteins are derived from direct coupling analysis 102 and statistical coupling analysis 103 among paired homologs, structural features are obtained by calculating the solvent-accessible surface area of available PDB 64 or ModBase 65 models using NACCESS 104 and docking features are the average interchain distance and surface occlusion per residue from a consensus of independent Zdock 105 trials. Slight alterations were made to accommodate SARS-CoV-2–human predictions. First, construction of multiple sequence alignment (MSA) for statistical coupling analysis and direct coupling analysis calculations require at least 50 species containing homologs of both interacting proteins. Therefore, coevolution features could not be calculated for interspecies interactions. Second, MSAs for conservation features typically only allow one homolog per species. Because viral species classifications are less precise and are often subdivided into unique strains (and because all higher-order ECLAIR classifiers require protein conservation features) we modified the MSAs for viral proteins to include homologs from various strains in a single species. The initial prediction results from ECLAIR are provided in Supplementary Table 1 . Interface prediction using guided HADDOCK docking Interface predictions for all 332 interactions reported previously 20 were made in two phases. In phase two, we leveraged high-confidence interface predictions from ECLAIR to perform guided docking in HADDOCK 45 , 46 . An introduction to protein–protein docking in HADDOCK is provided at . In brief, HADDOCK is designed to perform data-driven docking using (traditionally experimentally derived) priors about the interface. These data (for example scanning mutagenesis) often indicate sets of residues involved in the interface but no pairwise information linking interface residues between each protein. These residues (termed active residues) are used in conjunction with any neighboring surface residues (termed passive residues) to drive rigid body docking, by introducing a scoring penalty for any active residue on one protein not in proximity of an active or passive residue on the other. This approach is formalized as a set of ambiguous interaction restraints (AIRs) that evaluate the distances of each active residue to the active or passive residues on the other protein. The approach ensures that experimental priors about interface composition are enforced, but leaves the exact orientation and pairing of residues flexible to HADDOCK’s energy-based scoring function. To incorporate computational interface predictions from ECLAIR we use the standard HADDOCK protein–protein docking framework. Active residues are encoded as all high-confidence ECLAIR predictions at the surface (≥15% solvent-accessible surface area (SASA)). Passive residues are identified as all surface residues (≥40% SASA) within 6 Å of an active residue. For definition of surface residues, the 15% SASA cutoff provides consistency with our definition of interface residues, whereas the 40% SASA cutoff provides consistency with the typical recommendation in HADDOCK. All SASA calculations were carried out using NACCESS 104 and neighboring residues were selected using PyMol 98 . Following HADDOCK recommendations to reduce computational burden from using many restraints, we defined our AIRs using only the α-carbons and increased the upper distance limit for from 2 Å to 3 Å. All other HADDOCK run parameters were left at the default. In total, 1,000 rigid body docking trials were performed and the top-200-scored orientations were retained for subsequent iterations, refinement and analysis. For each interaction we identified available PDB or homology model structures to determine whether the interaction should be eligible for docking. Previous benchmark evaluations show that HADDOCK performs well using homology models, but that performance drops off for models produced from low sequence identity templates 106 . In all cases PDB models were prioritized over homology models. We next evaluated risks of using low-coverage structures for protein–protein docking; using structure fragments that completely exclude the true interface residues will produce false interface predictions. We aimed to minimize this risk while maximizing the dockable interactome by setting two conditions for determining structure eligibility. First, protein structures covering at least 33% of the total protein length were considered sufficiently large for docking. Second, protein structures at least 50 residues in length and containing at least one high-confidence ECLAIR-predicted interface residue to use as an active residue were made eligible. Inclusion of an ECLAIR-defined active residue gives us reasonable confidence that part of the interface is covered and therefore, true docked interface predictions should be possible. When multiple structures were available for one protein, ranking was based on the sum of ECLAIR scores for all residues covered by each structure; we always selected the available structure most likely to include the true interface. In total we performed guided HADDOCK docking on 138 out of 332 interactions. The remaining 194 interactions did not have reliable 3D models for both interactors. The top-scored docked conformation from each HADDOCK run was retained. The final docked interface annotations are provided in Supplementary Table 2 . Definition of interface residues We annotated interface residues from atomic-resolution docked models, using an established definition for interface residues 44 . The SASA for both bound and unbound docked structures was calculated using NACCESS 104 . We defined an interface residue as any residue that is both (1) at the surface of a protein (defined as ≥15% relative accessibility) and (2) in contact with the interacting chain (defined by a ≥1.0 Å 2 decrease in absolute accessibility). Human–pathogen co-crystal structure benchmark set We constructed a benchmark set of experimentally determined co-crystal structures to evaluate the performance of both our ECLAIR and guided HADDOCK docking interface predictions on interspecies interactions (Fig. 2a ). First, we parsed 165,567 PDB structures, identified all interacting chains by interface residue calculation and mapped PDB chains to UniProt protein IDs using SIFT 74 to identify a total of 33,242 unique protein–protein interactions. Using taxonomic lineages from UniProt we filtered this set to 7,738 interactions involving human proteins, of which 6,256 represented human–human intraspecies interactions and 1,482 represented interspecies interactions between humans and some other species. Finally, to provide the most relevant set of interactions that would be biologically similar to SARS-CoV-2–human interactions, we considered only interactions between human and viral proteins (346) or between human and bacterial proteins (163). We refer to this collective set of 509 co-crystal structures as our human–pathogen PDB benchmark set. The full list of structures and interface annotations for this benchmark set is provided in Supplementary Table 3 . To validate performance of ECLAIR predictions on the human–pathogen PDB benchmark, ECLAIR predictions were run as described above for SARS-CoV-2–human interactions. Evaluation of raw prediction probabilities was performed by AUROC in Python using scikit-learn and was compared against ECLAIR’s original test set containing 200 intraspecies interactions 44 . Precision and recall metrics were calculated based on ECLAIR’s binary definition for high-confidence versus non-interface predictions. To validate HADDOCK guided docking performance using our human–pathogen PDB benchmark, we compared performance with a raw HADDOCK docking protocol. Guided docking was performed as described for SARS-CoV-2–human interactions. No PDB protein chains from the human–pathogen benchmark were used during docking. For raw HADDOCK docking no experimental constraints (AIRs) were provided and the ranair and surfrest parameters in the run.cns were set to true. Using these parameters, each rigid dock generates one random AIR between one surface residue from each protein A and B, which is used to ensure that the two protein chains slide together during docking. Overall performance of protocols was evaluated based on precision and recall of the true interface (Fig. 2c ). Secondary evaluation was conducted based on RMSD in PyMol before refinement between the docked and co-crystal structures (Fig. 2d ). When multiple co-crystal structures were used to define the interfaces, the RMSD was reported as the average RMSD against all co-crystal structures. Compilation of sequence variation sets For analysis of genetic variation that may impact the viral–human interactome, two sets of mutations were compiled: (1) viral mutations and (2) human population variants. For viral mutations, we identified sequence divergences between SARS-CoV-1 and SARS-CoV-2 versions of each protein based on alignment. Representative sequences for 16 SARS-CoV-1 proteins were obtained from UniProt (Proteome ID UP000000354 ) 107 , 108 . Sequences for 29 SARS-CoV-2 proteins were reported previously 20 and based on GenBank accession code MN985325 (refs. 109 , 110 ). Notably, UniProt accession codes for the SARS-CoV-1 proteome report two sequences for the uncleaved ORF1a and ORF1a-b, which correspond to NSP1 through NSP16 in SARS-CoV-2. Sequence divergences were reported after pairwise Needleman Wench alignment 111 , 112 (using Blosum62 scoring matrix, gap open penalty of 10 and gap extension penalty of 0.5) between the corresponding protein sequences from each species. A total of 1,003 missense variants were detected among 23 SARS-CoV-2 proteins. No suitable alignment form a SARS-CoV-1 sequence was available for orf3b orf8 or orf10. Additionally orf7b, nsp3 and nsp16 were excluded because they were not involved in any viral–human interactions. The full list of SARS-CoV-2 mutations is reported in Supplementary Table 5 . We obtained human population variants for all 332 human proteins interacting with SARS-CoV-2 proteins from gnomAD 61 . We used gnomAD’s graphQL API to run programmatic queries to fetch all missense variants per gene. Details on performing gnomAD queries in this manner are available at . We used the Ensembl Variant Effect Predictor 113 to map gnomAD DNA-level single-nucleotide polymorphisms (SNPs) to equivalent protein-level UniProt annotations. After Variant Effect Predictor mapping, variants were parsed to ensure the reported reference amino acid and position agree with the UniProt sequence and roughly 4.4.6% of variants that did not match were dropped from our dataset because they could not reliably be mapped to UniProt coordinates. In total 127,528 human population variants were curated. The full list of human population variants from gnomAD is reported in Supplementary Table 4 . Log odds enrichment calculations To determine enrichment or depletion, ORs were calculated as described previously. 114 $$\mathrm {OR} = \frac{{a/c}}{{b/d}}$$ Where, a , b , c and d describe values in a contingency table between case and exposure criteria. For a particular application, where we are interested in the enrichment of viral mutations or human population variants (case, variant versus nonvariant) along predicted interaction interfaces (exposure, interface versus non-interface), we would have: $${a} = {\mathrm{number}}\,{\mathrm{of}}\,{\mathrm{variant}}\,{\mathrm{interface}}\,{\mathrm{residues}}$$ $${b} = {\mathrm{number}}\,{\mathrm{of}}\,{\mathrm{nonvariant}}\,{\mathrm{interface}}\,{\mathrm{residues}}$$ $${c} = {\mathrm {number}}\,{\mathrm{of}}\,{\mathrm{variant}}\,{\mathrm{noninterface}}\,{\mathrm{residues}}$$ $${d} = {\mathrm {number}}\,{\mathrm{of}}\,{\mathrm{nonvariant}}\,{\mathrm{noninterface}}\,{\mathrm{residues}}$$ Statistical tests for enrichment or depletion were performed by calculating the z -statistic and corresponding two-sided P value for the OR (unadjusted for multiple hypothesis testing). $$z = \frac{{\ln OR}}{{\sqrt {\frac{1}{a} + \frac{1}{b} + \frac{1}{c} + \frac{1}{d}} }}$$ All reported ORs were log 2 transformed to maintain interpretable symmetry between enriched and depleted values. To avoid arbitrary OR inflation or depletion from missing data, in all cases where the interface residues were predicted by molecular docking, the OR was altered to only account for positions that were included in the structural models used for docking. Curation of disease-associated variants To explore whether human proteins interacting with SARS-CoV-2 proteins were enriched for disease or trait-associated variants, three datasets were curated: HGMD 68 , ClinVar 69 and the NHGRI-EBI GWAS catalog 70 . Disease annotations for HGMD and ClinVar were downloaded directly from these resources and mapped to UniProt. To calculate enrichment of individual disease terms, we reconstructed the disease ontology from NCBI MedGen term relationships ( ) and propagated counts up through all parent nodes up to a singular root node. A meaningful subset of significantly enriched terms were reported using the most general term with no more significant ancestor term (Supplementary Table 7 , sheet 1). Raw enrichment values for all terms are also provided (Supplementary Table 7 , sheet 2). For curation of disease and trait associations from the NHGRI-EBI GWAS catalog ( ) 70 , lead SNPs ( P value <5 × 10 −8 ) for all diseases/traits were retrieved on 16 June 2020. Proxy SNPs in high linkage disequilibrium (LD) (parameters, R 2 > 0.8; pop, ‘ALL’) for individual lead SNPs were obtained through programmatic queries to the LDproxy API 115 , which used phase 3 haplotype data from the 1000 Genomes Project as reference for calculating pairwise metrics of LD. Both lead SNPs and proxy SNPs were filtered to retain only missense variants. In silico scanning mutagenesis and ΔΔG estimation To explore the importance of each SARS-CoV-2–human interface residue and the impact of all possible mutations along the interface, we performed in silico scanning mutagenesis. We used a setup provided by the PyRosetta documentation ( ) designed around an approach previously benchmarked to correctly identify nearly 80% of interface hotspot mutations 59 . For consistency, we replaced the PyRosetta implementation’s definition of interface residues (≤ 8.0 Å away from partner chain), with our definition described above. We encourage reference to the original well-documented demo for details, but in brief, we considered all interface residue positions and began by estimating the wild-type binding energy for the interaction. The complex state energy is calculated following a PackRotamersMover operation to optimize the side chains of residues within 8.0 Å of the interface residue to be mutated. The chains are separated 500.0 Å to eliminate any interchain energy contributions and energy for the unbound state is calculated the same way. The difference between these two values provides the binding energy for the wild-type (WT) structure. $${{{{\rm{\Delta}}}{{\rm{G}}}_{{\rm{WT}}}}} = {E}_{\mathrm {complex}} - {E}_{\mathrm {unbound}}$$ To estimate the binding energy for all 19 amino acid mutations possible at the given position, each mutation is made iteratively and the ΔG Mut is as above using the mutated structures. Finally, the change in binding energy from each mutation is the difference between these two binding energies. $${{\rm{\Delta}}}{{\rm{\Delta}}}{{\rm{G}}} = {\Delta}{G}_{Mut} - {\Delta}{G}_{WT}$$ The scoring function used for these calculations is as described previously 59 using the following weights: fa_atr = 0.44, fa_rep = 0.07, fa_sol = 1.0, hbond_bb_sc = 0.5, hbond_sc = 1.0. To account for stochasticity of the PackRotamersMover optimization between trials, all ΔΔG values are reported from an average of ten independent trials. To test whether a mutation had a significantly nonzero impact on binding energy, a two-sided z -test between the ten independent trials was performed. To account for average impact of other same amino acid mutations at other positions along the interface, each average ΔΔG was z -normalized relative to the rest of the interface and outliers were called at ≥1 × s.d. away from the mean. Mutations that passed both criteria were identified as interface binding affinity hotspots. No adjustments were made for multiple hypothesis corrections. Predicting ΔΔG from SARS-CoV-1 and SARS-CoV-2 divergences Estimates of the overall impact of the cumulative set of mutations between SARS-CoV-1 and SARS-CoV-2 were made based on the in silico mutagenesis framework modified to introduce multiple mutations at a time. We generated interaction models using the SARS-CoV-1 protein by applying all amino acid substitutions between the two viruses to initial docked models containing the SARS-CoV-2 protein. A minority of mutations that comprised insertions or deletions could not be modeled under this framework. The ΔΔG calculation here was identical to the single mutation ΔΔG described above, except that side-chain rotamer optimization involved all residues within 8.0 Å of any of the mutated residues. The ΔΔG values were calculated considering the SARS-CoV-1 as the wild-type such that a negative ΔΔG indicates that the interaction is more stable (lower binding energy) in the SARS-CoV-2 version of the interaction compared to the SARS-CoV-1 version of the interaction: $${{{\rm{\Delta}}}{{\rm{\Delta}}}{\rm{G}}} = {\Delta}{G}_{SARSCoV2} - {\Delta}{G}_{SARSCoV1}$$ To account for stochasticity between trials for these predictions (which notably had a larger impact likely due to the decreased constraints on rotamer optimization in these cases), this set of ΔΔG values was reported as an average of 50 trials. Outliers for overall binding affinity change from SARS-CoV-1 to SARS-CoV-2 were called based on similar criteria to the individual mutations, except the z score normalization was performed relative to all other interactions. Protein–ligand docking using smina To further prioritize 76 previously reported candidate drugs targeting human proteins in the SARS-CoV-2–human interactome 20 , we performed protein–ligand docking for, 30 interaction–drug pairs (involving 25 unique drugs) that were amenable to docking. For docking, we excluded any human protein targets whose structures were below 33% coverage. To prep for docking, 3D structures for all ligands were first generated using Open Babel 116 and the command: obabel -:”[SMILES_STRING]”--gen3d -opdb -O [OUT_FILE] -d Protein–ligand docking was executed using smina 87 with the following parameters. The autobox_ligand option was turned on and centered around the receptor PDB file with an autobox_add border size of 10 Å. To increase the number of independent stochastic sampling trajectories and increase the likelihood of identifying a global minimum, the exhaustiveness was set to 40 and the num_modes was set to retain the top 1,000 ranked models. To reduce real wall time, each docking process was run using five CPU cores (no impact on net CPU time). The final smina command used was as follows: smina -r [RECEPTOR] -l [LIGAND] --autobox_ligand [RECEPTOR] --autobox_add 10 -o [OUT_FILE] --exhaustiveness 40 –num_modes 1000 --cpu 5 --seed [SEED] Each protein–ligand docking command was repeated ten times (essentially the same as one trial with exhaustiveness set to 400) with a unique seed to saturate the ligand binding search space as thoroughly as possible. We note that a single run with exhaustiveness ranging from 30–50 is considered sufficient for most applications 87 . To retain candidate poses covering different low-energy binding sites, a final set of up to ten of the best-scoring poses with centers at least 1 Å away from one another was selected. Results described in this manuscript are reported based the top-ranked pose. Protein residues involved in drug binding sites were annotated using the same criteria used to define interface residues. The record type for all ligand atoms was first manually changed from HETATM to ATOM because NACCESS otherwise excluded ligand atoms from the solvent-accessible surface area calculations. Validation of smina docking to identify drug binding sites Past evaluation of smina shows competitive performance across numerous Community Structure-Activity Resources 87 , 88 . However, traditional docking evaluation tasks, focus on sampling and correctly scoring docked conformations within a single known binding site and may frequently restrict the docking space to a few angstroms bounding box around the known ligand conformation. The focus is on recovering precisely how a ligand orients within a binding site rather than identifying the binding site from the whole protein surface. Because this performance metric may not provide sufficient confidence in smina’s ability to identify a binding site from scratch (our application in this manuscript) we re-benchmarked smina’s performance using an established drug docking benchmark set containing 4,399 protein–ligand complexes representing 95 protein targets 88 . We defined true ligand binding site residues from the available crystal structure and evaluated the fraction correctly recovered by smina’s top-ranked dock across the full protein surface. Docking was performed as above and evaluated based on both re-docking (ligand docked back into the exact receptor structure it came from) and cross-docking (ligand docked into an alternate conformation of the receptor it came from) conditions. Because the conformation of the binding pocket from an alternate receptor may not perfectly accommodate the ligand, cross-docking is considered more difficult, but also more representative of real conditions when making new predictions. To provide a reference for whether smina selectively recovered the true binding site we calculated a baseline random expectation. Artificial binding sites were defined by selecting a single surface residue and its N nearest neighbors, where N is the number of binding site residues in the true binding site. The average recovery of the true binding site from all such artificial binding sites was used as the null expectation for each drug–target pair. Construction of plasmids for Y2H and co-IP Clones of all human proteins tested were picked from the hORFeome 8.1 library 117 . Clones for all SARS-CoV-1 and SARS-CoV-2 proteins tested were designed to match GenBank entries AY357076 and MN908947 , respectively. To construct plasmids for testing by Y2H, viral genes were PCR amplified and cloned into pDEST-AD and pDEST-DB vectors (for Y2H). For co-IP, Gateway LR reactions were used to transfer bait SARS-CoV-2 nsp1 protein into a pQXIP (ClonTech, 631516) vector modified to include a Gateway cassette featuring a carboxy-terminal 3× FLAG. Yeast two-hybrid screens Y2H experiments were carried out as previously described 76 , 81 , 118 to (1) confirm that SARS-CoV-2–human interactions previously detected by IP–mass spectrometry could be recapitulated in Y2H, (2) compare the occurrence of interactions using SARS-CoV-1 versus SARS-CoV-2 viral baits and (3) profile the disruption of SARS-CoV-2–human interactions by human population variants. In brief human and viral clones were transferred into Y2H vectors pDEST-AD and pDEST-DB by Gateway LR reactions then transformed into MAT a Y8800 and MAT α Y8930, respectively. For comparisons of interest, the viral–human interactions were screened in both orientations; namely viral DB-ORF MAT α transformants were mated against corresponding human AD-ORF MAT a transformants and vice versa. All DB-ORF yeast cultures were also mated against MAT a yeast transformed with an empty pDEST-AD vector to screen for autoactivators. Mated transformants were incubated overnight at 30 °C, before being plated onto selective Synthetic Complete agar medium lacking leucine and tryptophan (SC-Leu-Trp) to select for mated diploid yeast. After another overnight incubation at at 30 °C, diploid yeast were plated onto two sets of SC-Leu-Trp agar selection plates; one lacking histidine and supplemented with 1 mM of 3-amino-1,2,4-triazole (SC-Leu-Trp-His+3AT), the other lacking adenine (SC-Leu-Trp-Ade). After overnight incubation at 30 °C, plates were replica-cleaned and incubated again for 3 d at 30 °C for final interaction calling. Cell culture, co-immunoprecipitation and western blotting HEK 293T cells (ATCC, CRL-3216) were maintained in complete DMEM supplemented with 10% FBS. Cells were seeded onto six-well dishes and incubated until 70–80% confluency. Cells were then transfected with 1 µg of either empty vector, SARS-CoV-1 nsp1 or SARS-CoV-2 nsp1, respectively and combined with 10 µl of 1 mg ml−1 PEI (Polysciences, 23966) and 150 µl OptiMEM (Gibco, 31985-062). After 24 h incubation, cells were gently washed three times in 1× PBS and then resuspended in 200 µl cell lysis buffer (10 mM Tris-Cl, pH 8.0, 137 mM NaCl, 1% Triton X-100, 10% glycerol, 2 mM EDTA and 1× EDTA-free Complete Protease Inhibitor tablet (Roche)) and incubated on ice for 30 min. Extracts were cleared by centrifugation for 10 min at 16,000 g at 4 °C. For co-IP, 100 µl cell lysate per sample was incubated with 5 μl EZ view Red Anti-FLAG M2 Affinity Gel (Sigma, F2426) for 2 h at 4 °C under gentle rotation. After incubation, bound proteins were washed three times in cell lysis buffer and then eluted in 50 μl elution buffer (10 mM Tris-Cl pH 8.0, 1% SDS) at 65 °C for 10 min. Cell lysates and co-IP samples were then treated in 6× SDS protein loading buffer (10% SDS, 1 M Tris-Cl, pH 6.8, 50% glycerol, 10% β-mercaptoethanol and 0.03% bromophenol blue) and subjected to SDS–PAGE. Proteins were then transferred from gels onto PVDF (Amersham) membranes. Anti-FLAG (Sigma, F1804) and anti-PRIM2 (abcam, ab241990) at 1:3,000 dilutions were used for immunoblotting analysis. Cloning human population variants through site-directed mutagenesis Mutant clones containing human population variants were generated using site-directed mutagenesis as described previously 83 . In brief, wild-type G3BP2 was picked from the hORFeome 8.1 library 117 and used as a template for site-directed mutagenesis. Site-specific mutagenesis primers (Eurofins) for mutagenesis were designed using the webtool primer.yulab.org. To minimize sequencing artifacts, PCR was limited to 18 cycles using Phusion polymerase (NEB, M0530). PCR products were digested overnight with DpnI (NEB, R0176) then transformed into competent bacteria cells to isolate single colonies. To confirm successful mutagenesis single colonies were then hand-picked, incubated for 21 h at 37 °C under constant vibration and submitted for Sanger sequencing to ensure the desired single base-pair mutation (and no other mutations) had been introduced. Reporting Summary Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability Protein–protein interaction sets and drug–target candidate pairs were obtained from data reported previously 20 . Protein sequences were obtained from UniProt and GeneBank. Population variants were mined from gnomAD using their batch query API ( ). Disease and phenotypic variations were downloaded directly from HGMD, ClinVar and the NHGRI-EBI GWAS catalog. The data from these resources were modified and reformatted by some post-processing using custom Python scripts. Wherever restrictions on relevant data did not apply (for example, HGMD is restricted access) the formatted data are provided in the Supplementary Tables accompanying this manuscript or through the downloads page for our SARS-CoV-2–human interactome browser ( ). Homology modeling for SARS-CoV-2 proteins was performed through a custom script using Modeller (based on their multiple templates modeling example ). Protein structures either presented as raw structures or used as templates in homology modeling were obtained from the PDB. Homology models for human proteins were obtained from ModBase. Guided protein–protein docking and in silico mutagenesis were performed in HADDOCK and PyRosetta respectively using these structures. Relevant analysis summaries for these experiments are provided in the supplemental tables that accompany this manuscript. Raw.pdb files for the original undocked structures and for all docking trials are provided through our downloads page ( ). We believe that all data have been described; however, should any additional piece of data supporting the findings of this study later become of interest, the authors will strive to make it available upon request. Please address any additional requests or clarifications to S.D.W. ([email protected]) and H.Y. ([email protected]). Source data are provided with this paper. Code availability Scripts used for guided docking and relevant analyses are available on GitHub ( ) and Zenodo ( ). Source data are provided with this paper.

A computational tool allows researchers to precisely predict locations on the surfaces of human and COVID-19 viral proteins that bind with each other, a breakthrough that will greatly benefit our understanding of the virus and the development of drugs that block binding sites. The tool's developers have provided a user-friendly interactive web server that displays all of the protein structures, such that virologists and clinicians without computational skills can make use of the protein models to see if existing drugs, or those in development, fit into the proper binding sites. The study, published Nov. 29 in Nature Methods, describes the tool and uses it to predict how the SARS-COV-2 diverged structurally from SARS-COV-1 (which caused a SARS outbreak in 2002-04); how genetic variation of proteins in human populations may contribute to virus-human binding and higher risk of infection; and which existing drugs show promise for binding to targets on surfaces of human proteins. "Our computational tool allows you to see with unprecedented resolution where the viral proteins are binding on the human protein, and therefore, we can really understand what part of these proteins are key for these interactions," said Hiayuan Yu, the study's senior author and a professor in the Department of Biological Statistics and Computational Biology and the Weill Institute for Cell and Molecular Biology. Shayne Wierbowski, a graduate student in Yu's lab, is the paper's first author. A previous study by other scientists described interactions between COVID-19 and human proteins, for the purpose of repurposing human drugs to block the virus from binding. But binding interfaces are small compared to the overall protein's surface, and previous research has lacked the detailed resolution to understand exactly where drugs might block a binding site. "The tool we developed to predict protein-to-protein interfaces is the most accurate," Yu said, "and we can use it to make the most informed predictions for any interactions." The pandemic spurred a surge of research worldwide to understand the structure of SARS-COV-2, with scientists using advanced imaging technologies to reveal proteins that make the virus infectious. As a result, Yu and colleagues were able to validate their computationally predicted structures against those described by others using imaging technologies. The tool also allows researchers to predict how genetic variations in human proteins affect viral-protein interactions, as two people of similar health and age can have diverging responses to catching COVID-19, with some being asymptomatic and others showing dramatic negative reactions. "Because of our structural models, we can predict how mutations to proteins in individuals potentially affect viral interactions," Yu said. The results could one day shed light on whether some individuals may be at higher risk due to their genetics, which could prioritize them for monitoring, vaccines and treatments. Additionally, the tool will not only help clinicians develop drugs that precisely target human protein binding sites, it can also help reduce toxic or negative effects that could result when drugs bind to the wrong sites.

10.1038/s41592-021-01318-w

Computer

A system for automating robot design inspired by the evolution of vertebrates

Ryosuke Koike et al, Automatic robot design inspired by evolution of vertebrates, Artificial Life and Robotics (2022). DOI: 10.1007/s10015-022-00793-4

https://dx.doi.org/10.1007/s10015-022-00793-4

https://techxplore.com/news/2022-10-automating-robot-evolution-vertebrates.html

Abstract Devices","3":"Control, Robotics, Mechatronics"},"secondarySubjectCodes":{"1":"I21000","2":"I16013","3":"T19000"}},"sucode":"SC6"},"attributes":{"deliveryPlatform":"oscar"}},"Event Category":"Article"}]; window.dataLayer.push({ ga4MeasurementId: 'G-B3E4QL2TPR', ga360TrackingId: 'UA-26408784-1', twitterId: 'o47a7', ga4ServerUrl: ' imprint: 'springerlink', page: { attributes:{ featureFlags: [{ name: 'darwin-orion', active: false }], darwinAvailable: false } } }); window.initGTM = function() { if (window.config.mustardcut) { (function (w, d, s, l, i) { w[l] = w[l] || []; w[l].push({'gtm.start': new Date().getTime(), event: 'gtm.js'}); var f = d.getElementsByTagName(s)[0], j = d.createElement(s), dl = l != 'dataLayer' ? '&l=' + l : ''; j.async = true; j.src = ' + i + dl; f.parentNode.insertBefore(j, f); })(window, document, 'script', 'dataLayer', 'GTM-MRVXSHQ'); } } (function(w,d,t) { function cc() { var h = w.location.hostname; var e = d.createElement(t), s = d.getElementsByTagName(t)[0]; if (h.indexOf('springer.com') > -1) { e.src = ' e.setAttribute('onload', "initGTM(window,document,'script','dataLayer','GTM-MRVXSHQ')"); } else { e.src = '/static/js/lib/cookie-consent.min.js'; e.setAttribute('data-consent', h); } s.insertAdjacentElement('afterend', e); } cc(); })(window,document,'script'); (function(w, d) { w.config = w.config || {}; w.config.mustardcut = false; if (w.matchMedia && w.matchMedia('only print, only all and (prefers-color-scheme: no-preference), only all and (prefers-color-scheme: light), only all and (prefers-color-scheme: dark)').matches) { w.config.mustardcut = true; d.classList.add('js'); d.classList.remove('grade-c'); d.classList.remove('no-js'); } })(window, document.documentElement); (function () { if ( typeof window.CustomEvent === "function" ) return false; function CustomEvent ( event, params ) { params = params || { bubbles: false, cancelable: false, detail: null }; var evt = document.createEvent( 'CustomEvent' ); evt.initCustomEvent( event, params.bubbles, params.cancelable, params.detail ); return evt; } CustomEvent.prototype = window.Event.prototype; window.CustomEvent = CustomEvent; })(); if (window.config.mustardcut) { (function(w, d) { window.Component = {}; window.suppressShareButton = false; window.onArticlePage = true; var currentScript = d.currentScript || d.head.querySelector('script.js-entry'); function catchNoModuleSupport() { var scriptEl = d.createElement('script'); return (!('noModule' in scriptEl) && 'onbeforeload' in scriptEl) } var headScripts = [ {'src': '/oscar-static/js/polyfill-es5-bundle-51eb718839.js', 'async': false}, {'src': '/oscar-static/js/airbrake-es5-bundle-195baf30dc.js', 'async': false}, ]; var bodyScripts = [ {'src': '/oscar-static/js/app-es5-bundle-9e851d5bc0.js', 'async': false, 'module': false}, {'src': '/oscar-static/js/app-es6-bundle-208aa30589.js', 'async': false, 'module': true} , {'src': '/oscar-static/js/global-article-es5-bundle-e595d418d4.js', 'async': false, 'module': false}, {'src': '/oscar-static/js/global-article-es6-bundle-38f74de5cf.js', 'async': false, 'module': true} ]; function createScript(script) { var scriptEl = d.createElement('script'); scriptEl.src = script.src; scriptEl.async = script.async; if (script.module === true) { scriptEl.type = "module"; if (catchNoModuleSupport()) { scriptEl.src = ''; } } else if (script.module === false) { scriptEl.setAttribute('nomodule', true) } if (script.charset) { scriptEl.setAttribute('charset', script.charset); } return scriptEl; } for (var i = 0; i < headScripts.length; ++i) { var scriptEl = createScript(headScripts[i]); currentScript.parentNode.insertBefore(scriptEl, currentScript.nextSibling); } d.addEventListener('DOMContentLoaded', function() { for (var i = 0; i < bodyScripts.length; ++i) { var scriptEl = createScript(bodyScripts[i]); d.body.appendChild(scriptEl); } }); // Webfont repeat view var config = w.config; if (config && config.publisherBrand && sessionStorage.fontsLoaded === 'true') { d.documentElement.className += ' webfonts-loaded'; } })(window, document); } {"mainEntity":{"headline":"Automatic robot design inspired by evolution of vertebrates","description":"This paper proposes a novel method to design a robot by simultaneously improving its morphology and controller. The number of rigid parts of the robot and the layout of joints connecting them are represented by a rooted tree, which is called a discrete parameter. Meanwhile, parameters that can be represented by real values are called continuous parameters; these parameters include properties such as the length and the direction of each rigid part, as well as the weights and biases of the controller composed of a multilayer perceptron. For the discrete parameters, we propose an efficient improvement rule, which was established based on the actual evolution of vertebrates. For the continuous parameters, we apply the REINFORCE algorithm. By combining these two methods, we propose a method to simultaneously improve both the discrete and continuous parameters. The advantages of the proposed method are shown by comparison with other design strategies.","datePublished":"2022-09-09","dateModified":"2022-09-09","pageStart":"624","pageEnd":"631","sameAs":" Intelligence,Computation by Abstract Devices,Control,Robotics,Mechatronics","image":" Life and Robotics","issn":["1614-7456","1433-5298"],"volumeNumber":"27","@type":["Periodical","PublicationVolume"]},"publisher":{"name":"Springer Japan","logo":{"url":" Koike","affiliation":[{"name":"Kyoto University","address":{"name":"Department of Mechanical Engineering and Science, Graduate School of Engineering, Kyoto University, Kyoto, Japan","@type":"PostalAddress"},"@type":"Organization"}],"email":"[email protected]","@type":"Person"},{"name":"Ryo Ariizumi","affiliation":[{"name":"Nagoya University","address":{"name":"Department of Mechanical Systems Engineering, Graduate School of Engineering, Nagoya University, Nagoya, Japan","@type":"PostalAddress"},"@type":"Organization"}],"@type":"Person"},{"name":"Fumitoshi Matsuno","affiliation":[{"name":"Kyoto University","address":{"name":"Department of Mechanical Engineering and Science, Graduate School of Engineering, Kyoto University, Kyoto, Japan","@type":"PostalAddress"},"@type":"Organization"}],"@type":"Person"}],"isAccessibleForFree":false,"hasPart":{"isAccessibleForFree":false,"cssSelector":".main-content","@type":"WebPageElement"},"@type":"ScholarlyArticle"},"@context":" .cc-banner{-webkit-box-sizing:border-box;box-sizing:border-box;background-color:#01324b;color:#fff;position:fixed;bottom:0;left:0;right:0;line-height:1.5;z-index:99999}.cc-banner *{color:inherit!important}.cc-banner:focus{outline:none}.cc-banner a{color:#fff;text-decoration:underline}.cc-banner a:focus{outline:none;-webkit-box-shadow:0 0 0 3px #fece3e;box-shadow:0 0 0 3px #fece3e}.cc-banner a:active,.cc-banner a:focus,.cc-banner a:hover{text-decoration:none;color:inherit}.cc-banner__content{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-orient:vertical;-webkit-box-direction:normal;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;margin:0 auto;max-width:1320px;max-height:90vh;padding:12px}@media (min-width:680px){.cc-banner__content{padding:16px}}@media (min-width:980px){.cc-banner__content{padding-top:60px;padding-bottom:60px}}.cc-banner__title{font-size:18px;margin:0 0 12px}@media (min-width:980px){.cc-banner__title{margin:0 0 16px;font-size:28px}}.cc-banner__body{-webkit-box-flex:1;-webkit-flex:1 1 auto;-ms-flex:1 1 auto;flex:1 1 auto;overflow-x:hidden;overflow-y:auto;padding:0 0 16px;font-family:-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,Oxygen-Sans,Ubuntu,Cantarell,Helvetica Neue,sans-serif!important}@media (min-width:680px){.cc-banner__body{padding:0;margin:0 0 16px}}.cc-banner__copy{font-size:14px;margin:0}@media (min-width:980px){.cc-banner__copy{font-size:16px}}.cc-banner__footer{padding:12px;margin:0 -12px -12px;background-color:#012132;border-top:1px solid rgba(0,0,0,.3);-webkit-box-shadow:0 0 5px 0 rgba(0,0,0,.8);box-shadow:0 0 5px 0 rgba(0,0,0,.8);position:relative;font-family:-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,Oxygen-Sans,Ubuntu,Cantarell,Helvetica Neue,sans-serif!important}@media (min-width:380px){.cc-banner__footer{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-wrap:wrap;-ms-flex-wrap:wrap;flex-wrap:wrap;-webkit-box-align:center;-webkit-align-items:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:justify;-webkit-justify-content:space-between;-ms-flex-pack:justify;justify-content:space-between}}@media (min-width:680px){.cc-banner__footer{margin:0;padding:0;background-color:transparent;border-top:0;-webkit-box-shadow:none;box-shadow:none;-webkit-box-pack:start;-webkit-justify-content:flex-start;-ms-flex-pack:start;justify-content:flex-start}}.cc-banner__button{width:100%;white-space:nowrap}@media (min-width:380px){.cc-banner__button{width:auto}}@media (min-width:680px){.cc-banner__button{width:-webkit-fit-content;width:-moz-fit-content;width:fit-content}}.cc-banner__button:not(:last-child){margin-bottom:4px}@media (min-width:680px){.cc-banner__button:not(:last-child){margin-right:8px;margin-bottom:0}}@media (min-width:980px){.cc-banner__button:not(:last-child){margin-right:16px}}@media (min-width:380px){.cc-banner__button-accept,.cc-banner__button-reject{width:100%}}@media (min-width:380px) and (max-width:680px){.cc-banner__button-accept,.cc-banner__button-reject{width:100%}}@media (min-width:680px){.cc-banner__button-accept,.cc-banner__button-reject{width:-webkit-fit-content;width:-moz-fit-content;width:fit-content;min-width:240px}}.cc-banner__button-preferences{padding-left:0;padding-right:0}@media (min-width:380px){.cc-banner__button-preferences{-webkit-box-flex:0;-webkit-flex:0 0 auto;-ms-flex:0 0 auto;flex:0 0 auto;margin:auto}}@media (min-width:680px){.cc-banner__button-preferences{margin:0}}@media (min-width:380px) and (max-width:680px){.cc-banner__button-break{display:block}}.cc-banner--is-variant{line-height:1.3}.cc-banner--is-variant .cc-banner__content{padding:16px}@media (min-width:680px){.cc-banner--is-variant .cc-banner__content{padding:16px}}@media (min-width:980px){.cc-banner--is-variant .cc-banner__content{padding-top:20px;padding-bottom:20px}}@media (min-width:1320px){.cc-banner--is-variant .cc-banner__content{padding-top:40px;padding-bottom:40px}}.cc-banner--is-variant .cc-banner__title{font-weight:700!important;font-size:22px!important;margin:0 0 16px}@media (min-width:680px){.cc-banner--is-variant .cc-banner__title{font-size:24px!important;margin:0 0 20px}}@media (min-width:1320px){.cc-banner--is-variant .cc-banner__title{font-size:26px!important;margin:0 0 24px}}.cc-banner--is-variant .cc-banner__body{overflow-x:visible;overflow-y:visible;padding:0;margin:0 0 20px}@media (min-width:680px){.cc-banner--is-variant .cc-banner__body{padding:0;margin:0 0 24px}}@media (min-width:1320px){.cc-banner--is-variant .cc-banner__body{margin:0 0 30px}}.cc-banner--is-variant .cc-banner__copy{font-size:16px!important}.cc-banner--is-variant .cc-banner__copy:not(:last-child){margin:0 0 16px}@media (min-width:680px){.cc-banner--is-variant .cc-banner__copy:not(:last-child){margin:0 0 20px}}@media (min-width:1320px){.cc-banner--is-variant .cc-banner__copy:not(:last-child){margin:0 0 24px}}.cc-banner--is-variant .cc-banner__copy a{font-weight:700}.cc-banner--is-variant .cc-banner__footer{padding:0;margin:0;background-color:transparent;border-top:none;-webkit-box-shadow:none;box-shadow:none;position:static}.cc-banner--is-variant .cc-button{font-size:16px}.cc-banner--is-variant .cc-banner__button{padding:.5em 1em}.cc-banner--is-variant .cc-banner__button:not(:last-child){margin-bottom:8px}@media (min-width:680px){.cc-banner--is-variant .cc-banner__button:not(:last-child){margin-bottom:0;margin-right:16px}}.cc-button{font-size:13px;font-weight:700;padding:.5em 1em;color:#fff;border:1px solid #183642;background-color:#000;border-radius:3px;cursor:pointer;line-height:1.2}.cc-button:focus{outline:none;-webkit-box-shadow:0 0 0 3px #fece3e;box-shadow:0 0 0 3px #fece3e}.cc-button:hover{-webkit-box-shadow:0 -2px 4px 0 rgba(60,64,67,.15),0 2px 4px 0 rgba(60,64,67,.15);box-shadow:0 -2px 4px 0 rgba(60,64,67,.15),0 2px 4px 0 rgba(60,64,67,.15);text-decoration:underline}@media (min-width:680px){.cc-button{padding:.75em 1em;font-size:14px}}.cc-button--contrast{color:#000!important;border-color:#fff;background-color:#fff}.cc-button--outline{background-color:transparent;color:#000!important}.cc-button--contrast.cc-button--outline{color:#fff!important}.cc-button--link{color:inherit;text-decoration:underline;background-color:transparent;border-color:transparent}.cc-button--link:hover{text-decoration:none;-webkit-box-shadow:none;box-shadow:none}.cc-button--text{padding:0}@media (max-width:380px){.cc-button__truncated{border:0;clip:rect(0,0,0,0);height:1px;margin:-100%;overflow:hidden;padding:0;position:absolute!important;width:1px}}.cc-radio{position:relative}.cc-radio *{cursor:pointer}.cc-radio__input{width:22px;height:22px;position:absolute;top:0;left:0}.cc-radio__input:focus{outline:none}.cc-radio__label{padding-left:30px;font-size:14px;line-height:22px;margin:0;color:inherit}.cc-radio__label:after,.cc-radio__label:before{position:absolute;content:"";display:block;background-color:#fff}.cc-radio__label:before{width:22px;height:22px;top:0;left:0;border:2px solid;border-radius:50%}.cc-radio__label:after{top:6px;left:6px;width:0;height:0;border:5px solid;border-radius:50%;opacity:0}.cc-radio__input:focus+.cc-radio__label:before{outline:none;-webkit-box-shadow:0 0 0 3px #fece3e;box-shadow:0 0 0 3px #fece3e}.cc-radio__label--hidden{display:none}.cc-radio__input:checked+.cc-radio__label:after{opacity:1}.cc-radio__input:disabled{cursor:default}.cc-radio__input:disabled+.cc-radio__label{opacity:.5;cursor:default}.cc-preferences{-webkit-box-sizing:border-box;box-sizing:border-box;color:#111;font-family:sans-serif;overflow:auto;z-index:100000;position:fixed;background-color:rgba(5,10,20,.95);line-height:1.4;top:0;left:0;right:0;bottom:0}.cc-preferences:focus{outline:none}.cc-preferences *,.cc-preferences :after,.cc-preferences :before{-webkit-box-sizing:inherit!important;box-sizing:inherit!important}.cc-preferences h1,.cc-preferences h2,.cc-preferences h3,.cc-preferences h4,.cc-preferences h5,.cc-preferences h6{font-family:sans-serif;font-style:normal}.cc-preferences a{color:inherit;text-decoration:underline}.cc-preferences a:hover{text-decoration:none;color:inherit}.cc-preferences input{margin:0}.cc-preferences__dialog{position:relative;display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-orient:vertical;-webkit-box-direction:normal;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;padding:16px;margin:auto;max-width:960px;max-height:100vh}.cc-preferences__dialog>:last-child{border-bottom-left-radius:5px;border-bottom-right-radius:5px}.cc-preferences__close{position:absolute;border-radius:3px;top:24px;right:24px;padding:0;width:40px;height:40px;border:0;font-size:40px;line-height:1;cursor:pointer;background:transparent;font-family:Times New Roman,serif}.cc-preferences__close:focus{outline:none;-webkit-box-shadow:0 0 0 3px #fece3e;box-shadow:0 0 0 3px #fece3e}.cc-preferences__close-label{border:0;clip:rect(0,0,0,0);height:1px;margin:-100%;overflow:hidden;padding:0;position:absolute!important;width:1px}.cc-preferences__header{background:#fff;text-align:center;padding:16px;border-bottom:1px solid #d0d0d0;border-top-left-radius:5px;border-top-right-radius:5px}.cc-preferences__title{font-family:sans-serif;font-size:18px;font-weight:700;padding-right:24px;margin:0;color:#111}@media (min-width:480px){.cc-preferences__title{padding-right:0}}@media (min-width:980px){.cc-preferences__title{font-size:22px}}.cc-preferences__body{padding:16px;-webkit-box-flex:1;-webkit-flex:1 1 auto;-ms-flex:1 1 auto;flex:1 1 auto;min-height:200px;overflow-x:hidden;overflow-y:auto;background:#fff}.cc-preferences__footer{background:#fff;margin-bottom:0;padding:16px;border-top:1px solid #d0d0d0;-webkit-box-shadow:0 0 5px 0 rgba(0,0,0,.2);box-shadow:0 0 5px 0 rgba(0,0,0,.2);position:relative}@media (min-width:480px){.cc-preferences__footer{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-orient:horizontal;-webkit-box-direction:normal;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;-webkit-box-align:center;-webkit-align-items:center;-ms-flex-align:center;align-items:center}}.cc-preferences__footer>.cc-button{display:block;width:100%}@media (min-width:480px){.cc-preferences__footer>.cc-button{-webkit-box-flex:1;-webkit-flex:1 1 auto;-ms-flex:1 1 auto;flex:1 1 auto;width:-webkit-fit-content;width:-moz-fit-content;width:fit-content}}@media (min-width:980px){.cc-preferences__footer>.cc-button{width:-webkit-fit-content;width:-moz-fit-content;width:fit-content;-webkit-flex-basis:auto;-ms-flex-preferred-size:auto;flex-basis:auto}}@media (min-width:480px){.cc-preferences__footer>.cc-button:not(:first-child){margin-left:16px;margin-right:16px}}.cc-preferences__footer>.cc-button:not(:last-child){margin-bottom:8px}@media (min-width:480px){.cc-preferences__footer>.cc-button:not(:last-child){margin-bottom:0}}@media (min-width:980px){.cc-preferences__footer>.cc-button:last-child{margin-left:auto}}.cc-preferences__categories{list-style:none;padding:0;margin:0}.cc-preferences__category:not(:last-child){margin-bottom:16px;padding-bottom:16px;border-bottom:1px solid #ececec}.cc-preferences__category input{margin-right:10px}.cc-preferences__category-header{margin:0 0 8px;display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-pack:justify;-webkit-justify-content:space-between;-ms-flex-pack:justify;justify-content:space-between;-webkit-box-align:baseline;-webkit-align-items:baseline;-ms-flex-align:baseline;align-items:baseline}.cc-preferences__category-heading{margin:0;font-size:16px;font-weight:700;display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-align:baseline;-webkit-align-items:baseline;-ms-flex-align:baseline;align-items:baseline}.cc-preferences__category-description{margin:0 0 8px;font-size:14px}.cc-preferences__category-footer{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-box-pack:justify;-webkit-justify-content:space-between;-ms-flex-pack:justify;justify-content:space-between;-webkit-box-align:baseline;-webkit-align-items:baseline;-ms-flex-align:baseline;align-items:baseline}.cc-preferences__status{border:0;clip:rect(0,0,0,0);height:1px;margin:-100%;overflow:hidden;padding:0;position:absolute!important;width:1px}.cc-preferences__controls{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;margin:0}.cc-preferences__controls>:not(:last-child){margin-right:16px}.cc-preferences__always-on{font-size:14px;position:relative;padding-left:26px}.cc-preferences__always-on:before{content:"";display:block;position:absolute;left:0;top:2px;width:18px;height:10px;-webkit-transform:rotate(-45deg);transform:rotate(-45deg);border:solid;border-width:0 0 4px 4px;border-top-color:transparent;background:transparent}.cc-preferences__details-button{position:relative;font-size:14px;padding:2px 4px;font-weight:400;border:1px solid transparent;background-color:#fff;color:#000}.cc-preferences__details-button--show{background-color:#000;border-color:#000;color:#fff}.cc-preferences__details-button--show:before{content:"";display:block;position:absolute;right:50%;top:100%;width:0;height:0;margin-right:-8px;border-left:8px solid transparent;border-right:8px solid transparent;border-top:8px solid #111}.cc-preferences__details{font-size:14px;display:none;padding:12px 0 0;border-top:2px solid #000;margin:12px 0 0}.cc-preferences__details--show{display:block}.cc-preferences__cookie-list,.cc-preferences__provider-list{list-style:none;margin:0;padding:0}.cc-preferences__provider-list{-webkit-columns:170px;-moz-columns:170px;columns:170px}.cc-preferences__cookie-title{font-size:1em;margin:0}.cc-preferences__cookie-description{font-size:1em;margin:0 0 8px}.cc-preferences__cookie-domain,.cc-preferences__cookie-lifespan{color:#666;padding-left:4px;margin-left:8px;border-left:1px solid #999}body.cc-has-preferences-open{overflow:hidden;position:relative} .MathJax_Hover_Frame {border-radius: .25em; -webkit-border-radius: .25em; -moz-border-radius: .25em; -khtml-border-radius: .25em; box-shadow: 0px 0px 15px #83A; -webkit-box-shadow: 0px 0px 15px #83A; -moz-box-shadow: 0px 0px 15px #83A; -khtml-box-shadow: 0px 0px 15px #83A; border: 1px solid #A6D ! important; display: inline-block; position: absolute} .MathJax_Menu_Button .MathJax_Hover_Arrow {position: absolute; cursor: pointer; display: inline-block; border: 2px solid #AAA; border-radius: 4px; -webkit-border-radius: 4px; -moz-border-radius: 4px; -khtml-border-radius: 4px; font-family: 'Courier New',Courier; font-size: 9px; color: #F0F0F0} .MathJax_Menu_Button .MathJax_Hover_Arrow span {display: block; background-color: #AAA; border: 1px solid; border-radius: 3px; line-height: 0; padding: 4px} .MathJax_Hover_Arrow:hover {color: white!important; border: 2px solid #CCC!important} .MathJax_Hover_Arrow:hover span {background-color: #CCC!important} #MathJax_About {position: fixed; left: 50%; width: auto; text-align: center; border: 3px outset; padding: 1em 2em; background-color: #DDDDDD; color: black; cursor: default; font-family: message-box; font-size: 120%; font-style: normal; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 15px; -webkit-border-radius: 15px; -moz-border-radius: 15px; -khtml-border-radius: 15px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} #MathJax_About.MathJax_MousePost {outline: none} .MathJax_Menu {position: absolute; background-color: white; color: black; width: auto; padding: 5px 0px; border: 1px solid #CCCCCC; margin: 0; cursor: default; font: menu; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 5px; -webkit-border-radius: 5px; -moz-border-radius: 5px; -khtml-border-radius: 5px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} .MathJax_MenuItem {padding: 1px 2em; background: transparent} .MathJax_MenuArrow {position: absolute; right: .5em; padding-top: .25em; color: #666666; font-size: .75em} .MathJax_MenuActive .MathJax_MenuArrow {color: white} .MathJax_MenuArrow.RTL {left: .5em; right: auto} .MathJax_MenuCheck {position: absolute; left: .7em} .MathJax_MenuCheck.RTL {right: .7em; left: auto} .MathJax_MenuRadioCheck {position: absolute; left: .7em} .MathJax_MenuRadioCheck.RTL {right: .7em; left: auto} .MathJax_MenuLabel {padding: 1px 2em 3px 1.33em; font-style: italic} .MathJax_MenuRule {border-top: 1px solid #DDDDDD; margin: 4px 3px} .MathJax_MenuDisabled {color: GrayText} .MathJax_MenuActive {background-color: #606872; color: white} .MathJax_MenuDisabled:focus, .MathJax_MenuLabel:focus {background-color: #E8E8E8} .MathJax_ContextMenu:focus {outline: none} .MathJax_ContextMenu .MathJax_MenuItem:focus {outline: none} #MathJax_AboutClose {top: .2em; right: .2em} .MathJax_Menu .MathJax_MenuClose {top: -10px; left: -10px} .MathJax_MenuClose {position: absolute; cursor: pointer; display: inline-block; border: 2px solid #AAA; border-radius: 18px; -webkit-border-radius: 18px; -moz-border-radius: 18px; -khtml-border-radius: 18px; font-family: 'Courier New',Courier; font-size: 24px; color: #F0F0F0} .MathJax_MenuClose span {display: block; background-color: #AAA; border: 1.5px solid; border-radius: 18px; -webkit-border-radius: 18px; -moz-border-radius: 18px; -khtml-border-radius: 18px; line-height: 0; padding: 8px 0 6px} .MathJax_MenuClose:hover {color: white!important; border: 2px solid #CCC!important} .MathJax_MenuClose:hover span {background-color: #CCC!important} .MathJax_MenuClose:hover:focus {outline: none} .MJX_Assistive_MathML {position: absolute!important; top: 0; left: 0; clip: rect(1px, 1px, 1px, 1px); padding: 1px 0 0 0!important; border: 0!important; height: 1px!important; width: 1px!important; overflow: hidden!important; display: block!important; -webkit-touch-callout: none; -webkit-user-select: none; -khtml-user-select: none; -moz-user-select: none; -ms-user-select: none; user-select: none} .MJX_Assistive_MathML.MJX_Assistive_MathML_Block {width: 100%!important} #MathJax_Zoom {position: absolute; background-color: #F0F0F0; overflow: auto; display: block; z-index: 301; padding: .5em; border: 1px solid black; margin: 0; font-weight: normal; font-style: normal; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; -webkit-box-sizing: content-box; -moz-box-sizing: content-box; box-sizing: content-box; box-shadow: 5px 5px 15px #AAAAAA; -webkit-box-shadow: 5px 5px 15px #AAAAAA; -moz-box-shadow: 5px 5px 15px #AAAAAA; -khtml-box-shadow: 5px 5px 15px #AAAAAA; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')} #MathJax_ZoomOverlay {position: absolute; left: 0; top: 0; z-index: 300; display: inline-block; width: 100%; height: 100%; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)} #MathJax_ZoomFrame {position: relative; display: inline-block; height: 0; width: 0} #MathJax_ZoomEventTrap {position: absolute; left: 0; top: 0; z-index: 302; display: inline-block; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)} .MathJax_Preview {color: #888} #MathJax_Message {position: fixed; left: 1px; bottom: 2px; background-color: #E6E6E6; border: 1px solid #959595; margin: 0px; padding: 2px 8px; z-index: 102; color: black; font-size: 80%; width: auto; white-space: nowrap} #MathJax_MSIE_Frame {position: absolute; top: 0; left: 0; width: 0px; z-index: 101; border: 0px; margin: 0px; padding: 0px} .MathJax_Error {color: #CC0000; font-style: italic} .MJXp-script {font-size: .8em} .MJXp-right {-webkit-transform-origin: right; -moz-transform-origin: right; -ms-transform-origin: right; -o-transform-origin: right; transform-origin: right} .MJXp-bold {font-weight: bold} .MJXp-italic {font-style: italic} .MJXp-scr {font-family: MathJax_Script,'Times New Roman',Times,STIXGeneral,serif} .MJXp-frak {font-family: MathJax_Fraktur,'Times New Roman',Times,STIXGeneral,serif} .MJXp-sf {font-family: MathJax_SansSerif,'Times New Roman',Times,STIXGeneral,serif} .MJXp-cal {font-family: MathJax_Caligraphic,'Times New Roman',Times,STIXGeneral,serif} .MJXp-mono {font-family: MathJax_Typewriter,'Times New Roman',Times,STIXGeneral,serif} .MJXp-largeop {font-size: 150%} .MJXp-largeop.MJXp-int {vertical-align: -.2em} .MJXp-math {display: inline-block; line-height: 1.2; text-indent: 0; font-family: 'Times New Roman',Times,STIXGeneral,serif; white-space: nowrap; border-collapse: collapse} .MJXp-display {display: block; text-align: center; margin: 1em 0} .MJXp-math span {display: inline-block} .MJXp-box {display: block!important; text-align: center} .MJXp-box:after {content: " "} .MJXp-rule {display: block!important; margin-top: .1em} .MJXp-char {display: block!important} .MJXp-mo {margin: 0 .15em} .MJXp-mfrac {margin: 0 .125em; vertical-align: .25em} .MJXp-denom {display: inline-table!important; width: 100%} .MJXp-denom > * {display: table-row!important} .MJXp-surd {vertical-align: top} .MJXp-surd > * {display: block!important} .MJXp-script-box > * {display: table!important; height: 50%} .MJXp-script-box > * > * {display: table-cell!important; vertical-align: top} .MJXp-script-box > *:last-child > * {vertical-align: bottom} .MJXp-script-box > * > * > * {display: block!important} .MJXp-mphantom {visibility: hidden} .MJXp-munderover, .MJXp-munder {display: inline-table!important} .MJXp-over {display: inline-block!important; text-align: center} .MJXp-over > * {display: block!important} .MJXp-munderover > *, .MJXp-munder > * {display: table-row!important} .MJXp-mtable {vertical-align: .25em; margin: 0 .125em} .MJXp-mtable > * {display: inline-table!important; vertical-align: middle} .MJXp-mtr {display: table-row!important} .MJXp-mtd {display: table-cell!important; text-align: center; padding: .5em 0 0 .5em} .MJXp-mtr > .MJXp-mtd:first-child {padding-left: 0} .MJXp-mtr:first-child > .MJXp-mtd {padding-top: 0} .MJXp-mlabeledtr {display: table-row!important} .MJXp-mlabeledtr > .MJXp-mtd:first-child {padding-left: 0} .MJXp-mlabeledtr:first-child > .MJXp-mtd {padding-top: 0} .MJXp-merror {background-color: #FFFF88; color: #CC0000; border: 1px solid #CC0000; padding: 1px 3px; font-style: normal; font-size: 90%} .MJXp-scale0 {-webkit-transform: scaleX(.0); -moz-transform: scaleX(.0); -ms-transform: scaleX(.0); -o-transform: scaleX(.0); transform: scaleX(.0)} .MJXp-scale1 {-webkit-transform: scaleX(.1); -moz-transform: scaleX(.1); -ms-transform: scaleX(.1); -o-transform: scaleX(.1); transform: scaleX(.1)} .MJXp-scale2 {-webkit-transform: scaleX(.2); -moz-transform: scaleX(.2); -ms-transform: scaleX(.2); -o-transform: scaleX(.2); transform: scaleX(.2)} .MJXp-scale3 {-webkit-transform: scaleX(.3); -moz-transform: scaleX(.3); -ms-transform: scaleX(.3); -o-transform: scaleX(.3); transform: scaleX(.3)} .MJXp-scale4 {-webkit-transform: scaleX(.4); -moz-transform: scaleX(.4); -ms-transform: scaleX(.4); -o-transform: scaleX(.4); transform: scaleX(.4)} .MJXp-scale5 {-webkit-transform: scaleX(.5); -moz-transform: scaleX(.5); -ms-transform: scaleX(.5); -o-transform: scaleX(.5); transform: scaleX(.5)} .MJXp-scale6 {-webkit-transform: scaleX(.6); -moz-transform: scaleX(.6); -ms-transform: scaleX(.6); -o-transform: scaleX(.6); transform: scaleX(.6)} .MJXp-scale7 {-webkit-transform: scaleX(.7); -moz-transform: scaleX(.7); -ms-transform: scaleX(.7); -o-transform: scaleX(.7); transform: scaleX(.7)} .MJXp-scale8 {-webkit-transform: scaleX(.8); -moz-transform: scaleX(.8); -ms-transform: scaleX(.8); -o-transform: scaleX(.8); transform: scaleX(.8)} .MJXp-scale9 {-webkit-transform: scaleX(.9); -moz-transform: scaleX(.9); -ms-transform: scaleX(.9); -o-transform: scaleX(.9); transform: scaleX(.9)} .MathJax_PHTML .noError {vertical-align: ; font-size: 90%; text-align: left; color: black; padding: 1px 3px; border: 1px solid} Your Privacy We use cookies to make sure that our website works properly, as well as some ‘optional’ cookies to personalise content and advertising, provide social media features and analyse how people use our site. By accepting some or all optional cookies you give consent to the processing of your personal data, including transfer to third parties, some in countries outside of the European Economic Area that do not offer the same data protection standards as the country where you live. You can decide which optional cookies to accept by clicking on ‘Manage Settings’, where you can also find more information about how your personal data is processed. Further information can be found in our privacy policy . Accept all cookies Manage preferences Google Tag Manager (noscript) End Google Tag Manager (noscript) Skip to main content Advertisement Log in Menu Find a journal Publish with us Search Cart (function () { var exports = {}; if (window.fetch) { "use strict"; Object.defineProperty(exports, "__esModule", { value: true }); exports.headerWidgetClientInit = void 0; var headerWidgetClientInit = function (getCartInfo) { console.log("listen to updatedCart event"); document.body.addEventListener("updatedCart", function () { console.log("updatedCart happened"); updateCartIcon().then(function () { return console.log("Cart state update upon event"); }); }, false); return updateCartIcon().then(function () { return console.log("Initial cart state update"); }); function updateCartIcon() { return getCartInfo() .then(function (res) { return res.json(); }) .then(refreshCartState) .catch(function () { return console.log("Could not fetch cart info"); }); } function refreshCartState(json) { var indicator = document.querySelector("#ecommerce-header-cart-icon-link .cart-info"); /* istanbul ignore else */ if (indicator && json.itemCount) { indicator.style.display = 'block'; indicator.textContent = json.itemCount > 9 ? '9+' : json.itemCount.toString(); var moreThanOneItem = json.itemCount > 1; indicator.setAttribute('title', "there ".concat(moreThanOneItem ? "are" : "is", " ").concat(json.itemCount, " item").concat(moreThanOneItem ? "s" : "", " in your cart")); } return json; } }; exports.headerWidgetClientInit = headerWidgetClientInit; headerWidgetClientInit( function () { return window.fetch(" { credentials: "include", headers: { Accept: "application/json" } }) } ) }})() Search Search by keyword or author Search Navigation Find a journal Publish with us Home Artificial Life and Robotics Article Automatic robot design inspired by evolution of vertebrates Original Article Published: 09 September 2022 Automatic robot design inspired by evolution of vertebrates Ryosuke Koike 1 , Ryo Ariizumi 2 & Fumitoshi Matsuno 1 Artificial Life and Robotics volume 27 , pages 624–631 ( 2022 ) Cite this article 310 Accesses 1 Citations 55 Altmetric Metrics details Abstract This paper proposes a novel method to design a robot by simultaneously improving its morphology and controller. The number of rigid parts of the robot and the layout of joints connecting them are represented by a rooted tree, which is called a discrete parameter. Meanwhile, parameters that can be represented by real values are called continuous parameters; these parameters include properties such as the length and the direction of each rigid part, as well as the weights and biases of the controller composed of a multilayer perceptron. For the discrete parameters, we propose an efficient improvement rule, which was established based on the actual evolution of vertebrates. For the continuous parameters, we apply the REINFORCE algorithm. By combining these two methods, we propose a method to simultaneously improve both the discrete and continuous parameters. The advantages of the proposed method are shown by comparison with other design strategies. Access provided by MPDL Services gGmbH c/o Max Planck Digital Library Working on a manuscript? Avoid the common mistakes 1 Introduction When we build a robot, we typically first design its morphology, and then design a controller to calculate the appropriate input signals for the specific morphology. High performance can only be achieved by a well-designed morphology with a finely tuned controller. Nevertheless, most improvement methods for robots focus only on improving the controller design, while the morphological design is given by humans. With this approach, even if the global optimum of the controller design is found, the solution is a local one in the sense that it is only for a given robot morphology. Therefore, by simultaneously improving the morphology and controller designs, we can expect to achieve higher performance than has been possible with conventional methods. The most challenging issue in achieving simultaneous improvement is that the design parameters consist of both discrete and continuous parameters. Most of the popular controllers can be tuned by adjusting real-valued parameters such as gains. Several morphological properties can also be expressed by real values. For example, the length of rigid parts and the positions and the directions of joints can be expressed by real values. We call these real values continuous parameters. However, some properties have more appropriate representation than real values. Typical examples are the number of rigid parts and the layout of the joints connecting them, which can be represented by a rooted tree. We call this rooted tree a discrete parameter. Reinforcement learning and many other improvement methods are suitable for continuous parameters, but not for a discrete parameter of this type, so it is not easy to apply them to robot design problems. Furthermore, it is difficult to evaluate a specific choice of discrete parameters for a given task because the behavior of the robot cannot be determined independently from continuous parameters. Therefore, various typical theories for the discrete improvement problem [ 1 ] are not applicable, and a new method that can overcome these difficulties is needed. Several prior studies have challenged the problem of improving the morphology and controller design. An earlier work [ 2 ] on this type of problem uses evolutionary strategies to simultaneously improve both discrete and continuous parameters. When using evolutionary strategies, simple controllers with a small number of tunable parameters are used to simplify the problem. However, this makes it difficult to achieve a complex controller [ 2 ]. In [ 3 ], the continuous parameters of both morphology and controller are designed using the REINFORCE algorithm [ 4 ]. Their method achieves high performance that is not possible with the morphology designed by humans, i.e., Ant-v2 [ 5 ], which is a simulated four-legged robot often used in AI research, but it does not handle the discrete parameters. Some authors [ 6 , 7 ] succeeded in improving the morphology and controller design by restricting the robot morphology to those composed of predefined parts. They employed model predictive control to assure the optimality of the controller. However, the controller relies on a perfect physical model, including factors with low reproducibility such as collision detection and random noise. Recent research [ 8 ] has shown the efficacy of using evolutionary algorithms for improving the morphology design and Proximal Policy Optimization (PPO) algorithm for controller design improvement. By this combination, high performance in various environments can be achieved. However, many different simulations must be run in parallel, which requires very powerful computers. In this study, we propose a new method to improve the robot’s morphology and its controller designs simultaneously. By this approach, we aim to obtain a better performing robot along with an appropriate controller than a human-designed robot with a well-tuned controller. The morphology and controller designs of a robot are represented by discrete and continuous parameters. In the actual evolution of vertebrates, a particularly large change in skeletal structure was the evolution from fins to limbs, where the number of bones decreased [ 9 ]. The vertebrates that acquired limbs adapted to various environments without significantly increasing the number of their bones, and evolved into humans, seals, bats, and so on [ 10 ]. By referring to this example of skeletal change, we adopt the strategy of changing the discrete parameters only to decrease the number of rigid parts. For continuous parameters, we apply the REINFORCE algorithm. We perform simulation experiments that aim to design a legged robot that can move on a flat surface as fast as and as energy-efficiently as possible. Simulation experiments show that the proposed method successfully improves both the discrete and the continuous parameters simultaneously. Simulation experiments also suggest that performance drops temporarily when discrete parameters are changed, but performance is quickly recovered thanks to the transfer learning-like effect of improving continuous parameters. Moreover, the advantages of the proposed method are shown by comparison with other design strategies. In this study, \({{\mathbb {R}}}\) denotes the set of all real values. The function \(\mathrm{Sigmoid}:{{\mathbb {R}}}\rightarrow (0,1)\) is defined by $$\begin{aligned} \mathrm{sigmoid}(x)=\frac{1}{1+e^{-x}}. \end{aligned}$$ (1) We define \(E[X|\alpha ]\) as the expected value of the random variable X under the assumption of condition \(\alpha\) . For vector \(\mu\) and matrix \(\varSigma\) , let \({{\mathcal {N}}}(\mu ,\ \varSigma )\) denote the multivariate normal distribution. For a vector x , use \(\mathrm {diag}\ x\) to denote the diagonal matrix with each element of x ordered from the top left. 2 Problem settings 2.1 Morphology of a robot This paper deals with robots that move in 3D space (Fig. 1 a). Each robot has only one sphere-shaped part with a fixed size, which we call the torso. The other rigid parts should be long and slender to some extent because they are expected to serve mainly as linkages. Therefore, all rigid parts except the torso are assumed to be capsule-shaped, i.e., cylinders with spherical ends. For the convenience of simulation, all these rigid parts are assumed to pass through each other without colliding. Meanwhile, the joint has one degree of freedom and can rotate within a limited angle. Each joint takes a torque input at each step. Fig. 1 An example of a morphology of a robot. ( a ) The visual appearance of a robot in a physical simulation. ( b ) The discrete parameter \({{\mathcal {G}}}\) of the robot shown in ( a ) Full size image 2.2 Controller of a robot The controller of the robot is composed of a multilayer perceptron (MLP). Inputs to the MLP are state variables, which are set depending on the task. The candidates of the inputs will include the angles and angular velocities of the joints, robot posture, and velocity. The output from the MLP is the input to each joint, i.e., the power given to the motors. The number of nodes in the middle layer of the MLP is fixed, while the numbers of nodes in the input and the output layers are variable. 2.3 Discrete parameter We represent the robot topology design by an undirected graph, in particular a rooted tree. An example is given in Fig. 1 : the rooted tree in Fig. 1 b shows the discrete parameter of the robot shown in Fig. 1 a. Each node in the tree corresponds to a rigid part and each edge in the tree corresponds to a joint between the rigid parts. The torso corresponds to the root node. Let the node set, the edge set, and the tree be denoted as \({{\mathcal {V}}}\) , \({{\mathcal {E}}}\) , and \({{\mathcal {G}}}=\{{{\mathcal {V}}},{{\mathcal {E}}}\}\) , respectively. 2.4 Continuous parameters There are two types of real-valued parameters. The first type is the parameters of the controller, i.e., the weights and biases of the MLP introduced in Sect. 2.2 . We refer to them as continuous control parameters. Let \(\phi \in {{\mathbb {R}}}^{N_\phi }\) be a flattened vector of continuous control parameters, where \(N_\phi\) is the number of them. The second type is the properties of robot items. We consider the length and the direction of rigid parts, the position of joints, and the direction of rotation as real-valued properties. We prepare a real-valued vector \(\psi \in {{\mathbb {R}}}^{N_\psi }\) to describe these real-valued properties, where \(N_\psi\) is the number of the properties. For \(\psi _i\) , we introduce fixed constants \(c_i,\ d_i\in {{\mathbb {R}}}\) to calculate \(x_i = c_i+d_i\mathrm{sigmoid}(\psi _i)\) . Then, \(\psi _i\) is regarded as the parameter to be learned and \(x_i\in (c_i, c_i+d_i)\) is regarded as the real-valued property. This pre-processing is introduced to restrict the range of the properties. We refer to the elements of the real-valued vector \(\psi \in {{\mathbb {R}}}^{N_\psi }\) as continuous morphological parameters. The continuous control parameters and continuous morphological parameters are combined into a single real-valued vector \(\theta =[\phi ^{{\mathsf {T}}}, \psi ^{{\mathsf {T}}}]^{{\mathsf {T}}}\in {\mathbb R}^N\) , where \(N=N_\phi +N_\psi\) . The elements of \(\theta\) are called continuous parameters. 2.5 Objective The robot design problem can be formulated as an improvement problem. Once a set of discrete and continuous parameters \(({{\mathcal {G}}},\ \theta )\) is selected, the set can be evaluated via physics simulations. In each step of the simulations, the input signal is computed and the next state is determined. A reward, which expresses how good the state is, is also calculated in each step. The reward given to the robot is designed based on the task; for example, if we want a robot to learn to move forward, we can give it a reward of the distance it moves forward for each step. In one episode, the robot simulates a specified number of steps. An expected cumulative reward in one episode for a robot whose discrete and continuous parameters are \({{\mathcal {G}}}\) and \(\theta\) , respectively, can be expressed as \(R({{\mathcal {G}}},\ \theta )\) . The goal of improvement is to find \({{\mathcal {G}}}\) and \(\theta\) that maximize \(R({{\mathcal {G}}},\ \theta )\) . 3 Evolution of vertebrates To get a hint on how to update the tree \({{\mathcal {G}}}\) , we refer to the evolution of real animals. Since the evolution of animals is very diverse, it is difficult to find a common law for all animals. Therefore, we will focus on vertebrates only. Moreover, we believe that rigid parts of robots should mainly play the role of transmitting force, which is similar to the role of vertebrate bones. Hence, we focus on the evolution of vertebrate bones. In the evolution of vertebrates, one of the most dramatic changes in bone structure was the evolution of fins to limbs. During this evolution, the number of bones in fins, which had many bones, was greatly reduced and evolved into limbs, which had only a few bones [ 9 , 11 ]. Subsequently, limbs were adapted to terrestrial environments and became capable of performing various functions without a significant increase in the number of bones. For example, the arms of humans, the forelimbs of seals, and the wings of birds (or bats) are all derived from a common ancestor and have similar bone structures, which is called a homologous structure [ 10 ]. Meanwhile, their functions are diverse, such as holding objects, crawling in water, and flying. Although it seems that the evolution from fish fin to limbs is forced by the difference in the environments, the functionality of the fin does not necessarily depend on the complexity of its skeleton: for example, Cetacea (whales and dolphins), which are believed to once be a terrestrial mammal, have fins suitable for swimming but their skeletons are similar to those of other mammals [ 12 ]. Furthermore, there are many examples of decreasing in the number of the bones [ 9 , 11 , 13 ]. From the above viewpoint, we hypothesize that there is a tendency of decreasing the number of bones in the vertebrates’ evolutions. Based on our hypothesis, we limit the update of \({{\mathcal {G}}}\) to only those methods that reduce the number of nodes. 4 Method 4.1 Improvement of a discrete parameter Based on the discussion in Sect. 3 , we believe that it is reasonable to limit the update of \({{\mathcal {G}}}\) to only those methods that reduce the number of nodes. More concretely, we prepare \({{\mathcal {G}}}\) with a sufficient number of rigid parts in prior and remove the “less important” leaf nodes from it. There are several possible indicators for determining “less important” rigid parts. The indicator should be determined based on the task and the environment. In this paper, we consider an indicator tuned to the walking task. Since the main role of rigid parts corresponding to leaf nodes is to transmit forces to the body through contact with the world, rigid parts that do not have frequent contact with the world can be regarded as “less important.” Therefore, we count the number of contacts with the world through several simulations. Note that, with random noise given to the inputs and outputs, we run several simulations for each pair of the robot morphology and its controller. If the number of contacts is smaller than a certain threshold, we remove the leaf node and the corresponding edge. Conversely, for the addition of rigid parts, it is difficult to design an indicator of the high possibility of performance improvement. We avoid this problem by updating only to decrease the number of rigid parts. Moreover, we can expect a relationship that the “importance” per rigid part increases as the number of rigid parts decreases. Therefore, if the number of rigid parts continues to decrease, we will eventually reach a situation where every rigid part is “important.” Once a balance is achieved, no further changes will occur. Consequently, this strategy has the advantage of making it easier to decide when to terminate the learning. 4.2 Improvement of continuous parameters In this subsection, we give a method to search for a \(\theta\) such that \(R({{\mathcal {G}}},\ \theta )\) is maximized while \({{\mathcal {G}}}\) is fixed. We use the REINFORCE algorithm shown in [ 3 , 4 ]. This algorithm gives a way to search for two real vectors \(\mu =[\mu _1,\ldots ,\mu _N]^{{\mathsf {T}}}\) and \(\sigma =[\sigma _1,\ldots ,\sigma _N]^{{\mathsf {T}}}\) instead of \(\theta\) , and samples \(\theta\) from a multivariate normal distribution \({{\mathcal {N}}}\left( \mu ,\ (\mathrm {diag}\ \sigma )^2\right)\) . The objective function \(f({{\mathcal {G}}},\ \mu ,\ \sigma )\) for a pair of \({{\mathcal {G}}}\) , \(\mu\) , and \(\sigma\) can be expressed as follows: $$\begin{aligned} \begin{aligned} f(&{{\mathcal {G}}},\ \mu ,\ \sigma )\\&= E\left[ R({{\mathcal {G}}},\ \theta )\ \big |\ \theta \sim {{\mathcal {N}}}\left( \mu ,\ (\mathrm {diag}\ \sigma )^2\right) \right] . \end{aligned} \end{aligned}$$ (2) To improve \(\mu\) and \(\sigma\) using the gradient method, we can update as $$\begin{aligned} \begin{aligned} \mu&\leftarrow \mu + \alpha \nabla _\mu f({{\mathcal {G}}},\ \mu ,\ \sigma ) ,\\ \sigma&\leftarrow \sigma + \alpha \nabla _\sigma f({{\mathcal {G}}},\ \mu ,\ \sigma ) \end{aligned} \end{aligned}$$ (3) with learning rate \(\alpha >0\) . Let the population size be M . In other words, sample \(\theta ^j\sim {{\mathcal {N}}}\left( \mu ,\ (\mathrm {diag}\ \sigma )^2\right)\) independently for each \(j\in \{1,\ldots ,M\}\) and take M vectors \(\theta ^1=\left[ \theta _1^1,\ldots ,\theta _N^1\right] ^{{\mathsf {T}}}, \ldots , \theta ^M=\left[ \theta _1^M,\ldots ,\theta _N^M\right] ^{\mathsf T}\) . Let b and s be the sample mean and sample standard deviation of \(R\left( {{\mathcal {G}}},\ \theta ^1\right) ,\ldots ,R\left( {{\mathcal {G}}},\ \theta ^M\right)\) , respectively. According to the paper [ 4 ], the following approximations hold: $$\begin{aligned} \begin{aligned} \alpha \nabla _{\mu _i} f({{\mathcal {G}}},\ \mu ,\ \sigma )&\approx \frac{\gamma }{M}\sum _{j=1}^M \frac{R\left( {{\mathcal {G}}},\ \theta ^j\right) -b}{s} \left( \theta _i^j - \mu _i\right) , \\ \alpha \nabla _{\sigma _i} f({{\mathcal {G}}},\ \mu ,\ \sigma )&\\ \approx \frac{\gamma }{M}&\sum _{j=1}^M \frac{R\left( {{\mathcal {G}}},\ \theta ^j\right) -b}{s} \frac{\left( \theta _i^j - \mu _i\right) ^2 - \sigma _i^2}{\sigma _i} , \end{aligned} \end{aligned}$$ (4) for some \(\gamma > 0\) . We use this \(\gamma\) as the learning rate instead of \(\alpha\) . More precisely, in Eq. ( 4 ), the angle between (the vector formed by the left-hand side) and (the vector formed by the expected value of the right-hand side) is ensured to be equal to or less than 90 degrees. Therefore, using Eqs. 3 and 4 , we can update \(\mu\) and \(\sigma\) according to the gradient method. 4.3 Simultaneous improvement of discrete and continuous parameters We combine the methods of Sects. 4.1 and 4.2 to give a method for simultaneously improving discrete and continuous parameters. Both the methods in Sects. 4.1 and 4.2 utilize the results of several simulations for updates. Because of this similarity, these methods can be naturally combined. For the k th generation, the following operations are performed. First, the continuous parameter \(\theta ^j\in {{\mathbb {R}}}^N,\ j\in \{1,\ldots ,M\}\) is sampled from the multivariate normal distribution \({{\mathcal {N}}}\left( \mu ,\ (\mathrm {diag}\ \sigma )^2\right)\) . Next, we obtain the cumulative rewards \({{\mathcal {R}}}^1({{\mathcal {G}}},\ \theta ^j),\ldots ,{{\mathcal {R}}}^L({{\mathcal {G}}},\ \theta ^j)\) through L simulations. Using $$\begin{aligned} R({{\mathcal {G}}},\ \theta ^j) \approx \frac{1}{L}\sum _{l=1}^L {{\mathcal {R}}}^l({{\mathcal {G}}},\ \theta ^j)\ , \end{aligned}$$ (5) \(R({{\mathcal {G}}},\ \theta ^j)\) is calculated approximately. Furthermore, \(\mu\) and \(\sigma\) are updated by Eqs. 3 and 4 . Finally, the number of contacts with the world of the rigid part corresponding to each leaf node is checked. If it is less than the threshold C , \({{\mathcal {G}}}\) is updated by removing v from \({{\mathcal {V}}}\) . The above process is iterated K times. The pseudocode for this method is shown in Algorithm 1. Let \({{\mathcal {G}}}_{\mathrm{before}}\) and \({{\mathcal {G}}}_{\mathrm{after}}\) be the discrete parameters before and after the change in a generation, respectively. Since \(\theta\) is learned for \({{\mathcal {G}}}_{\mathrm{before}}\) , it would be worthless when the discrete morphology changes to \({{\mathcal {G}}}_{\mathrm{after}}\) . In other words, the expected cumulative reward for the set \(({{\mathcal {G}}}_{\mathrm{after}},\ \theta )\) is expected to be smaller than that for the set \(({{\mathcal {G}}}_{\mathrm{before}},\ \theta )\) . Simulations in Sect. 5 confirm that this expectation is correct. However, in our method, only the nodes of \({{\mathcal {G}}}_{\mathrm{before}}\) that have a small impact on the environment are changed. Because of this, the values of \(\theta\) that are suitable for \({{\mathcal {G}}}_{\mathrm{before}}\) and \({{\mathcal {G}}}_{\mathrm{after}}\) are expected to be similar. Hence, \(\theta\) have a transfer learning-like effect when learning with \({{\mathcal {G}}}_{\mathrm{after}}\) , resulting in faster convergence to the desired value. 5 Simulation experiments 5.1 Proposed method for walking task Table 1 Hyperparameters in the simulation experiment Full size table Fig. 2 \({{\mathcal {G}}'}\) , i.e. the initial value of \({{\mathcal {G}}}\) Full size image We applied the proposed method to design a robot that is suitable for walking task. Each hyperparameter is defined in Table 1 . Let the initial value of \({{\mathcal {G}}}\) be \({{\mathcal {G}}'}\) , which is shown in Fig. 2 . The robot is learned to move on a flat plane in a 3D space. The inputs to the controller are the height of the torso from the ground, the sine and cosine values of the yaw angle of the torso to the target direction, the velocity of the torso in the x, y, and z directions, the roll angle of the torso, the pitch angle of the torso, the current time, and the angle and angular velocity of each joint. Note that the yaw angle, roll angle, and pitch angle of the torso are measured with respect to the coordinate system fixed in the torso. All the angles are zero at the initial step of the simulation. Let the position of the torso at time step t be \((r_x(t), r_y(t), r_z(t))\) , the total number of joints be \(N_{\mathrm {joints}}\) , torque of the joint i be \(\tau _i\) , the angular velocity of joint i be \(v_i\) , and the weight constants \(K_r>0\) and \(K_\tau >0\) . The reward r ( t ) at time step t was defined as: $$\begin{aligned} r(t) = 1 + K_r(r_x(t)-r_x(t-1)) - K_\tau \sum _{i=1}^{N_{\mathrm {joints}}} \tau _i v_i. \end{aligned}$$ (6) The cumulative reward \({{\mathcal {R}}}\) is calculated as \({{\mathcal {R}}} = \sum _{t=1}^{T} r(t)\) , where the total number of time steps T was 1000, except the cases where the torso touches the ground. If the torso touches the ground, the episode is terminated at that moment. The settings for input, reward, and the number of steps follow the Ant-v2 environment settings [ 5 ] used in OpenAI Gym [ 14 ]. When updating the discrete parameter, we counted the number of contacts of each rigid part. However, we excluded episodes where the simulation was terminated within 300 steps from the count. If the number of non-excluded episodes is less than 50, we do not update the discrete parameter for that generation. We use an MLP with two intermediate layers, 128 and 64 nodes in each of them. The physics simulation was implemented using PyBullet [ 15 ], the Python bindings of the Bullet Physics SDK. The simulation of the robot corresponding to each \(j\in \{1,\ldots ,M\}\) is performed in parallel. We used a CPU with 128 virtual cores (AMD Ryzen Threadripper 3990X), and it takes about 110 hours to compute 9000 generations. Since the time required for a physical simulation is highly dependent on the number of rigid parts, this time can be greatly reduced by reducing the number of nodes in \({{\mathcal {G}}}'\) . There is a tradeoff between the time required for learning and the diversity of the resulting discrete parameters. Fig. 3 The final shape of the robot obtained by the proposed method. ( a ) The Appearance of the robot. ( b ) The discrete parameter \({{\mathcal {G}}}\) Full size image Fig. 4 The graph plots the expected cumulative reward \(R({{\mathcal {G}}},\ \theta ^j)\) and the number of nodes in \({{\mathcal {G}}}\) obtained using the proposed method for each generation. The 3 solid lines refer to the maximum, the average, and the minimum expected cumulative rewards in \(\{R({{\mathcal {G}}},\ \theta ^j)\ |\ j\in \{1,\ldots ,M\}\}\) Full size image As a result of learning, a robot shown in Fig. 3 a was obtained. The corresponding discrete parameter is also shown in Fig. 3 b. This robot has three particularly long legs, which form a support polygon. The other legs that are in contact with the ground near the torso move particularly rapidly and are thought to be the main source of propulsion. They move their legs quickly in small motions. Their motions do not resemble the trot, pace, or any other movements that we are familiar with. The expected cumulative reward obtained during the learning process is shown as a graph in Fig. 4 . As described in Sect. 4.3 , the expected cumulative reward drops discontinuously in the generation where the number of nodes in \({{\mathcal {G}}}\) changes. However, immediately after that, the expected cumulative reward tends to increase significantly, as can be seen in the area around generation 6000. 5.2 Comparison of improvement methods In this section, we improve the robot using various methods for the same simulation environment as Sect. 5.1 and compare the results. The following four methods are used for comparison: A. Only continuous control parameters are learned using the method of Sect. 4.2 . For fixed morphological parameters, we use OpenAI Ant-v2 [ 5 ]. There are four legs consisting of two capsule-shaped rigid parts extending from the torso, with a total of eight joints (Fig. 5 a). B. Only the continuous control parameters and continuous morphology parameters are learned using the method of Sect. 4.2 . For the initial values of the morphological parameters, the same parameters as in method A are used. This method of updating continuous parameters using REINFORCE is similar to prior work [ 3 ]. C. In addition to the method B, the nodes of the discrete parameter are deleted and added randomly. Change the discrete parameter every 100 generations. The type of change is chosen between “increase” and “decrease” with a probability of 50 percent each. In the case of “increase,” a new node is added to a randomly selected node (including torso), and in the case of “decrease,” a randomly selected leaf node is deleted. For the initial value of morphological parameters, the same parameters as in method A are used. D. All continuous and discrete parameters are learned, using the proposed method shown in Sect. 4.3 . This is equivalent to the simulation experiment in Sect. 5.1 . Fig. 5 The final shape of the robot obtained using each method Full size image Fig. 6 The expected cumulative reward obtained using each method. The plotted data indicate the average value for each generation Full size image Figure 5 shows the shape of the robot obtained from the simulation, and Fig. 6 shows the expected cumulative reward for each generation. Regarding the performance of method A, from the results of related studies [ 3 ], we conclude that the result is close to the true optimal. Method B increases the expected cumulative reward faster than method A and is saturated at about the 6,000th generation. To obtain better performance than method B, it would be necessary to learn the discrete parameter. Method C is not very efficient in learning. In the present simulation experiment, method D produced the best final performance. The disadvantage of proposed method D is that learning in the early stages is slower than that of method B. This is due to giving a discrete parameter as the initial value which is not efficient for walking. Methods A, B, and C used the discrete parameter known to be reasonable as initial values. In contrast, method D used the discrete parameter that was clearly not suitable for walking as initial values (Fig. 2 ). Nevertheless, method D performed as well as or better than the other methods. Therefore, it can be said that the discrete parameter was effectively learned by method D. 6 Conclusion In this study, we approached the problem of finding a robot morphology and controller design that would achieve the maximization of the reward in a given environment. The robot design was modeled with the discrete and continuous parameters. For the discrete parameter, we proposed and applied an efficient improvement rule. For the continuous parameters, we applied the REINFORCE algorithm. Finally, to simultaneously improve these parameters, we proposed a method by combining these two methods. The simulation experiment using the proposed method showed the successful simultaneous improvement of discrete and continuous parameters. Another simulation experiment was also performed using the existing method, and it was shown that our method performed as well as or better than other methods. Although the validity of our method was shown by the task of designing a good robot for locomotion on a flat surface, we have not shown the validity to other types of tasks. To apply our method to other tasks, it is necessary to set another criterion to judge which part is less important. We are planning to apply our method to other types of tasks such as grasping in future work. Another future work is to elaborate on the simulator to make it possible to design a realizable robot. Note that, in the simulation, we did not consider the mass distribution and the actuator size for the sake of simplicity. However, they are necessary for a resulting robot to be realizable.

Researchers at Kyoto University and Nagoya University in Japan have recently devised a new, automatic approach for designing robots that could simultaneously improve their shape, structure, movements, and controller components. This approach, presented in a paper published in Artificial Life and Robotics, draws inspiration from the evolution of vertebrates, the broad category of animals that possess a backbone or spinal column, which includes mammals, reptiles, birds, amphibians, and fishes. "The automatic robot design is a completely novel research project for the Matsuno Lab, the laboratory led by Fumitoshi Matsuno, and this is the first paper published for this project," Ryosuke Koike, one of the researchers who carried out the study, told TechXplore. "Its primary objective was to design a good-performing robot for a given task. Since there are innumerable possible combinations of robot morphologies and controllers, it is impossible to reach the best robot by manual human exploration. Therefore, we realized that it is necessary to establish a method for automatically designing robots using computers." Essentially, instead of developing a single robot design for a specific application, Koike and his colleagues wished to explore the possibility of creating a system that can automatically produce robot designs with specific morphologies and characteristics. If such a system worked efficiently, it could eventually simplify the development of task-specific robotic systems. The most notable feature of the automatic robot design system is that it is partly inspired by the evolution of vertebrates. More specifically, Koike and his colleagues hypothesized that as they evolved over centuries, the number of bones in the bodies of vertebrates tended to decrease. Their system follows a so-called "improvement rule" that is based on this hypothesis. More specifically, it was trained to design robots by removing unnecessary rigid parts from a previously developed complex robot. This in turn allows it to identify more effective morphologies and controller components for the newly designed robot. "Current walking robots often have morphologies resembling vertebrates, such as dogs," Koike explained. "By mimicking the evolution of vertebrates, we might design robots with performance comparable to these practical robots. Furthermore, we hope that, ideally, morphologies that are more evolved than today's vertebrates will emerge and perform better." Koike and his colleagues evaluated their automatic robot design method in a series of tests, where they compared it with other robotic design strategies. They found that the method's vertebrate evolution-inspired improvement rule led to the creation of simplified and yet interesting robot designs. "The discrete factors of robot morphology—how many rigid parts a robot needs and how they should be connected—are difficult to handle theoretically, so they have largely depended on the designers' experience and intuitions," Koike said. "Or, in recent studies considering automatic robot design, they have often been explored in an almost exhaustive way. We made the search more efficient by setting a simple rule: monotonically reduce the number of rigid parts." In the future, the automatic robot design method introduced by this team of researchers could be implemented in real-world settings to speed up and improve the design of robots. In addition, this work could inspire the development of other systems based on the save vertebrate evolution hypothesis. In their next studies, Koike and his colleagues plan to explore the potential of other possible approaches for automatically designing robots, which will not necessarily be inspired by the evolution of vertebrate. In addition, they recently created an alternative method that tackles some of the limitations of their robot design approach, which will soon be presented in a new journal paper. "So far, we only confirmed this method's effectiveness on walking tasks and found that it takes a long time to learn and only slightly outperforms existing methods," Koike added. "Humans invented the wheel to move efficiently across flat terrain. As the scope of robots' morphology expands, even the wheel should be invented automatically. If a better mechanism than the wheel exists, automatic robot design should even be able to invent it. Furthermore, as the scope of tasks expands, it should invent better aerial vehicles, ships, industrial robots, and so on."

10.1007/s10015-022-00793-4

Medicine

Dietary changes could help reduce pregnancy complications in women with type 1 diabetes

Alexandra J. Roth-Schulze et al, Type 1 diabetes in pregnancy is associated with distinct changes in the composition and function of the gut microbiome, Microbiome (2021). DOI: 10.1186/s40168-021-01104-y

http://dx.doi.org/10.1186/s40168-021-01104-y

https://medicalxpress.com/news/2021-10-dietary-pregnancy-complications-women-diabetes.html

Abstract Background The gut microbiome changes in response to a range of environmental conditions, life events and disease states. Pregnancy is a natural life event that involves major physiological adaptation yet studies of the microbiome in pregnancy are limited and their findings inconsistent. Pregnancy with type 1 diabetes (T1D) is associated with increased maternal and fetal risks but the gut microbiome in this context has not been characterized. By whole metagenome sequencing (WMS), we defined the taxonomic composition and function of the gut bacterial microbiome across 70 pregnancies, 36 in women with T1D. Results Women with and without T1D exhibited compositional and functional changes in the gut microbiome across pregnancy. Profiles in women with T1D were distinct, with an increase in bacteria that produce lipopolysaccharides and a decrease in those that produce short-chain fatty acids, especially in the third trimester. In addition, women with T1D had elevated concentrations of fecal calprotectin, a marker of intestinal inflammation, and serum intestinal fatty acid-binding protein (I-FABP), a marker of intestinal epithelial damage. Conclusions Women with T1D exhibit a shift towards a more pro-inflammatory gut microbiome during pregnancy, associated with evidence of intestinal inflammation. These changes could contribute to the increased risk of pregnancy complications in women with T1D and are potentially modifiable by dietary means. Video abstract Background The gut microbiome provides essential metabolites, vitamins, co-factors and hormones, protects against pathogenic microorganisms and has a key role in the development of the immune and other systems [ 1 , 2 ]. Changes in the composition of the gut microbiome are associated with ageing, environmental conditions, life events and disease states [ 2 , 3 , 4 ]. In pregnancy, women undergo significant physiological changes, but only recently has the gut microbiome been studied in this context [ 5 , 6 ]. Koren et al. [ 5 ] sampled the gut microbiome in the first and third trimesters and found that the taxonomic composition in the first trimester was similar to that of non-pregnant women but in the third trimester the abundance of Actinobacteria and Proteobacteria phyla increased along with an overall decrease in bacterial richness (alpha diversity). In studies in germ-free mice, they observed that inoculation with third compared to first trimester feces led to greater weight gain, insulin resistance and gut inflammation and suggested this was an adaptive proinflammatory response to defend the fetus from pathogens and provide it with nutrients. In contrast, after analysing fecal samples weekly across pregnancy, DiGiulio et al. [ 6 ] found no significant temporal differences in diversity or composition of the gut microbiome. These contrary findings and the dearth of studies warrant further investigation of the gut microbiome in pregnancy. Type 1 diabetes (T1D) is an autoimmune disease in which insulin-producing β cells in the islets of the pancreas are destroyed by T lymphocytes leading to insulin deficiency [ 7 ]. In pregnancy, T1D is associated with systemic and intra-uterine markers of sub-clinical inflammation and higher risks of complications for mother and fetus [ 8 , 9 , 10 ]. Alterations in the bacterial gut microbiome have been reported in T1D, mainly in children at high risk and at diagnosis (reviewed in [ 11 ], [ 12 , 13 , 14 , 15 , 16 , 17 ]). They include a decrease in alpha diversity (richness) [ 12 , 13 , 14 ] and in the abundance of lactate- and butyrate-producing and mucin-degrading bacteria [ 13 , 14 , 15 , 16 , 17 ], and an increase in the abundance of the Bacteroides genus [ 13 , 14 ]. Functionally, these compositional changes are reflected by a decreased abundance of genes encoding related metabolic pathways and enzymes, e.g. butyryl-coenzyme A (CoA)-CoA transferase [ 15 ] and butyryl-CoA dehydrogenase for butyrate synthesis [ 16 ]. These changes are not necessarily specific for T1D but nevertheless, they may have clinical consequences, including in pregnancy. Gut butyrate is a key determinant of gut health and regulator of gene expression and homeostatic immunity [ 18 , 19 , 20 ]. It is the major energy source for the colonic mucosa, induces the synthesis of mucin and it promotes gut epithelial integrity, preventing ‘gut leakiness’. In the non-obese diabetic (NOD) mouse model of T1D, dietary butyrate supplementation promoted an increase in regulatory T cells and a decrease in the incidence of spontaneous diabetes [ 20 ]. Increased gut leakiness has been described in established T1D [ 21 ] and recently by ourselves in association with gut microbiome changes in children with islet autoimmunity who progressed to T1D [ 22 ]. Gut leakiness with translocation of toxins and dietary antigens into the blood may result in systemic inflammation, reported with T1D in pregnancy complicated by pre-eclampsia [ 10 ]. Because a consensus about the gut microbiome in pregnancy is lacking, even in the absence of T1D, we applied shotgun whole metagenomic sequencing (WMS) to analyse the gut microbiome across pregnancy in women with and without T1D participating in the Australia-wide Environmental Determinants of Islet Autoimmunity (ENDIA) study. Results Study population Fecal samples were collected between February 2013 and October 2017 from women enrolled in the ENDIA study, a prospective, pregnancy-birth cohort study that follows 1500 Australian children who have a first-degree relative with T1D [ 23 ]. Thirty-five women (36 pregnancies) with T1D and 31 women (34 pregnancies) without T1D had each provided from one to three fecal samples across pregnancy (total 134 samples) for analysis by shotgun WMS (Fig. 1 ). Table 1 summarizes and compares characteristics of the T1D and non-T1D pregnancies. Fig. 1 Fecal samples obtained in pregnancy. n: number of samples; T1: trimester 1; T2: trimester 2; T3: trimester 3; T1D: women with type 1 diabetes; Non-T1D: women without T1D Full size image Table 1 Summary of characteristics of non-T1D and T1D pregnancies Full size table Whole metagenomic sequencing The WMS dataset, 47,766,763 ± 10,956,057 (mean ± SD) paired-end reads per sample, was obtained using an Illumina NovaSeq 6000. Raw reads (SRA accession: PRJNA604850) were pre-processed using KneadData bioBakery tool [ 24 ] to eliminate human DNA sequences and filter sequences with poor quality which on average removed 6% of the reads. After quality control and read filter steps, 44,940,628 ± 10,572,188 (mean ± SD) paired-end reads per sample were obtained ( Excel file E0 ). Taxonomic diversity and composition of the gut microbiome in women with and without T1D during pregnancy Sequences were analysed with MetaPhLan2 implemented within the HUMAnN2 pipeline. Overall, 340 bacterial species were identified, with an average of 93 ± 13 (mean ± SD) species per sample. The top 25 most abundant species accounted for more than 50% of the gut microbiome composition of each subject in any given trimester (Figure S1 ). Alpha diversity (observed richness or number of species) per sample was calculated and generalized estimating equations (GEE) were applied to test for differences between women without and with T1D, and between trimesters, and to determine if there was an interaction between T1D status and trimester. No differences were found in richness due to T1D status or time, or interactions (Figure S2 , Excel file E1 ). For analysis of beta diversity, Bray-Curtis coefficients were calculated between sample pairs, ordinated and plotted by principal coordinate analysis (PCoA) for each taxonomic level (Figs. 2 , S3 and S4 ). To test for differences in beta diversity, a repeated-measure aware permutational analysis of variance (RMA-PERMANOVA) of the Bray-Curtis coefficients was performed on proportional log transformed data. This revealed a significant interaction between T1D status and time at all taxonomic levels. Therefore, differences between women with and without T1D were assessed within trimesters. No significant differences were detected in trimesters 1 and 2. However, differences were significant at the strain ( P = 0.002), species ( P = 0.001), genus ( P = 0.070) and family ( P = 0.034) levels in trimester 3 ( Excel file E2 ). Fig. 2 Beta diversity analysis by T1D status. PCoA ordination plots based on Bray-Curtis distances between samples at the strain and species taxonomic levels separated by trimesters in pregnancy. T1D: women with type 1 diabetes (red); Non-T1D: women without T1D (blue) Full size image To rule out the possibility that these results were influenced by the difference in sample size between trimesters 1 and 3, we performed a sensitivity analysis by subsampling trimester 3 to the size of trimester 1 ( n = 23), using samples of trimester 3 from the same women in trimester 1, and repeated the beta diversity analysis. Similar to the complete trimester 3 dataset, differences were significant at the strain ( P = 0.003), species ( P = 0.003), genus ( P = 0.043) and family ( P = 0.047), but also phylum ( P = 0.09), taxonomic levels ( Excel file E2 ). Differences in beta diversity reflect differences in taxonomic composition. To identify differences in specific taxa between women with and without T1D in pregnancy, differential abundance was analysed in limma. Only taxa for which the prevalence (i.e. proportion of samples with those taxa) was above 50% in at least one group and with a log2 fold-change (logFC) greater than 0.5 or less than − 0.5 were considered. Across all trimesters, the species Bacteroides caccae (FDR 0.03) and its unique strain (unclassified) in the dataset (FDR 0.03), as well as the order Enterobacteriales (FDR 0.07) were increased in women with T1D (Fig. 3 ; Excel file E3 ). On the other hand, species Bacteroidales bacterium ph8 (FDR 0.034) and its strain ( GCF000311925) (FDR 0.03), the genus (FDR 0.08) and family (FDR 0.08) to which Bacteroidales bacterium ph8 belongs, and the order Bifidobacteriales (FDR 0.07), were decreased in women with T1D (Figure 3 ; Excel file E3 ). Fig. 3 Means and standard errors of the log2-transformed fitted values shown as a point in each trimester for differentially abundant taxa in women with (red) and without (blue) T1D. * in the top right corner denotes a significant difference (FDR < 0.1) between groups throughout pregnancy; * between points denotes a significant difference (FDR < 0.1) between groups in that trimester Full size image Differences between women with and without T1D were also assessed within trimesters. In trimesters 1 and 2, taxa were not significantly different. However, several differences were found in trimester 3, in which the unique strain (unclassified) of Bacteroides caccae (FDR 0.004), the species Bacteroides caccae (FDR 0.004), the species Bacteroides vulgatus (FDR 0.04) and its unique strain (unclassified) (FDR 0.04) and Bacteroides uniformis (FDR 0.04) were increased in women with T1D, while the species Bacteroidales bacterium ph8 (FDR 0.01) and its strain ( GCF000311925; FDR 0.005), and the genus (FDR 0.08) and family (FDR 0.08) of Bacteroidales bacterium ph8 and the order Bifidobacteriales (FDR 0.07), were decreased (Fig. 3 ; Excel file E3 ). A significant Spearman correlation ( R 2 > 0.4) was found between B . caccae and B . vulgatus ( R 2 = + 0.43; adj. P = 0.013). A sensitivity analysis of differential abundance was also applied to the subset of trimester 3 samples referred to above: 13 species, 13 strains, 2 genera, 3 families, 2 orders and 3 phyla were detected as differentially abundant. From these, Bacteroides caccae and Bacteroides uniformis , an unclassified strain of Bacteroides caccae and the order Bifidobacteriales were also detected in the larger dataset of trimester 3 samples. Differential abundance results are summarised in Excel file E3 . In order to identify the bacterial species that were most abundant within the Enterobacteriales and Bifidobacteriales orders, we plotted the average relative abundance in women with and without T1D (Figure S5 A). Escherichia coli was the most abundant species within Enterobacteriales and, together with an unclassified species of the genus Escherichia , accounted for almost the complete abundance of this order. In addition, a significant Spearman correlation was found between E . coli and Coprococcus sp. ART55_1 ( R 2 = − 0.6, adj. P = 0.09). Bifidobacterium adolescentis and Bifidobacterium longum were the most abundant species within Bifidobacteriales (Figure S5 A). A lmer test applied to test differences in the abundance of these four species between women with and without T1D revealed that the abundance of E . coli in trimester 3 and of B . adolescentis in trimester 1 were significantly different between women with and without T1D ( P = 0.01 for both; Figure S6 ). Effect of gestation time and other factors on the gut microbiome during pregnancy No significant differences in alpha diversity were detected in women with or without T1D according to time, analysed either by days of gestation ( P value 0.5) or by trimester ( P values > 0.6), i.e. as continuous or categorical variables, respectively ( Excel file E1 ). Due to the significant interaction between T1D status and time, differences in beta diversity across time (days or trimesters) were assessed separately in women with and without T1D ( Excel file E2 ). Differences were detected only at the strain ( P value 0.03) and species ( P value 0.06) levels in women without T1D with time as continuous variable ( Excel file E2 ). However, in women with T1D, differences in beta diversity across days of gestation and between trimesters were significant at all taxonomic levels except order and phylum ( Excel file E2 ). These observations suggested that the microbial community structure across pregnancy is less stable in women with T1D. Therefore, we sought to identify differentially abundant taxa across trimesters separately within each group. Throughout pregnancy, in women with T1D, the abundance of an unclassified species of the family Peptostreptococcaceae (FDR 0.02), the species Odoribacter splanchnicus (FDR 0.098), the genus Prevotella (FDR 0.066) dominated by the species Prevotella copri (Figure S5 B) and the phylum Verrucomicrobia (FDR 0.043) decreased, while an unclassified strain of species Streptococcus thermophilus (FDR 0.099) and the species Streptococcus thermophilus (FDR 0.04) and family Porphyromonadaceae (FDR 0.092) increased ( Excel file E4 ; Figure S7 ). In women without T1D, an Anaerostipes hadrus GCF000332875 strain (FDR 0.038) and species Anaerostipes hadrus (FDR 0.059), an unclassified strain of Haemophilus parainfluenzae (FDR 0.001) and species Haemophilus parainfluenzae (FDR 0.003), genus Haemophilus (FDR 0.004), family Pasteurellaceae (FDR 0.002), strain (GCF000218445 [FDR 0.04]) and species of Lachnospiraceae bacterium 1157FAA (FDR 0.055) and an unclassified species of Veillonella genus (FDR 0.083) decreased during pregnancy ( Excel file E4 ; Figure S7 ). Furthermore, in women without T1D strains Ruminococcus sp. 5139BFAA GCF000159975 and Lachnospiraceae bacterium 3157FAACT1 GCF000218405 (FDR 0.063 and 0.075, respectively) and their corresponding species (FDR 0.065 and 0.075), unclassified strains of Streptococcus thermophilus (FDR 0.06) and Bifidobacterium_animalis (FDR = 0.06) and their corresponding species (FDR 0.06 for both) increased throughout pregnancy ( Excel file E4 ; Figure S7 ). As expected, women with and without T1D differed in serum 1,5-anhydroglucitrol (1,5-AG), a marker of short-term glycemic control [ 25 ] (Table 1 ), but in women with T1D, serum 1,5-AG was related to beta diversity only at the phylum level ( Excel file E2 ). Mode of delivery had an effect on the beta diversity only at the family level ( Excel file E2 ). No significant associations were found between beta diversity and age at conception, body mass index (BMI), parity, carbohydrate or fibre intake ( Excel file E2 ). However, a difference was observed in the microbiome composition at the strain and species levels according to the human leukocyte antigen (HLA) class II type ( Excel file E2 ). The model used to test for differences in beta diversity between women with and without T1D was adjusted for HLA type. HLA type accounted for 3.2% of the variation [ R 2 ] in beta diversity in trimester 3 ( Excel file E2 ). After controlling for this effect, T1D status explained 2.9% of the variation and the difference in beta diversity between women with and without T1D women was statistically significant ( P = 0.004) ( Excel file E2 ). Finally, even though for the differential abundance analysis an adjustment for HLA type was included in the model, an additional analysis was performed to detect differences in the abundance of specific taxa due to HLA type and to verify that the taxa that were detected as differentially abundant due to T1D status were not affected by HLA type. Differences due to HLA type were detected only between HLADR34 and HLADR3X and DR4X for the abundance of strain Eubacterium ramulus GCF000469345 and species Eubacterium ramulus in trimester 1 and an unclassified strain of species Eubacterium rectale and species Eubacterium rectale in trimester 3, which were decreased in women with HLA DR34 Excel file E3 . None of the taxa identified as differentially abundant due to T1D status were significantly affected by HLA type. Validation of differentially abundant species by qPCR To validate the findings from metagenomic sequencing, we analysed the relative abundance of two of the top-ranked differentially abundant bacteria, Bacteroides caccae and Bacteroides vulgatus , in the same cohort of T1D and non-T1D mothers in trimester 3. Relative abundances obtained by metagenomic sequencing and qPCR were strongly correlated (Spearman R = + 0.91 and + 0.74 for B . caccae and B . vulgatus , respectively). By fitting linear models in lmer with conception age, BMI, parity and HLA type introduced as fixed effects, and ‘woman ID’ and processing batches as random effects, qPCR confirmed the increase in relative abundance of B . caccae ( P = 0.00005) and B . vulgatus ( P = 0.04) in women with T1D (Figure S8 ). Functional annotation of gut microbiome taxa Sequences processed with HUMAnN2 were annotated, complete metabolic pathways quantified, gene abundances calculated and regrouped into KO (Kegg Orthology) and MetaCyc reaction functional categories. A total of 451 complete pathways, 5628 KO and 3204 MetaCyc reaction categories were obtained. No significant interaction in richness was detected between factors T1D status and time. In the model in which time was considered as a continuous variable, richness was significantly higher in women with T1D for all three functional categories (Figure S9 , Excel file E1 ). For beta diversity, the interaction between T1D status and time was significant. Therefore, differences between groups were assessed within each trimester, but were significant for the three functional categories only in trimester 3 (Fig. 4 ; Excel file E2 ). Fig. 4 Beta diversity analysis by type 1 diabetes (T1D) status. PCoA ordination plots based on Bray-Curtis distances between samples for a pathways, b KOs and c MetaCyc reactions in trimester 3. T1D: women with type 1 diabetes (red); women without T1D (blue) Full size image Women with and without T1D displayed significant differences in the abundance of a number of features identified in pathways, KO and MetaCyc categories; these are comprehensively listed in Supplementary Excel files E5-10 . Selected functions, namely LPS production, vitamin K2 synthesis, vitamin B6 synthesis, vitamin B12 synthesis, short-chain fatty acid (SCFA) synthesis and mucin degradation, and the principal bacterial species contributing to these functions, are summarised in Table 2 . Examples of bacteria contributing to a functional feature are shown in Figures S10 and S11 . Of interest, a pathway (PWY1269: CMP-3-deoxy- d -manno-octulosonate biosynthesis I), 17 KO gene categories and two MetaCyc reactions (DARAB5PISOM-RXN and UDPGLCNACEPIM-RXN) involved in the synthesis of bacterial lipopolysaccharides (LPSs) were enriched in women with T1D ( Excel file E5-E7 ; Table 2 ; Fig. 5 , Figure S10 A). Seven pathways and 6 KO categories involved in vitamin K2 synthesis were also increased in women with T1D ( Excel files E5-E6 ; Table 2 ; Fig. 5 ; Figure S10 B). In addition, two KO categories increased in women with T1D in trimester 3 were involved in antibiotic tolerance (K03771) and biofilm formation and (K18831) ( Excel file E6 ). Table 2 Pathways and enzymes differentially abundant in T1D women Full size table Fig. 5 Means and standard errors of the log2-transformed fitted values shown as a point in each trimester for differentially abundant functional features in women with (red) and without (blue) T1D. One example for each of six broad categories is shown: lipopolysaccharide (LPS) production (CMP–3–deoxy– d –manno–octulosonate synthesis [PWY–1269]), vitamin K2 synthesis (superpathway of menaquinol–8 synthesis [PWY–5838]), vitamin B6 synthesis (pyridoxal 5′–phosphate synthase [K06215]), vitamin B12 synthesis (adenosylcobalamin salvage from cobinamide [COBALSYN–PWY]), short-chain fatty acid (SCFA) production (3–hydroxybutyryl–CoA dehydrogenase [K00074]) and mucin degradation (beta– N –acetylhexosaminidase [K01207]). * in the top right corner denotes a significant difference (FDR < 0.1) between groups throughout pregnancy; * between points denotes a significant difference (FDR < 0.1) between groups in that trimester Full size image The enzyme pyridoxal 5′-phosphate synthase (K06215) involved in the deoxyxylulose 5-phosphate (DXP)-independent pathway for vitamin B6 synthesis, one pathway, five KO categories and six metaCyc reactions related to vitamin B12 (cobalamin) synthesis and five pathways, 11 KO categories and 13 MetaCyc reactions involved in SCFA synthesis, including pyruvate and acetyl-CoA production and butyrate synthesis from acetate or lactate, were decreased in women with T1D ( Excel file E5-E7 ; Table 2 ; Fig. 5 ; Figure S11 ). The abundance of beta- N -acetylhexosaminidase (K01207) involved in the degradation of mucin was also significantly reduced in women with T1D, but again only in trimester 2 (Table 2 ; Excel file E6 ; Fig. 5 ; Figure S11 ). Identification of bacterial clusters based on differentially abundant functional features Differentially abundant functional features derived from HUMAnN2 were contributed not by a single species but rather a combination of species. Therefore, relative abundances of the principal contributing species in each of the six selected functions could be grouped into clusters (Table 2 ). For functions with three or more features, only principal contributors to at least three features were considered. For each cluster, a linear model was fitted with lmer using the same design as for the differential abundance analysis. This confirmed that women with T1D had an increased abundance of bacterial clusters contributing to production of LPS and synthesis of vitamin K2 and a decreased abundance of bacterial clusters contributing to synthesis of vitamins B6 and B12, production of SCFA and degradation of mucin (Figure S12 ). Markers of gut pathology Because the composition and function of the gut microbiome of women with T1D was suggestive of a pro-inflammatory state, we sought evidence for gut inflammation in women with T1D. Fecal calprotectin, released from neutrophils and monocytes, is a marker of intestinal inflammation that may result in increased epithelial permeability [ 26 ]. Serum intestinal fatty acid-binding protein (I-FABP) is a marker of intestinal epithelial damage [ 27 ]. Fecal calprotectin and serum I-FABP were measured in trimester 3 in 61 women (32 with T1D) and 55 women (27 with T1D), respectively. Fecal calprotectin was increased in women with T1D compared to those without T1D (112 ± 148 vs. 36 ± 28 [mean ± SD] mg/kg: P value 0.04; Mann-Whitney test). Serum I-FABP was also increased in women with T1D compared to women without T1D (587 ± 235 vs. 314 ± 185 [mean ± SD] pg/mL: P value 0.0003; Mann-Whitney test) (Figure S13 ). However, these markers did not significantly correlate (Spearman R > 0.4) with any of the individual taxa that were differentially abundant between T1D and non-T1D women. Discussion The gut microbiome in pregnancy has previously been analysed in two studies, by Koren et al . [ 5 ] and DiGiulio et al. [ 6 ], who pyrosequenced the 16S rRNA gene V1–V2 and V3–V5 regions, respectively, but with different conclusions. To our knowledge, the current study is the first to also include women with T1D, who have a higher frequency of complications and evidence of systemic and intra-uterine inflammation in pregnancy [ 8 , 11 , 12 ] that could conceivably be related to the gut microbiome. Koren et al. [ 5 ] compared single samples from trimesters 1 and 3 from 91 pregnancies and reported a decrease in alpha diversity and ‘remodelling of the gut microbiome’ by the third trimester, specifically a decrease in the abundance of taxa in the genus Faecalibacterium that generate the anti-inflammatory SCFA butyrate [ 18 ] and an increase in taxa in the phylum Proteobacteria recognised to be pro-inflammatory [ 28 ]. On the other hand, DiGiulio et al. [ 6 ] by weekly sampling of 49 women found no significant changes in diversity or composition across pregnancy. Similar to DiGiulio et al. [ 6 ], we observed no differences across pregnancy in alpha diversity but found differences in beta diversity at the strain and species levels in women without T1D and at all taxonomic levels in women with T1D. In addition, particularly in women with T1D, we saw changes in the relative abundance of specific taxa across pregnancy with progression to a more pro-inflammatory microbiome, similar to Koren et al. [ 5 ]. The taxonomic differences between women with and without T1D were reinforced by functional annotation, revealing differential abundance in enzymes and pathways as pregnancy progressed. These differences could not be attributed to demographic or other factors, including diet. It is important, however, to keep in mind that our findings are based on DNA analysis and they might not necessarily reflect changes at the RNA or protein level. In examining differential abundance, we observed two main patterns: (1) taxa that were differentially abundant between women with and without T1D across all trimesters and (2) taxa that were similar in abundance in women with and without T1D in the first trimester but decreased or increased to be differentially abundant in trimester 3. In the first category, B . caccae and the order Enterobacteriales were increased in women with T1D across all trimesters. Within Enterobacteriales , Escherichia coli was the most abundant species and was enriched in women with T1D. Products of E . coli including lipopolysaccharide [ 29 ] and microcin [ 30 ] promote intestinal inflammation, intestinal permeability and low-grade systemic inflammation, and are implicated especially in the pathogenesis of inflammatory bowel disease [ 31 , 32 ]. Moreover, an increase in these facultative anaerobes may displace obligate anaerobic bacteria that produce SCFAs [ 32 ], supported by the negative correlation between the abundance of E . coli and Coprococcus sp. ART55_1, further accentuating inflammation. In women with T1D, this may contribute to the decrease in abundance of the genus Prevotella comprising almost entirely Prevotella copri , a species that produces succinate and the SCFAs propionate and acetate known to be associated with improved glucose metabolism [ 19 , 33 ]. Furthermore, in women with T1D, E . coli contributed to an increased abundance of enzymes involved in antibiotic tolerance (K18831) and biofilm formation (K03771). Bacterial biofilms confer increased tolerance to antibiotics and host immune responses [ 34 ] and may provide E . coli with a protective advantage over other more sensitive bacteria that compete for the same resources in the gut including SCFA-producers, which were less abundant in women with T1D. In the second category (taxa that became differentially abundant by trimester 3), we observed that three species from the Bacteroidales order, B . caccae , B . uniformis and B . vulgatus , were increased in women with T1D. The genus Bacteroides was reported to be more abundant in children with islet autoimmunity compared to healthy controls [ 11 , 12 ]. Bacteria from the Bacteroides ( B . caccae , B . vulgatus , B . uniformis , B . dorei , B . fragilis , B . faecis , B . finegoldii , B . thetaiotaomicron , B . xylanisolvens , B . stercoris and B . ovatus ) and Alistipes ( A . finegoldii , A . onderdonkii and A . shahii ) genera, all belonging to the Bacteriodales order, formed part of the LPS bacterial cluster which was enriched in women with T1D especially by trimester 3. Twenty-nine functional features related to the production of SCFAs were decreased in women with T1D. The SCFAs bacterial cluster was composed of Faecalibacterium prausnitzii , Eubacterium rectale , Anaerostipes hadrus , Lachnospiraceae _ bacterium 5 1 63FAA , Ruminococcus torques , Roseburia intestinalis and Roseburia inulinivorans all of which belong to the class Clostridia and are major butyrate producers [ 35 ]. Butyrate prevents gut inflammation and promotes gut barrier function [ 19 ]. In addition, the enzyme beta- N -acetylhexosaminidase (K01207), which degrades mucin [ 36 ], contributed by Eubacterium rectale , E . siraeum , Ruminococcus bromii , Bifidobacterium adolescentis , B . bifidum and Roseburia intestinalis , was decreased in women with T1D, but only in trimester 2. Degradation of mucins produces oligosaccharides, and acetate and propionate, which together then stimulate mucus production and enhance epithelial integrity [ 37 ], preventing ‘gut leakiness’ and translocation of toxins and dietary antigens into the systemic circulation. Because mucin degradation was lower in women with T1D, the stimulus to mucin production would be less. This would also be contributed to by the lower abundance of butyrate-producing bacteria observed in women with T1D. Thus, the gut microbiome in women with T1D exhibits pro-inflammatory features likely to be associated with low-grade systemic inflammation. Women with T1D bacterial functions associated with vitamin K2 (menaquinone; MK-7) synthesis were increased and those associated with synthesis of the B-group vitamins B6 (pyridoxine) and B12 (cobalamine) were decreased. Mammals cannot synthesize these vitamins and must acquire them from the diet or gut microorganisms [ 38 ]. A small study based on metagenomic sequencing of the gut microbiome [ 39 ] observed a similar increase in the vitamin K2 superpathway (PWY-5838) in people with type 2 diabetes. Vitamin K2 is required for blood clotting and bone health [ 40 ] but why its synthesis by gut bacteria is increased in diabetes is unclear. All B-group vitamins contribute to regulation of immunity-inflammation and their deficiency has been associated with inflammatory disorders [ 41 ]. Vitamin B6 deficiency has been associated with inflammatory markers in population-based studies [ 42 ] and is reported to be common in T1D [ 43 , 44 ]. Of interest therefore, we found that the key enzyme in B6 synthesis, pyridoxal 5′-phosphate synthase (K06215), was decreased across pregnancy in women with T1D. Vitamin B12 has several anti-oxidant properties [ 45 ] and is required for conversion of succinate to propionate by Prevotella [ 46 ]. The majority of women with and without T1D reported taking multi-B group vitamins from early pregnancy (Table 1 ) and in the third trimester plasma B6 and serum B12 did not differ significantly between women with and without T1D (Table 1 ). Nevertheless, the relative deficiency of these vitamin-synthesizing gut bacteria in women with T1D could contribute to other alterations in the gut microbiome, underscoring the importance of dietary supplementation in this group of women. Our findings reveal that the composition of the gut microbiome not only changes across pregnancy but in a distinct way in women with T1D. By the third trimester, women with T1D exhibited a more pro-inflammatory and catabolic gut microbiome profile, reflected by an increase in LPS-producing bacteria and a decrease in SCFA-producing bacteria. These changes may account for the increase in calprotectin (marker of gut inflammation) and I-FABP (marker of gut epithelial integrity) we observed in women with T1D, known to be associated with impaired epithelial barrier function and leakage of LPS and other bacterial products leading to low-grade systemic inflammation. We suggest that systemic inflammation secondary to changes in the gut microbiome in T1D may contribute to the increased risk of pregnancy complications in T1D. Furthermore, a pro-inflammatory gut microbiome in the mother may impact the infant postnatally. In an elegant study in mice, Aguero et al. [ 47 ] found that transient exposure to an auxotrophic E . coli mutant in the intestine of germ-free mothers in pregnancy accentuated innate immune development in the intestine of their germ-free offspring. This effect was mediated by the transfer, in part via maternal antibodies, of a range of E . coli products across the placenta and in the mother’s serum and milk. Thus, with a single gut bacterium, the mother primed the immune system of her offspring, before their exposure to the external environment [ 47 ]. If T1D mothers with an increased abundance of E . coli and other LPS-producing gut bacteria better prime innate immunity in their offspring, this could protect against potentially diabetogenic infections in early life [ 48 ] and account for the lower risk of T1D in infants with a maternal compared to paternal proband [ 49 ]. Conclusions The gut microbiome changes across pregnancy but these changes are distinct in women with T1D. They include an increase in bacteria with pro-inflammatory properties, a decrease in bacteria with anti-inflammatory properties and a decrease in bacteria that synthesize essential vitamins, which together may lead to low-grade gut inflammation, epithelial barrier dysfunction, increased epithelial permeability and low-grade systemic inflammation. These are features of the gut microbiome ‘dysbiosis’ observed in a wide range of diseases [ 50 ], some of which have shown clinical benefit in response to probiotic and other dietary interventions. The relationship of these changes to the increased risk to mother and fetus in the T1D pregnancy requires further investigation. Intervention by dietary means to promote a less pro-inflammatory gut microbiome could potentially benefit both mother and fetus. Methods Participants and study design This study involved 70 pregnancies in women participating in the ENDIA pregnancy-birth cohort study [ 23 ], 36 in women with established T1D on daily insulin treatment and 34 in healthy women with no history of gestational diabetes. The main criterion for participation in ENDIA was an unborn child with a first-degree relative with T1D. Table 1 provides summary statistics for participants, on 66 women with 70 pregnancies (four sibling pairs). Therefore, four women were included twice in the study population (each with two pregnancies). The unit of observation is the pregnancy , and therefore observations from the same mother but different pregnancies have been included as separate observations, as characteristics might change between pregnancies. Trimesters were categorised according to gestational age: T1 0–99 days; T2 100–196; T3 197–274. Women provided written informed consent and were enrolled into the study between 2013 and 2016 at one of eight clinical sites. Up to three study visits occurred during pregnancy, ideally one in each trimester. The study was approved by a Human Research Ethics Committee (HREC) at each clinical site, with the Women’s and Children’s Health Network HREC in Adelaide acting as the lead HREC under the Australian National Mutual Acceptance Scheme (reference number HREC/16/WCHN/066). ENDIA is an observational study registered on the Australia New Zealand Clinical Trials Registry (ACTRN1261300794707). Maternal and paternal demographics, medical history, past-pregnancy history, pre-pregnancy weight, assisted conception status and plurality of pregnancy were recorded at the first opportunity. Standardized questionnaires were offered at each pregnancy visit to record pregnancy complications, antibiotic and supplement usage, maternal and household smoking, household composition and pet ownership. Maternal diet during pregnancy was measured at the third trimester visit using a validated 74 item food frequency questionnaire, Dietary Questionnaire for Epidemiological Studies version 2 (DQESv2) [ 51 ]. Even though this was administered only in the third trimester, evidence for stability of dietary intake over the course of the pregnancy was obtained from a separate, purpose-built ENDIA Pregnancy Lifestyle Questionnaire administered before each of the three study visits during pregnancy. This assessed consumption of milk (dairy and non-dairy), caffeinated and decaffeinated tea and coffee, caffeine-containing soft drinks, dairy products, soy, gluten containing cereals (wheat, barley and rye) and non-gluten containing cereals (rice, corn and oats). Analysis across the study visits revealed that on 86% of occasions respondents reported either the same unit or within one-unit difference between visits 1–2, visits 1–3 and visits 2–3. Magnitude changes of four or five units were reported on < 2% of occasions. This supports the DQESv2 as being reflective of the whole pregnancy period. Women were advised to take multi B-group vitamin supplements from as early as possible in pregnancy. Sample collections and analyses Serum 1,5-AG, a measure of glucose control in pregnancy [ 25 ], was measured by GlycoMark (Nippon Kayaku Co. Ltd., New York, NY, USA) in a single batch. Serum vitamin D3 was measured with a Liaison Analyser by the DiaSorin method (DiaSorin, Turin, Italy). Plasma vitamin B6 was measured by the Chromsystems HPLC-based assay (Chromsystems Instruments & Chemicals, Gräfelfing, Germany). Serum vitamin B12 was measured by the Abbot Architect Chemiluminescent Microparticle Immunoassay. Fecal samples were collected in accordance with our validated collection-processing-storage method [ 52 ]. Briefly, samples were captured in a toilet using the Easy Sampler device (Co-Vertec Ltd, Waterlooville, UK) then transferred into a sterile 70 mL collection jar. Participants were instructed to store the sample in the refrigerator prior to transport to the laboratory in an insulated bag within 24 h. Samples were divided into aliquots with a sterile spatula in a Biosafety Level 2 cabinet, then stored at − 80°C. A total of 134 fecal samples were collected from the 70 pregnancies with either two or three samples collected longitudinally in each pregnancy (Fig. 1 ). DNA was extracted from fecal samples at the Walter and Eliza Hall Institute of Medical Research (WEHI) with the MoBio PowerSoil kit (MoBio Laboratories, Carlsbad, CA) as per manufacturer’s instructions. Fecal calprotectin (micrograms per kg) was measured in 31 and 26 samples collected in the third trimester from women with and without T1D, respectively, by quantitative, enzyme-linked immunoassay (CALPROTMOslo, Norway) according to the manufacturer’s instructions. Human serum intestinal fatty acid binding protein (I-FABP) (picograms/mL) was measured in 26 and 26 samples collected in the third trimester from women with and without T1D, respectively, by a commercial ELISA kit (Enzyme-Linked Immunosorbent Assay; Hycult Biotech, the Netherlands) according to the manufacturer’s instructions. HLA DR typing was performed on DNA in saliva collected with OG-500 Oragene DNA tubes (DNA Genotek, Ontario, Canada) by TaqMan-based PCR-typing and imputation from three single-nucleotide polymorphisms (rs3104413, rs2187668 and rs9275495), as described previously [ 53 ]. Whole metagenome sequencing and generation of taxonomic and functional profiles Whole metagenome sequencing (WMS) libraries were generated and sequenced with the 2 × 150 bp paired-end chemistry on two separate runs of an Illumina NovaSeq 6000 (Illumina, San Diego, California, USA) sequencer at the Ramaciotti Centre for Genomics (UNSW, Sydney, Australia. ). Sequencing data were quality controlled, and reads aligning to the human genome were removed using KneadData (v0.6.1) [ 24 ]. For the functional analysis, filtered reads classified using Kraken2 with the standard database [ 54 ], were further processed using HUMAnN2 (v0.11.1) [ 55 ] with the UniRef90 database to generate functional annotations (i.e. genes and metabolic pathways) and define the metabolic potential of the microbial communities. A functional profile (i.e. function-per-sample counts matrix) for metaCyc [ 56 ] complete metabolic pathways was obtained. In addition, two functional profiles were generated by grouping genes into KO [ 57 ] and MetaCyc-reactions functional categories using the humann2_regroup_table command. As part of the HUMAnN2 pipeline, MetaPhLan2 (v 2.7.5) [ 58 ] was used on the reads filtered with KneadData, to detect and quantify individual species with a library of clade-specific markers (ChocoPhlAn database) and generate whole-metagenome-based profiles at strain, species, genus, family, order and phylum taxonomic levels. Taxonomic and functional profiles were imported into the phyloseq [ 59 ] package in R [ 60 ]. An abundance filter was applied to remove all taxa and functional categories with a relative abundance across all samples of < 0.01%. Alpha diversity (diversity within microbial communities) was obtained from the number of observed taxa (richness) using the function estimate richness from the R package phyloseq. Beta diversity (diversity between microbial communities) was determined with phyloseq (function distance, method=‘bray’) on proportional log transformed data. This function calculates Bray-Curtis coefficients, which measure the distance between communities based on the taxa/functions that they contain and their abundances. The data were visualised using principal coordinates analysis (PCoA) plots in phyloseq. Statistical analyses For continuous responses, where appropriate, the summary tables present mean and standard deviation derived from fitting a linear mixed model. The model fit for each continuous response adjusts for the fact that the observations from women with more than one pregnancy are not fully independent but may be correlated. For some response variables, the assumption of normally distributed residuals was not met. In these analyses, the response variable was transformed using a square root or log transformation, as appropriate. For transformed responses, the back transformed means and approximated standard deviations are presented. A Wald’s test is used to determine whether the groups are significantly different. For categorical responses, summary tables show numbers and percentages. The percentage was calculated using the total number of pregnancies or samples as the denominator. To determine whether the distribution of observations between groups for categorical data were similar or not, a generalized linear mixed model was fitted, with a random effect for woman (i.e. subject). Such models adjust for correlated women observations. To determine whether groups were significantly different, the change in deviation of the final mode (i.e. a likelihood ratio test), which includes and excludes the treatment term, was examined. A pre-set P value of 0.05 was used as a cut-off for determining statistical significance for all models. Data analysis was performed in R (R Core Team, 2018 [ 61 ];), with the R packages lme4 v1.1.21 [ 62 ], car v3.0.3 [ 63 ], predictmeans v1.0.4 [ 64 ] and nnet v7.3.12 [ 65 ]. For testing differences in alpha diversity between groups of interest, GEEs [ 66 ] were applied using the R function geeglm from package geepack v1.2-1 ([ 67 ]; parameter family set to default ‘Gaussian’) to account for possible correlation of multiple measurements within a woman over time. The default empirical (robust or ‘sandwich’) estimator was used to ensure that estimates are robust to misspecification of the correlation structure. The model used for the regression included T1D status and time (i.e. two models were tested considering time as a continuous [gestational days] or categorical [trimesters] variable) as well as their interaction term (T1D × days or trimester; to test if differences in alpha diversity between women with and without T1D change across days or trimesters) and was adjusted for sample processing batches (which includes sequencing run), conception age, BMI, parity and HLA type. Mean-centred values were used for gestational days, conception age and BMI to ensure that the model coefficients are meaningful. Differences in beta diversity were evaluated by PERMANOVA using Bray-Curtis dissimilarities with the Adonis function from the vegan [ 68 ] R package. For tests that included multiple samples across trimesters from the same participant (i.e. longitudinal analysis), a modified version of Adonis, which performs a RMA-PERMANOVA test [ 32 ], was employed. This statistical model included T1D status and time with their interaction adjusted as in the alpha diversity model (i.e. adjusting for sequencing run, conception age, BMI, parity and HLA type). In addition, interactions between time and other factors were also tested as described in the results section. When an interaction was significant (i.e. FDR < 0.1), statistical analysis was performed within trimester (i.e. when testing for differences between women with and without T1D) or by separating data from women with and without T1D (i.e. when testing for differences in time). Differential abundance of taxa from MetaPhlan2 and gene categories and metabolic pathways from HUMAnN2 was analysed with the R package limma [ 69 ]. First, taxonomic relative abundances from MetaPhlan were multiplied by the library size of each sample, whereas for the functional analysis of data generated with HUMAnN2 counts (CPM) were used. Taxonomic and functional data were filtered using the filterByExpr function with default parameters with an additional general abundance filter that removed all those taxa or functions with a relative abundance across all samples of < 0.01% and a prevalence filter that removed those taxa present in less than 33 samples (i.e. ~ 25% of the samples). Library sizes were normalized using the trimmed mean of log expression ratios with singleton pairing (TMMwsp) method [ 70 ] in edgeR which is expected to perform better for data with a high proportion of zeros. Counts were transformed to log2-counts per million (logCPM), voom precision weights were calculated and limma linear models were fitted while allowing for loss of residual degrees of freedom due to exact zeros using the voomLmFit function [ 71 ] [ 72 ]. Here, ‘women IDs’ were considered as blocks to calculate the consensus correlation and account for multiple measurements while estimating contrasts statistics using the contrasts.fit function and empirical Bayes moderated t statistics. Since we have samples for all possible combinations of T1D status and trimester, this is a factorial design. Therefore, in order to build our model, factors T1D status and trimester were combined into a single factor with six levels and the comparisons of interests were defined as contrasts. In addition, the model design included an adjustment for sample processing batches, conception age, BMI, parity and HLA type. For this, groups means were computed by mean-correcting covariates and factors before performing the test: for numerical covariates, the mean was subtracted and for factors the contr.sum function was used (e.g. contrasts(ExtractionBatch) <- contr.sum(levels(ExtractionBatch). The mean abundance of each taxon in each group is computed by limma when fitting the model and are contained in the coefficients component (fit$coef). The standard errors are obtained as following: fit$stdev.unscaled[,] * sqrt(fitc$s2.post). Note that after normalization, the data are on the log2 scale. For testing for differences in specific taxa due to HLA type, the limma model included, similarly to the previous model, ‘women IDs’ as blocks and adjustments for sample processing batches, conception age, BMI and parity. Four contrasts were fitted: (1) High (DR34) vs. low (DRXX, DR3X and DR4X) risk HLA types, (2) DR34 vs. DR3X and DR4X grouped into a factor, (3) DRXX vs. DR3X and DR4X grouped into a factor and (4) DR34 vs. DRXX. P values were adjusted with the Benjamini and Hochberg method to control the FDR. FDR < 0.1 were considered significant. Taxa or functions significantly different with an abundance logFC greater than 0.5 or less than − 0.5 and present in at least 50% of the samples in either of the groups being compared were regarded as biologically significant. For identifying the ‘principal bacterial contributors’ to each differentially abundant function, first, the HUMAnN2-generated files with functions (Kegg orthology, MetaCyc reaction and complete pathways) stratified by contributing species were obtained. Next, the functions of interest with contributing species were disaggregated into individual files using grep. Finally, limma was applied as explained above to each subset of function with contributing species and only species with a larger log2-FC in the group of interest for the specific function were considered principal contributors belonging to the same bacterial cluster. The significance of the difference between measurements of serum markers was tested with a Wilcoxon rank sum test equivalent to the Mann-Whitney test using the wilcox.test function from the stats R package with parameter paired set to FALSE. Real-time quantitative PCR analysis The qPCR reaction comprised 10 μL Sybr Green GoTaq qPCR Master Mix (2×) (Promega), 0.3 mM of each primer, 8.4 μL of water and 1 ng of DNA in 20 μL. Assays were performed in triplicate using the QuantStudio 12K Flex Real-Time PCR System (Thermofisher) with the following protocol: one cycle at 95 °C for 10 min, followed by 40 cycles of a two-stage temperature profile at 95 °C for 15 s and 60 °C for 1 min. Primers were Bacteroides vulgatus (BV-1) 5′-GCATCATGAGTCCGCATGTTC-3′, BV-2 5′-TCCATACCCGACTTTATTCCTT-3′; Bacteroides caccae (BaCA-1) 5′-GGGCATCAGTTTGTTTGCTT-3′, BaCA-2 5′-GAACGCATCCCCATCTCATA-3′; universal 16S V4 primers Univ-1 5′-GTGYCAGCMGCCGCGGTAA-3′, Univ-2 5′-GGACTACNVGGGTWTCTAAT-3′. Standard curves were generated by 2-fold dilutions ranging from 10 to 0.02 ng of a pooled human fecal DNA. Data from each triplicate fell within a 0.5 threshold cycle (Ct); outliers (> 1 standard deviation) were removed before calculating the average Ct of each sample. Amplification efficiency (E) was determined from the slope of the standard curves for each primer pair using the formula E = (10−1/slope)-1. Efficiencies ranged from 97 to 102%. The abundances ( N ) for Bacteroides vulgatus and Bacteroides caccae were determined relative to the total bacterial load measured with the universal 16S primers, where N ( B . vulgatus ) = (Efficiency_ B . vulgatus + 1) Ct_ B . vulgatus , N ( B . caccae ) = (Efficiency_ B . caccae + 1) Ct_ B . caccae , N universal = (Efficiency_16S universal + 1) Ct_Universal , B . vulgatus relative abundance = N ( B . vulgatus )/ N Universal, B . caccae _relative abundance = N ( B . caccae )/ N Universal. Relative abundances were log10-transformed and used as input for the regression models. The association between the relative abundance of Bacteroides caccae and Bacteroides vulgatus and T1D status was determined using a linear mixed effects model (lmer) with conception age, BMI, parity and HLA type as fixed effects, and ‘woman ID’ and processing batches as random effects. Availability of data and materials The demultiplexed raw datasets supporting the conclusions of this study can be accessed in the NCBI SRA (project number PRJNA604850). All the python commands used to run HUMAnN2 and the R code used to perform statistical analyses are available at GitHub ( ) as R markdown coding and knitr html files along with the necessary R objects which contain taxonomic and functional profiles with metadata. Abbreviations AG: Anhydroglucitrol ASV: Amplicon sequence variants BMI: Body mass index BCAA: Branched chain amino acid CoA: Coenzyme A CPM: Counts per million ENDIA: Environmental determinants of islet autoimmunity FDR: False discovery rate GEE: Generalized estimating equations HLA: Human leukocyte antigen HPLC: High-performance liquid chromatography HREC: Human research ethics committee KO: Kegg orthology LCBD: Local contribution to beta diversity LogFC: Log2 fold-change LPS: Lipopolysaccharide NOD: Non-obese diabetic OTU: Operational taxonomic unit PCoA: Principal components analysis PWY: Pathway (Metacyc) RMA-PERMANOVA: Repeated measure-aware permutation analysis of variance SCFA: Short-chain fatty acid SD: Standard deviation T: Trimester TCA: Tricarboxylic acid T1D: Type 1 diabetes TMM: Trimmed mean of log expression ratios WEHI: Walter and Eliza Hall Institute of Medical Research WMS: Whole metagenomic sequencing

A recent study by WEHI and ENDIA has found type 1 diabetes is associated with changes in the gut microbiome during pregnancy and could contribute to complications in both the mother and baby. The research found a link between type 1 diabetes and changes in the gut microbiome that are associated with intestinal inflammation, which could account for the increase in pregnancy complications in women with the condition. The study revealed pregnant women with type 1 diabetes had a decrease in "good" gut bacteria that normally protect against inflammation and an increase in 'bad' gut bacteria that promote intestinal inflammation and damage to the intestinal lining. These changes could contribute to the increased risk of pregnancy complications seen in women with type 1 diabetes and could potentially be modified by dietary changes. Gut health linked to pregnancy complications The research was part of the ENDIA (Environmental Determinants of Islet Autoimmunity) study, investigating genetic and environmental factors that may contribute to the development of islet autoimmunity and type 1 diabetes in children. The observational study has recruited 1500 babies from pregnancy who have an immediate relative with type 1 diabetes and is following them through childhood. WEHI clinician-scientist Professor Len Harrison, who led the research, said women with type 1 diabetes have a higher frequency of complications in pregnancy. "We decided to study the gut microbiome because there was evidence of systemic and intra-uterine inflammation in pregnancy for women with type 1 diabetes that could conceivably be related," he said. Together with colleagues in bioinformatics at WEHI, as well as ENDIA partners, the study team undertook whole genome sequence analysis of samples from pregnant women. The samples were taken from groups of women at different stages of pregnancy. "In women with type 1 diabetes, we observed changes in their gut microbiome, including a decrease in 'good' gut bacteria and an increase in 'bad' gut bacteria," he said. "The 'good' bacteria make substances that prevent inflammation and the 'bad' bacteria release substances that activate the immune system to trigger inflammation." "We are now investigating if these changes are linked to the higher rate of complications during pregnancy in women with type 1 diabetes." Professor Harrison said changes in the gut microbiome of the mother could have implications for the pregnancy and for the health of the baby. "It is also possible that the changes we observed in the mothers with type 1 diabetes might have a lasting influence on the baby which continues after birth," he said. Changing diet Professor Harrison said the next stage of the project was to identify markers that would determine which women with type 1 diabetes might benefit from safe interventions during pregnancy, including dietary changes. "We believe that if these women made some safe dietary modifications it could help to restore the health of their microbiome and lower their risk of complications during pregnancy. This is what we are investigating now," he said. "We will also look into the immune system of their babies at birth to see what impacts that impaired maternal gut health has on the baby after birth."

10.1186/s40168-021-01104-y

Biology

Geneticists track the evolution of parenting

Christopher B. Cunningham et al. Ethological principles predict the neuropeptides co-opted to influence parenting, Nature Communications (2017). DOI: 10.1038/ncomms14225 Journal information: Nature Communications

http://dx.doi.org/10.1038/ncomms14225

https://phys.org/news/2017-02-geneticists-track-evolution-parenting.html

Abstract Ethologists predicted that parental care evolves by modifying behavioural precursors in the asocial ancestor. As a corollary, we predict that the evolved mechanistic changes reside in genetic pathways underlying these traits. Here we test our hypothesis in female burying beetles, Nicrophorus vespilloides , an insect where caring adults regurgitate food to begging, dependent offspring. We quantify neuropeptide abundance in brains collected from three behavioural states: solitary virgins, individuals actively parenting or post-parenting solitary adults and quantify 133 peptides belonging to 18 neuropeptides. Eight neuropeptides differ in abundance in one or more states, with increased abundance during parenting in seven. None of these eight neuropeptides have been associated with parental care previously, but all have roles in predicted behavioural precursors for parenting. Our study supports the hypothesis that predictable traits and pathways are targets of selection during the evolution of parenting and suggests additional candidate neuropeptides to study in the context of parenting. Introduction The selective pressures that lead to the evolution of parental care are well documented 1 . Parental care typically evolves to minimize unusually stressful or hazardous environments for offspring 2 , 3 , 4 . Although this hypothesis is widely supported 4 , parental care is not the only evolutionary solution to adverse conditions. Moreover, it may not be the most likely response as the evolution of parenting reflects changes in multiple behavioural inputs, involving many pathways 5 . At a minimum, the evolutionary transition from asociality (social interactions limited to mating) to subsociality (extensive social interactions between parents and offspring involving parental care) is predicted to require modification of several contributing behaviours including tendencies for dispersal, feeding, mating, aggression and tolerance of social interactions 1 , 2 , 3 . Caring parents no longer disperse from a mating site, they provision food rather than feed themselves, they pause reproduction and mating, they show aggression to protect offspring and shared resources rather than their own resources, and they tolerate the presence of others and increased social interactions 1 , 2 , 3 , 4 , 6 . The early ethological literature therefore predicts that parental care evolves only when there are suitable behavioural and ecological precursors present within the evolutionary ancestor, such as nest building, defensive postures and aggression, and potentially shared resources 2 , 3 . The early predictions of the specific constituent behaviours were made without reference to the mechanistic changes that would be required. Implied, however, is that repurposing existing traits involves changes in the timing and direction of interactions. This suggests a potential mechanism: that changes in timing of gene expression are involved in the evolution of derived behaviour 1 . If true, we can use the predicted behaviours to be modified from non-parenting to parenting to infer the underlying mechanisms. We specifically hypothesize that modifying behaviours affecting parenting will result from altered gene expression rather than the evolution of novel genes. This hypothesis is a natural extension of Wright’s theory of nearly universal pleiotropy 7 , suggesting that genes gain functions when used in novel contexts, and the ubiquity of changes of gene regulation that are seen during evolutionary transisitions 8 , 9 , 10 . Moreover, this is consistent with previous work that uses the nature of selection to predict the genetic changes underlying the evolution of social behaviour 1 , 6 , 11 , 12 . For example, when parenting, animals are typically selected to be unreceptive to mating. We therefore predict that mechanisms that influence mating will be altered. Following this logic, overall we predict that parenting will involve changes in expression of genes that influence mating, feeding, aggression, and increased tolerance for social interactions as these are the behaviours modified as lineages evolve from asocial to subsocial 1 , 2 , 3 . However, following Wright 7 , we also predict polygenic changes rather than one or few genes. Thus, we need to use techniques that can detect multiple changes. Studies of changes in gene expression have been revolutionized by an ability to assay overall transcriptional changes in many genes simultaneously. However, transcriptomics is not a particularly powerful method for identifying changes in expression of many genes that code for proteins that influence behaviour, such as neuropeptides, that have low gene expression 13 , highly restricted sites of release 14 , and can be hard to detect with transcriptomic studies that are not highly tissue specific 15 . Proteomics provides a complementary approach that overcomes some of these limitations and provides a method to target protein categories of interest. Adopting a complementary approach is necessary because neuropeptides strongly influence the social behaviour of animals 16 and many neuropeptides are likely to be associated with parenting. One of the most studied neuropeptides, oxytocin, is necessary for parenting across the animal kingdom 17 . There is a causal relationship between the neuropeptide galinin and parental care in mice 18 . In the burying beetle Nicrophorus vespilloides , neuropeptide F receptor is differentially expressed between parenting and non-parenting states 19 . The importance of neuropeptides is expected, as parenting individuals must undergo many rapid shifts in behaviour. Neuropeptides can exhibit their influence within minutes, have highly localized effects targeting very select neural circuits, or have highly widespread effects targeting many and diffuse neural circuits 14 . Here we test the hypothesis that a transition from a non-parenting state to a parenting state will reflect differences in neuropeptides known to be generally associated with mating, feeding, aggression and increased social tolerance and that neuropeptides influencing other traits will not change during parenting. To test this, we estimated the abundances of neuropeptides of the burying beetles N. vespilloides sampled from a solitary and parenting states. N. vespilloides adults are normally solitary but switch to parenting in the presence of appropriate resources. Parenting in this species is extensive and elaborate. Adult beetles are stimulated to parent after they locate a vertebrate carcass and bury it. Parents remain on this carcass and provide indirect care by removing the fur or feathers and forming a nest within the carcass. They also repeatedly coat the carcass with excretions that retard microbial growth. Eggs are laid in the surrounding soil, hatch and larvae crawl to the crypt where they interact with one or both parents. Direct parental care involves feeding larvae pre-digested carrion by regurgitation for the first two days of larval life ( Fig. 1 ). Parenting occurs for 75% of larval development, yet lasts only days 20 . N. vespilloides is also molecularly tractable with a published genome 21 , allowing for efficient proteomic work. Figure 1: A female burying beetle feeding her offspring. In this species, a parent spends around 72 h preparing a carcass, after which larvae hatch and arrive at the carcass. Once larvae arrive, parents spend a further 72 h feeding larvae (with peak parenting 12–24 h after larval arrival), and then disperse around 100 h after larvae first arrive on the carcass. Larvae disperse fully grown around 125 h after their arrival on the carcass. As shown here, feeding involves direct mouth-to-mouth contact and a transfer of pre-digested carrion from the parent to the offspring. Photograph by AJM. Full size image In this study we first identified peptides and neuropeptides from the brains of adult female N. vespilloides collected during three different behavioural and social states: virgin and solitary, actively caring and post-caring and solitary. We next examined abundances of neuropeptides in the different states and found that 8 of 18 changed in abundance in at least one state, with 7 increasing in abundance during parenting. Consistent with our hypothesis, these 7 are known to function in pathways of the behaviours that ethologists predicted change during the evolution of parenting: feeding, mating, aggression and social tolerance. Importantly, no neuropeptides that have functions outside of these behaviours changed in abundance. Our work supports the notion that ethological principles can be used to a priori identify candidate genetic pathways and molecules that influence complex behaviours. Results Identification of neuropeptides in Nicrophorus vespilloides We identified 133 peptides in the brains of N. vespilloides belonging to neuropeptide proteins ( Supplementary Table 1 ). We found very few peptides identified in one state but not others. Specifically, actively parenting individuals exclusively displayed two peptides from FMRFamide (FMRFa; DKGHFLRF and GDLPANYEMEEGYDRPT) and a single peptide from Neuropeptide-like 1 (NPLP-1; KESYDDDYYRMAAF). No Apis -NVP-like (NVP) peptides of the sequence FLNGPTRNNYYTLSELLGAAQQEQNVPLYQRYVL were found in actively parenting samples. These 133 peptides allowed us to identify 18 neuropeptide proteins that were present in at least one behavioural state (virgins, actively parenting and post-parenting individuals; Supplementary Table 1 ). Twelve were represented in all three behavioural states, while pheromone biosynthesis activating neuropeptide (PBAN), short neuropeptide F (sNPF) and natalisin (NTL) were absent in post-parenting individuals, diuretic hormone 47 (DH 47 ) was only found in actively parenting individuals, and crustean cardioactive peptide (CCAP) was only found in post-parenting individuals. Virgins showed a higher level of variability than the other two behavioural states. Ion transport peptide (ITP) was detected in a single biological replicate (a virgin), and is therefore not included in any further analyses. Changes in neuropeoptides associated with parenting Having defined these neuropeptides, we tested for changes in the overall abundances of all neuropeptides across behavioural states using a multivariate analysis of variance (ANOVA). We found statistically significant differences in the overall abundance between the states ( F 2,9 =28.476; P =0.0001; Fig. 2 , Table 1 ). The nature of this multivariate difference is best illustrated by creating linear combinations of the neuropeptides with principle components analysis. Five principle components had eigenvalues greater than 1 and are presented in Table 1 . The first PC describes overall abundance and explains 41% of the variance, although SIFamide (SIFa) contributed very little to this vector and CCAP, consistent with being present in only one state, was opposite in sign. The second principle component explains 20% of the variance and, using a criterion of loadings of 0.3 or greater 22 , contrasts changes in SIFa with myoinhibiting peptide (MIP) and RYamide (RYa). No obvious interpretation is suggested to us by this pattern. The remaining PCs describe 10% or less of the variance, and present no obvious interpretation of the contrasting loading for each neuropeptide. Consistent with the majority of the differences arising due to overall abundance, Fig. 2 illustrates that the three behavioural states separate primarily along the PC1 axis. Figure 2: Principal component analysis of neuropeptide relative abundances. Graph of the association between abundances and three non-parenting and parenting behavioural states of Nicrophorus vespilloides (red: virgin, black: actively parenting, blue: post-parenting). Principal component analysis (PCA) based on four biological replicates of each behavioural state, with eight individual brains pooled to form a biological sample. Ellipses show the 95% confidence area of each group. Full size image Table 1 Principal component analysis (PCA) of neuropeptide abundance of virgins, actively caring and post-caring Nicrophorus vespilloides females. Full size table We followed these multivariate analyses of variance (ANOVAs)with a priori defined univariate comparisons to examine how the relative abundances of specific neuropeptides changed. The results of individual ANOVAs are presented in Supplementary Table 1 and here we describe the Tukey–Kramer honest significant difference post hoc pairwise of the behavioural states where the overall ANOVA was significant. In general, when neuropeptide abundance changed, it increased within actively parenting individuals. NPLP-1 had higher abundance in actively parenting compared with post-parenting ( P =0.0063), as did tachykinin (TK) ( P =0.020), FMRFa ( P =0.0023), sulfakinin (SK) ( P =0.0087), PBAN ( P =0.023), NVP ( P =0.043) and NTL ( P =0.044). FMRFa was also more abundant when individuals were actively parenting compared with virgins ( P =0.011). Sulfakinin (SK) had higher abundance in virgins ( P =0.026) compared with post-parenting, as did NTL ( P =0.018). CCAP differed from all others in having higher abundance in post-parenting compared with either virgins ( P =0.046) or actively parenting ( P =0.046). Although not reaching the level of conventional statistical significance, two neuropeptides that showed a strong trend toward differential expression were RYa ( F 2,9 =4.033, P =0.056) and myosuppressin (MYO; F 2,9 =3.611, P =0.071). Both were most highly expressed in actively parenting individuals. The remaining neuropeptides showed no strong trends ( Supplementary Table 1 ). Discussion Our goal was to test the prediction that the mechanisms involved in the evolution of parental care reside in pathways reflecting behavioural precursors predicted by ethological principles. Ethologists predict that parenting involves modification of pre-existing behavioural traits including mating, feeding, aggression, resource defence and social tolerance 2 , 3 . We suggest that this implies that the mechanistic underpinnings of these behaviours must also be altered. Specifically, we predicted that the timing of expression of neuropeptides associated with the behaviours will be altered, and that this will be reflected in the abundance of neuropeptides that influence mating, feeding, aggression and social tolerance in different behavioural states. To test this, we examined peptide abundance, with the prediction that the neuropeptides that have changed abundance during parenting function in feeding, mating, aggression and social interactions in organisms that do not display parental care. We profiled these changes from brains of the burying beetle N. vespilloides , which provides direct care by regurgitating food to dependent offspring. We identified 18 neuropeptides in the brain of N. vespilloides , which is consistent with other studies of non-model organisms 23 , 24 , 25 . Of these, the abundance of eight neuropeptides changed during parenting, all but one increasing during parenting. Although this is not evidence of causality, it is a strong correlation consistent with a priori predictions. Although conclusive evidence for our hypothesis requires functional manipulations or comparative analyses, these results support our initial prediction derived from how behaviour evolves. Parenting across species typically involves a pause of mating, feeding others, appropriately directed aggression for defence and social interactions 2 , 3 , 4 . If our predictions are correct, then the neuropeptides that are more abundant when parenting will function in these behaviours in other taxa. Moreover, the neuropeptides that do not show changes in abundance should not have known functions in these behaviours. The eight neuropeptides that differed in abundance during parenting ( Supplementary Table 1 ) support our prediction. In other taxa, both FMRFa 26 , NTL 27 and SK 28 influence mating. Feeding behaviour and food intake are influenced by NVP 24 and SK 28 , 29 . Aggression and resource defence are influenced by TK 30 , 31 and SK 28 . NPLP-1 (ref. 32 ), TK 23 , 32 and PBAN 33 all influence social interactions. Of the 11 neuropeptides that were not differentially expressed, many have poorly understood functions (for example, ITG, RYa, MIP, MYO 34 , 35 , 36 ) or function outside the predicted pathways (CCAP, DH 31 , DH 47 , IDL, ITP 36 ). Three of these neuropeptides have the potential to function in the predicted pathways were sNPF 34 , 36 , and ITG-like 24 , which influences feeding, and SIFa 34 , 37 , which influences mating. Critically, there were no results consistent with the null hypothesis that pathways are unpredictable; none of the differentially expressed neuropeptides we identified in this study function solely outside the predicted pathways. Thus, although we do not identify every known neuropeptide, those that we can identify fit our prediction. As a corollary to our predictions, our results support the idea that like candidate gene studies 12 , hypotheses about the pathways that are co-opted are likely to be more robust than hypotheses about specific neuropeptides when examining analogous behaviour in novel species. Our study suggests three areas for further consideration to understand the mechanisms underlying parental care. First, we suggest that knowing or predicting the behavioural modules that provide the substrate for behavioural evolution provides insights into proximate mechanisms by also providing predicted pathways. Here we associated changes in protein abundance, but gene expression changes are also potentially predictable using this logic 19 , 38 . This can be tested further in other behaviours where the selective pressures and targets are known and therefore the underlying behavioural traits that are predicted to change can be identified a priori. Second, we provide information about specific neuropeptides that appear to underpin parental care and these can be examined in other subsocial organisms. Comparative studies will help us move beyond correlations. Finally, by specifying the behavioural and genetic pathways expected to be co-opted when parenting evolves, we can then identify particularly influential molecules that deserve further examination in N. vespilloides . Functional studies are desperately needed for organisms outside the genetic model species, and our work suggests several candidates. Among those neuropeptides we have identified, both tachykinin and sulfakinin influence nearly all the pathways thought to be co-opted during the evolution of parenting 28 , 29 , 30 , 31 , 32 and deserve further investigation in comparative or functional contexts. Methods Experimental design We used female N. vespilloides derived from an outbred colony we maintain at the University of Georgia, Athens. The colony was founded with beetles originally captured from Cornwall, UK and is subsidized yearly with new beetles from the same location. Thus, the colony is outbred. Larvae that disperse from a carcass were allocated to individual 9 mm diameter 4 cm deep circular containers filled with 2.5 cm potting soil. After emergence to adult, beetles were fed once weekly with decapitated mealworms ad libitum . Once larvae dispersed, larvae, pupae and adults had no further social interactions with other burying beetles until adults were paired for mating. We maintained all beetles in a common room at 22±1 °C on a 15:9 h light:dark cycle. To examine how neuropeptide expression changed with transitions of behavioural state, we collected age-matched females in three behavioural and social states: virgin (no social experience, no mating, no reproductive resource and no parenting), actively parenting (social experience, mated, reproductive resource and actively parenting), post-parenting (social experience, mated, reproductive resource and past parenting experience). Full descriptions of each behavioural state can be found in Roy-Zokan et al . 39 We collected virgins directly from their individual housing boxes. We collected actively parenting females directly from the carcass cavity where offspring are fed. We collected post-parenting females 9 days from the start of a breeding cycle after they had been isolated for 24 h. We collected all beetles at 19–22 days post-adult eclosion and all beetles were fed 1 day before their collection or before their pairing to standardize feeding status. We performed dissections in ice-cold 1 × phosphate-buffered saline (National Diagnostics, Atlanta, GA, USA) and completed them within 4 min. We placed single brains into 0.6 ml Eppendorf tubes with 30 μl of ice-cold acidified acetone extraction buffer (40:6:1 (v/v/v) acetone: H 2 O: Concentrated HCl). We did not collect the retro-cerebral complex (corpora allata–corpora cardiaca). Once collected, we stored samples at −80 °C until extraction. We pooled eight brains into a single biological replicate by removing brains and their associated acetone extraction buffer to a single 2.0 ml low protein binding Vivacon 500 tubes (Sartorius AG, Göttingen, Germany). We pooled eight brains; this number was based on preliminary studies that confirmed that eight reliably provided sufficient material. We collected four biological replicates per behavioural state. We sonicated each biological replicate with a Sonicator S-4000 (Misonix, Farmingdale, NY, USA) fitted with a 1/8″ tip (#419) set to an amplitude of 20 for a total of 60 s sonication with 15 s pulses followed by 15 s rest on ice. We then centrifuged replicates at 16,000 g for 20 min at 4 °C with a 5810-R Eppendorf centrifuge. We collected the supernatant into a new Vivacon tube and repeated the extraction with the same volume of buffer and sonication protocol. We pooled and extracted all replicates at the same time without ordering. We stored samples at 4 °C until LC-MS/MS analysis. We analysed our biological replicates with a Finnigan LTQ linear ion trap mass spectrometer (Thermo-Fisher Scientific, Waltham, MA, USA) and an 1100 Series Capillary LC system (Agilent Technologies, Santa Clara, CA, USA) with an ESI source with spray tips built in-house. The extraction buffer was vacuum-dried off of all biological replicates with a VirTis Benchtop K Lyophilizer (SP Scientific, Warminster, PA, USA) and biological replicates were suspended in 11 μl of buffer A (5% acetonitrile/0.1% formic acid/10 mM ammonium formate) and 8 μl of each replicate were injected into the LC column. The peptides were separated using a 200 μm × 150 mm HALO Peptide ES-C18 column packed with 5 μm diameter superficially porous particles (Advanced Materials Technology, Wilmington, DE, USA). The gradient used for each replicate was 5–75% buffer B (80% acetonitrile/0.1% formic acid/10 mM ammonium formate) for 120 min at a 2 μl min −1 flow rate. The settings for the mass spectrometer included taking the five most intense ions from each full mass spectrum for fragmentation using collision-induced dissociation and the resulting MS/MS spectra were recorded. Our biological replicates from the three treatments were interspersed with each other for LC-MS/MS analysis. All chemicals were LC-MS or molecular biology grade. Neuropeptide identification and analysis We converted the resulting RAW spectra using Trans Proteomic Pipeline (Seattle Proteome Center, Seattle, WA, USA). MS/MS spectra were then imported into MASCOT (v2.2.2; MatrixScience, Boston, MA, USA) and searched against all annotated proteins from the published N. vespilloides genome 21 to produce peptide-spectrum-matching scores only. We set search parameters as: enzyme, none; fixed modifications, none; variable modifications as oxidation (M), acetyl (N terminus), pyroglutamic acid (N terminus glutamine), and amidation (C terminus); maximum post-translational modifications, 6; peptide mass tolerance,±1,000 p.p.m.; fragment mass tolerance,±0.6 Da, tolerances set by the machine. We identified proteins with ProteoIQ (v2.6.03; default setting; Premier Biosoft, Palo Alto, CA, USA), which filters and uses MASCOT peptide-spectrum matching scores to statistically validate proteins identifications using the PeptideProphet Protein Probability scoring algorithm 40 . We identified proteins, peptides and assigned spectral counts using all biological replicates within each behavioural state. We only tallied the ‘top-hit’ for each spectrum as a further restriction on quantification. We also used ProteoIQ to estimate abundance of neuropeptides after the secondary validation of protein identities. This analysis produces a list of peptides assigned to each identified protein and from this we looked for qualitative differences in the presence/absence of peptides across the behavioural states for peptides that had at least three spectra and were not truncated forms of a larger observed peptide from a particular protein. We excluded peptides from proteins that were only observed in a single behavioural state. We then calculated normalized spectral abundance factor (NASFs) for all proteins within each biological replicate using the protein length for the NASF length correction factor 41 . Only peptides with at least two spectra within one biological replicate were quantified. Neuropeptide proteins were extracted from the overall protein list after establishing their identity within the published N. vespilloides gene set with a Tribolium castaneum neuropeptidome 42 augmented with neuropeptides identified and described from other insects 27 , 43 , 44 , 45 , 46 . We confirmed each neuropeptide’s identity using NCBI’s non-redundant insect protein database. We also assessed whether each neuropeptide had a predicted signal peptide using SignalP (v4.1; ref. 47 ) with a D-cutoff value of 0.34 (ref. 48 ). To test the hypothesis that changes in neuropeptide expression can be predicted a priori, we first performed a multivariate ANOVA to establish that there was an overall difference in the neuropeptide composition between treatments. A significant multivariate analysis allows for univariate a priori contrasts using ANOVA without inflating the Type I error 22 . We followed this multivariate test with univariate tests (ANOVAs) for difference of individual neuropeptide abundance, testing for the effect of behavioural state on abundance. Where the ANOVA was significant, we performed post hoc tests of differences in the pairwise means of the behavioural states using Tukey–Kramer honest significant difference tests, which allow us to compare pairs of behavioural states while controlling for FDR. All statistical analyses were conducted with JMP Pro (v11.0.0, Cary, NC, USA). Raw abundances, the NASF values, of each neuropeptide from every sample were used to calculate composite abundances using principal component analysis. We used R (v3.3.1) using the prcomp function after scaling the raw abundance data for each detected peptide to mean zero and unit variance to calculate PCAs. Visualizations were prepared in R using ggbiplot (v0.90; github.com/vqv/ggbiplot). Data availability Raw mass spectral data are available at ProteomeXchange (PXD005460; proteomecentral.proteomexchange.org ). Additional information How to cite this article: Cunningham, C. B. et al . Ethological principles predict the neuropeptides co-opted to influence parenting. Nat. Commun. 8, 14225 doi: 10.1038/ncomms14225 (2017). Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

University of Georgia researchers have confirmed that becoming a parent brings about more than just the obvious offspring—it also rewires the parents' brain. The study, published this month in Nature Communications, finds that the transition from a non-parenting state to a parenting state reflects differences in neuropeptides generally associated with mating, feeding, aggression and increased social tolerance. Neuropeptides are small proteins that allow neurons in the brain to communicate with each other; they also influence behavior. The team's research-tested on an insect, the burying beetle Nicrophorus vespilloides-provides a predictive framework for studying the genetics of parenting and social interactions. The burying beetle is intimately involved in raising its children, including regurgitating food to its begging offspring. "We tested the idea that we could predict the genetic pathways involved in parenting based on old predictions from ethologists in the 1960s and 1970s," said the study's lead author Allen Moore, Distinguished Research Professor and head of the department of genetics. "When [burying beetle] parents feed their babies, they are feeding others rather than themselves and so genes that influence food-seeking behavior are likely to be involved." Behavioral scientists predicted that genetic changes occur over time to develop parenting in a species. Based on this hypothesis, Moore's team sequenced and assembled the genome of the burying beetle and measured the abundance of neuropeptides. They theorized that behaviors related to parenting stemmed from alterations in existing genes rather than the evolution of new ones. By looking at parenting and non-parenting beetles, their tests indicated that neuropeptides changed in abundance during parenting. "When new traits evolve, evolution tends to modify existing genetic pathways rather than create new genes," Moore said. The research, Moore said, suggests that many of the genes influencing parenting will be the same across many species. The commonality among organisms will help researchers identify genetic pathways important to parenting. "It is exciting science when you take a step toward predicting the genetic changes involved in a behavior as complicated as parental care," he said. "And it was pleasing to collaborate with colleagues in genetics and Complex Carbohydrate Research Center, which allowed us to apply techniques that wouldn't otherwise be available to test our ideas."

10.1038/ncomms14225

Medicine

Weather-forecast tool adapted to evaluate brain health of oxygen-deprived newborns

Lina F. Chalak et al. Novel Wavelet Real Time Analysis of Neurovascular Coupling in Neonatal Encephalopathy, Scientific Reports (2017). DOI: 10.1038/srep45958 Journal information: The Lancet , Scientific Reports

http://dx.doi.org/10.1038/srep45958

https://medicalxpress.com/news/2017-04-weather-forecast-tool-brain-health-oxygen-deprived.html

Abstract Birth asphyxia constitutes a major global public health burden for millions of infants, despite hypothermia therapy. There is a critical need for real time surrogate markers of therapeutic success, to aid in patient selection and/or modification of interventions in neonatal encephalopathy (NE). This is a proof of concept study aiming to quantify neurovascular coupling (NVC) using wavelet analysis of the dynamic coherence between amplitude-integrated electroencephalography (aEEG) and near-infrared spectroscopy in NE. NVC coupling is assessed by a wavelet metric estimation of percent time of coherence between NIRS S ct O 2 and aEEG for 78 hours after birth. An abnormal outcome was predefined by a Bayley III score <85 by 18–24 m. We observed high coherence, intact NVC, between the oscillations of S ct O 2 and aEEG in the frequency range of 0.00025–0.001 Hz in the non-encephalopathic newborns. NVC coherence was significantly decreased in encephalopathic newborns who were cooled vs. non-encephalopathic controls (median IQR 3[2–9] vs.36 [33–39]; p < 0.01), and was significantly lower in those with abnormal 24 months outcomes relative to those with normal outcomes (median IQR 2[1–3] vs 28[19–26], p = 0.04). Wavelet coherence analysis of neurovascular coupling in NE may identify infants at risk for abnormal outcomes. Introduction Despite hypothermia therapy for hypoxic neonatal encephalopathy (NE), 50% of treated newborns have disabilities at 12–18 months of age 1 , 2 . The asphyxia insult impairs fetal cerebral blood flow (CBF) and is manifested postnatally by a distinctive neonatal encephalopathy (NE), which is usually classified after birth using the clinical modified Sarnat stages as mild, moderate and severe NE 3 . Recent reports suggest new evidence of cognitive impairment in a subset of 30–50% of “mild NE” cases, who are currently not recognized or offered any therapies 4 , 5 . While adjuvant therapies are being sought, an important focus of research is to recognize which neonates are in need of therapies to improve outcome 6 , 7 . The challenge in the field of neonatal brain injury has been the difficulty to clinically discern the mild-moderate NE severity within the short therapeutic window after birth. A sensitive and specific physiological marker that directly assesses neurovascular function in real-time, and is a marker of clinical outcomes, is critically needed to guide therapies 8 , 9 , 10 . The neurovascular unit consists of neurons, astrocytes, endothelial cells of blood–brain barrier (BBB), myocytes, pericytes and extracellular matrix components which play an important role in the delivery of oxygen and nutrients to the brain and the maintenance of cerebral circulation homeostasis 9 , 11 , 12 . Individual metrics of electroencephalography 6 , 13 , 14 , 15 and near-infrared spectroscopy (NIRS) 16 , 17 have been used independently to assess the function of neurovascular unit, but the ability to study coupling in real time of NIRS and EEG in HIE has received little attention 11 , 17 , 18 , 19 These methods demonstrated that a severe asphyxia insult leads to impaired function, but the extent of dysfunctional neurovascular coupling (NVC), and its impact on clinical outcome of newborns with encephalopathy cannot be currently assessed in real time due to limited methodology 20 , 21 . Intrinsic oscillating activities at multiple time scales have been described in cortical neurons across mammalian species 22 . These oscillations span from the very low to ultra-fast frequencies and although difficult to measure, are functionally relevant, phylogenetically conserved across species, and play important roles in facilitating synaptic plasticity and brain functional connectivity 22 . Cerebral blood flow (CBF) and brain tissue oxygenation also manifest oscillations at multiple time scales which occur simultaneously with changes in neuronal activities suggesting NVC 23 . Our previous work demonstrated a new time-frequency domain wavelet analysis of the dynamic relationship between spontaneous oscillations in mean arterial pressure and NIRS-measured cerebral tissue oxygen saturation (S ct O 2 ) allowed quantification of cerebral autoregulation at the bedside in newborns with encephalopathy 24 , 25 . We use the novel wavelet analysis to quantify NVC at the bedside, as defined in a global context to reflect the relationship between each of 1) NIRS S ct O 2 , a surrogate of CBF under steady state conditions, and 2) the amplitude integrated EEG (aEEG), which measures ensemble neuronal activity. The aims in this proof of concept study are to assess the feasibility of using the wavelet analysis to quantify NVC at the bedside, and its potential to stratify NE severity and identify newborns at high risks for abnormal outcomes following neuroprotective therapies. Results Patient characteristics Twenty inborn newborns with fetal metabolic acidosis were evaluated and screened during the one year study period. Of those, ten newborns had simultaneous aEEG and NIRS recordings which were suitable for wavelet analysis. The study was approved by UT Southwestern IRB and parental consent was obtained. These ten newborns comprised eight with moderate to severe encephalopathy who received hypothermia and two with no encephalopathy who did not meet criteria for cooling and served as references. Patient characteristics are summarized in Table 1 . Neuromonitoring was initiated at 12 ± 2 hours for duration of 60 ± 6 hours with no artifacts during recording. No scalp edema or any other complications were noted and no leads were replaced during the duration of the recording. No infant had seizures during the recording period. Only one infant had clinical seizures within 6 hours of life prior to enrollment. All cooled infants were categorized to have moderate encephalopathy, except one who was severe. All infants with a normal outcome had continuous patterns on aEEG. Sleep wake cycles (SWCs) were identified by aEEG in eight infants, occurring every 60–90 minute as summarized in Table 1 , two infants showed no SWC. MRI was available on all patients and was performed at a median age of 7 days. The MRI was abnormal in 4 neonates with white matter injury and watershed infarcts, one had added basal ganglia involvement. Abnormal outcome with Bayley III scores <85 at 18–24 months occurred in six of the ten infants and performance in each domain is summarized in Table 1 . Table 1 Clinical Characteristics. Full size table Wavelet findings Significant in-phase coherence (p < 0.05) between S ct O 2 and aEEG was identified by the wavelet analysis in the two non-cooled references, in the wavelet scale of 16–64 minutes equivalent to a very low-frequency (VLF) range of 0.00025–0.001 Hz. Similar findings were observed in the cooled infants with a normal Bayley outcomes at 18–24 months post hypothermia therapy ( Fig. 1 ). In contrast, wavelet-based S ct O 2 →aEEG coherence showed decreased NVC for cooled infants with abnormal outcomes. Figure 2 , highlights a case example of low NVC where the infant subsequently had white matter injury on MRI day 7 of life, and a Bayley III cognitive score of 70 at 18 months of age. Of note, the observed frequency of NVC coherences was outside the range of SWC frequencies summarized in Table 1 , and occurred even in the two cases where no SWC was observed, hence the analysis of NVC did not appear to be influenced by the sleep-awake cycles. Figure 1: Wavelet-based S ct O 2 -aEEG coherence showing intact neurovascular coupling (NVC). This extract is from a cooled neonate with normal outcome (MRI showed no injury and Bayley III >85. ( A ) An enlarged segment of the real-time S ct O 2 and aEEG data. ( B ) Squared wavelet coherence, , where the x-axis represents time, the y-axis represents scale in minute representing the range of frequencies, and the color scale represents the magnitude of R 2 . Significant coherence between the S ct O 2 and aEEG is seen in a very low-frequency (VLF) range of 0.00025–0.001 Hz. Full size image Figure 2: Wavelet-based S ct O 2 -aEEG coherence showing low neurovascular coupling. This infant had white matter injury on MRI. The Bayley III scores at 18 months are: cognitive 71, language 70, and motor 85. ( A ) An enlarged segment of the real-time S ct O 2 and aEEG data. ( B ) Squared wavelet coherence, , where the x-axis represents time, the y-axis represents scale of frequencies. No significant areas of coherence are seen through the range of time and frequencies studied. Full size image Figure 3 Describes the percentage of significant coherence plotted for each individual patient with or without abnormal outcomes at 18–24 months. According to this figure, the most distinct differences between infants with normal vs. abnormal outcomes were observed in a very low-frequency (VLF) range of 0.00025–0.001 Hz (wavelet scale s = 16–64 minutes). The boxplot distribution Fig. 4 shows significant differences in NVC coherence between groups. NVC coherence was significantly lower in newborns with abnormal outcomes compared to normal outcomes (median IQR 2[1–3] vs 28[19–26], p = 0.04). Figure 3 Individual data of significant S ct O 2 →aEEG in-phase coherence from newborns quantified in: ( A ) normal outcome group (n = 4), and ( B ) abnormal outcome group (n = 6). The two non-cooled are labeled by + and • in the left panel. Significant differences between normal vs. abnormal groups ( p < 0.05) was observed in the frequency range of 0.00025–0.001 Hz, highlighted by gray shade. Full size image Figure 4: Area under the curve (AUC) of significant SctO2-aEEG coherence in infants categorized by the 18–24 developmental outcomes. X axis shows infants divided into two groups: normal (n = 4) and abnormal outcomes (n = 6) based on Bayley <85 at 18–24 months. Y axis is AUC for significant NVC coherences. Boxplot (median, 25% and 75% percentiles) for the % NVC coherence AUC over the frequency range of 0.00025–0.001 Hz, p = 0.01 by Exact Wilcoxon Rank Sum test. Full size image Discussion This proof of concept study demonstrated the feasibility of using wavelet analysis of the dynamic coherence function between S ct O 2 and aEEG to assess NVC at the bedside in asphyxiated newborns with encephalopathy. The preliminary findings suggest that wavelet measures of NVC in the first 72 hours of life are associated with long-term outcomes following NE, with and without hypothermia therapy. The studies of spontaneous oscillations has always been entwined with neuronal network functionality and self-organization 26 . Understanding the physiological mechanisms of self-emerging oscillations of brain neuronal activity not only provide insight into brain function, but also may assist in the recognition of newborns with HIE who need added therapies to hypothermia. In the normal brain at term, a tight temporal and spatial coupling exists between neuronal activation and CBF 27 . Coupling between NIRS-measured S ct O 2 and EEG activity ensures a rapid spatial/temporal increase in the CBF in response to neuronal activation under normal conditions 28 . The capillary endothelial cells, astrocytes and neurons, together forming a neurovascular unit, are involved in the tightly coupled regional blood flow in response to local metabolic demands 29 . The underlying physiological mechanisms of NVC are complex, but hyperemic responses appear to be mediated by astrocytes 30 , 31 , 32 , occurring within seconds of localized brain activity. The resulting increased local CBF is likely to be larger than the concomitant oxygen consumption, resulting in increased NIRS saturation 33 . The prevailing concept of NVC, introduced over a century ago by Roy and Sherrington 34 , has been applied mostly in the modern-day neuroimaging studies of brain function with positron emission tomography (PET) and functional MRI in adult subject 35 . However, these techniques are either invasive and therefore cannot be applicable in the fragile newborn, or susceptible to movement artifact and therefore do not permit a real time analysis of continuous dynamics. Moreover, none of these methods can measure the multi-frequency aspect of neurovascular coupling. Oscillatory coupling of neuronal networks using NIRS and EEG has only been described prior in a cohort of healthy newborns 28 . The latter study involved non-sick preterm newborns and established coupling via intermittent recordings of integrated EEG and NIRS during normal gestational growth 28 . However, it is not known how these observations apply in the normally developing newborn apply to sick newborns with NE. The new findings of significant coherence between the oscillations in S ct O 2 and aEEG (in the VLF range of 0.00025–0.001 Hz) suggest the presence of intact NVC in large time scales in this group of term newborns with NE and normal outcomes. These slow rhythms have also been shown in adults to synchronize large spatial domains affecting connectivity, repair and functions 36 , and the present study findings would suggest they are also likely to be impaired with NE associated with abnormal outcomes. The wavelet analysis of NVC in this study may provide new insights into the regulation of cerebral hemodynamics and neuronal function. The sophisticated analysis is not limited to a specific monitoring device, but can be adapted to a multitude of non-invasive devices (such as continuous BP monitors, NIRS, and aEEG) that are all available at the bedside. While high NIRS cerebral saturation and suppressed aEEG background neuronal activity have been each separately linked to abnormal outcomes in HIE 17 , 19 , prior studies have provided only description of such uncoupling, without quantifiable metrics. The wavelet coherence analysis of NIRS-aEEG signals integrates the diagnostic information obtained from each device individually to permit continuous evaluation of coupled cortical activity and perfusion 37 , 38 , 39 . This study demonstrated that the large time-scale functional oscillations of the neurovascular unit can be measured non-invasively and continuously in asphyxiated newborns. The findings are novel, as the dynamic coherence relationship between changes in NIRS signal and neuronal activity has not been previously established in hypoxic NE or during hypothermia using non-invasive devices at the bedside. Study strengths include (1) a prospective design, (2) a real time wavelet analysis of continuous recordings over long time durations which allowed quantification of a wide range of VLF oscillations, and (3) use of standardized long term developmental outcomes in in newborns with encephalopathy. The study ensured maintaining an optimal temperature at 33.5 °C, steady state conditions, and absence of high impedance or artifacts during the wavelet recording. This proof of concept study is limited by a small number of patients, allowing only exploratory analyses. The NVC monitoring of the parietal channel might be limited in providing information regarding deep structures like the basal ganglia 40 , but was selected as it is representative of the global injury covering the watershed white matter that is usually seen in HIE. In conclusion, the study demonstrated the feasibility of using the novel wavelet coherence analysis of spontaneous oscillations in S ct O 2 and aEEG to quantify the NVC in newborns with encephalopathy. The new wavelet bundle approach, once validated in large NE cohorts promises to significantly impact the early NE stratification and prediction of neurocognitive outcomes. The ability to prospectively monitor the global neurovascular unit functions in real time, via novel approaches as delineated in this study, could in the future provide a paradigm shift to the field of neonatal neuro-critical care and improve the selection of candidates in neuroprotection studies. Methods Subjects This study included inborn infants ≥36 weeks of gestation and birth weight ≥1800 grams who were admitted to the neonatal intensive care unit (NICU) at Parkland Hospital of Dallas and had perinatal asphyxia with metabolic acidosis and a clinical exam showing encephalopathy in the first six hours from June 2012 to June 2013. Exclusion criteria included the presence of congenital anomalies, imminent death, or transfer to another facility. The study was approved by the Institutional Review Board of the University of Texas Southwestern Medical Center. Informed consent was obtained from both parents before enrollment. All methods were performed in accordance with the relevant STROBE guidelines and regulations and all patient data is de-identified. Perinatal acidemia was determined by measuring blood gases in umbilical arterial blood that is obtained from a double clamped section of umbilical cord at birth. Specifically, criteria were as defined by the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network study of whole body hypothermia 41 . A neurological examination was performed by one study investigator (LC) within 6 hours of birth using the NICHD classification for modified Sarnat staging to establish the severity of encephalopath 42 . The reference comprised newborns with fetal acidosis screened by a detailed neurological examination, who had no abnormalities and did not meet the criteria to receive hypothermia. The hypothermia group was comprised of newborns with a composite exam consistent with moderate or severe encephalopathy who did receive hypothermia therapy. Encephalopathy was defined as the presence of either moderate or severe signs in at least 3 of the following 6 categories as per NICHD cooling trials: (1) level of consciousness (moderate is lethargic, severe is stupor or coma), (2) spontaneous activity (moderate is decreased activity, severe is no activity), (3) posture (moderate is distal flexion or complete extension, severe is decerebrate), (4) tone (moderate is hypotonia, severe is flaccid), (5) primitive reflexes (moderate is a weak suck [or incomplete Moro reflex], severe is an absent suck [or absent Moro reflex]), and (6) autonomic nervous system 41 , 43 . Whole body hypothermia was started within 6 hours after birth and achieved by placing the newborn on a cooling blanket (Blanketrol II, Cincinnati Sub-Zero) and maintaining the esophageal temperature at 33.5 °C by the blanket servomechanism for 72 hours. Monitoring Amplitude Integrated EEG (aEEG) Electrical brain activity was monitored by using an amplitude-integrated EEG monitor (Natus Medical, San Carlos, CA, USA). Five hydrogel electrodes were placed by the same research personnel (DV) according to the international 10–20 system predefined criteria: ground, P3, C3, C4 and P4. The skin was prepared using Neo-Prep gel, and the impedance was monitored during the study for the analysis to include only recordings with impedance <5 kΩ. Electro cortical activity was classified according to a standardized classification as (1)Continuous: Continuous activity with lower (minimum) amplitude around 5 to 10 mcV and maximum amplitude of 10 to 25–50) mcV; (2)Discontinuous: Discontinuous background with minimum amplitude below 5 mcV, and maximum amplitude above 1 mcV; (3) Burst-suppression: Discontinuous background with minimum amplitude without variability at 0 to 1 mcV and bursts with amplitude >25 mcV; (4) Low voltage: Continuous background pattern of very low voltage (below 5 mcV); (5) Inactive, flat: Primarily inactive (isoelectric below 5 mcV) 44 . The EEG signals as per aEEG manufacturing standards were amplified and bandpass-filtered (1 to 50 Hz) to minimize artifacts and electrical interferences, rectified, and then peak to peak amplitudes were measured and smoothed using a moving average of 0.5 seconds 45 . The resulting upper margin amplitude, lower margin amplitude and bandwidth (BW) for the selected P3–P4 channel were quantitatively calculated and displayed using Brainz Analyze Research (v1.5) software (Natus Medical). In addition, cyclical pattern typically occurring about every 60 to 90 minutes indicating sleep wake cycles (SWC) with broadened band width (discontinuous activity) representing quiet sleep and narrow band width (continuous activity) representing wakefulness or active sleep were checked and highlighted. Near-Infrared Spectroscopy (NIRS) The INVOS (Somanetics) spatially-resolved NIRS oximeter was used (INVOS 4100–5100; Somanetics, Troy, MI). A neonatal probe (neonatal, SomaSensor) containing a light-emitting diode and two distant sensors was placed on the infant’s head, in close proximity to the P3 and P4 aEEG leads. The sampling rate of NIRS signal was every 30 seconds. Data pre-processing The S ct O 2 and aEEG were synchronously recorded every minute with a Vital Sync™ system (Somanetics Corporation, Troy, Michigan). Both the S ct O 2 and aEEG data were first inspected to identify artifacts, which were removed by linear interpolation between neighboring data points, followed by a second-order polynomial de-trending to remove the slow drifts of each time series. Wavelet coherence analysis Wavelet transform coherence is a time-frequency domain analysis, which characterizes the cross-correlation between two time series at multiple time scales and over time 46 . Details of this method were recently published 47 and a brief description is provided in the online supplement . The concept is to use a time varying, squared cross-wavelet coherence, R 2 , between two time series in multiple time scales without a priori assumptions of linearity and stationarity 48 . R 2 ranges between 0 and 1 and can be conceptualized as a localized correlation coefficient in the time-frequency domain between any two pre specified variables. The two variables selected for evaluation of NVC were the NIRS S ct O 2 and aEEG in this study. The statistical significance of the estimated coherence between the two times series against noise background was assessed based on theMonte Carlo method 49 . The wavelet coherence metrics include the relative phase, Δ φ , and the squared wavelet coherence, R 2 , in the interrogated range of frequencies ( f wt = 0.97/ s ; s represents wavelet scale in seconds) and over the time of the recorded S ct O 2 and aEEG durations. NVC is represented as the significant in-phase coherence (Δ φ ≈ 0) between the S ct O 2 and aEEG oscillations. Therefore to quantify the results, we first select a phase range of interest. Within the selected phase range, the percentage of significant coherence, P(s ), is quantified as the percentage of time during which the S ct O 2 →aEEG coherence is statistically significant from the noise background ( p < 0.05) by using the Monte Carlo simulation. P(s ) is a function of wavelet scale, s , or Fourier frequency, f wt . Neurodevelopmental outcomes Brain MRI was performed on day 7–10 using conventional T1- and T2-weighted spin echo sequences to evaluate for injuries in the basal ganglia, cortical or watershed areas, by using the NICHD classification 50 . Neurological examination and psychometric testing were conducted 18 to 24 months of age based on the Bayley Scales of Infant Development III (Bayley III for short) in three domains: cognitive, language and motor 1 , 51 . All study infants completed follow up including the non-cooled controls. An abnormal outcome was predefined by abnormal MRI scores >1 and Bayley III scores <85 in any of the cognitive, language or motor domains. Examiners of neurodevelopmental outcomes were blinded to the aEEG and NIRS results. Statistical analysis This Wavelet method generates a large ensemble of surrogate data set pairs with the same power spectrum as the input datasets. Wavelet coherence is estimated for each pair. Then the coherence of input datasets is statistically tested against those of surrogate datasets with a null hypothesis that the signal is generated by a stationary noise background. The statistically significant squared cross-wavelet coherence, R 2 , against noise background and the corresponding phase between the time series of S ct O 2 and aEEG were determined, and the percentage of time durations during which R 2 was statistically significant ( p < 0.05) was calculated relative to the whole recording period for each infant. Exact Wilcoxon Rank Sum test were used to compare the wavelet coherence (% of significant coherence) across NE severity in cooled infants (moderate-severe NE combined since only 1 severe NE) vs. controls as well as between the infants with normal outcome vs. abnormal outcome groups. This included between-group comparisons for each wavelet scale at a significance of p < 0.05, and a comparison of area under the curve (AUC) over the selected frequency range at a significance of p < 0.01. For this pilot study, multiple comparison corrections were not applied. Additional Information How to cite this article: Chalak, L. F. et al . Novel Wavelet Real Time Analysis of Neurovascular Coupling in Neonatal Encephalopathy. Sci. Rep. 7 , 45958; doi: 10.1038/srep45958 (2017). Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

UT Southwestern Medical Center pediatric researchers have harnessed an analytical tool used to predict the weather to evaluate the effectiveness of therapies to reduce brain injury in newborns who suffer oxygen deprivation during birth. The analytical tool, called wavelet analysis technology, is best known for predicting long-term weather patterns, such as El Nino. UT Southwestern researchers say this same analytical tool can help improve assessment and treatment of newborns with asphyxia, which is when the baby's brain is deprived of oxygen due to complications during birth. The non-invasive method produces real-time heat maps of the infant's brain that doctors can use to determine whether therapies to prevent brain damage are effective. "These are babies to whom something catastrophic happened at birth. What this technology does is measure physiologic parameters of the brain - blood flow and nerve cell activity - to produce a real-time image of what we are calling 'neurovascular coupling.' If there is high coherence between these two variables, you know that things are going well," said Dr. Lina Chalak, Associate Professor of Pediatrics at UT Southwestern and lead author of the study. The wavelet analysis correlates information from two non-invasive technologies that are currently used on a day-to-day basis in neonatal intensive care: amplitude EEG and near infrared spectroscopy. The approach combines the results from these commonly done tests in a sophisticated way and creates a new proxy measure of brain health called neurovascular coupling. When neuronal activity and brain perfusion are synchronized - as indicated by large areas of red on heat maps created by this method - treatment is working well. About 12,000 newborns experience oxygen deprivation (asphyxia) during birth in the U.S. each year, according to a 2010 article in Lancet. This can occur for a number of reasons, such as the cord being wrapped around the baby's neck, a difficult breech birth, or the separation of the placenta from the uterus too soon, Dr. Chalak explained. These infants are at high risk of developing serious consequences such as cerebral palsy, epilepsy, and cognitive deficits. No treatments were available until about 10 years ago when a national study in which UT Southwestern participated, showed that reducing the baby's core temperature could counteract the impact of birth asphyxia for some infants. The cooling blankets are now standard treatment, but only about half of babies treated with a cooling blanket benefit. Until the adaptation of the wavelet technology, doctors couldn't determine which infants were benefitting from cooling treatment and which babies may need additional therapies, which are being developed. Dr. Lina Chalak, Associate Professor of Pediatrics at UT Southwestern Medical Center and a specialist in treating birth asphyxia, uses wavelet analysis of amplitude EEG and near infrared spectroscopy to create a proxy measure of brain health she calls 'neurovascular coupling.' Credit: UT Southwestern Wavelet analysis information also can help in evaluation of new therapies. Dr. Chalak plans to use wavelet analysis as part of the HEAL study, a large clinical trial to determine the effectiveness of erythropoietin, a hormone that promotes the formation of red blood cells, to treat newborns with asphyxia. Wavelet analysis will be used to evaluate infants who are part of the HEAL study at the Neonatal Intensive Care Unit at William P. Clements Jr. University Hospital, Parkland Hospital, and Children's Medical Center. Wavelet technology also may help determine which children should be treated. "Of the babies who are oxygen-deprived, some don't qualify for cooling because their brain damage or encephalopathy is judged to be mild. Yet some of these children have adverse outcomes. This technology may help us identify who needs cooling," said Dr. Rashmin Savani, Chief of Neonatal-Perinatal Medicine and Professor of Pediatrics and of Integrative Biology, who holds The William Buchanan Chair in Pediatrics. It could also lead to other discoveries. "Understanding of brain blood flow regulation and its impact on brain function in newborns with asphyxia using this novel technology has great potential for developing new sensitive biomarkers for clinical diagnosis, treatment, and prognosis for these sick babies, which is desperately needed in the field," said co-author Dr. Rong Zhang, Associate Professor of Neurology and Neurotherapeutics, and of Internal Medicine, and a member of the Peter O'Donnell Jr. Brain Institute at UT Southwestern. The research appears in Scientific Reports. Other UT Southwestern researchers who contributed to this paper are Dr. Beverley Adams-Huet, Assistant Professor of Clinical Sciences and of Internal Medicine; Diana Vasil, research nurse; and Dr. Takashi Tarumi, Instructor in Neurology and Neurotherapeutics, along with Dr. Fenghua Tian, Assistant Professor of Biomedical Engineering at UT Arlington. The research was funded by grants from the National Institutes of Neurological Disorders and Stroke (NINDS), part of the National Institutes of Health. Dr. Chalak said the new tools represent a paradigm shift for physicians who treat oxygen-deprived newborns. "Hopefully, the future will be bright for babies who suffered asphyxia, differing from the bleak prognoses of the past."

10.1038/srep45958

Chemistry

Alternative process for converting white phosphorus promises more sustainability in the chemical industry

Maximilian Donath et al, Direct conversion of white phosphorus to versatile phosphorus transfer reagents via oxidative onioation, Nature Chemistry (2022). DOI: 10.1038/s41557-022-00913-4 Journal information: Nature Chemistry

https://dx.doi.org/10.1038/s41557-022-00913-4

https://phys.org/news/2022-04-alternative-white-phosphorus-sustainability-chemical.html

Abstract The main feedstock for the value-added phosphorus chemicals used in industry and research is white phosphorus (P 4 ), from which the key intermediate for forming P(III) compounds is PCl 3 . Owing to its high reactivity, syntheses based on PCl 3 are often accompanied by product mixtures and laborious work-up procedures, so an alternative process to form a viable P(III) transfer reagent is desirable. Our concept of oxidative onioation, where white phosphorus is selectively converted into triflate salts of versatile P 1 transfer reagents such as [P(L N ) 3 ][OTf] 3 (L N is a cationic, N-based substituent; that is, 4-dimethylaminopyridinio), provides a convenient alternative for the implementation of P–O, P–N and P–C bonds while circumventing the use of PCl 3 . We use p-block element compounds of type R n E (for example, Ph 3 As or PhI) to access weak adducts between nitrogen Lewis bases L N and the corresponding dications [R n EL N ] 2+ . The proposed equilibrium between [R n EL N ] 2+ + L N and [R n E(L N ) 2 ] 2+ allows for the complete oxidative onioation of all six P–P bonds in P 4 to yield highly reactive and versatile trications [P(L N ) 3 ] 3+ . Main Phosphorus is an essential building block of all biological life. It not only fuels the global food economy but has also found a plethora of additional applications in industry 1 . Phosphate rock is the natural but non-renewable resource of this highly important element. The majority of phosphate rock is used for the production of phosphoric acid (H 3 PO 4 ) via the wet process using concentrated sulfuric acid (Fig. 1 ) 2 , 3 , 4 , 5 , 6 . This process provides a product that is mainly used for fertilizers (~90%) 2 , 3 , 4 , 5 , 6 . The industrial reduction of phosphate rock is termed the electrothermal process and yields white phosphorus (P 4 ), which is the major phosphorus source for industrial and academic applications, for example, pharmaceuticals, flame retardants, battery ingredients, fertilizers, herbicides, metals and electronics etchants, as well as additional phosphorus fine chemicals 2 , 3 , 4 , 5 , 6 . Fig. 1: Routes into value-added phosphorus chemicals from phosphate rock. Transformation of phosphate rock via the wet process to phosphoric acid (H 3 PO 4 ) and via the thermal process to white phosphorus (P 4 ) with subsequent (oxy)chlorination to PCl 3 , OPCl 3 or P 4 O 10 . The green box outlines the present work: oxidative onioation of P 4 to give [P(L N ) 3 ][OTf] 3 , a viable alternative to PCl 3 . An element compound R n E is oxidized to an element oxide R n EO. In the onio-ligand assisted deoxygenation, the R n EO is deoxygenated using an oxygen scavenger in the presence of an N-based ligand (L N ) to give [R n E(L N ) 2 ][OTf] 2 , a weak adduct that undergoes an equilibrium dissociation into cation [R n EL N ][OTf] 2 and uncoordinated ligand L N . This weak adduct is regarded as an oxidative onio-reagent for the desired P 4 oxidation. The element compound R n E is retrieved to close the cycle. L N is a suitable nitrogen-based Lewis base (dimethylaminopyridine, pyridine and quinoline) and R n E is a p-block-element compound (triphenylarsane and phenyl iodide); Tf 2 O = trifluoromethanesulfonic anhydride; Me 3 SiOTf = trimethylsilyl triflate. Equations are not balanced. Full size image Among the range of key phosphorus intermediates, PCl 3 is still the most important P(III) source. The production of this very important bulk chemical includes the use of chlorine gas (Cl 2 ), which is widely considered to be problematic; however, as yet, there is no alternative 2 , 3 , 4 , 5 , 6 for industrial and laboratory applications. PCl 3 is used in subsequent transformation reactions such as hydrolysis, alcoholysis, salt metathesis and Grignard reactions to obtain value-added phosphorus chemicals. These reactions often suffer from low selectivity and/or require an additional base in some cases (for example, to capture HCl). In this regard, the functionalization of white phosphorus into chemically useful products without the use of chlorine gas 7 , 8 , 9 , 10 , 11 , 12 , 13 has received much interest from the research community, including efforts to develop ecologically friendlier chemistry routes 5 , 6 , 14 . Recent contributions in this field include the light-induced catalytic functionalization of P 4 or its transformation using tributyltin hydride 15 , 16 . However, such reactions are extremely challenging and still very rare due to the selective and complete cleavage of all six P–P bonds of the reactive P 4 tetrahedron and, in turn, the selective formation of new phosphorus-element bonds. A selective transformation of white phosphorus to a single P(III) compound that is a viable alternative to PCl 3 is thus highly desirable. The issues accompanying white phosphorus processing guided our research in the direction of developing an alternative route towards an easy-to-handle P(III) precursor from which a variety of value-added phosphorus compounds might be synthesized. During our research on N-heterocyclic phosphanes, such as pyrazolylphosphanes, we realized the substantial advantage of utilizing pyrazolyl substituents (or related N-based substituents) as extremely amenable leaving groups on phosphorus, rendering these pyrazolylphosphanes as synthetic equivalents of a P 1 synthon 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 . We have previously demonstrated that certain N-heterocyclic phosphorus compounds are highly reactive, but allow for controlled reactions in which these compounds serve as P 1 building blocks and where the side product can either be recycled or is of synthetic importance 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 . The results discussed in this work converge on the concept of oxidative onioation, where P 4 is systematically functionalized into a P 1 compound featuring labile N-based substituents (Fig. 1 , green box). Starting with a p-block-element compound R n E (for example, Ph 3 As or PhI), the respective element oxide R n EO is prepared by oxidation. In the next step, an onio-ligand-assisted deoxygenation of R n EO with an N-based ligand (L N ) and a suitable oxygen scavenger (for example, Tf 2 O or Me 3 SiOTf) yields an onio-substituted triflate salt [R n E(L N ) 2 ] 2+ , which is a weak adduct that undergoes an equilibrium dissociation into cation [R n EL N ] 2+ and uncoordinated ligand L N . The oxygen of R n EO is accordingly transferred into the corresponding anion of the salt [R n E(L N ) 2 ][OTf] 2 . This weak adduct [R n E(L N ) 2 ] 2+ is regarded as an oxidative onio-reagent for the desired P 4 oxidation, rather than elemental Cl 2 . Thus, the combination of the nucleophile L N and an electrophilic oxidizing cation 25 , 26 , 27 [R n EL N ] 2+ allows for the oxidative onioation of white phosphorus in which R n E is retrieved and the onio-substituted phosphorus trication [P(L N ) 3 ] 3+ is formed. The synthetic concept of oxidative onioation is an alternative approach for the chlorine-free oxidation of white phosphorus. Notably, a report was recently published about the use of trimetaphosphate [P 3 O 9 ] 3− , which is produced from phosphoric acid, for conversion into value-added chemicals without the need for P 4 as an intermediate altogether 28 . In this regard, the synthesis of P(III) compounds from P 4 has been decribed as a bottleneck in modern (sustainable) phosphorus chemistry 29 . However, given the fact that there are currently no alternatives to using P(III) compounds in syntheses 2 , 3 , 4 , 5 , 6 , 29 , we believe that the efficient, direct conversion of white phosphorus is still the most accomplishable route to value-added phosphorus chemicals 1 , 5 , 6 , 14 . In this Article we report the direct oxidative onioation of white phosphorus using Lewis-acidic compounds Ph 3 As(OTf) 2 ( 1 ) or PhI(OAc) 2 ( 9 ) in combination with the weak base triphenylarsane (Ph 3 As) or nitrogen-based Lewis bases such as quinoline, pyridine, 4-dimethylaminopyridine (DMAP) and 1,10-phenanthroline. NMR spectroscopic and theoretical investigations illustrate the dependency of the Lewis-acidic compound 1 together with the triflate anion and different Lewis bases (such as Ph 3 As or nitrogen-based Lewis bases) having a substantial impact on the outcome in the reaction with white phosphorus. Although bicyclo[1.1.0]tetraphosphane compound [Ph 3 As–P 4 –AsPh 3 ][OTf] 2 ( 3 [OTf] 2 ) with a P 4 -butterfly core is formed via the reaction of P 4 with 1 and the Lewis base AsPh 3 , the complete and selective degradation of white phosphorus is feasible via the reaction of P 4 with 1 and a nitrogen-based Lewis base L N . Detailed mechanistic investigations revealed that the transformation of white phosphorus proceeds over the bicyclo[1.1.0]tetraphosphane dication [L N –P 4 –L N ] 2+ ( 15 2+ ; L N = DMAP), which is successively degraded to form salt P(L N ) 3 [OTf] 3 ( 11 [OTf] 3 ; L N = DMAP) 30 . The reactive P–N bonds of 11 [OTf] 3 render this compound a versatile P 1 synthon and a viable alternative to PCl 3 , thus allowing for P–O, P–N and P–C bond-formation reactions in a single step. Results and discussion Oxidative onioation of P 4 We recently reported on the synthesis of the highly Lewis-acidic compound Ph 3 As(OTf) 2 ( 1 ) from the reaction of Ph 3 AsO and Tf 2 O (trifluoromethanesulfonic anhydride) and its use to synthesize polyphosphorus frameworks from PCl 3 (ref. 31 ). We hypothesized that 1 , together with a suitable ligand, would form an adduct that is able to transform white phosphorus by oxidative onioation. In this regard, we reacted 1 with the weak base Ph 3 As, which, after work-up, gives analytically pure diarsonium compound 2 [OTf] 2 in a moderate yield of 60% (Fig. 2 ). The structural connectivity of dication 2 2+ was confirmed by X-ray analysis of suitable single crystals of 2 [OTf] 2 , revealing a strong interaction of the triflate anions with the As atoms indicated by short As···OTf distances (Supplementary Fig. 4 ). Notably, the Burford group has reported on a related reaction of Ph 3 Sb(OTf) 2 with trialkyl phosphanes, giving diphosphonium ditriflates of type [R 3 PPR 3 ][OTf] 2 (ref. 32 ). In a subsequent reaction, P 4 was added to a colourless CH 3 CN solution of in situ formed 2 [OTf] 2 and, after 15 h, the 31 P NMR spectrum of the yellow reaction solution showed the formation of the bicyclo[1.1.0]tetraphosphane derivative 3 2+ ( δ (P A ) = −327.4 ppm; δ (P X ) = −174.7 ppm; 1 J AX = −158 Hz) 33 next to P 4 ( δ (P) = −523.4 ppm; Extended Data Fig. 1a ). With a calculated Gibbs energy of Δ G = −3.4 kcal mol −1 for this reaction, rapid equilibration between the two species can be anticipated (Fig. 2 ). As a result, any attempt to isolate butterfly compound 3 [OTf] 2 as a pure material remained unsuccessful. When the ratio of the oxidizing reagent ( 1 ) and the ligand (AsPh 3 ) is increased in the reaction with P 4 , no further degradation of the P 4 tetrahedron or butterfly compound 3 2+ is observed. To our surprise this reaction does not proceed when the anion is changed to the weakly coordinating anion BArF − [tetrakis(3,5-bis(trifluoromethyl)phenyl)borate], thus indicating the necessity of the triflate anion in the initial activation step of P 4 (Fig. 2 and Extended Data Fig. 1b ). The calculated Gibbs energy of the reaction using BArF − is Δ G = 6.5 kcal mol −1 . Fig. 2: Formation of butterfly compound 3[OTf] 2 via oxidative onioation of P 4 . a , Reaction of 1 with AsPh 3 to give diarsonium ditriflate 2 [OTf] 2 and subsequent reaction with 1 equiv. of P 4 to butterfly compound 3 [OTf] 2 . Conditions: (i) PhF, r.t., 15 min, (60%); (ii) CH 3 CN, 50 °C, 15 h. b , Triflate-initiated equilibrium between 2 2+ and 4 + and cooperative activation of P 4 with 4 + and Ph 3 As to give butterfly compound 3 [OTf] 2 . The formation of 3 [BArF] 2 from 2 [BArF] 2 (BArF − = tetrakis(3,5-bis(trifluoromethyl)phenyl)borate) is hampered, underlining the necessity of the triflate anion in this reaction. Δ G values are given in kcal mol −1 (blue). Calculations were carried out using DFT-optimized geometries applying the BP86 37 , 38 , 39 functional with a dispersion correction (D3) 40 and COSMO (dichloromethane) on a def2-TZVP basis set 41 . Full size image Fig. 3: Onio-ligand-assisted deoxygenation reactions of compounds Ph 3 AsO and PhIO. a , Reaction of 1 with 2 equiv. of DMAP or pyridine to give weak adduct salts 5 [OTf] 2 and 6 [OTf] 2 , respectively, and reaction of 1 with 1 equiv. of quinoline to form compound 7 [OTf] 2 / 8 [OTf]. Conditions: (i) CH 2 Cl 2 , r.t., 30 min ( 5 [OTf] 2 , 92%); (ii) CH 2 Cl 2 , r.t., 5 min ( 6 [OTf] 2 88%); (iii) CH 2 Cl 2 , r.t., 1 h, (70%). b , Formation of the iodine-based weak adduct 10 [OTf] 2 from the reaction of compound 9 with DMAP and Me 3 SiOTf or from the reaction of PhIO with DMAP and Tf 2 O. Conditions: (iv) –2 Me 3 SiOAc, CH 2 Cl 2 , 1 h (95%); (v) CH 3 CN, r.t., 45 min, not isolated. Δ E (blue) was calculated using DFT-optimized geometries of 7 [OTf] 2 and 8 [OTf] applying the BP86 37 , 38 , 39 functional with a dispersion correction (D3) 40 and COSMO (dichloromethane) on a def2-TZVP basis set 41 . Full size image It can be assumed that the As–As bond in diarsonium compound 2 [OTf] 2 is heterolytically cleaved by the free triflate anion to give Ph 3 As and triflyl-substituted arsonium compound 4 [OTf] (refs. 34 , 35 ), which is calculated to be an equilibrium reaction with a calculated Gibbs energy of Δ G = −2.4 kcal mol −1 . We believe that Ph 3 As and arsonium compound 4 [OTf] cooperatively react with P 4 to give butterfly 3 [OTf] 2 (Fig. 2b , dashed box). As the combination of Ph 3 As together with 1 seems too weak to fully degrade P 4 , in the process of the oxidative onioation stronger N-based Lewis bases were applied. We therefore continued our investigation on the formation of suitable weak adducts and reacted 1 with 2 equiv. of DMAP and observed the clean formation of the penta-coordinate adduct 5 [OTf] 2 , which is isolated in very good yield (>88%) after work-up (Fig. 3a ). The structural connectivity of dication 5 2+ was confirmed by X-ray analysis of suitable single crystals of 5 [OTf] 2 (Supplementary Fig. 20 ). To prove the weak adducts equilibrium, a 1:1 mixture of 5 [OTf] 2 and DMAP in CD 2 Cl 2 was investigated by means of 1 H and 13 C{H} NMR spectroscopy (Supplementary Figs. 14 and 15 ). The NMR spectra revealed only one set of resonances for the DMAP ligands, suggesting a rapid exchange at ambient temperature and thus an equilibrium in solution. Although salt 6 [OTf] 2 (L N = pyridine), which can be isolated in 92% yield, reveals a similar behaviour in solution when compared to 5 2+ , the corresponding penta-coordinate compound with quinoline as Lewis base is not isolable (Fig. 3a ). Instead, the reaction of 1 with quinoline in variable stoichiometries exclusively gives the 1:1 adduct 7 2+ as the triflate salt. Variable-temperature NMR studies of the 1:2 reaction of 1 and quinoline show only one set of resonances and thus a fast exchange of the quinoline ligands can be assumed (Supplementary Figs. 31 – 34 ; for a detailed description see Supplementary Section 2.11 ) 36 . The space-filling model of 7 2+ reveals that the accessibility of the As atom is considerably reduced for a second quinoline ligand, preventing the formation of a bipyramidal complex (Supplementary Fig. 138 ). Further density functional theory (DFT) calculations 37 , 38 , 39 , 40 , 41 reveal that, by incorporation of the triflate anion, the As atom exhibits a penta-coordinate geometry where the quinoline ligand and the triflate occupy the axial position. In this regard, penta-coordinate compound 8 + is energetically favoured by Δ E = −8.3 kcal mol −1 compared to 7 2+ (Extended Data Fig. 2a ). We assume that both triflate anions influence the underlying equilibrium and thus have to be considered in the reaction of 1 with Lewis bases (and P 4 ) by the formation of contact ion pairs 42 , 43 . Furthermore, thermochemical calculations on the reaction of 1 with quinoline towards the formation of 8 + show that the calculated Gibbs free enthalpies are negative at low temperatures (Δ G 193K = −2.0 kcal mol −1 ), but considerably positive at high temperatures (Δ G 343K = +10.5 kcal mol −1 ; Extended Data Fig. 2b ). These investigations support the assumption that the simultaneous presence of electrophilic 1 and a suitable Lewis base L N should provide the necessary platform for the hitherto unknown oxidative onioation of white phosphorus via a cooperative process. When P 4 is added to a concentrated reaction solution of 6 equiv. 5 [OTf] 2 in CH 2 Cl 2 and stirred at 50 °C in a sealed flask, the molten white phosphorus dissolves completely within 20 min, and a colourless, microcrystalline solid precipitates ( 11 [OTf] 3 ) in the course of a reaction time of 3 h (Fig. 4a and Supplementary Fig. 39 ). The 31 P NMR spectrum of the off-white material dissolved in CD 3 CN reveals a singlet resonance at δ (P) = 102.1 ppm that is substantially low-field-shifted compared to P 4 ( δ (P) = −523 ppm), which is in agreement with the reported chemical shift of cation 11 3+ (ref. 30 ). The structure was confirmed by X-ray analysis of suitable single crystals (Fig. 4b ) 30 . The 31 P NMR spectrum of the supernatant of the reaction remains silent in the area of P 4 , suggesting complete consumption (Supplementary Fig. 45 ). AsPh 3 is recovered in up to 85% and used to resynthesize 1 , thus closing the loop in this process. A few crystals of the triflate salt of (L N ) 2 POP(L N ) 2 4+ (L N = DMAP; δ (P) = −106.4 ppm), which is related to the Hendrickson reagent 44 , 45 , could be isolated from the reaction filtrate (Extended Data Fig. 3 ). It is assumed that its formation results from the degradation reaction of 11 3+ involving the triflate anions (vide infra). Notably, the reaction of 5 [BArF] 2 with P 4 does not lead to the formation of the BArF salt of 11 3+ (Supplementary Fig. 48 ). Fig. 4: Selective oxidative onioation of white phosphorus (P 4 ) to P 1 transfer reagents. a , Reaction of white phosphorus with weak adducts 5 [OTf] 2 , 6 [OTf] 2 , 8 [OTf] and 10 [OTf] 2 (shown in the dashed box) to give the N-substituted P 1 transfer reagents 11 [OTf] 3 , 12 [OTf], 13 [OTf] and 14 [OTf] 3 , respectively. Compounds AsPh 3 and PhI can be recovered from the reactions and thus re-used to synthesize the weak adducts. Conditions: (i) –6 AsPh 3 , CH 2 Cl 2 , 50 °C, 5 h (89%); (ii) –6 PhI, CH 2 Cl 2 , 50 °C, 2 h (85%); (iii) – AsPh 3 , – CF 3 S–SCF 3 , CH 2 Cl 2 , r.t., 24 h; (iv) + quinoline, – AsPh 3 , – CF 3 S–SCF 3 , CH 2 Cl 2 , 50 °C, 3 h; equations are not balanced; (v) –6 AsPh 3 , CH 2 Cl 2 , 75 °C, 3 h (63%). b , Molecular structures of the P 1 compounds 11 3+ in 11 [OTf] 3 · CH 2 Cl 2 , 12 + in 12 [OTf] · 0.67 MeNO 2 and 14 3+ in 14 [OTf] 3 · 2 CH 3 CN. Thermal ellipsoids are displayed at the 50% probability level and hydrogen atoms and counterions are omitted for clarity. Full size image Surprisingly, a different reaction outcome is observed for the L N bases pyridine or quinoline. When P 4 is added to solutions of 1 and pyridine or quinoline in CH 2 Cl 2 , oxo-species 12 + and 13 + (L N : 12 + = pyridinio, 13 + = quinolinio) are obtained as triflate salts after work-up (Fig. 4a ). Investigation of the 31 P NMR spectra of the reaction mixtures revealed that only small amounts of P 4 remain unreacted (Supplementary Figs. 49 and 55 ). The 31 P NMR spectra of the main products reveal a singlet resonance at δ (P) = −15.2 ppm ( 12 + ) and δ (P) = −14.6 ppm ( 13 + ), which is in accordance with the reported chemical shifts 46 . Single crystals of 12 [OTf] suitable for X-ray analysis confirm the structural connectivity and thus the further oxidation of the P atom by an oxygen transfer (Fig. 4b ). Formation of oxo-species 12 + and 13 + apparently results from the deoxygenation reaction of the triflate anions, which is supported by the observation of fluorine-containing side products of the triflate ions in the corresponding 19 F NMR spectra of the reaction mixtures (Supplementary Figs. 50 and 56 ). The resonances observed in the 19 F NMR spectrum of the filtrate can be assigned to uncoordinated triflate ions ( δ (F) = −78.9 ppm), bis(trifluoromethyl)disulfane (CF 3 S) 2 ( δ (F) = −46.2 ppm) and CF 3 Cl ( δ (F) = −27.4 ppm) 47 . The formation of disulfane (CF 3 S) 2 results from the deoxygenation reaction and the concomitant reduction of the sulfonyl groups of the triflate anions. These results clarify that the basicity, rather than steric requirements of the Lewis base (pyridine versus quinoline), might be the reason for the involvement of the triflate anion in the reaction (vide infra). Given the high electrophilicity of peronio-substituted P(III) compounds 26 , the deoxygenation reaction may be rationalized by a strong interaction of the triflate anion with a highly charged cation being formed in the course of the reaction. A proposed mechanism is given in the Supplementary Information (Supplementary Scheme 2 and Supplementary Section 2.18 ). Seeking an alternative in the chlorine-free functionalization of P 4 without using the toxic arsenic-based reagent 1 , we decided to test the use of a cationic iodine(III) reagent. Compound 10 [OTf] 2 (ref. 48 ), which is conveniently prepared from commercially available compounds (Fig. 3b ), readily reacts with P 4 at 50 °C in CH 2 Cl 2 and gives an off-white suspension within 3 h (Fig. 4a ). The 1 H and 13 C NMR spectra of the reaction filtrate reveal the characteristic resonances for phenyl iodide (PhI), which is recovered in 87% yield, and the investigation of the solid material by means of 31 P NMR spectroscopy shows the formation of the targeted P(III) derivative 11 [OTf] 3 . The advantages of the latter reaction are (1) lack of a requirement for toxic arsenic-based reagents and (2) the use of commercially available starting materials for the formation of compound 10 [OTf] 2 , that is, PhI(OAc) 2 , PhIO, MeSiOTf or Tf 2 O and DMAP (Fig. 3b ) 48 . Inspired by the complete conversion of P 4 to a single P 1 compound, we expanded the concept of the oxidative onioation of P 4 using a strong chelating Lewis base. We reacted a solution of 1 and 1,10-phenantroline in CH 2 Cl 2 with P 4 in an exact 6:8:1 stoichiometry at 75 °C for 3 h, whereupon a pale yellow precipitate was formed. After work-up, the colourless material was identified as phosphoranide derivative 14 [OTf] 3 (Fig. 4a ). The 31 P NMR spectrum of 14 [OTf] 3 displays a singlet resonance at δ (P) = 41.9 ppm, which is shifted to a slightly lower field with respect to that of the comparable derivative [P(bipy) 2 ] 3+ (bipy = 2,2′-bipyridine; δ (P) = 33.9 ppm) 30 . Crystals of the solvate 14 [OTf] 3 · 2 CH 3 CN suitable for X-ray analysis were obtained by slow diffusion of Et 2 O into a solution of 14 [OTf] 3 in CH 3 CN at −30 °C (Fig. 4b ). The selective formation of 14 [OTf] 3 directly from P 4 illustrates that the concept of oxidative onioation is also well applicable to chelating N-based Lewis bases (L N ), so a variety of other derivatives featuring phosphorus in different oxidation states should generally be accessible (this is part of ongoing investigations). Mechanistic investigations To gain further insights into the reaction mechanism of the oxidative onioation of white phosphorus, time-dependent NMR spectra of 1 and DMAP in the reaction with white phosphorus were performed to elucidate the formation of possible intermediates. However, the only phosphorus-containing compounds in the respective 31 P NMR spectra of these reactions were 11 3+ and unreacted P 4 (Supplementary Fig. 71 ). As bicyclo[1.1.0]tetraphosphane (P 4 butterfly) derivatives are generally considered key intermediates in P 4 activation reactions 7 , 8 , 9 , 10 , 11 , 12 , 13 , we added DMAP and 1 to solutions of butterfly compound 3 [OTf] 3 and P 4 and investigated the reaction mixtures by means of 31 P NMR spectroscopy. The successive addition of DMAP (1 equiv., 3 equiv., 6 equiv., 12 equiv.) to a reaction mixture of P 4 , 6 equiv. 1 and 12 equiv. AsPh 3 led, indeed, to the degradation of P 4 and butterfly compound 3 2+ and ultimately to the formation of trication 11 3+ (Fig. 5a and Extended Data Fig. 4 ). From the 31 P NMR spectra of the reaction mixtures with different amounts of DMAP, several intermediates can be identified from which a possible mechanism can be derived (Fig. 5b and Supplementary Section 4.2 ). The mechanism is supported by DFT calculations, revealing an overall Gibbs energy of Δ G = −30.9 kcal mol −1 for the formation of 11 [OTf] 3 . It is important that the triflate anions are included in the calculations, thus regarding the respective compounds as contact ion pairs (Extended Data Fig. 5 ). Notably, the proposed bonding motifs of the intermediates are known from degradation reactions of P 4 and have been described previously 10 , 49 . Fig. 5: Mechanistic investigation of the successive degradation of white phosphorus with weak adduct 5[OTf] 2 . a , Degradation of butterfly compound 3 [OTf] 2 with x equiv. DMAP ( x = 1, 3, 6, 12) in the presence of an excess of Ph 3 As and 5 equiv. 1 (Ph 3 As(OTf) 2 ), justifying a butterfly-type intermediate in the first step of the oxidative onioation of P 4 . b , Proposed mechanism of the successive degradation of white phosphorus over several, intermediary formed oligo-phosphorus compounds ( 16 [OTf] 4 , 17 [OTf] 6 , 18 [OTf] 5 and 19 [OTf] 4 ). Calculations were carried out using DFT-optimized geometries applying the BP86 37 , 38 , 39 functional with a dispersion correction (D3) 40 and COSMO (dichloromethane) on a def2-TZVP basis set 41 . Δ G values are given in kcal mol −1 (blue). 5 [OTf] 2 = (L N ) 2 AsPh 3 [OTf] 2 . Full size image The first step in the degradation of P 4 is the formation of the butterfly compound 15 [OTf] 2 , which is indirectly confirmed by the generation/degradation of the arsonio-substituted compound 3 [OTf] 2 . In the next step, bicyclo[1.1.0]tetraphosphane derivative 15 [OTf] 2 reacts with a second equivalent of 5 [OTf] 2 to give monocyclic intermediate 16 [OTf] 4 , which transforms into iso-tetraphosphane 17 [OTf] 6 on further reacting with another equivalent of reagent 5 [OTf] 2 . This iso-tetraphopshane intermediate can be identified in the 31 P NMR spectrum and gives rise to an A 3 X spin system ( δ (P A ) = −34.4 ppm, δ (P X ) = 60.8 ppm, 1 J PP = 41 Hz; Supplementary Fig. 74 ). In the next step, the first equivalent of product compound 11 [OTf] 3 is formed, accompanied by the formation of triphosphane 18 [OTf] 5 , which shows an AX 2 spin system in the 31 P NMR spectrum ( δ (P A ) = 105.2 ppm, δ (P X ) = 121.6 ppm, 1 J PP = 79 Hz; Supplementary Fig. 75 ). Triphosphane 18 [OTf] 5 is further degraded to diphosphane 19 [OTf] 4 ( δ (P) = 75.1 ppm; Supplementary Fig. 72 ) under release of a second equivalent of 11 [OTf] 3 . The last step involves the fragmentation of diphosphane 19 [OTf] 4 to 2 equiv. 11 [OTf] 3 . Application of P(DMAP) 3 [OTf] 3 (11[OTf] 3 ) as an alternative to PCl 3 The triflate salt of trication 11 3+ features phosphorus in oxidation state +III with three reactive P–N bonds, which renders this molecule a non-volatile and comparatively less corrosive P 1 synthon for the formation of value-added phosphorus chemicals in one single step (Fig. 6 ). In this regard, an important class of phosphorus chemicals (which normally require the use of PCl 3 ) comprise trialkyl- and triarylphosphites P(OR) 3 . These compounds are used for the production of inter alia pesticides, phosphonates or antioxidants in polymers 50 , 51 . We therefore targeted the synthesis of these compounds via alcoholysis of 11 [OTf] 3 with different alcohols. When 11 [OTf] 3 is reacted with 3 equiv. of diverse alcohols, the symmetrically substituted phosphites 21a – g and asymmetrically substituted phosphites 21h , 21i are formed (Fig. 6 (i)(ii)). A general observation is that the reactions are fast in most cases and do not need laborious reaction conditions. The volatile phosphites were not isolated, however, and their formation (>94%) is derived from 31 P NMR spectroscopy. The work-up of the heavier phosphites is performed by extraction from the reaction mixture for separation from the stoichiometrically formed dimethylaminopyridinium triflate salt ( 20 [OTf]), using a non-polar solvent and subsequent removal of the solvent under reduced pressure. Owing to the presence of DMAP in 11 [OTf] 3 , addition of a base is not needed, which is a further advantage compared to the conventional synthesis of phosphites from PCl 3 (refs. 52 , 53 ). Fig. 6: Application of 11[OTf] 3 as an alternative to PCl 3 in P–O, P–N and P–C bond-formation reactions. Formation of P–O bonds: reaction of 11 [OTf] 3 with alcohols to give different phosphites 21a – i . Conditions: (i) +3 ROH, –3 20 [OTf]; a : R = Me: pentane, r.t., 30 min (98%) a ; b : R = Et: pentane, r.t., 20 min (95%) a ; c : R = i Pr: pentane, r.t., 5 min (94%) a ; d : R = Ph: pentane, r.t., 15 min (98%) b ; e : R = n -decyl: benzene, r.t., 2 h (97%) b . f = 2-ethylhexyl: pentane, r.t., 2 h (97%) b . g : R = 2,4-ditertbutylphenyl: pentane, r.t., 22 h (95%) b ; (ii) +2 ROH (15 min), + R 1 OH, CH 3 CN, r.t., – 3 20 [OTf]; h : R = n -decyl, R 1 = 2,4-ditertbutylphenyl (83%) a ; i : R = 2,4-ditertbutylphenyl, R 1 = n -decyl (96%) b ; hydrolysis of 11 [OTf] 3 to phosphorous acid 22 ; (iii) +3 H 2 O, –3 20 [OTf], CH 3 CN, r.t., 30 min (98%) b . Formation of P–N bonds: reaction of 11 [OTf] 3 with dimethylpyrazole to tripyrazolylphosphane 23 . Conditions: (iv) +3 dimethylpyrazole, –3 20 [OTf], Et 2 O, r.t. 1.5 h (96%) b . Formation of P–C bonds: reaction of 11 [OTf] 3 with Grignard reagents to give different phosphanes 24a – g . Conditions: (v) +3 R 2 –Mg–X, –3 MgXOTf, –3 DMAP; a : R 2 = Me: +3 MeMgBr (3 M, in Et 2 O), toluene, r.t., 30 min, (>99%) a ; b : R 2 = i Pr: +3 i PrMgCl (2 M, in THF), CH 2 Cl 2 , r.t., 2 h, (87%) a ; c : R 2 = n -C 14 H 29 : +3 n -C 14 H 29 MgCl, toluene, r.t., 16 h, (97%) b ; d : R 2 = Cy: +3 CyMgCl (1 M, in 2-Me-THF), CH 2 Cl 2 , r.t., 2 h, (79%) b ; e : R 2 = Ph: +3 PhMgBr (1 M, in Et 2 O), toluene, r.t., 3 h, (84%) b ; R 2 = 2-pyridyl: +3 2-pyridyl-MgCl, CH 2 Cl 2 , r.t., 12 h, (71%) b . R 2 = 6-Br-2-pyridyl: +3 6-Br-2-pyridyl-MgCl, CH 2 Cl 2 , r.t., 1 h, (79%) b . Reaction of 11 [OTf] 3 with trimethylsilylcyanide to give tricyanophosphane 25 : (vi) +3 Me 3 SiCN, –3 26 [OTf], CH 2 Cl 2 , r.t., (85%) b . Reaction of 11 [OTf] 3 to give tris(phenylethynyl)phosphane 27 : (vii) +3 phenylacetylen, –3 20 [OTf], CH 2 Cl 2 , r.t., six days (71%) b . a Not isolated; formation (in %) is derived from 31 P NMR spectroscopy. b Isolated. Full size image Additionally, 20 [OTf] is generally recovered in high yield (>95%). The reaction of 11 [OTf] 3 with 3 equiv. H 2 O in CH 3 CN accordingly gives phosphorous acid (H 3 PO 3 ; 22 ) after 30 min of reaction time (Fig. 6 (iii)). The isolation of 22 from the crude product can be achieved by extraction using Et 2 O. Phosphorous acid ( 22 ), which is normally synthesized from PCl 3 , is an important basic chemical for the production of lead phosphite, which is a stabilizer in polyvinyl chloride polymers 2 , 3 , 4 , fungicides and glyphosate, the world’s most widely used and commercially valuable chemical for crop protection 5 , 6 , 54 . To further prove its synthetic potential, 11 [OTf] 3 was reacted with 3,5-dimethylpyrazole, leading to a complete consumption of the starting materials within 90 min and an almost quantitative formation of tripyrazolylphosphane ( 23 ) (Fig. 6 (iv)). Compound 23 has recently been used as a precursor for the synthesis of indium phosphide nanoparticles 55 and as a P 1 synthon in various reactions 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 . As alkyl- and aryl-substituted phosphanes (R 3 P) have been used in numerous applications both in academia and industry, for many years (for example, in catalysis 56 , as ligands for transition metals 57 or in the Wittig olefination 58 ), we investigated the reaction of 11 [OTf] 3 towards Grignard reagents. Indeed, the reaction of 3 equiv. of the corresponding Grignard derivative R 2 MgX (R 2 = alkyl, aryl; X = Cl, Br) with 11 [OTf] 3 in CH 2 Cl 2 or toluene at room temperature leads to the formation of the corresponding phosphanes 24a – g in yields ranging from 71% to >99% (Fig. 6 (v)). Generally, these reactions proceed smoothly and allow functional groups, as demonstrated by the formation of tripyridylphosphanes ( 24f ), as well as small alkyl groups, as demonstrated by the clean formation of trimethylphosphane ( 24a ). The latter is typically synthesized by the corresponding Grignard reaction with trimethylphosphite ( 21a ), as the direct reaction with PCl 3 leads to scrambling reactions, lowering the overall yield 59 . Tricyanophosphane ( 25 ) has recently received attention as a precursor for the synthesis of carbon phosphonitride (C 3 N 3 P) extended solids, which are assumed to possess interesting photocatalytic and optoelectronic properties 60 . When 11 [OTf] 3 is suspended in a solution of Me 3 SiCN in CH 2 Cl 2 , rapid dissolution of 11 [OTf] 3 is observed. Subsequent investigation of the reaction mixture by means of multinuclear NMR spectroscopy revealed the clean and quantitative formation of P(CN) 3 ( 25 ) ( δ (P) = −140.0 ppm) and the silylated DMAP derivative 26 [OTf] (Fig. 6 (vi)). Compound 26 [OTf] represents an onio-substituent transfer reagent, which itself has considerable synthetic potential. Application of 26 [OTf], for example, in the peronio-substitution of p -chloranil, allows for the synthesis of strong, organic oxidizing reagents 61 . As terminal alkynes reveal a rather acidic proton, these substrates are also suitable candidates for the generation of a P–C bond. Thus, the reaction of 11 [OTf] 3 with 3 equiv. phenylacetylene in CH 2 Cl 2 gives tris(phenylethinyl)phosphane ( 27 ) with concomitant formation of 20 [OTf]. Compound 27 , which reveals a distinct chemical shift at δ (P) = −88.5 ppm in the 31 P NMR spectrum 62 , is readily isolated by extraction from the dried reaction mixture using Et 2 O. Conclusion In summary, we report the concept of oxidative onioation, where P 4 is systematically functionalized into P 1 compounds, namely 11 [OTf] 3 , 12 [OTf], 13 [OTf] and 14 [OTf] 3 . Compound 11 [OTf] 3 can be used as a versatile phosphorus precursor for the synthesis of value-added compounds via P–O, P–N and P–C bond-forming reactions. Thus, starting with the most industrially relevant phosphorus source P 4 , we have developed a convenient and high yielding process to produce a variety of value-added phosphorus derivatives in only two steps via the concept of oxidative onioation. The oxidative onioation starts with a p-block-element based compound R n E (for example, Ph 3 As or PhI), which is oxidized to the respective element oxide R n EO. In the next step, an onio-ligand-assisted deoxygenation of R n EO with an N-based ligand (L N ) and a suitable oxygen scavenger (for example Tf 2 O or Me 3 SiOTf) yields an onio-substituted cation [R n E(L N ) 2 ] 2+ , a weak adduct that undergoes equilibrium dissociation into cation [R n EL N ] 2+ and uncoordinated onio-ligand L N . The mechanism for the functionalization of P 4 has been thoroughly investigated experimentally and computationally. It was found that the first step in the functionalization of white phosphorus is the formation of a bicyclo[1.1.0]tetraphosphane (P 4 butterfly) intermediate ( 15 [OTf] 2 ), which is successively degraded to [P(L N ) 3 ] 3+ (L N = DMAP) when the oxidation reagent 1 and the Lewis base L N are present. A detailed understanding of the underlying principles forms the basis for the future development of alternative processing routes towards value-added phosphorus-containing chemicals from white phosphorus. Further oxidation reagents as well as the direct electrochemical functionalization of white phosphorus are currently under investigation. Herein, we have introduced an alternative to the hitherto still most relevant and cheap P(III) source, PCl 3 , and believe that this work represents a valuable contribution to the development of new ideas in the use and functionalization of P 4 . Methods Caution P 4 is toxic and highly pyrophoric and should be handled, manipulated and quenched with corresponding caution. General Information Materials and methods are included in the experimental section of the Supplementary Information alongside the relevant characterization data. All manipulations were performed in a glovebox or using Schlenk techniques under an atmosphere of purified argon or nitrogen. Dry, oxygen-free solvents were distilled either from CaH 2 or from potassium. Deuterated solvents were purchased from Merck, Deutero or Eurisotop. All solvents were stored over molecular sieves (4 Å: CH 2 Cl 2 , C 6 H 5 F, C 6 H 4 F 2 , n -pentane, Et 2 O, CD 2 Cl 2 ; 3 Å: CD 3 CN, CH 3 CN). All glassware was oven-dried at 160 °C before use. DMAP, Ph 3 As, 1,10-phenanthroline and 3,5-dimethylpyrazole were purchased either from Sigma Aldrich or ABCR and were sublimed before use. Sodium tetrakis[3,5-bis(trifluoromethyl)phenyl]borate was purchased from Sigma Aldrich and used without further purification. Me 3 SiCN, phenylacetylene, pyridine (dried over CaH 2 ), quinoline (dried over CaH 2 ) and all alcohols were purchased from Sigma Aldrich, ABCR or Acros Organics and were degassed and distilled or sublimed before use. Grignard reagents phenylmagnesium bromide (1 M, in THF), methylmagnesium bromide (3 M, in Et 2 O), isopropylmagnesium chloride (2 M, in THF) and cyclohexylmagnesium chloride (1 M, in 2-Me-THF) were purchased from Sigma Aldrich or ABCR. 2-Pyridylmagnesium chloride, 6-bromo-2-pyridylmagnesium chloride and n -tetradecylmagnesium chloride were synthesized via textbook procedures from the corresponding alkyl- or aryl-chloride with magnesium strings. Me 3 SiOTf and Tf 2 O were donated by Solvay and were degassed and distilled before use. White phosphorus (P 4 ) was donated by Thermphos International BV and dried with Me 3 SiCl and sublimed before use. Ph 3 As(OTf) 2 ( 1 ) 31 and PhI(DMAP) 2 [OTf] 2 ( 10 [OTf] 2 ) 48 were prepared as described in the literature. NMR spectra were measured on a Bruker AVANCE III HD Nanobay 400-MHz UltraShield equipped with a BBO probe ( 1 H, 400.13 MHz; 13 C, 100.61 MHz; 31 P, 161.98 MHz; 19 F, 376.50 MHz) or on a Bruker AVANCE III HDX, 500-MHz Ascend equipped with a BBO(F) ProdigyCryo probe ( 1 H, 500.13 MHz; 13 C, 125.75 MHz; 31 P, 202.45 MHz; 19 F, 470.59 MHz). Reported numbers assigning atoms in the 13 C spectra were indirectly deduced from the cross-peaks in 2D correlation experiments (HMBC, HSQC). Chemical shifts are referenced to δ (Me 4 Si) = 0.00 ppm ( 1 H, 13 C, externally), δ (CFCl 3 ) = 0.00 ppm (externally) and δ (H 3 PO 4 , 85%) = 0.00 ppm (externally). Unless stated otherwise, all NMR spectra were measured at 300 K. Chemical shifts ( δ ) are reported in ppm. Coupling constants ( J ) are reported in hertz. The designation of the spin systems is performed by convention. The furthest downfield resonance is denoted by the latest letter in the alphabet and the furthest upfield by the earliest letter. Melting points were recorded on an electrothermal melting point apparatus (Büchi Switzerland, Melting Point M-560) in sealed capillaries under a nitrogen atmosphere and are uncorrected. Infrared and Raman spectra were recorded at ambient temperature using a Bruker Vertex 70 instrument equipped with a RAM II module (Nd:YAG laser, 1,064 nm). The Raman intensities are reported in percent relative to the most intense peak and are given in parentheses. An ATR unit (diamond) was used for recording infrared spectra. The intensities are reported relative to the most intense peak and are given in parentheses using the following abbreviations: vw, very weak; w, weak; m, medium; s, strong; vs, very strong. Elemental analyses were performed on a Vario MICRO cube Elemental Analyzer by Elementar Analysatorsysteme in CHNS modus. Data availability The data supporting the findings of this study are available within the paper and its Supplementary Information . All structures have been deposited with the Cambridge Crystallographic Data Centre (CCDC) and can be accessed free of charge via under the numbers 2105521 ( 2 [OTf] 2 ), 2019149 ( 5 [OTf] 2 ·CH 2 Cl 2 ), 2105523 ( 5 [BArF] 2 ), 2019148 ( 6 [OTf] 2 ), 2019145 ( 7 [OTf] 2 ), 2019147 ( 11 [OTf] 3 ·CH 2 Cl 2 ), 2061965 ( 12 [OTf]·0.67 MeNO 2 ), 2105520 ((L N ) 2 POP(L N ) 2 [OTf] 4 ·4 CH 3 CN), 2019146 ( 14 [OTf] 3 ·2 CH 3 CN) and 2061966 ( 24g ).

Chemists at the Technische Universität Dresden have developed a new, more sustainable process for synthesizing numerous important everyday chemicals from white phosphorus. The new process has the potential to establish innovative, more resource-efficient processes in the chemical industry. The groundbreaking results, which are the product of more than a decade of intensive research, have now been published in the journal Nature Chemistry. The chemical element phosphorus (P) is one of the essential building blocks of all biological life and, based on it, a function-giving component of many products: in medicines, food products or fertilizers. In nature, phosphorus occurs exclusively in bound form as phosphate in the earth's crust. However, continental deposits are finite and are estimated to last for only a few more decades. For industrial use, phosphates are converted into the so-called white phosphorus by laborious chemical processes. Alongside red, black and violet phosphorus, white phosphorus is the most important modification of the element in industrial terms and, to date, is still an irreplaceable starting point for the production of many pharmaceuticals, flame retardants, battery electrolytes, herbicides and other phosphorus fine chemicals. For the production of phosphorus-containing everyday chemicals, the white phosphorus is mostly converted by chlorination with chlorine gas to phosphorus trichloride (PCl3); a corrosive and toxic liquid, which is of central importance for the chemical industry as a large-scale industrial intermediate. However, the production and use of PCl3, which has so far been without alternative, is highly problematic. The chemist Prof. Jan J. Weigand of the TU Dresden and his team have now succeeded in specifically converting white phosphorus (P4) into an alternative and much less problematic phosphorus intermediate reagent. In this process, the use of chlorine gas can be completely omitted. Instead, the process chemicals needed to convert the white phosphorus are recyclable. "Economic factors still stand in the way of industrial application of the process, however a rethink is currently taking place here due to necessary, more sustainable aspects in the chemical industry. The more resource-conserving and efficient use of finite raw materials and the development of sustainable processes in many areas of chemistry are of the utmost importance. This work is a decisive breakthrough in phosphorus chemistry and of great importance for the further development of more sustainable and environmentally friendly processes," affirms Dr. Kai Schwedtmann, one of the two first authors of the publication. Prof. Weigand and his group are currently developing further concepts with the aim of completely eliminating the need to use white phosphorus or PCl3 for the synthesis of pharmaceuticals, flame retardants, battery electrolytes, herbicides and other phosphorus fine chemicals: "In order to meet the greatest challenges of our time, a rethink must also take place in the chemical industry. We want to make a small contribution to this with our research by developing a 'blueprint' for a more modern and more sustainable phosphorus chemistry."

10.1038/s41557-022-00913-4